Cognitive Perspectives on Word Formation (Trends in Linguistics. Studies and Monographs)

Citation preview

Cognitive Perspectives on Word Formation

Trends in Linguistics Studies and Monographs 221

Editors

Walter Bisang (main editor for this volume)

Hans Henrich Hock Werner Winter

De Gruyter Mouton

Cognitive Perspectives on Word Formation

Edited by

Alexander Onysko Sascha Michel

De Gruyter Mouton

ISBN 978-3-11-022359-0 e-ISBN 978-3-11-022360-6 ISSN 1861-4302 Library of Congress Cataloging-in-Publication Data Cognitive perspectives on word formation / edited by Alexander Onysko, Sascha Michel. p. cm. ⫺ (Trends in linguistics. Studies and monographs ; 221) Includes bibliographical references and index. ISBN 978-3-11-022359-0 (hardcover : alk. paper) 1. Grammar, Comparative and general ⫺ Word formation. I. Onysko, Alexander. II. Michel, Sascha. P245.C635 2010 4151.92⫺dc22 2010002362

Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.d-nb.de. ” 2010 Walter de Gruyter GmbH & Co. KG, Berlin/New York Printing: Hubert & Co. GmbH & Co. KG, Göttingen ⬁ Printed on acid-free paper Printed in Germany www.degruyter.com

Acknowledgments

Tlie making of C'ogtritive Pevspect~venon CVord Forn~atiorzwould not have been possible without the suppol-t of a number of people. First of all, we would like to thank all tlie contributors for their stimulating work and excellent cooperation tliroughout tlie publication process. In particular, their readiness to participate in the iniernal review inspired a lively discussion across topics and helped to consolidate some of the findings emerging from tlie different concerns of word formation research. We would like to thank the series editors of Trendr in Linguistic,, (Walter Bisang, Hans Henrich Hock, and Werner Winter) for expressing their interest in this volume and for granting room for its publication in the series. We owe a substantial part of this decision to an anonymous reviewer who commented in detail on t11c volume and its contributions. This feedback not only led to an eutension of individual ideas and inspired revisions on various levels, but it also proved valuable for giving the contributors and tlie editors a chance to take a clear stance on their work. Any remaining inaccuracies of course remain in the hands of the authors and tlie editors. Then we would particularly like to thank Birgit Sievert and her dynamic team at the publishing house for guidance and support in all aspects of publication. The personal way of conimunication offers a delightful basis of cooperation. Finally, Alexander Onysko would like to express his gratitude to Roswitlia and Manfred Onysko for their iuianinious support in all projects and ideas. He would particularly like to extend heartfelt thanks to Marta Degani for sharing her keen thoughts on the introduction and on some of the chapters in tlie volurne. Beyond that, it is her invaluable presence that so profoundly accompanied the period of the making.

A Ie.xander 0~1y.sko (Lienz, Austria) SuLrchuMichel (Koblen~,Germany)

Table of contents

Acknowledgments ............................................................................ v Introduction: Unravelling the cognitive i n word formation ................ 1 .4/exunu'er 011y.vkounu' Sc~schuMichel Part I: Theory and interfaces in word formation Word-forination or word forination? The forination of complex words in Cognitive Linguistics .................................................... kfartintr L a n p r ~u11d Giinther L ~ w ~ I ) c T !

29

On the viability of cognitive morphology for explaining language change ................................................................................ Livio (3ac.i~

75

(Ke)defining component structures in morphological constructions: A Cognitive Grammar perspective .................................................... 97 Cie~hal-dB vun Huji.\.siee~ Between phonology and morphology ............................................. 127 Nilke Elsen The role of graininatical gender in noun-form,'1 t'ion: A diachronic perspective froin Norwegian ...................................... 147 Philiyy., C'onze// Acljective + Noun constructions between s~ ntax and word formation in Dutch and German ............................................ 195 Mcri1hiu.s ITzining

Part II: Theory and processes of word f~rmation Setting limits on creativity in the productioii and use of metaphorical and metonymical compounds .................................... 2 19 KCko Be~~czes

Casting the conceptual spotlight: Hybrid compounding in German as an example of head-frame internal specifier selection ................ 243 A1cx~1ni~ic~~c.r 0ny.sko Does love come from to love or to love from love'? Why lexical motivation has to be regarded as bidirectional ................................ 30 1 Birgit (I~nhreit Doz4hler-upper nouns: A challenge for usage-based ~nodelsof language? ........................................................................ 335 Bert C'trppelle The intluencc of morphological structure on the processing of German prefixed verbs ................................................................ 375 .Judith Heicie, Antje Lorenz, And!+ Mc>inungor.~rriu' Frcrnk Bzdrchert Harvesting and understanding on-line neologisms .......................... 399 Tony Veule trnd C'ristiwu Burnuriu

Sub-ject Index ...................................................................................42 1 Contributors ..................................................................................... 429

Introduction: Unravelling the cognitive in word formation

Alexander Onysko and Sascha Michel

1. On word and formation The notion of word, as fi~zzyas it is, appears as tlie central concept for understanding and investigating language. In tlie early 20"' century, when Ferdinand de Sai~ssureelaborated on the arbitrariness of signifier and signified, words and otlier linguistic i~nitswere essentially characterized by tlicir symbolic nature. Thc fact that words are merely abstract labels for categorizing objects, qualities, and actions in tlie world has two immediate consequences. On the one liand, abstract conceptual reference to the perceived world creates associative space for cognitive processes sucli as simulation, projection, and concept manipulation (e.g., recombining, blending, adjunction, and focal highlighting), which, on the surface of language, can inspire the formation of words and expressions. On the otlier hand, the symbolic quality of language creates a conceptual distance between an utterance and what it stands for. Language operates on its own dimension connected but at the same time removed from the real world. Iconic motivation mainly exists internal to its system when established form-meaning units are combined to express derivative concepts.' This happens, for example, in compounding as conceptual units are created out of existing ones (e.g., ~v.shir?g+ machine 4 washing machine). The referential remoteness between language and the perceivable world lias been expressed in the ob-jectivist dilemma of using language as a tool for scientific reasoning. Pal-titularly when it comes to tlie description of language, the q~~estion arises of how the object of analysis can be described by its own means. r I l ~ eonly way out of this dilemma is to postulate a higher level of abstraction, i.e., a referentially specific metalanguage in order to avoid the inherent circularity in tlie description and analysis of language. This ob-jectivist line of reasoning lias remained an undercurrent in the modern study of language as structuralist, generativist, pragmatic, and f~~nctionalist theories of language emerged. Since from an evolutionary perspective language functions as a pragmatic tool of social interaction to

improve chances of survival, assuming an essentialist stance of reasoning ~n tlie description of language is, lio\vever, removed from the biological reality and necessity of Iiuman interaction and language use. As sucli, the dilemma of metalanguage and object language exists as an artefact of liuman reasoning, as an attempt to transcend the boundaries of expression by c r e a t i ~ ~a gfill-tlier level of reference. What appears as central to language, however, is tlie fact that it is the product of human cognition. Thus, explorations into the nature of language seem closer to a Iiutnan kind of reality if cognitive processes are considered for language analysis. 11; along these lines, we disclaim a rigid objectivist approach to tlie definition of word as a basic segment of language with its many facets of cliaracterization,hthe cognitive aspect of a word as representing a conceptual unit emerges as a pivotal criterion. In cognitive ternis, a word can be described as a symbolic label of mental categories referring to (in)anirnate ob-jects, to states, actions, conditions and qualities as they are perceived by and conventionally construed in the Ii~~man mind in interaction with the social and natural environment. According to this definition, words can also label compleu mental categories. For example, the word lover expresses an agentive role (encapsulated semantically in tlie suffix -er) that is tied to the state of love. Tliis simple example indicates the general creative capability of tlie human mind to construct and label new concepts, also by combining existing mental categories. In tlie history of linguistic research, sucli creative processes have come to be referred to as processes of word formation.' In tlie structuralist tradition, Marcliand (1969) provided a conipreliensivc description of mord formation processes such as compounding, affixation (derivation), conversion, abbreviation, and blending. The structi~ralist classifications continued to be used as a terminological spine in the field up to the present day even though word formation researcli has become modulated concomitant with the general strands of linguistic theory. Inspired by Coseriu's tlieot-4 of str~icturalsemantics, Lipka (cf. 2002 for a recent comprehensive treatment) developed a lexicological approach to word formation. Tliis path was also partly followed by Kastovsky (cf. 1992). Under the influence of generative grammar, research on word formatton has focused on principles and rules (cf. the lexicalist approach sulntnarized by Scalise and Guevara 2005). This path was first explored by Halle (1973) and later eupanded in the foundational work of Aronoff (1976). Ever since, generative approaches have dominated research on word formation and morphology as a variety of landmark publications in

IJnrczvelling the cognitive in word /orn?(~tion 3

the tield demonstrate (cf., among others, Katamba 1993, Kiparsky 1983, Lieber 1992, Scalise 1984, Spencer 199 1 , Stump 200 1). In co~ijnnctionwith structural types of word formation, basic generative principles (such as affixal strata, cyclic rules, the Right-Hand Head Rule, the Elsewhere Principle) have also shaped tlie contents of major handbooks and introductions to word formation and morphology (cf. Bauer 1983, Plag 2003, Booij 2005, Spencer and Zwicky 2001). The basic tenet unifying these works is tliat word formation is rule-governed. Thus, Plag states in his fairly recent introduction to English word formation that answers to the questions of how words are related to each other and how languages allow speakers to coin new words can be found "i~nderthe assunlption that language is a rule-governed system" (Plag 2003: 1). Booij highlights tlie notion of L'rule-governed creativio" tliat accounts for tlie formation of new words (2005: 6), and, at tlie end of his book, Bauer summarizes that "it has been assumed throughout tliis book tliat word-formation is rule-governed, but that rules are complex and far from obvious" (1983: 293). tle goes on to argue in favour of word formation rules by observing that The best evidence tliat rules rather than analogies are at work seems to me to lie in the fact of lexicalization of word-formation processes and in the existence of unacceptable forms for which there is an obvious analogy. If it is true that bluen, analogous to redden, is impossible, this suggests that rules are applying. Unfortunately, it is not absolutely clear that such a form is totally impossible. (1983: 295) The closing remark in this quote actually unmasks tlie potential fallacies of a categorical rille-based approach. Can tlic assumption about tlie rulegoverned obstruction of the word hlurri be supported by the mere fact that it has not yet been registered as an English word even though it coheres to a generally productive pattern of formation (i.e., deadjectival verbal suffixation)? For bluen, one could argue tliat the rare combination of vowels at the syllable boundaries between stem and suffix might cause a dispreferred phonological sequence for a speaker of English; however, the semantic function could overlay this formally uncommon sequelice and in an adequate context lead to tile creation and i~nderstandingof hluen. Of course, this example is not to be taken as a denial for any tjpe of regirlarities or patterned structi~resin language and word formation. Rather, it is a token of tlie tendency in generativist, structuralist, and opti~nality theoretic terminology to depict language as a very rigid and inflexible system consisting of discrete elements that can be combined according to certain mechanisms. The notion of rule is central to tliis mechanistic view of

4

Ale,~underOnysko nnd S a ~ c h nMichel

language. The term rule, however, can evoke a series of associations tliat reinforce a dichotomous view on how language functions. Thus, a rule implies an imposing authority, a rille clearly divides between what is right and wliat is wrong (i.e., wliat is good and what is bad), and in the Iii~man view of tlie world transcending rules licenses punishment of tlie transcendee (be it agents or words). While this is an extreme depiction of possible associations to the notion rule, and it is true tliat descriptively minded stri~cti~ral and generative linguists acknowledge tlie basic dynamic and fluctuating nature of language and apply rule in a non-prescriptivist but schematic sense, the metaphorical images created in scholarly discourse bear a trace of a mechanistic view of language. The attempt of trying to tie language in a corset of rules could thus be viewed as an expression of tlie human need and struggle for exercising control over their most precious gift. As an illustration of structi~ralistimagery in the description of word formation, in his lucid book The Grunimur of Words, Booij speaks of morphemes as "the atoms of words" (Booij 3005: 27; also cf. Plag's similarly structuralist depiction of words as syntact~catoms, see note 2). By a similar metapliorical token, Booi-j goes on to refer to the process of determining tlie structure of words as "words can be chopped into smaller pieces" (2005: 27). On the other hand, Booij also acknowledges the fact tliat "morphology contributes to the uider goals of cognitive science that explores the cognitive abilities of human beings" (2005: 24), and he devotes a full chapter in his book on morphology and mind to discussing issues of the mental lexicon, the rule vs. storage dichotomy, connectionist modelling of language (cf. Rumelliart and McClelland 1986), and Bybee's network approach (cf. 1985, 2001). This includes the notion of exemplar-based models according to which new words are created in analogy to stored exemplars, e.g., seatcupe coined in analogy to luncl~cape.In the end, Booij proposes a "morpl~ological race" scenario which is based on parallel processing of stored word forms and rule-governed formations (2005: 25 I). A parallel processing architecture of word formation which allows for redundancy in mental processing is also in line with recent findings in psyclloI~nguisticresearch (cf. Libben'~postulate of wluxlwiutlon o f opl?ortunlty for both morpl~ologicaldecomposition and holistic processing of stored words in a speaker's mind, 2006 6). Apart from Booij, the importance of a cognitive perspectibe on processes of word formation has also been implied in Plag's discussion on the role of tlie mental lexicon in word storage and productivity (2003: 47-5 I).

Most outspokenly, Schniid calls for a cognitive ti~rnin word formation research and outlines the central concerns of a cognitively inspired view: In1 Zentrum des Interesses stehen die Fragen nach dem kognitiven Prozess der Konzeptbildung (s. Kap. 4). den kognitiven Funktionen von Wortbildung und Wortbildungsniustern sowie die Mechanismen der Protilierung von Konzepten in komplexen Lexemen (2005: 103) [The questions of central concern are those targeted at the cognitive process of concept formation (see chapter 4), at cognitive functions of word formation and patterns of word formation, and at mechanisms of conceptual protiling in cornplex lexemes]

These recent voices stressing the impol-tance of investigating word formation in the light of cognitive processes can be interpreted from two general perspectives. First of all, they indicate tliat a structural approach to the architecture of words and a cognitive view are not incompatible. On the contrary, both perspectives try to work out regularities in language. What sets them apart is the basic vision of how language is encapsulated in tlie mind and the ensuing choice of terminology in the description of the processes. While a generativist, a structuralist. and by a similar token an optilnality theoretic view (cf. Ackeina and Neelernan 2001) assumes innate governing modules that formulate rules of language production and a separate word store as tlie mental lexicon, cognitive linguists tend to see structures and regularities emerging from a network of interconnections guided by general cognitive processes instantiated in language use. Crucially, cognitive linguistics metapliorizes language as holistic, non-modular, emergent, and depicts language as consisting of interconnected adaptive prototypical categories (cf. Givon 2005: 46-48), as based on image schemas, semantic spaces and frames (cf. Fillmore 1982, 1985), grammatical constructions (e.g., Croft 2001, Goldberg 2006), metaphorical mappings extending from embodied and cultural experiences (cf. Lakoff and Joh~ison 1980, Kovecses 2005), figure-ground relations (cf. Langacker 1999, 2008), and as grounded in patterns of usage (cf. Bybee 2001). As such, cognitive linguistics takes a constructionist position on language arguing in terms of lexical networks, emergent schemas, and associative patteriis tliat create possibilities of expression in contrast to a generativist-structuralist viewpoint of setting boundaries on language production via constraints operating on binary logic. In this way, cognitive linguistics concedes closely to tlie self-organizing nature of humans and their language whereas generativist-structuralist perspectives represent external boundaries as given in the institutionalized order of human interaction.

6

/fIexunder On,v?yrkonnd Snrchu h.llc,hrl

Apart from the unifying concern of these diverse schools of tliouglit in trying to grasp the way language functions, the recurrent mention of the importance of a cognitive perspective on word formation among structuralist accounts stresses the necessity for tlie field of cognitive linguistics to expand its endeavours of language description onto productive processes pertaining to tlie word level of language. This evokes tlie issue whether cognitive linguistics lias developed the appropriate theoretical notions to take on tliis task. In order to dwell on tliis question, a brief overview of ma-jor theoretical developments in the field is necessary.

2. The advent of cognitive linguistic thought Originally, cognitive lingit istics gained its impetus from three related tlieoretical developments and findings in the mid seventies and early eighties. Eleanor Kosch's empirical investigations of tlie internal associative structure of certain sets of categories (e.g., bird, fifr-u it,vehicle,fiirnitur.e, weapon cf. 1975, 1978) proved that humans associatively construct their categories around central members, the so-called prototypes. Rosch's findings have inspired a prototypical conception of lexical semantics which lias readily been embraced as a model of semantic category structures in cognitive linguistics. 'Today, a prototypical view of tlie semantic structure of categories is still prevalent even thoi~glithe dynamic nature of prototypes has been emphasized. Tlius, Stnitli and Samuelson (1 997) provide evidence for variable categories that are constructed on-line during language production according to speakers' needs. In a similar fashion, Geeraerts concludes his discussion on the notion of prototypicality by stating that "prototype formation may be influenced by other factors than purely conceptual ones" (2006: 46). Tlius, "stylistic, sociolinguistic, [and] connotational expressivity rather than purely conceptual needs may determine the flexible use of a category" (2006: 46). Complementary to prototype theory, Charles Filltnore developed the notion of frame semantics (cf. 1982, 1985), which Croft and Cruse describe in its essence as an attempt to account for "a concept as a complexly structured body of interconnected knowledge" (2004: 9 1). Such interconnected knowledge is activated in association with any concept in the mind of a speaker. Thus, dependent on discourse context and intention, the semantic frame of HOUSE can involve a set of associations to encyclopaedic and experientially connected parts of a liouse such as batliroom, kitchen, staircase, and so on. Frame semantics inspired tlie foundation of related notions

IJt~ruvellingthe cognitive in word formution

7

such as u'orf~uinsapplied in Cognitive Grammar (1,angacker 2008: 44-47) and Ideulized ('ognitive Mode1.s (ICM, Lakoff 1987: 68-74), which describe a mode of organizing knowledge by its idealized kernel, i.e., by its decontextualized default interpretation. The third major foundation of cognitive linguistics was laid down in the groundbreaking work of L,akoff and Johnson ( 1 980) on the cognitive interpretation of metaphor and metonymy. Built upon the central function of and experience one kind of thing in terms of anmetaphor to ~~nderstand other ( 1980: 51, Lakoff and Johnson argue that metaphorical imagery is fundamental to our conceptual system, as such based on internalized experiences and perceptions, and structured by a basic set of metaphorical pro-jections that underlie human discourse. In their early work they distinCONCEI'P , I S ENTITY), orienguished between ontological (e.g., AHs.r~~c.1. tational (e.g., C;OOl> Is UP, BAD IS DOWN), and structural types of metaphors More (e.g., ANGI'K IS A ~0'1'F L ~ J I DIN A C'ON.I'AINEK, I,OVF, IS A JOURNEY). recently, Lakoff and Johnson abandoned this distinction since all metaphors are structural and ontological and many metaphors emerge from basic orientational image schetnas (2003: 264). Instead, they stress the profundity of primary metaphor as postulated in Grady's work (1997), which establishes a basic level of metaphorical thought grounded in primaly, image-schematic experiences. As a further development of metaphor research, cross-cultural variation of conceptual metaphors has been acknowledged, particularly in Kovecses' work, which stresses cultural influences on metaphorical conceptions and considers metaphor as a linguistic, conceptual, socio-cultural, neural, and bodily phenomenon (2005: 8-9). Closely related to cognitive metaphor, conceptual metonymy developed into a separate branch of research particularly focussing on linguistic expressions that conceptually higliliglit part of a concept which is expressed linguistically while reference is made to tlie whole entity. For detailed treatments of metonymic processes see the works of Barcelona (2003, 2005) and the volumes of Barcelona (2003), Panther and Thornburg (2003), and Panther and Radden ( 1999). While these foundational theories of cognitive linguistics primarily provided insight into lexical semantic issues, they also crucially stimulated further theory construction. Thus, drawing on tlie metaphor of semantic frames, Fauconnier (1985) developed a model of mental spaces that has recently been proven as a valid explanatory tool for describing grammatical phenomena of hypotlietical space building in English conditional clauses (cf. Dancygier and Sweetser 2005). Apart from that, cognitive linguistics saw the emergence of two major grammatical frameworks: Cognitive

8 .ille.~underOnysko cmd Swchu Michel Grammar and Construction Grammar. The latter developed from initial concerns to treat idiomatic phrases as constructions, i.e., uniform representations of all grammatical knowledge, and grew into several strands of theories (cf. Kay and Fillmore 1999, Goldberg 2006, Croft 2001). 'These theories of Construction Grammar are unified by three central principles: the independent existence of constri~ctionsas symbolic units, the i~niform representation of grammatical structures, and the taxonomic organization of constructions in grammar (Croft and Cruse 2004: 265). In his model of Cognitive Grammar, Langacker (2008) conceives of the essence of language as based on a series of cognitive principles and fi~nctions,such as tlie existence of image schemas (preconceptual schematized patterns of activity derived from bodily sense experiences), the construction of Idealized Cognitive Models (i.e., frames) based on prototypical classifications, tlie perception of figure and ground (or the corresponding trajector and landmark), conceptual profiling of active zones through focal attention, and grounding of events in discourse context. His conception of language is also in line with usage-based appl-oaclies to language structure (cf. Bybee 1985, 200 1 ). Tlius, in Cognitive Grammar, patterns of language emerge via entrenched, i.e., frequently employed, schemas and constructions. For example, Laligacker describes compound forrnatioli in English as a constructional schema of specijier - specified, e.g., a tublecloth is a cloth that is specified for its use as covering a table (2008: 164-1 74). Apart from tlie widening of cognitive linguistic approaches to mainly grammatical phenomena, the field has recently experienced the advent of another tlieory that caters for the way the human mind creates novel meaning by drawing from previously constructed reference frames and conceptual units. This is conceptual blending theory as put forward most comprehensively in Fauconnier and Turner (2002). In a nutsliell, conceptual blending theory builds on the notion of mental spaces wliicli form a conceptual integration network. 'The structure of the network consists of input spaces which feed associable meaning components into a blended space. Generally, Fauconnier and Turner distinguisli four types of conceptual blending networks: simplex networks, mirror networks, single scope and double scope networks (2002: 120-135). To give just one example of the latter, most creative type of conceptual blending, Fauconnier and Turner describe the conlpuler desklop as a double scope blending in which the input spaces of cornpuler and de,tklop have different organizing frames. Tlius, regular computer commands (e.g., delete a digital document) are mapped onto office work (e.g., throwing a piece of paper in a trashcan) (2002: 13 1 ).

Unr-uvelling[he cognitive m word fortnution

9

In view of these developments in cognitive linguistic thoughts,' two immediate questions arise as soon as word forination knocks on the door of research on language and cognition. First of all. are the theoretical frameworks that structure the field of cognitive linguistics capable of investigating processes of word internal structuring? Secondly, and related to the first question, what kinds of attempts have been made so far to explain word fortnation from a cognitive vision of language? To briefly address the first question, it appears as if the various basic tenets of cognitive linguistics are applicable to issues pertaining to the word level. Thus, the notion of prototype and periphery could be employed to distinguish between different types of productive processes (e.g., preferences atnong different types of derivational suffixes). Furtlier~nore,frame semantics in combination with contiguity among sense elements of a frame (cf. Koch 1999) shed some light on the issue of cornpound formation, and particularly conceptual metaphor theory and conceptual blending theory represent insightful frameworks that add to the understanding of meaning construction in newly created terms and compounds. On a general level, processes of word formation also tie in with tlie neurological grounding of a cognttive linguistic conception of language. Thus, word formational processes can be conceived of as constructional schemas and patterns that emerge in usage-based mental networks (cf. Bybee 1985, 200 1 ). As far as Construction Grammar and partly also Cognitive Grammar is concerned, the neglect of word for~nationis, thus, not motivated by a theoretical void but seems related to a more stringent interpretation of grammar as primarily syntactic phenomena in line with traditional views. In his discussion of specification schenias, Langacker actually creates a link to the issue of compound formation (see above). This cursory glance at the tool kit of cognitive linguistic theories already alludes to their possible function of disentangling the internal composition of words and the productive processes of coining new terms. In spite of this potential,i processes o r word fortnation have remained a fairly neglected branch of study in the field of cognitive linguistics. This general disregard is also reflected in tnajor introductory books and volurnes on cognitive linguistics which remain silent on this issue (cf. Croft and Cruse 2004, Evans and Green 2006) or merely tentatively point out its relevance and call for further research in this area (cf. Dirven and Verspoor 1998, Ungerer and Schmid 2006). Thus, it is not surprising that only recently first efforts have been made to apply and to develop cognitive views capable of accounting for processes of word formation.

10

i-llc..vanu'er.Ot1y.rko and Smchu Michel

As part of tlie most recent handbook of cognitive linguistics (Geeraerts and Cilyckens 2007), Ungerer provides a first summary of cognitive research into word formation (2007: 650-675). According to him, studies have mostly tried to explore traditional types of word formation such as compounding, affixation, processes of abbreviation, and blending. Apart from diminutives and contrastive prefixation, the Englisli suffix -er has earned considerable attention among derivational processes. Thus, Panther and Thornburg (2001) postulate a prototypical agentive function of -er, wliich by metaphorical and metonymic extension develops into a radial set of meanings including instrumental, causal, locative, and patient senses. A silnilar line of reasoning has also been given earlier in Dirven and Verspoor, who describe tlie derivational schema (verb (V) + -er) as "a human or other source tliat is fi~nctionallylinked to the event in V" (1998: 65). Kyder (1999) criticizes a pure argument structure analysis of -er suffixation and, instead, proposes a series of constraints acting on tlie event schema of tlie base concept in order for it to undergo tliat type of word formation. Thus, base event scliemas preferably refer to single events, must be able to bear durative and habitual aspect, and must be specific and liiglily entrenched. 111view of the different accounts on -er suffixation, the volume at hand adds yet another facet to tlie discussion, wliich is tlie peculiar process of double suffixation apparent in doubler-upper nouns. Apart from some first explorations into the metonymic nature of conversion (cf. Dirven 1999), and the schematic structure of blends (cf. Kernmer 20031, nominal compounding and attributive nominal specitication (cf. Sweetser 1999, Coulson 200 I) have been subject of both psycholinguistic investigations (cf. Libben and Jarema 3006) and cognitive linguistic research on word formation at large. The latter has particularly drawn from conceptual blending theory as applied in a study on creative compounding in English (cf. Benczes 2006). Cognitive aspects of compounding and nominal specification also feature prominently in the present volume in wliich three contributions pick up loose ends in the area of compounding. In general, the overview of research so far alludes to tlie fact that cognitive linguists have largely tackled issues of word formation by applying established theories in the tield. In a similar vein, Tuggy employs the basic tenet of scliematicity from Cognitive Grammar to focus on constructional scliemas of compound formation and pluralization (2005: 233-265). His all-out schematic vision of language is built on layers of scliemas ranging from general, liiglier-level ones to more specific, lower-level types. Thus, the compound octopu,c curry is constructed as a nested structi~reof compounding scliemas ranging from a general schema of (NI Thing associated

Unrcrvclling tlie cognitive in word fomcition

1I

with N: Thing) to the ingredient schema in wliicli (N,is an ingredient of N-)(2005: 253). The postulate of language as purely schematic allows a holistic interpretation of language wliicli defies any nodular distinctions between syntax and morphology and likewise between inflection and wordformation. In addition, constituent ordering becomes irrelevant as the question of how pieces are combined matters more than their order or type (2005: 258). While the formation of schemas is indeed central to the nature of language, Tuggy's approach to word formation remains a perspectival sketch which calls for more tine-grained schematic analyses of word formation patterns. So, interesting questions remain open to further research such as what is the internal conceptual and formal structure of sche~natic units, how is the combination of schematic units motivated, and why are certain schematic combinations more productive than others? In some of his recent work, Tal~ny(cf. 2007) propounds a theory of recombinance which is constitutive of conibinatorial processes evident in lexical and phrasal-idiomatic expression of concepts. Recombinance consists of a continuum bounded by two poles of productivity: recombination and emergentness. While recombination explains regular semantically and fortnally transparent schemas of word formation (e.g., -er agentive suffixation), emergentness describes unpredictable formal - setnantic mappings which are processed as autono~nous gestalts (e.g., idiomatic phrases). Talmq's thoughts on these issues are picked up more extensively in the opening chapter of this volume, which discusses his theory as a viable process of lexical creation. Finally, looking beyond the core of cognitive linguistic theories, Stekauer's onomasiological depiction of word formation emerges as a relevant approach since it provides a functional-cognitive response to '-the onesided formalism'' of mainstream generative accounts of word fortnatioll (2005: 21 2). Inspired by the Prague School of functional-structuralism, he does not emphasize the cognitive Iiolistic account of word formation, i.e., its enlergence from a network of schematic representations. Instead, Stekauer postulates that word fortnation is an independent component of linguistics, which is crucially related to extra-linguistic reality and the naming demands of speakers in a speech community. In his view, the naming demand is a cognitive process based on conceptual activation and semantic processing which build the onomasiological structure as the basis for the act of naming (2005: 2 16). The expression of an onolnasiological stimulus is then carried out on an onomatological level of morpheme - seme mapping and is t'inally marked by phonological conventions (e.g., assignment of initial stress in compounds).

12

/Ilexandcr Onj>.5koond Suschu hlichcl

Altliougli Stekauer's onomasiological theory of word formation is not explicitly addressed among the individual contributions of this volume, his model strikes two base notes that reverberate tlirougliout tlie book. First of all, cognitive and structuralist reasoning are reconcilable if structuralist categories are grounded in cognitive processes of schematicity and conceptual combination. Secondly, the formation of words is in its essence a cognitive plienomenon. Accordingly. research on various types and processes of lexical creation needs to be based on cognitive principles in order to adequately account for a speaker-oriented, usage-based view of word formation.

3. Outline of the volume The discussion so far has shown that in modern linguistics tlie field of word formation has traditionally been rooted fimily in structuralist and generativist theories. These approaches have established taxonomies of word formation types and various sets of rules which, for example, try to account for the combination of affixes and for issues of lieadedness in morphologically complex constructions. Structuralist, generativist, and optimality theoretic views of word formation have continued to dominate the scene of research up until today, and only recently first alternative approaches have been put forward. Initially, cognitive linguistics has kept silent on questions of word formation as the field developed primarily along insights into conceptual categorization, frame semantics, metaphor theory, mental spaces, and early Cognitive Gra~iimar.Ever since Bybee postulated the relation between usage frequency and entrencliment of morpl~ologicalprocesses (cf. 1985), tlie ground has been prepared for casting cognitive perspectives on issues of analyzing word internal structures and productive patterns. In line with fitrtlier developments of cognitive linguistic theories in metaphor and metonymy research, conceptual blending, Construction Grammar, and the entrenchment of a schematic vision of language, the set of tools for analyzing word formation have increased, and, a i shown above, tlie first investigations have come underway. To rephrase Ungerer's claim (cf. note 5 ) . however, the potential of cognitive explorations into the issue have only sparsely been realized, and there is a lack of studies utilizing diverse empirical methods for achieving deeper conceptual insights into word formation. In light of this need, tlie present volume is the first one to take a substantive step towards illuminating diverse aspects of word formation from

Unru~vllingthe cognitive in word$~rrnation

13

cognitive perspectives. For this purpose, the book brings together contributions of tlie 2"" International Cognitive Linguistics Conference of tlie Gernian Cognitive Linguistics Association in Munich in 2006 and a selection of invited papers by scholars working on issues of word formation and cognitive linguistics. Tlie concerns of part of tlie contributions are of a theoretical nature such as tlie discussion of recombinance as a model of word formation and tlie postulate of a taxonomy of word formation processes as construction types. A few papers deal witli the interface between word formation and syntactic processes, word formation and inflection, and plionology and word formational patterns. Tlie remaining studies in the volume are concerned witli investigating phenomena related to traditional types of word formation (compounding, derivation, and conversion), and they contribute to reframing our understanding of these processes. In ~nethodologicalterms, the majority of tlie papers take a data-driven approach to analysis including corpus linguistic investigations, elicited data, psycholinguistic experiments and computational linguistic application. A few follow an introspective path of reasoning based on tlie discussion of selected examples (e.g., creative co~npoundsand word coinages in fantasy writing). Tlie metliodological pluralism falls in line witli recently uttered methodological concerns that cognitive linguistics should strive to employ tlie full range of methods for analyzing language, from introspection to corpus evidence and empirical data. This would strengthen interdisciplinary links to psyclioling~iistics,discourse analysis, and corpus linguistics (cf. Gibbs 2007: 2-17; Grondelaers, Geeraerts and Speelman 2007: 167). As mentiolied in the previous section, structural linguistics and cognitive considerations are not necessarily irreconcilable in their basic concerns. In fact, this volume intends to contribute to a vision of tlie two as being complernentary rather than diametrically opposed. '1 hus, many studies go from traditional terminology and established categories to novel, cognitive ways of analyzing patterns of word formation. Language data and examples are mainly drawn from diverse Germanic languages (Afrikaans, Dutch, English, German, Lurembourgisli, and Norwegian), and, in one case, from Italian. While we are aware that tlie range of papers would allow for several ways of arrangement, we have decided to highlight their nature along tlie following two lines: Part 1 subsumes primarily theoretical contributions on basic cognitive processes of word formation, on the interplay between diaclirony and synchrony in word formational analysis, on a classitication of word formation types, and on issues on the interfaces between word forina-

14

Alexunder Onv.,ko und Saschm Mchel

tion and inflection, word formation and phonology, and word formation and phrasal constri~ctions.Part 2 comprises papers focussing on established processes of word formation (compounding, derivation, and conversion) including an empirical psycholinguistic and a computational linguistic approach. In detail, Part 1 commences witli Martina and Giintlier Lampert's contribution on WordTfi)rn~a/ion or worcl,fOrrnation?The formation qf'conlplex ~z~or*dv in Cognitive Z,ingui.~tic.,.This paper raises the fundamental issue of how word forniatioli can be couched in a cognitive li~iguisticframework. To answer this question, tlie authors first of all provide an overview of usage-bascd approaches to morphology and word formation, in which they discuss tlie notions of morpliological constructions, tlie dichotomy of storage vs. computation, and the view of ~norphologicalconstructions as scliemas, as perpetuated in Bybee's network type approach to ~norphological structure. As a viable alternative for describing compositional morpliology, the authors then follow Talmy's notion of recombinance, which is characterized by tlie interplay of recombination, i.e., semantically transparent (re)arrangements of linguistic units, and emergentness, i.e., unpredictable, arbitrary and, thus, gestalt-like combinations. After giving a detailed accourit of I'almy's theory wit11 examples from the lexical network of emotion, tlie authors conclude witli an application of recombinance to traditional types of word formation in order to stress the essential similarity across the range of compositionality. In tlie next chapter, Livio Gaeta retlects On the viubilit~c?j'cognitive 111orph010,qy~ fi)r expluining langzrcrge chcmge. His discussion high lights tlie importance of considering tlie diachronic dimension in cognitive analyses of language since purely synchronic evidence based on "naturally" occurring cognitive processes (e.g., metaphor and metonymy) can lead to oversimplified and misleading explanations. He demonstrates the c r ~ ~ c i interal action between diaclirony and synchrony by closely investigating two examples of language change: the passive auxiliary ginn in Luxernbourgisli, and the developtnent of agentive/instrumental suffixes in Romance languages. I:or ginn, an image schema analysis combined with historical evidence show that presumed homonymy can actually be disentangled as grammaticalization following abstraction from polysemous senses. As far as agentive/instrumental suffixes are concerned, tlie reverse conclusion appears to hold truc since tlie universal claim of cognitively grounded polyselny extensions of agentive suffixes is not borne out in Italian where alternative explanations of borrowing, sound change, and ellipsis account for liornonymy of individual agentive and instrumental suffixes.

Unr.crvel/ingthe cognitive in word,fi)rrntrlion

15

The third paper of the volume aims to detine component structures in ~norphologicalconstructions in Afrikaans from the perspective of Cognitive Grammar. Based on a set of valence factors comprising schematicity, autonomyldepende~~ce,promiscuity, and constituency, Gerhard Van Huyssteen proposes a classification of morphological component structures into autonomous stems, dependent stems, affixes, and paramorphemes. While conceding lo cognitive principles, this taxonomy offers the necessary clarity of categorization, which facilitates its application for the computational annotation of morphological composition in Afrikaans lexis. This approach could serve as a rnodel of morphological analysis in other languages. Chapter 4 opens up the discussion of interface issues in word formation. In Beween pho17ology und moi-pholop, Hilke Elsen provides a rich array of data from child language, the terminology of chemistly, and names in fantasy and science tiction novels to argue for, on the one hand, the gestaltlike processing of lesser known sound sequences, and, on the other hand, for semanticization of recurrent sound shapes into morphemic units. These processes can be aligncd on a perceptual continuum ranging liom sound shape to recurring groups of sounds to segmental morphemes. Thus, her contribution stresses the interwoven nature of phonological and morphological phenomena which dispute their analytical segmentation as separate levels of language. Philipp Conzett's contribution investigates The role of grurni?ialical in Norwegian based on a comparison of historigender in 11out7~forn1ution cal and recent language data. His study sheds light on the interface of word formation and inflection and is a further contribution that considers diachronic evidence in its comprehensive analysis. The author approaches the issue from a bidirectional perspective, stressing that patterns of word formation can function as a base of assigning gender to nouns and that gender assignment also plays an active role in the formation of novel nouns. By comparing all gender marked non-compound nouns in an Old Norse and in a modern Norwegian dictionary, Conzett reaches the conclusion that gender patterns (formalized as gender specific suffixes or simple change of gender in the determiner) represent a productive means of word formation. This is indicated by the functions of deriving gender specific referents (e.g., masculine and feminine pairs of agentive nouns), and of marking semantic differences of same surface forms, i.e., to dissociate instances of lexical homonymy. Thc last chapter of Pai-t I probes the transitional area between word formation and syntax as exemplified in ad-jective and noun constructions in

Dutch and Gerinan. Mattliias Hiining points out that both in Dutch and German lexicalized phrases (e.g., suurer. Regeri 'acid rain') mirror regular syntactic groups in ternis of stress placement, inflection of tlie attributive ad-jective, and their orthographic conventions. However, lexicalized phrases also exhibit the typical compound features of bearing a naming (labelling) function arid of disfavouring modification of the adjectival specifier. When comparing tlie creation of compounds and syntactic groups in Gertnan and Dutch, kluning assumes tliat tlie trend towards miverbation in German is related to its synthetic character while in Dutch, which has comparably fewer inflectional processes, adjective-noun phrases are preferred as naming elements. The paper's concern with compounding anticipates Part 2, which shifts the perspective onto individual processes of word creation. Thus, as the first contribution of the second part, Reka Benczes pursues the immanent questions of why people coin inetaphorical and metonymic noun-noun compounds at all and how transparency of meaning and creativity in this type of compounds can be explained in a cognitive linguistic framework. In answering tlie first of tliese questions, Benczes discusses the role of compactness and vividness, context, memorability, analogy, and remotivation as motivating factors of compound creation. Her analysis is supported by tlie application of conceptual blending theory which shows how coricept~~al traits of tlie compound elements are integrated to form the conceptual (blended) space of the compound. When it comes to the issue of semantic transparency vs. creativity, Benczes argues from a Langackerian schematic perspective. She specifically applies the notions of elaboration and extension to correlate creative extension with transparency of meaning. To add yet another facet to the cognitive construal of compounding, Alexander Onysko's study provides evidence in support of the view that the formation of co~npoundsis concept~rallystructured. A detailed investigation of hybrid norninal compounds in German (e.g., Ahen~lshow,Fir~nenrn~unuger) unveils that tlie specifying element in a compound emerges from the semantic frame of the head, which acts as tlie conceptual base of compound formation. His analysis of corpus data also shows tliat conceptual blending operates on the base of head-frame internal specifier selection. As such, conceptual blending is only active in the case of metaphorical and metonymic compounds. llis study concludes with a detailed appendix containing the individual types of hybrid compounds and a set of conceptual maps illustrating head frames and their contiguous sense elements as specifiers in co~npoundformation.

Ui~vuvetlingthe cognitive in wordfi,rmation

17

After the focus on compounding as a central process of lexical creation, Birgit Umbreit's contribution topicalizes the crucial aspect of directionality in conversion. Before proposing her own theory on directionality in conversion pairs, Ilnibreit ~ ~ n v e ithe l s weaknesses of the claim for unidirectionality in certain patterns of word formation. She discusses common methods of trying to account for directionality in lexical creation whicli are etymology, usage frequency, semantic dependency and specificity, and the fortnal criterion of overt analogy. Each of these seems insufficient to explain tlie direction of conversions. As an alternative, Umbreit offers a bidirectional view, whicli is supported by empirical evidence from tlie cognitive organization of word-families, by data from L2 acquisition, and by evidence of speaker judgements on motivation. The latter, however, tend to show preferences for certain r o ~ ~ t eofs conversion whicli empliasizes the existence of a continuum of bidirectionality framed between two opposite poles of unidirectional motivation. In Doubler-upper nouns: A cl?allenge ,fi,r usuge-bused modeLs of language?, Bert Cappelle investigates the rare but regular plienomenon of double -er derivations (e.g., dropper-inner, a as her-upper). His detailed analysis of the phenomenon sets out with a quantitative corpus investigation based on tlie British iVutionu1 Chrpus (BNC), tlie CTor~)u,s c!f'C'onteniporary Anlerican Eu?gli.r.h(COCA), and tlie World Wide Web, all of which confirm tlie rare but steady presence of double -er derivations. Previous research on tlie phenomenon has uncovered a few reasons for its occurrence: plionological expressiveness, tlie misleading assumption of the unavailability of adverb - verb compounds, and the fact that double suffixation depends on the interim status of adverbs and prepositions as head or non-head of the construction. 'l'hus, fuzzy lieadedness exerts double force for -er suffixation in doubler-upper nouns: suffixation of tlie semantic head (left hand element) and suffixation of tlie structural head (right hand element). This last observation is also crucial for Capelle's approach to the issue, wliicli adds a few other principles of formation s~lchas the phonaestlietical jingling effect of double -er suffixation. In the end, tlie author comments on the fact tliat the capabilities of language users to create this structure without prior input would favour a rule-based account over a usage-based view. However, double -er suffixation, could also be abstracted from a more general schema of -er suffixation so tliat a usagebased view could actually explain tlie formations. Continuing with derivational processes, Judith Heide, Antje Lorenz, Andre Meinunger, and Frank Burchert's contribution sheds light on The i??fluenceqf'n~orphol(>gicwl structure on the proce.ssing of Geri~~cm prefixed

1 8 Ale.xc~nu'er.Onj>.~ko and Suschu Michel verh,~from an experimental psycholinguistic perspective. The particular focus of their study is on inseparable ver-prefixed verbs (e.g., vergehen 'to forgive', verluufin 'to get lost'), wIiic11 call forth the question whether the prefix is processed separately from the root, i.e., whether tlie prefix is represented autonomously in the mental lexicon. To test this hypothesis, tlie authors conducted a lexical decision task with masked ~norphologicalpriming, measuring response accuracies, response latencies, and priming effects for related and unrelated root and root+en primes. By discussing their findings in relation to previous research, tlie authors conclude that ver-verbs are decon~posedeitlicr as a right branching structure (e.g., ver + Iuufin) or as a flat structure (e.g., vpr + u r t ~+ en 'to impoverisl~') dependent on whether the root exists as a separate item in the mental lexicon or not. The final contribution of the volume takes an applied computational approach, called Zeitgeist, to finding neologisms as blends in Wikipedia. For classifying novel blends semantically, Veale and Butnariu have designed Zeirgeisi to first of all seek out neologisms that represent a formal blend of two different lexical items. As a next step, the cross-references given in corresponding Wikipedia articles are used as semantic information of the blend, and this information allows for an integration of the neologisms in the lexical-semantic network of WordNet. The relation between headwords in Wikipedia and lexical entries in WordNet is built on two semantic connectives: x isu y (as in ,superhero is a hero) and x hedges y (as in ~yintroni c ~hedges electronics), which is a case of taxonomic coordination. After exemplifying various schemas arising from tlie two connectives, Veale and Butnariu present the results of how precise tlie neologisms extracted from Wikipedia are interpreted semantically. From a cognitive perspective, Zeitgeist sl~owsthat many of the sense elements entailed in neologisms are instantiated in tlie cross-references given in the corresponding Wikipedia entry. As the brief previews of the individual contributions show, the volume provides a rich array of topics emerging under the umbrella of cognitive linguistic thought and established patterns and processes of word formation. The various studies add to a yet marginal body of research in cognitive word formation and, thus, are intended to help substantially increase awareness about the benefits of applying cognitive linguistic thoughts for investigating processes of lexical creation.

(Inravelling /he cognitive in ~~ordfi,rn7ution 1 9

Notes 1.

2.

3.

4.

5.

'Traces of iconic mappings between the world and language exist in onomatopoeic imitation, in the phenomenon of reduplication, and as secondary iconicity in selected morphological and syntactic principles (e.g., sequential order, cf. De Cuypere 2008). Plag, for example, distinguishes four properties of words: (a) they have a part of speech specification, (b) they are syntactic atoms, (c) they usually carry one main stress, and (d) they are indivisible units (2003: 8). We favour the unhyphenated spelling of word formation to emphasi~eits conceptual unity as an inclusive combinatorial process on the lexical level rather than as a traditional discipline of linguistics. In the volume, however, contributors are free to follow their own preferences, which lead to variation in the spelling of word,fi)rmution and ~~ordTfi,rrnation across contributions. With the exception of Martina and Giinther Lampert's contribution, this variation in spelling is deemed irrelevant for the claims and observations put forward in the volume. Cf. also The Handbook of Cognitive L2inguistics (Geeraerts and Cuyckens 2007) for a detailed overview of basic theoretical concepts. approaches to language analysis, and for the relation between cognitive perspectives and other theoretical strands of linguistics. As far as the contribution of cognitive linguistics to word formation is concerned, Ungerer stresses that "it [cognitive linguistics] can provide both the theoretical background and the empirical tools to complete a process that had already been set going: the semanticization of word-formation analysis (2007: 65 1 ). Despite this potential, lie concludes his survey by affirming that "current cognitive research in word-formation is still very much in its initial stages" and that "the application of most empirical methods has been too selective for a proper evaluation of their usefulness and should be supported by further studies" (2007: 67 1).

References Ackema, Peter, and Ad Neelenian Beyond Mor-pholofi)?.Oxford: Oxford University Press. 2004 Adams, Valerie An iniroc/uciion to ti~odernEnglish 1~ord,fi1rn7atior7. London: Long1973 man.

20

-1lexunu'er Onwko und Su\c hu &fiche1

Aronoff, Mark 1976 Word Forniation in Generative C;rtmimur. Cambridge, Mass.: MIT Press. Barcelona, Antonio 2003 Metonymy in Cognitive 1,inguistics: An analysis and a Sew modest proposals. In n/lotivu/ion in Lunguuge: Sludies in Honor c?fCiiniet* Rudden, Hubert Cuyckens, Thomas Berg, Kene Dirven, and KlausUwe Panther (eds.), 223-255. AtnsterdamlPhiladelphia: Benjamins. 1005 The multilevel operation of metonymy in grammar and discourse, with particular attention to metonymic chains. In Cognitive Linguistiex: lnlernul Dynunlics utid lnterdi.sciplinury /nterudion, Francisco J . Ruiz Mendoza Ibafiez and M. Sandra PeAa Cervel (eds.), 3 13-352. Berlin!New York: De Gruyter. Barcelona, Antonio (ed.) 2003 Metuphor nnd iMetonylr1.y at the C'ros.srouds: A cogniiive perspective. BerlinINew York: De Cruyter. Baiter, Laurie 1983 English Word-t;ornlu/ion. Catlibridge: Cambridge University Press. Benc~es,Reka 2006 CYreu/ivcconipounding it7 Et~gli.rh:The semuntic.~of' ~ ~ e t u p l ~ o r i c u l Amstera ~ d t??e/ot?ymi~~~/ 17oz1n-n0un con?hina/inn.s. dam1Philadelphia: Ben.janiins. Booij, Gert The Gramnlur o f Word.s. Oxford: Oxford University Press. 3005 Bybee, Joan I,. 1985 Morphology. AmsterdamiPhiIadeIptiia: Beri.jarnins. Phonology (mu' Lunglruge Use. Cambridge: Cambridge University 2001 Press. Coulson, Seana Set7iuniic Leups: Frunie-Shijting cmd C'onceptuul Blending in hleun200 1 Cambridge: Cambridge University Press. itig (Iori~trz~ction. Crofi, William Radical C'on.strmciion Grumniur: S,w~uciic Theory in Typological 200 1 Perspective. Oxford: Oxford University Press. Croft, William, and Alan Cruse Cognitive 1ingzri.stic.s. Cambridge: Cambridge University Press. 2004 Cuyckens, Hubert, Rene Dirven, and John R. Taylor (eds.) Cognitive Approuche~1 0 Lexic~ISeniuriiic~.BerlinJNew York: De 1003 Cruyter. Cuyckens, Hubert, Thomas Berg, Rene Dirven, and Klaus Panther (eds.) Motivutiori in Luriguc~ge.AmsterdamIPhiladeIphia: Benjamins. 2003

Unruvellin~y/he cognitive in word,fi,rrncrtion 2 1

Dancygier, Barbara, and Eve Sueetser 2005 Men/uI Spucc.s it1 C;rummur-: C'ondi~ionalC'on.strzrctions. Cambridge: Cambridge University Press. De Cuypere, I,udovic ZOO8 Lirniting the I~.onic.:Ft-otn the ~t~e~utlieoreticul ,fi)undutions to the c,l-ecxtivc possibilities of iconicitv in lunguuge. AmstcrdamPhiladelphia: Benjamins. Dirven, Rene 1999 Conversion as a conceptual metonymy of event schemata. In Metonyn7y in lur7guuge nnd ~houghl,Klaus-Uwe Panther and Giinter Radden (eds.), 273-287. AmsterdamiPhiladelphia: Benjamins. 2005 Ma-jor Strands in Cognitive Linguistics. In Cognitive Lingz~isfics: It7tcrntrl Djjnar~ic.~ ( ~ n dInterdis~iplinury Ir7teruction, Francisco J . Ruiz de Mendoza Ibafiez and M. Sandra Pefia Cervel (eds.), 17--68. BerlinINew York: De Gruyter. Ilirven. Rene, and Mar-joli.jn Verspoor 1998 Cognitive Explorutions of' Languugc ut7d Linguistics. Amsterdam1 Philadelphia: Benjamins. Evans, Vyvyan, and Melanie Green Cognitive Lit7guisric.s. An In/rodirction. Edinburgh: Edinburgh Uni3006 versity Press. Fauconnier, Gilles Men/ul Spuces: A.s/,ec/s of' h4euning Construelion in Nutrrrul LUII1985 guuges. Cambridge, Mass.: MIT Press. Fauconnier, Gilles. and Mark Turner 2002 T/7e Wuy We Think: C'onceptuul Blending und /he Mind's Hiddm C'otnplexxilies. New York: Basic Books. Fillmore, Charles J. 1982 Frame semantics. In 1,inguistics in the morning cultn. The 1,inguistic Society of Korea (ed.), I 1 1 137. Seoul: Hanshin. F'r-aines und /he semur7tic.v of undet,,stunding. Quuderr7i di ..rernunticu 1985 6, 2 2 2 254. Geeraerts, Dirk Words ond ()/her IVonclet-s: Pripers on Lexicul and Sen7antic Topics. 2006 BerlinINew York: De Gruyter. Geeraerts, Dirk, and Hubert Cuyckens (eds.) 2007 Hunc~!huok oj' ( ' o g ~ i t i v e I,ing~ii,stic.s. Oxford: Oxford University Press. Gibbs. Raymond W. Jr. Why cognitive linguists should care inore about elnpirical methods. 2007 In ~ l f e t h o d .it? ~ c o g ~ i ~ i vIit7g~istic~. e Monica Gonzalez-Marquez, Irene Mittelberg, Seana Coulson, and Michael J. Spivey (eds.), 2-18. AmsterdalnlPhiladelphia: Benjamins.

22

AIexun~Ier0nj:rko and Suschu Michel

Givon, Talmy Context us Other Minds: The Prugniulics c?f'Sociality,C'ognition und 2005 C'om~nunicuiion.AmsterdamIPhiladelphia: Benjamins. Goddard, Cliff (ed.) 2006 Ethnoprrig~natics:understonding a'i.rcourse ill cultmral context. BerIinINew York: De Gruyter. Goldberg, Adele 2006 Con.structions at Ifirk. The Nature c?f'Generalizution in Languuge. Oxford: Oxford University Press. Gonzalez-Marquez, Monica, Irene Mittclberg, Seana Coulson, and Michael J. Spivey (eds.) 2007 hkthods in cognitive 1inguislic.s. A~nsterdamiPhiladelphia: Benjamins. Grady, Joseph 1997 Foundations of' Meaning: Primaty Metaphors and Primary Scenes. PhD. diss., Department of Linguistics, University of California, Berkeley. Grondelaers. Stefan, Dirk Geeraerts, and Dirk Speelman A case for cognitive corpus linguistics. In Methods in cognitive lin2007 guistics, Monica Gonzalez-Marquez, Irene Mittelberg, Seana Coulson, and Michael J. Spivey (eds.), 1 4 9 1 6 9 . Amsterdamiphiladelphia: Ben.jamins. Halle, Morris Prolegomena to a Theory of Word Formation. Linguistic lnyuiyy 4, 1973 3-16. Janssen, Theo, and Gisela Kedeker (eds.) Cognitive Linguistics: F'o~rndation.~, Scope, and M e t h o d ~ l o ~ qBer. 1999 IinINew York: De Gruyter. Kastovsky, Dieter The formats change -- the problems remain: Word-formation theory 1992 between 1960 and 1990. In Thirij)years oflinguistic evolution, MarBenjamins. tin Piitz (ed.), 285-3 10. A~listerdamiPhiladeIphia: Kataniba, Francis 1993 Morphology. London: Macmillan. Kay, Paul, and Charles Fillmore 1999 Grammatical constructions and linguistic generalizations: the "What's X doing Y?" construction. Language 75, 1-33. Kemmer. Suzanne Schemas and lexical blends. In Motivutiorl it7 Language. Hubert 2003 Cuyckens, Thomas Berg, Rene Dirven, and Klaus Panther (eds.), 6997. AmsterdamIPhiladelphia: Benjamins.

Unravelling [he cognitive in word forniutiori

23

Kiparsky, Paul 1983 Word formation and the lexicon. In Proceedirigs of the 1982 Mid.America Linguistics C-onf2rence. Frances Ingemann (ed.), 3-32. University of Kansas. Koch, Peter 1999 Frame and contiguity: on the cognitive bases of metonymy and certain types of word formation. In Metonymy in Languqqe und Thought, Klaus-Uwe Panther and Gunter Radden (eds.), 139-167. Amsterdam/Philadelphia: Benjamins. Kovecses, Zoltan 2005 Metaphor in Culfz~re. Cambridge: Cambridge University Press. Lakoff, George 1987 Wornen, Fire, und Dangerous 7hirig.s. Chicago: University of Chicago Press. Lakoff, George, and Mark Johnson 1980 Metul~horswe live by. Chicago: University of Chicago Press. 2003 Metaphors we live by. [Afterword]. Chicago: University of Chicago Press Langacker, Ronald 1999 Gramrnur und Concel~tuulizution.BerlinINew York: De Gruyter. Cognitive Grurnrnur. Oxford: Oxford llniversity Press. 2008 Libben, Gary, and Gonia Jarema (eds.) The Representation und Proci?s.cing of Cornpound W0rd.s. Oxford: 2006 Oxford University Press. Lieber, Rochelle 1992 Deconstructi~gMorphology. Chicago: University of Chicago Press. Lipka, Leonhard Erlgli.sh 1e.xicology. Lexicul .structzre, word semantics and ~jord2002 ,fi)rmution. Tiibingen: Narr. Marchand, Hans The Cutegories and Types of Present-Day English Word-Formution. 1969 Miinchen: C.H. Beck. Panther, Klaus-Uwe, and Linda Thornburg (eds.) 2003 Metonyniy and Prugrnutic Infcrencing. AmsterdamIPhiladelphia: Benjamins. Panther, Klaus-Uwe, and Linda Thornburg A conceptual analysis of English -er nominals. In Applied Cognitive 2001 Linguislics 11: Languuge Pcdugogy, Martin Putz, Susanne Niemeier, and Rene Dirven (eds.), 149-200. Berlinmew York: De Gruyter. Panther, Klaus-Uwe, and Gunter Radden (eds.) hletonyrny in Lunguugc and Thought. Amsterdam. Philadelphia: 1999 Benjamins.

24

Ale.~~n&r Onysko und Sascha Michel

Plag, lngo 2003

Word-Formution in English. Cambridge: Cambridge University Press. Radden, Giinter, and Klaus-Uwe Panther (eds.) 2004 Studies in Lingz/istic 1\4o/iva/ion.Berlinmew York: De Gruyter. Rosch, Eleanor 1975 Cognitive Representations of Semantic Categories. .Journal qf' Experimental P.syc.hology: Gmerul 104. 193-233. 1978 Principles of Categorization. In C'ognifion and Cutegorization, Eleanor Rosch and Barbara B. Lloyd (eds.), 2 7 4 8 . I-lillsdale, NJ: Lawrence Erlbaum. Rumelhart, David E., and James L. McClelland 1986 On Learning the Past Tenses of English Verbs: Implicit Rules or Parallel Distributed Processing. In Parallel Distributed Processing, James L. McClelland, David E. Rumelhart, and the PDP Research Group (eds.), 2 16-27 1. Cambridge: Cambridge University Press. Ryder. Mary Ellen 1999 Bankers and blue-chippers: An a c c o ~ ~of n t -er formations in presentday English. English Lungzruge und Linguistics 3,269-297. Saussure, Ferdinand de 1916 Cours de lingui.s!ique generule. Bally, Charles and Charles Albert Sechehaye (eds.). Paris: Payot. Scalise, Sergio 1984 Generative Morphology. Dordrecht: Foris. Scalise, Sergio, and Emiliano Guevara 2005 The Lexicalist Approach to Word-Formation. In Handbook of WordFormation, Pavol ~ t e k a i ~ eand r Rochelle Lieber (eds.), 147-1 87. Dordrecht: Springer. Schmid, Hans-Jorg Engli~chehI(>rphologieund Wortbildur~g:Eine Einfuhrung. Berlin: 2005 Erich Schmidt Verlag. Smith, Linda B., and Larissa K. Samuelson Perceiving and remembering: category stability, variability and de1997 velopment. In Knowledge, concepts and categories, Koen Lamberts and David Shanks (eds.), 16 1--195. Hove: Psychology Press. ~tekauer,Pavol An onon~usiological theory of' English word-formation. Amster1998 damiPhiladelphia: Benjamins. Stekauer, Pavol, and Rochelle Lieber (eds.) Hundl7ook qf' Word-Formation. Dordrecht: Springer. 2005 Spencer, Andrew 199 1 Morphologicul Theory. Oxford: Blackwell.

Unrcrvelling the cognitive in wordjbrmution

25

Spencer, Andrew, and Arnold M. Zwicky (eds.) The Hcrndhook c?fMorphology. Oxford: Blackwell. 200 1 Stump, Gregory lnfIec/ionul Morl~holoa-. Cambridge: Cambridge University Press. 200 1 Sweetser, Eve 1999 Compositionality and blending: semantic composition in a cognitively realistic framework. In C-ognitive Linguistics: Foundations, Scope, and hfilhodology, Theo Janssen and Gisela Redeker (eds.), 129-! 62. BerlinINew York: De Gruyter. Talmy, Leonard 2007 Reconibinance in the evolution of language. In CLS 39-2: The Pariels. Papers ,fi-on1 the 39"' A ~ M L IMeeting U/ of the Clhicogo Linguistic Society, Jonathan E. Cihlar, Arny L,. Franklin and David W. Kaiser (eds.), 26-60. Chicago: CLS. Tuggy, David 2005 Cognitive Approach to Word-Formation. In Handbook of' WordFormation, Pavol ~ t e k a u e rand Rochelle Lieber (eds.), 233-265. Dordrecht: Springer. Ungerer, Friedrich 2007 Word-Formation. In Hutidhook of'C'ognitive Linguisfics, Dirk Geeraer-ts and Andre Cuyckens (eds.), 650-675. Oxford: Oxford University Press. Ungerer, Friedrich, and Hans-Jorg Schrnid An Introu'llc/ion lo Cognitive Linguistics. 2"" ed. Harlow: Longman 2006 Pearson.

Part I

Theory and interfaces in word formation

Word-formation or word formation? The formation of complex words in Cognitive Linguistics

Martina Larnpert and Giinther Larnpert

I.

Introduction

To sustain a minimum of conceptual coherence within Cognitive Linguistics, attempts at stock-taking or meta-reflection may appear advisable from time to time,' especially so as Cognitive Linguistics is faced with an allpervasive "empirical imperialism" (Geeraerts 2006: 34) that might eventually seek to completely discard its standard methods of introspection and analytical thought (see Talmj 2007a,b for a recent defense of introspection)."e fi~llyendorse a remark made some time ago in the first volume of Cognitive LinguisticL$that it is indeed a "serious metatheoretical question whether we are basically all talking about the same thing or whether substantially different cognitive processes or organizational principles are being inferred from [the] data" (Brugman 1990: 265-266). We thus take George Lakoff s dictum seriously that "[w]ithout agreement on initial premises, arguments about conclusions will be pointless" (1990: 46). This article, then, is less about facts, but more about essences, commitments, and theories. With the present volume being devoted to cognitive linguistic perspectives on word-formation,' we will be addressing the theoretically relevant issue whether a specific 'site' for studies into morphological compositionality can be found in Cognitive Linguistics at all. So our ob-jective is to go beyond treating the notoriously difficult-to-define concept of wordformation merely as a convenient label in view of solving practical problems in the first place. Since most current data-oriented studies in Cognitive Linguistics explicitly or implicitly position the~nselveswithin the internally heterogeneous family of construction grammars, we will, in section 2, discuss the theoretical implications of the construction grammar approach for a study of word-f~rrnation.~ And if it can be agreed upon that word-formation has always been an interface level of language, caught between lexicon and syntax, words and rules, free combination and lexicalization, we notice that these are still prevailing queries within Cognitive Linguistics. While con-

30

Martima Lanzperl und Giinfher Lunzpert

struction graminars (including Ronald 1,angacker's Cognitive Grammar) tend to investigate all coinplex expressions as kinds of constructional idioms' (see Booij 2007a: 8 3 ) , Leonard Taliny has, in his recent account of recombinance (Talmy 2007a). expressed a inore differential stance on tliese issues that might turn out to be foundational for any cognitive theory of morpliology and word(-)formation. In the third section of the present article we will thus set out to present and assess Talmy's views, especially against the background of the two rnost influential usage-based models of morpliology, tliose of Langacker and Bybee, but also witli a side-glance at other related cognitive approaches to morpliology or word-formation that do not specifically anchor theinselves within the Cognitive Linguistics paradigm. In illustrating tliese approaclies we will not fall back on invented examples, but will instead liave recourse to an emergent lexical network witli ctnotion as its morphological base, wliicli has expanded over the last few The enloyears at an amazing rate (cf. littp:ll\~w\?i.~rba~~dictiona~y.co~n). tion-network instantiations to be cited in this article liave all been taken from Web sources (including examples froin blog language). While WebI by many corpus linguists for their based data are still looked L I ~ O Icritically messiness, their 'anarchic' character has proved a major asset not only in the present context. Thus, the Web appears to provide satisfactory access to language phenomena and realistic insight into the dynamics of language as represented by nonces, their spreading use, and their ongoing entrenchment as well as instit~~tio~lalization."

2. Usage-based approaches to morphology and word-formation 2.1. 'Pre-theoretical' accounts Maybe it comes as no surprise that the existing state-of-the-art surveys of word-formation in Cognitive Linguistics (Taylor 2002, Sclimid 2005, Tuggy 2005, Ungerer 2007) tend to use morpliology or word-formation in terms of the traditional umbrella notions, as convenient descriptive labels without assigning them any decisive theoretical import. Take Ungerer's statement as an exernplary case: "Word-formation is one of tliose linguistic terms that may be unsatisfactory on a Inore theoretical level, but that are immensely ilsefi~lwhen one tries to survey processes of extending the lexicon" (2007: 650). Key proble~nsof word-formation such as analyzability, compositionality vs. lexicalization (or, alternatively, coinp~~tatioii vs. stor-

Worcl-Formation or word,fi,rma/ion

31

age), institutionalization (echoed in 1,angacker's entrenchment concept), productivity, and the formation of novel expressions are ultimately considered non-specific of word-formation itself, either dealt with in an ad-lioc fashion as cross-domain plienoliiena, or investigated with respect to individual word-formation patterns. Of coursc researchers apply, in their specific analyses, the well-known tools and concepts from the repertoire of Cognitive Linguistics, but they often seem not to be aware tliat the concepts in question have not been specifically developed for studies in wordformation. A brief glance at existing surveys will imtiiediately substantiate this assessment: Ungerer, for instance, sees "the semanticization of wordformation analysis" as tlie most decisive asset for Cognitive Linguistics to stimulate further research (2007: 651). The envisaged semanticization becomes most obvious in tlie principles relevant for the formation of new gestalt-like (liiglier-level) conceptual identities on tlie basis of at least two input concepts (especially in compounding) and in the concept-modifying functions of affixation (see Schmid 2005: 105). Accordingly, Cognitive Linguistics approaches to word-formation have investigated (and will most likely continue to do so) metaphoric and metonymic concept extensions (Benczes 2006 and this volume, Geeraerts 2002, Panther and Thornburg 2001, Dirven 1999), figure-ground alignment (Schmid 2005), scliematization (Ti~ggy2005, Ryder 1991, 1994, 1999, Farrell 200 1 ), conceptual fusion or integration in terms of the blending framework (Sweetser 1999, Kemmer 2003), and form-meaning iconicity (Ungerer 2002), to mention sonie persistent topics.' In any case, the ultimate goal of the proposed semanticization of word-formation research under the aegis of Cognitive 1,inguistics is "to provide more comprehensive and consistent descriptions of individual word-formation phenomena" (Ungerer 2007: 651). For Ungerer, and obvioi~slynot only for him, word-formation appears to essentially imply some sort of 'constructiveness.' Since complex niorphological structures usually involve encoding the same kind of semantic extensions tliat are found in clialiges of meaning in simplex itenis, the only difference between semantic extensions in simplex struct~lresand those involved in the formation of complex ones, lies not in the increase of semantic components, but, prototypically at least, in an increase (or, less frequently, a decrease) in tlie formal complexity of expressions. Given this 'constructive' view, a nonce-formation sucli as ewioter would indeed have to be seen as -extending' tlie meaning of the (verbal) base emote by adding -er; if one adopts the constructional schema view to be sketched below, however, ptlloter would as well be conceivable as 'elaborating' or 'instantiating' the

32

Mcrrtintr Larnpert und Gzmrher L o n ~ p e r t

schematic meaning of -cr, where the elaboration site of the latter is completely exhaustive of one constituent strilcture of the resultant structure.

3.3. Morphological constructions From the perspective of the constri~ction-basedframeworks that have become so fashionable in Cognitive Linguistics, again including Cognitive ram mar,' it woi~ld definitely be odd to assume a concept of wordformation, for language users' knowledge is captured entirely in terms of a vast structured inventory of sy~nbolicunits-entrenched (i.e., conventionalized and routinized) pairings of phonological form and semantic representation; such pairings are, in turn, syntagtnatically combinable "to form progressively more elaborate symbolic structures" (Langacker 1987: 82). Among the resultant complex structures there would also be polymorphemic words, the very home domain of word-formation. All these complex symbolic units of language, including those that are only partially lexically filled (V + -able) and those that may even entirely consist of abstract patterns (N + N), are now termed con.structions; in being constructions, they are "fully parallel in all immediately relevant aspects" to other multi-word structilres like phrases, idioms, and even sentences: There is then "no fi~ndamental distinction between morphological and syntactic constructions" (Langacker 1987: 82), an assumption which follows from an axiom that Langacker in fact takes to be foundational for all constri~ction grammars: "Lexicon and gralntnar are not distinct components, but form a continuum 01' constructions" (2005a: 102). Consequentially, in such a mono-stratal, non-derivational approach, the domain of wordTfOrmafion will readily dissolve into myriads of hierarchically organized niorphologicul con.s/ructions (cf. Langacker 1987: 82), linked in networks of inheritance. One may note that the idea of subsuming morphology and syntax under the all-embracing concept of the cons/ruc/ion is actually not recent: It was in fact already suggested in Bloomfield's classic Lungucxge (1933: 169): Whenever two (or, rarely, more) forms are spoken together, as constituents of a complex form, the grammatical features by which they are cornbincd, make up a construction. Thus, the grammatical features by which duke and -12.5.5 combine in the fortn u'llches.5, or the grammatical features by which poor .John and run u~uu-ycombine in the fortn poor .John run L I M >make U ~ , up a construction.

Word-Fornia/ion or word fbrrnation

33

Accordingly, this early structuralist conception of the construction is not confined to the clause level but inclitdes 'smaller' units provided they are complex. But, as has often been noted before,') invoking the construction concept does not guarantee unequi\ocal usage of the term. In one view, it will embrace only coniplex assemblies consisting of more than one constituent: Langacker (e.g., 2005a: 168), for instance, assumes the minimal construction to be symbolically complex; other strands within construction grammar even consider nionomorphernic symbolic units to qualify as constructions (see Croft 2001 : 17, Evans and Green 2006: 693, Dqbrowska 2009), and, beyond that, one may categorize the (complex) formal and the semantic poles of any linguistic sign as construction^.'^' From a constructional point of view, there would thus appear no principled differences among linguistic structures of increasing formal complexity, as the following table reveals: Tahle I. Categories and exalnples of constructions, see Goldberg (2006: 5). The original examples are here replaced by items fi-om the erno(tion)-network. Example Morpheme un-, -1es~ Word emo, ernollon uizernotlonal, ernotron1e.s.rne.s~ Complex word Colnplex word (partially [N-sJfor regular plurals: emotrons filled) LMOED, i.e., every monlenl of every day Idiom

Category

Idiom (partially filled)

Away-Construction Transitive Passive

erno ;something> out 'sit and sulk about one's love life while cranking winey emomusic' Ving /he N away: emoing the night away Sub.i V Obj: ernotionuli;e, ernotifi Subj aux VPpp (PP by): 11 was emoed all over hv tlie azrthor.

In order to obliterate sucli discrepant views, Bergen (2007)" proposes a general, cognitively oriented conceptualization in which constructions are only those form-meaning pairings "that psycholinguistic evidence shows

people store". Only under these premises would it really make sense to other claim that constructions reflect &nowledge of grammar in lolo-"in words, it's constructions all the way down" from the sentence to the morpheme (Goldberg 2003: 223). In light of the professedly usage-based orientation of the approach, one might fancy it should rather read 'constructions all the way up' (see Talmy's view below), for the acquisition of constructions is generally presumed to proceed 'bottom-LIP' through simple exposure to usage-events; but, methodologically speaking, constructions are usually broken 'down' into their lower-level constituent constructions." And as the basic units of this exposure are no longer words, which may beco~necombinable to for111syntactic structures, but constructions all over, the mental resi~ltof the acquisition process would have to be a huge conl -i .s fruclicon. In the 'classical' conceptualization of the construction in Construction Grammar (CxGrammar), however, the label is to be accorded only to those linguistic patterns where some aspect of their form or function is not strictly predictable from their component par-ts or from any other constructions acknowledged to exist (see L,akoff 1987, Goldberg 1995, 2006, 2009). However, in her most recent surveys, Goldberg eliminates the nonpredictability constraint: "In addition, patterns are stored as constructions even if they are fully predictable as long as they occur with sufficient frequency" (2006: 5). Subscribing to such a position will have significant consequences for the demarcation of the traditional domains of morphology and word-formation: A boundary would then have to be assumed to exist between complex expressions that can be claimed to have some independent cognitive status (the constructions proper, as it were), and others whose form and meaning is conceived of as derivable by some general statements (reluctantly called -rules,' maybe).

2.3. Storage orlvs. computation The separation of constructions from non-constructions is accomplished in a different way by yet another strand of construction-based approaches where it is argued that all complex units may be stored, and not computed by rules (even if they are fully predictable) as long as they are used with sufficient frequency and are conventionalized, i.e., when they have become 'entrenched' as cognitive routines (see Langacker 2005a: 161 and 1987: 59, I00 as well as Bybee 1985, 1998, 2001 ). If it must be considered axiomatic that the notorious I-ule-listfallacy (1,angacker's term) is best avoided,

Word-Formation or word,formurio~i 3 5 a minimal construction would simply consist of two component strilctures that are filsed to form a composite structure'! More technically, then, composite structures of any type are obtained via unification of their components by superimposing corresponding elements and merging (or 'integrating') their specifications (see, e.g., Langacker 2005a: 168, Langacker 2007, chapter 3). I11 Langacker's view, composite structures are never strictly reducible to their components, for "no component routine is manifested in precisely the form it would have in isolation" (2000: 4). Compositionality is thus on principle only /~urtiul;the meanings of components typically do not exhaust the meaning of the composite, which has to be regarded as an entity in its own right. It is a gestalt that is more than the silrn of its parts and where the components only niotivute the whole; or, conversely, where the component structures may be thought of as just "categorizing those facets of the colnposite to which they belong" (Langacker 1987: 466). The colnpositional value of a composite, conceivable as the regular compositional function of two components, usually does not give a complete account of the fully specified contextual meaning of a composite, as, say, printer. But, for Langacker, this does not entail that the opposite extreme [nust prevail, namely "that the expression is semantically opaque, listed in the lexicon as an unanalyzable unit" (1987: 456). Ultimately, then, Langacker's ( 1987: 466) view on co~npositionality amounts to considering it a case of categorization: The meaning of a composite structure receives "systemic motivation" from its components' structures, but may diberge from the colnpositional value by means of extension (which is different from elaboration or instantiation of the compositional value) or specialization. If it is more the rule than the exception that complex expressions fall short of full compositionality, the building-block metaphor (see Lakoff and Johnson 1980: 204) that is seen to i~nderliethis view of compositionality turns out to be inadequate: "[llf the component structures are stacked together to form the composite structure, how can the latter have content that is either absent in the former or conflicts with their specifications?" (1,angacker 1987: 463) What is more, the same issues arise in syntax (or phraseology; see Cries 2008); and accordingly, Langacher (2000: 20) treats morphology and syntax in a strictly parallel fashion: What we intuitively feel to be the meaning of a complex syntactic structure (e.g. a clause or a sentence) is usually more elaborate than its compositional value, or else diverges from it. The only real difference is that syntactic

structures are less likely than morphological ones to recur and establish themselves as conventional units with extra-compositional meanings.'' Consequentially, Langacker (2000: 20) suggests to replace the buildingblock metaphor (entailing an i~ndesired 'constructive' view of grammar) with the metaphor of scaffolding: "[C]omponent structures are seen as scaffolding erected for the construction of a co~nplexexpression; once the cotnplex structure is in place (established as a unit), the scaffolding is no longer essetntial and is eventually discarded" (Langacker 1987: 461). Yet this shift of metaphors does not imply that colnplex expressions ulti~nately bar their components from being perceived, they are not to be seen as unanalyrable wholes. Indeed, as Langacker (1987: 46 1) notes, The notion of a complex lexical item would be a contradiction in terms: the n nit status characteristic of lexical items would entail their immediate and automatic loss of analyzability, removing any grounds for considering them to be complex; all fixed expressions would therefore constitute single morphemes, regardless of size or any resemblance to other units. Unequivocal as these statements may appear, yielding Cognitive Grammar to be clearly non-reductionist, they are partially retracted when Langacker, likewise ~unequivocally,maintains: "The building blocks of gralntnar are minimal sylnbolic units, i.e. morphemes" (1987: 97). Diminished systemic motivation in a non-compositional colnposite may lead to the gradual loss of analyzability (instantiated in propeller for instance), but such loss is generally not predictable. Analyzability is, in Langacker's framework, considered to refer to co-activation of components, "the likelihood that a cotnposite structure will be construed in relation to the component structures, irrespective of their degree of compatibility" ( 1 987: 464). An extreme application of this stance on compositionality and analyzability would be to regard words like doctor, hanlmer, .father, ~lolher,hroiher, and si.cier as composites, recognizing the presence of the common -erl-or s ~ ~ f f iwithout x, the residue entailing a similar (morpheme-) status. I 0 Far-fetched as this analysis may seem at first glance, the composite reading of history to yield her~tory" is a well-known attested example, as is the analysis ofpeddler in terms of a composite, which, in turn, is the pre-condition Ihr the reading of to peddle as a back-formation involving the deletion of a presumed component; see also the 'case' of enlolion below. As may be predictable, filndamentally different views on conceptualizing the notional value of the construction have other profound effects on potential studies in word-fortnation, especially when it comes to the traditional basic 'building-block' of uord-lorniation (and morphology, for that

CZ'ord- Forniulion or. ~.vor.d,/brtnutioll 3 7

matter), the n~orphen~e. As we have seen, 1,angacker's (1987: 452-457) non-constructive Cognitive Grammar strongly proposes to discard the building-block metaphor,'x so that, in line with Bybee's (1 985, 1998, 200 1 ) usage-based approach to morphology, I0 morphemes are considered epiphenomena that can only be obtained through abstracting them out of strings of naturally occurring discourse by observing similarities and recording frequency effects. Such a discovery procedure might gain additional plausibility just by the very fact that in the niajority of known languages most morphemes are bound to other lexical material and do not occur on their own. Should one prefer to adopt the widest-ranging construction concept by which all form-meaning pairings are to be conceived of as constructions, the morpheme would indeed be [he minimal construction. Finally, if the preferred option is Bergen's cognitive definition of the construction in terms of 'storage,' it becomes an elnpirical problem (perhaps difficult to solve in an uncontested manner) to determine what will count as a construction; in the case of bound morplietnes like -ah/. or -er one would inev~tablyhave to engage in the still unsettled debate on wug-test types of evidence (see Taylor 2002: 3 12-3 15). Maybe all this will appear as only a matter of perspective (as a way of steering clear of too much generativity and constructiveness in Cognitive Linguistics); it will lead to defining morphology in a different way, though, as does, for inslance, Taylor (2002: 282), echoing the approach long pi~rsuedin the work of Joan Bybee: Morphological analysis is not so much a question of breaking up a complex form into its building-block components, it is a matter of whether a given form shares commonalities (phonological andlor semantic) with other forms in the language. Conversely, creating a complex form is not so much a question of putting together its component parts, but of creating a form in accordance with existing constructional schemas.

2.4. Morphological constructions as schelnas Taylor's remarks express the conviction that the regular formation of novcl linguistic items is definitely not achieved by means of applying 'rules.' The only process sanctioned is that of schematization, with .schernu definable as "a cognitive representation comprising a generalization over perceived similarities among instances of usage", emerging "via repeated acti\ation of a set of co-occurring properties" (Barlow and Ke~niner2000: xxiii)."' Unlike combinatorial rules, which are claimed to have a separate

identity (see. c.g., Pinker 1999: 6-8). constructional schenias are always extractcd from acti~allinguistic structures-which is meant to imply that the only difference between a schema and its instantiations lies in degrees of specificity: The scherna is more abstract, and the structures instantiating it receive more specific descriptions. Any composite structure in which at least one component is not specified lexically may then be addressed as a con.vtructionu1schema. A typical example of a high-level schema would be the well-known word-for~nation pattern for deverbal nouns in English: [[XI" erIN 'one who Vs' (see Rooij 2002). 'Old-style' word-lhrmation patterns (like compounding, aft'ixatio back-formation, etc.) are now seen to emerge in a higlily piece-meal fashion, associated with specific lexical items first (by means of the 'starting small' principle), and only then gradually being generalized to become scliematic." If constructional schemas may be considered "templates" for the construction and evaluation of novel expressions" (Langacker 2005a: 170), the constructional schema (at whatever level) becomes the basic site for any studies of word-formation in Cognitive Linguistics (see Tuggy 2005). And indeed many data-oriented sti~dicsinvoke schemas at all levels of specificity, so that it might be legitimate to raise the question what is to count as an adequate level of 'granularity' for such schema-based studies. Moreover, all constructional scliemas (including morphological ones) identify their constituents via constraint satisfaction, or, expressed in more positive terms, by licensing or sanctioning them. These constituents may of course be constructions or constructional subscliemas themselves, so that constructions that satisfy the constraints on another construction's constituent can unify with it. This process is usually describcd in terms of superimposition or integration. Some might argue, however, that schemas as they have been conceived of here are just notional variants of rules, especially if they are also to cover the formation of non-idiosyncratic, i.e., 'regitlar' complex expressions; in fact, Pinker and Ullman (2002: 458), ~najorproponents of the Words-and-Rules theoly, claim that their approach "is compatible with constraint- and construction-based theories of language, as long as they allow for variables and combinatorial operations" - as does Rooij's constructional morphology (scc, e.g., 2002, 2005, 2007a,b).I3 But for other construction grammarians, especially for those following 1,angacker. constructional schemas are definitely not rules but symbolic units themselves:"' The scliemas do not only define how components are to be combined, but the constructional schema contributes its own (contextually determined) properties to the complex assemblies; while rules, as has been

Word-Formation or word,fi,rnfation

39

pointed out, are (maybe mistakenly) assumed to be of a different kind of entity than the elements the) combine. Bybee ( 1 995, 200 1 : 126-1 28), on tlie other hand, introduces two types of schemas (witli both types being "templatic" in nature), source-oriented scliemas and proU'~c.1-orier~le schemas: While the former are obviously tlie equivalents of the classical generative 'rules,' the latter rather correspond to Langacker's constructional scliemas, and they are also the prevalent ones in Bybee's (1995: 428) netmork-model of morphology: "[M]orpliological properties of words, paradigms and morphological patterns once described as rule emerge from associations made among related words in lexical representation". Words (including complex words) that are entered in tlie lexicon are "related to other words via sets of lexical connections between identical a ~ similar ~ d phonological and semantic features"; these connections are responsible for yielding an "internal morphological analysis of complex words". Tlie words that are found in tlie lexicon are no longer "broken up'' into their constituent morpliemes, but the morpliological structure "emerges from tlie connections the> make witli other words in tlie lexicon" (Bybee 1995: 428-429). Connections may be seen to exist between what is traditionally referred to as a base form and a complex form (cut - cut,,), as well as between related coniplex forms (cuts, hats, ruts, mcrts, cup.\). Morpliological relations are constituted by parallel sets of plionological and semantic connections, if they are repeated across multiple sets of words. Schemas, in turn, 'emerge' when sets of words "having similar patterns of semantic and phonological connections reinforce one another and create emergent generalisations" (Bybee 1995: 430), and novel formations are obtained by means of schema extension, which is, following Bybee (1995: 430), dependent on two factors: "(i) the defining properties of the scliema and (ii) its strengths. the latter property being derivable from the number of items that reinforce the schema"; schemas then come, as has been mentioned before, in two types, "corresponding to the two ways that morphologically complex forms can relate to other forms". Tlie first type, tlie socalled "source-oriented scl~emas", are "generalisations over pairs of basic and deribed forms" (for instance, believe - unbelievable), so that scliemas of this type "correspond roughly to generative rules, since they can be tliouglit of as instructions for liow to modify one form in order to derive another" (Bybee 1995: 430). l'he second type, "product-oriented scliemas", refers to "generalisations over sets of complex or 'derived' forms", as in climber, printer, propeller, and ruler. They reveal what common features the derived forms have, but they give no information as to what operations

40

Martina Lanipert und Gutither Latnper.t

are required to produce tlicm. In other words, "product-oriented schemas are generalisations over non-basic forms rather than generalisations about tlie relation of a non-basic form to some underlying stem or base form" (Bybee 3995: 443); membership in classes formed by product-oriented schemas is not based on prototypes but on family resemblances. Sy~nbolicrules, on the other hand, "correspond to a source-oriented schema in that it specifies tlie input shape and details the procedure for changing it" (Bybee 1995: 443). Morpliological regularities can now be formulated either in ter~nsof a source-oriented or a product-oriented schema. As an example, Bybee (200 1 : 128) mentions the regular past-tense formation in English: A better candidate for a source-oriented generalization would be an affixation process, such as that used to forln the regular Past Tense in English. In most cases, the productive use of this schema appears to add It/, /dl or 11dIto an input form, the base stem. However, there is some indication that even this fairly agglutinative affixation may be conceptualized as a productoriented schema - that is. a schema that simply says, 'a Past verb ends in It/, /dl or ildl,' rather than 'add 'ti, id/ or / ~ d to i a verb to form the Past.'

Since Bybee's model allows maximal redundancy in representations, it accepts both source-oriented and product-oriented scliemas (wliicli may cast doubt again on presumptions tliat tlie difference is that great). But in contradistinction to tlie Words-and-Rules theory, there is no rules component that is supposed to be separate from tlie lexicon. Only one module is deemed necessary for niorpliology, and tliat is tlie lexicon: "Morpliological structure and organisation emerge from connections made among related stored items. Even forms produced by combination are produced in tlie lexicon by accessing a stem and ~~tilising a schema tliat is available in tlie lexicon" (Bybee 1995: 452). At this point it is essential to note that we have so far (more or less implicitly) dealt with compositionality only in terms of .,ernantic compositionality. The same issues, however, arise with regard to (mor)plionology, and they will of course be handled in the same spirit: As in semantic compositionality, symbolic rules are not assumed to exist, but only schemas that may be abstracted from actual strings to serve as categorizing templates. Just as morphemes arc abstracted from words by means of observing recurring con~monalities,segments like phonemes will then emerge as distinct (cognitive) entities only by means of plionological decontextualization. If tlie decontextuali~edsegments have enougli in common "and occur in enough distinct environments, a schematized segment arises which em-

bodies their commonality" (Langackcr 2000: 44). How all this may find a practical application has again been amply demonstrated in Bybee's cognitive phonology (see especially 1994 and 2001; see also Langacker 2000: 43-46). As regards the notorlous words-and-rules debate, Langacker (2000: 58), more outspoken than anybody else, rejects the applicability of classic 'constructionist' rules, considered "algorithmic operations on strings of discrete symbols". Rules in Langacher's (2000: 58) sense of the term are "extracted regirlarities with some kind of enduring cognitive presence"; in short, they can only be schemas of a particular kind not assumed to be stored as fixed entities that might operate on other stored units to yield some 'output,' but conceived of as cognitive routines, or, more adeqi~atelypcrhaps, as "recurrent patterns of mental (ultimately neural) activation" (Harlow and Kem~ner2000: xii). T ~ L Iit\ ,is not accidental that both Langacker's Cognitive Grammar view on 'morphology' and Bybee's network approach display close ties with connectionist models where information simply resides in patterns of connection weights (see Bjbee and McClelland 2005 and Bybee 2006). If the matter were coerced to become an either-or-decision, Langacker (and Rybee) would clearly have to be located on the side of 'words' only, and, to play with ~ o r d c ,word-formation would just be an instance of (complex) word formation. I here is, however, an alternative approach within Cognitive Linguistics: As we read it, Len Talmy has recently submitted a more sophisticated and more balanced account of compositionality that goes both beyond the mords vs. rules dichotomy and beyond the 'constructions only' view. In particular, his framework allows for a cognitively plausible integration of linguistic conipositionality across all 'levels' of language, from the initial combination of phonetic features into phonemes through morphemes and \\lords to the complex rearrangement patterns characteristic of syl~tau !rl addition, Palmy's suggestions (cf. ralmy 2006, 2007a) are conclusivelj backed up by drawing on evidence from both the evolution of language and the cognitive sciences, to converge in a new neural model that maj \+ell accou~itfor the combinatorial principles pervasive in language Since Talmj's niodel may not yet be well-known, the following section will present an elaboration of its major points that relies more on quotation than may be ususal. In fact, Ialmy is most appropriately introduced by himself'. because, in his case, wording is anything but random. -

42

Mavtina Lotnpert and Gunther Lampert

3. Recombinance in language Leonard Talmy (2007a)'~ introduces and elaborates on i-ecornhinuncc, a fundamental cognitive mechanism that, during the evolution of language, has come to pervasively figure among the representation of information in language. Recombinance, the key concept of the present section, is framed, in most general terms, as a subordinate concept related to digitalness, which proves to be the major distinctive design feature of and the fundamental type of representation in (spoken'") language (cf. Talmy 2007a: 5 1) - a set of properties that categorially sets spoken and written language apart from the predominantly analog nature of both non-linguistic cognitive systems and of signed language. Digitalness features as a gradient category (i.e., not involving binary representation) "cumulatively built up" (Talmy 2006: 2) from the factors of 1.

degree of discreteness or granularity, ranging from non-digital gradience to distinctly chunked elements at its digital pole; 2. degree of categoriality, applying to distinctly chunked elements only, that, at its 'less digital' pole, corresponds to discrete steps along a parameter and comprises, at the higher end, qualitatively distinct categories with separate identities in their own right; 3 . degree of recombination, in turn only applying to qualitatively distinct categories with their own identities, and these categories systematically combine with each other in alternative arrangements; 4. degree of emergentness, applying to alternative arrangements only, either resulting in patterns that directly follow from the process of arranging at tlie less digital pole or, in its most advanced expression of digitalness, constituting new higher-level entities with their own identities (see Talniy 2006: 2, l l and 2007a: 3 9 4 1 ). That is, a high-end value for the first factor, its more digital pole, "enables tlie second factor to play a role, and so on progressively to the circumstance in wliicli digitalness is manifested to its greatest extent with a highend value for the fourth f'dctor" (Taltny 2007a: 40). In its most advanced form, digitalness is manifested in recombinance, which, both as distinctive and basic design feature of language, is intended to capture "the assembling of discrete units into a new higher-level unit with its own identity" (Talmy 3007a: 26). It is of immediate concern in the present context that, although recombination and elnergentness are conceivable as distinct types of properties in language, they are at the same time systematically related

in terms of gradience in, or extent of, digitalness, and together they constitute the cognilive rn~~clinni.\iii of reconibinnnce. As a general cognitive mechanism, recombinance appears to characteristically cut across all the traditional language 'levels' including "compositional morpliology" (Talmy 2007a: 49) and tlie "expanded lexicon" (Talmy 2006: 15). Considering tlie term recombinanc.e, it should be emphasized that tlie prefix re- is indeed intended to coentail that 're'-cornhinu~zce be conceptualized as a filndamentally transient niechnni,\n~,wliicli, of necessity, takes place every time anew in any instance of language production and perception, i.e., reis ~neantto testify to the essentially dynamic character of recombinance.

3.1. Recombination and emergentness Within tlie liierarchicallq organized and extensive category of digitalness in language, recombinance becomes manifest as a system that sliows up in two states of digitalness, recombination and enzergentness. To elaborate on the specifics of these properties as they become relevant in linguistic compositionality, the following observations are crucial: Reconzbination (or, Agree c!f reconlhiriutioi~)constitutes "a system in wliicli discrete categorial units with distinct identities-drawn with various selections and in various numbers from an available inventory-combine in alternative arrangements of a certain type in accordance with a system of constraints on possible arrangements" (Talniy 2006: 1 I). More specifically, recombination presupposes that its resultant entities, qualitatively distinct categories in their own right, are in accordance with a specific system of constraints on their type of arrangement; and, of major importance for the analysis to follow, "this factor is referred to in terms of 'degree' because a system of arrangements can range from the simple to the complex in type and constraint" (Talmy 2007a: 4 1 ). The occurrence of tliis system, in turn, proves to be the only requirement for tlie next stage, emergentness. En~ergentne~ts (or, degree oJ emergen/nens), in contrast, qualifies as the cognitive pattern in which discrete categorial units with particular identities, the 'outputs' of the third factor (degree of recombination). participate in a system of alternative arrangements; and tliis time these new resultant entities "bear no systematic relation to each other due to any commonalities among the identities or arrangements of those component units, hence, tlie higher-level identities are not predictable from the components' identities or arrangements-that is, they are 'arbitrary"' (Talmy 2006: 12).

44

Murtinu Lampert and Gzinther Lan7per.l

Emergentness, to reiterate, is thus exclusively exhibited by those combinations representing novel higher-level ide111itieswliose.forn1s und/or nleanings ure neither predictuhle from their conllwnents nor frotn the novel gestult.'7 As before, the qualification of 'degree' is indicative of the fact that "examples of emergentness seem to differ as to how fi~llythe higherlevel form functions as a wholly distinct new cognitive entity with its own identity" (Talmy 2007a: 41). This extra specification is of major relevance in accounting for 'borderline' cases when exemplars of compositionality are seen to 'oscillate' in their (perceived) status of recombinance, perhaps even fro111 language user to language user, between recombination or emergence; a brief investigation into the compositionality of emotion, in the context of such new coinages as, say, enloticon and emo-ticon, or emo(u)ge will elaborate on this argument (see section 3.4 below). Conceptualized in this vein. emergentness would thus appear to correspond to the defining criterion of a Goldberg-type of construction (repeated here): "Any linguistic pattern is recognized as a construction as long as some aspect of its form or function is not strictly predictable from its component parts or from other constructions recognized to exist" (2006: 5 ) . It may, however, be worth noting that Talmy (2000, I : 23-24) has expressed some reservation against the construction concept in general, perhaps as its inherent readiness for immediate proliferation may eventually render the concept vacuous or even gratuitous. Recombinance, then, specifies two major distinct categories of coniplex linguistic representations: first, those recotnhinutionul structures of variable complexity that result from 'additive' or 'computational,' in any case predictable, combinations of semantically and/or formally more simple items, and, second, emergent noncorrelated structures that cannot be readily analyzed in terms of an additive (predictable) computation of their formal constituents and/or semantic components, but are rather perceived as 'wholes' or gestalts, with a clear tendency toward conceptual autonomy (see below). It might be suitable to introduce, for matters of convenience, some 'manageable' terminology as suggested in M. Lampert (2009). To account for this basic categorial distinctness of combination vs. emergence, linguistic compositionality is assumed to be characterized, at its 'low end,' by structurally transparent and referentially predictable con~positions,while its 'high end' specifies emergent, that is, opaque structures such as idioms, whicli are here proposed to be designated as coniposites. The system of recornbinance exhibits a liierarcl~icallynested layering that becomes manifest on four successive forrnul levels of structure (see

Word-Forrnuto or. u~ord,f~rmution45

3.2) and two .semantic level,\ (see below 3.3) within the digital organization in language, with each for~naland semantic level specified by its own type of arrangement and sets oS constraints on the arrangements. Talmy (e.g., 2006: 13) is indeed very explicit in view of the respective 'amount' of either pole in linguistic structuring: Only some levels of recombination may also exhibit emergentness. As may become evident in the sketch to follow, predictable combinations by far dominate in (spoken) language, and, what is more, the 'higher' the level of recombinance, the more recombination and the less arbitrariness will be found (cf. Talmy 2006: 24-25). It should also be noticed that Talmy conceives of recombinance as a cognitive mechanism proceeding essentially from 'bottom' to 'top'; so, to echo Construction Grammar's slogan, there will be 'recombinance all the waq LIP.'This conceptualizatio~i,then, evokes (again) the metaphor of the building-block (in contradistinction to, say, Langacker's and Bybee's categorization models above). It is indeed Talmy's general approach for all the (cognitive) systems in language lie has reconstructed so far to assume basic or primitive features or Sactors that re-combine to yield categories, and categories to yield ever Inore abstract scliemas and systems. The same principles are seen in fact to hold for recombinance as well."

3.2. Formal levels in recombinance While even the more radical construction grammars do not go 'lower' than the word or morpheme as their constitutive baseline and thus exclude submorpliernic structures from the range of constructions (see, however, Langacker 2000: 43-46), Talmy adduces convincing evidence for the 'early' existence of recombinance. On a first level, linguistic entities re-combine in arrangement^'^ to yield distinctive novel identities, i.e., Phonetic features with certain identities, drawn with a certain selection in a certain number from a language's phonetic inventory itself a subset of a universal inventory - combine to constitute a higher-level entity, a phoneme, with a particular new identity (2006: 13). -

Different combinations of phonetic features constitute different phonemes, which are arbitrary and not predictable from their constituents-hence the 'output' are emergent entities. This instance of recombinance may be taken to represent the simplest type of arrangement, i.e., cooccurrence, with the only constraint on arrangement being compatibility (cf. Talmy 2007a: 45, 2006: 13)."'

46

blurtinu Lunlpert und Gur~lherLurnprrt

To trace the relevant 'transition' from the submorphemic to the morpheme stage of recombinance, on a second level of organization, plionemes, emergent entities of tlie first level, 'feed into' (cf. Talmy 2006: 13) morphemes, again emergent entities with pal-ticular new identities most typically associating a cognitively distinct and autonomous concept that is arbitrary and "can only be remembered, not inferred" (Talmy 2006: 24). Talrny illustrates the non-correlative and discontinuous nature of phoneme and morpheme identities (or '-phonemic-semantic decoupling" Talmy 2007a: 46) by having recourse to the three phonemes /k, t, ad, whicli are recombined to produce tlie morphemes cut, tuck, and act; also cat, cold, cut, ctrn represent emergent and arbitrarj entities lacking an) systematic correlation between tlieir constituent entities and tlieir 'results,' i.e., between phoneme and morpheme identities, though of course sharing the s case: As initial constituent phoneme /k/. To adduce an o b v i o ~ ~'transition' for some restrictions in view of emergence and arbitrariness, an exception to full discontinuity is found in phonemes recombining into morphemes. It is the effect of sound symbolism that is at stake when, "e.g., an initial Is11 in English morpliernes may tend toward some association witli a meaning of (semi-)liquids and tlieir attendant pliysical properties, as in: ~1o~~l~c1urry/.cl~~gislu~shl~luiwl.~liu'c/.~/ick/ c/ip/,slog/,slurp"(Talmy 2006: 25 ). For- a general assessment of Talmq 's approach, it is significant to note that botli tlie phoneme and morpheme concepts are conceived of in terms of building-blocks that recombine to form ever more complex entities. Leaving these two lower levels aside for the major concerns of the present article, we note tliat, at the third formal level of recombinance, emergentness is again a pervasive feature when morphetiies recombine to yield ilJioms. Employing, as before, a totally analogous wording in the description of the process, Talmy underscores the identity in tlie combinatorial cognitive meclianism already operative across the previous plionetic-toplionological and plionological-to-morpliogical 'interfaces': "[Mlorplienies witli certain identities, drawn witli a selection in a certain number from a language's morphemic inventory - i.e., its lexicon - combine to constitute a liigher-level entity, an idiom with a particular identity" linked to "a particular new associated concept", and "different combinations of morphemes constitute different idionis" (2006: 14). When morphemes recombine to yield idioms, both recombination and emergence do play a role: While in semantic terms, idioms may well qualify as composites (in the terniinology suggested above), located at tlie emergent pole of the recombinance gradient, they are rather Janus-like as regards tlieir forrnal structure. When, for example, con,sideruble is associated with the meaning

'fairlq great,' it qualifies as a "morpliological idiom" (Talmy 2006: 16, see also Jackendoff 2002: 17 I): its formal structure, however, clearly displays recombinational characteristics in tliat consider and -able are added up to yield tlie resultant for~nalstructure con~iderable,which, in turn, exhibits emergent qualities (i.e., it is perceived as a gestalt with a novel formal identity). This is tlie paradoxical character of idioms, also observed in Jackendoff (2002: 171): "[Alpart fro111 their semantics, idiolns are like stored regular morphological c~mbinations".~' Note also that considerable will display recombinational characteristics tlirougliout when its 'literal' ~neaning'able to be considered' is associated, turning it into a conlposition in the terminology suggested here. With such "morpliological idioms" being Inore peripheral cases, morpliemes recombine to yield "phrasal idioms" as in turn up, turn down, ftrrn in, turn out etc. (see Talmy 2006: 14-1 6). It shoi~ldbe added that, in thir, context, Tal~nyaddresses the issue of "motivation", another fundamental cognitive principle, exerting, as on the previous levels before, constraints on full discontinuity; lie observes that "aspects of semantic or metaphoric content or structure in tlie component morphemes ... remain as comparable aspects in an idiom" as when "down alone has a metaphorically associated meaning of 'negative, unfavorable' which remains in tlie idiom turn U'Owv7 'reject." (2006: 25). As to constraints on arrangement at the forlnal level, the prevalent strategies in operation are "largely those of lnorphology and syntax, i.e., tlie usual compositional kind for morphemes in nonidio~naticconstructions" (Talmy 2006: 14) - both in morpliological and phrasal idioms, in morpliemes within a word and morphemes as well as words within an expression; i.e., the cognitive mechanism of recombinance is easily seen to generalize over an idiom concept tliat is not restricted to tlie phrasal level only. Another point worth reiterating is that in such a view idioms do not only involve semantic idiosyncrasies but also formal specifics. Apart from their characteristic qualities of semantic emergentness, it is due to the type of constraints they exhibit tliat morphological idiolns like considerable are assigned to a separate for~nallevel of recombinance. One might argue, then, tliat a new 'interface' becomes manifest on grounds of formal (non-)analjzability tliat is now seen to plausibly account for the above notes on the presumed 'compositionality' of hammer or sister and, more importantly, lends viability to the option of ndive language users' immediate ability to re-analyze words like enlotion as consisting of enlo- + -tion at least when a sufficiently unequivocal context as attested in erno-ticon or eniooge would cue such a reading (see 3.4 for some detail).

48

Muutinu Lun7per.t und G'lmthcu Lunlpeut

I lie final fourth level of ~ t r ~ ~ c t urecombinance ral is represented by morphemes, bound as well as free. urid idiomatic combinations of these that recombine to yield words and expressions. So this stage also falls short of corresponding to thc traditional layering of linguistic domains, especially if a distinct and unified domain of word-formation is to be envisaged. Apart from any morpliological and phrasal idio~iis-those that are the 'output' oS the third (idiom) level - this level is unequivocally characterized by recombinational (correlated and predictable), but non-emergent, types of arrangement only. While syntax (instantiated in expression,) may well count as "tlie system of recombination par excellence" (Talmy 2007a: 45), this cliaracterizatio~~ is deemed to hold for "compositional morphology" as we I I : [Tllie compositions of morphology and syntax do not yield a higher-level entity with a novel identity. Of course, short of this, they do exhibit recombination. 'They involve discrete units combined in arrangements under a system of constraints, ones that govern the slotting of the discrete units in to discrete positions within a structured framework. (Talmy 2007a: 50) Recombination is thus the prevalent formal strategy of a language to generate, on tlie basis of its "expanded lexicon", complex words or expressions. It sliould be noted, hobvcver. tliat Talmy differentiates between the formation of words and expressions on grounds of their implementing difi >rev/ cons/ruint,r on patterns of recombination: Arrangement properties Sor words, applying to "bound morphemes (and idiomatic combinations of these) [are] mostly contiguous": and as to their specifics of arrangement, they are constrained by the principles of n~orphology,while expressions, building on "free morphemes and words (and idiomatic combinations of these)" (Talmy 2006: 15), also implement tlie contiguous type of arrangement but are constrained bq principles prevalent in syntax.32 Recall, however, that these differences refer to structi~ralspecifications only. The focus is here solely on distinct formal properties of (types of) arrangements and their respective constraints. Any decisive assessment as to a separate domain of word-formation in language (in the original sign-based structuralist reading) would also have to take into consideration semantic recombinance, whose 'values' will of course have to be added up to provide the final result in each instance (see section 4 for some arguments). At the same time, it should be emphasized that statements of this kind will only be justified h r individual cases; tliat is, as has already been noted, any such proposition proves an empirical question in the first place.

CVord-Fortnatit,nor word formrrtion

49

3.3. Semantic levels of recombinance 111 the semantic structure of language, only two levels of recombinance, uitli t u o or three distinct types of arrangement properties," are available: F'irst, semantic components recombine to yield the meaning of morphemes or idioms, again implementing both reconibinational and emergent strategies. While the 'procedure' of selection is basically identical to that of the st!-uctural dimension, the itwentory to be selected from is either open or closed.'%ore specifically, tlie semantic components forming the meaning of an open-class morpheme (or also an idiom) are generally not drawn from any defined inventory (cf. Talmq 2006: 16). as is documented by the vast productive lexical network of enlo(/ion), e.g., all-erno(tional), ufieremotion, de-emotih, emotery, enloti-comlui,\.$eur, emotinzuge, eMofion, emolionalesque, enlolion-fieling, eMOTIVe, extra-ert~olionallexlraenlofive, mi.temotion. out-emotionalize, o\jer-emote, senii-emolion1e.c.~(to ~ n c n t ~ ojust n a few examples; for a more comprehensive list see M. Lampert 2009). Conversely, the semantic components constituting the meaning of a closed-class morpheme are drawn from (a language's subset of) a roughly closed universally available inventory, as -tion in the compositional reading of enrotion or - u p in emocrge. Also, semantic recombinance differs from formal structures in arrangement properties as "tlie same components can combine in different arrangements to constitute different meanings" ('Talmy 2006: 16). Here, the major type of arrangement is concurrence with relationships among the components ranging from simple conjunction to complexly patterned schernas like disjunction or hierarcliical nesting; and as to constraints on arrangement, semantic recombinance requires general conceptual compatibility of components on both a local and a global basis within a schema, wit11 their specifics, however, being only "little understood" (Talmy 2006: 16). 'To align Talmy's view on the notorious problem of assigning emergentness status to a morpheme's meaning with those sketched in section 2, some remarks may be in order. As to tlie meaning of morphemes (constituted by a combination of semantic components), their emergent character appears a matter of debate. In one ('constructivist') view, the meaning of a morpheme will be non-emergent since it is, after all, largely equivalent to tlie meanings of the semantic components in their particular relationships. The competing argumentation conceives of meaning as emergent and comes in two versions: For tlie first.

50

hfurtrncr Lumpert und (;z~tit/~er LUIIIIIL'I.~ In any given language, the meaning of a morpheme is a stable 'prepackaged' assembly, a fixed bundling of particular components in a specific arrangement like a phoneme of the language, with its fixed featural components and unlike an expression, whose form and meaning are constructed on the spot (Talmy 2006: 17).

The second ('holistic') perspective holds that "cognitively, the meaning of a morpheme might have an autonomous unity, apart from whatever semantic components underlie it" (Talmy 2006: 17); it may come as no surprise tliat Talmy is definite about tlie validity of the first emergent option: His example is the "complexly patterned" spatial schema of closed-class p u ~ f tliat "equals" a specific selection of semantic components drawn from a cross-linguistical ly (or even universally) available set of categories (2006: 17-1 8). In this context a note on Talmy's view on cognitive autonomy appears suitable (not least with reference to the link-up of organizing systems). In general, higher-level constructs associating distinct novel cognitive identities in tlieir own right exhibit, to different degrees, gestalt-like properties that correspond to tlieir respective 'profiles' of cognitive autonomy, wliicli is an emergent attribute they inherit from systetnatically interacting with another basic organizing system or cognitive faculty - attention. To just elaborate somewhat on the most basic principle (leaving aside difference in hind and arbitrariness), predicting tliat attention be directed to 'global,' i.e., Iiiglier-order gestalt features in tlie tlrst place, while backgrounding 'local,' i.e., lower-level, features, Talrny (2007a: 5 5 , cf. also 2006: 24) refers to basic principles of gestalt perception as tlie rationale underlying such fundamental differences in degree of emergentness. This observation is most obviously reminiscent of the Global Precedence Effect confirmed for visual perception in Navon (1 977)'': [Tlhe separate identity of the higher-level entity seems to be promoted to the degree that awareness of or attention on the lower-level components is reduced relative to that of the higher-level entity itself. Thus attention on the phonemes that make up a morpheme would seem to usually be low compared to that on the morpheme itself. ... Such wide attentional ratios would seem to advance the independent status of the morpheme ... By contrast, it would seem that awareness of the morphemes that make up an idiom do not greatly fade away relative to awareness ofthe whole idiom.;6 'Thus, by virtue of focusing attention on either level of semantic recombinance, the individual morphemes should in fact be accessible. (And it is this principle that will be proposed to account for some notorious issues in

hord-formation such as have been 'played on' in the title of the present vs. ~.t~ordfimnution: tliis effect may also be seen to article: ~~or~d-forw~ation i~nderliethe 'decompositionality~of emotion, see 3.4 below.) Finally, the second type of semantic recombinance yields complex lingi~istic items exhibiting recombinational but non-emergent patterns of compositionality as when meanings of morphemes and idioms 'feed into' the meanings of complex words and expressions: [Tlhe meanings of words and of idioms with certain identities drawn with a certain selection in a certain number from a language's inventory of morphemes and idioms--i.e., its expanded lexicon--combine in different arrangements to constitute the meaning of a co~nplexword or an expression (Talmy 2006: IS). Note that, as before, these complex types of meaning require differential treatment, for they involve different patterns of semantic compositionality. For one, "the meaning of bound morphemes (and of idiomatic cornbinations of these)" rearrange, by "semantic co~npositionalitywithin a word", into tlie meaning of complex words, "whetl~erin or out of correspondence with any morphological compositionality"; and such patterns of semantic compositionality are constrained only by "internal coherence of the concept" (Talmy 2006: 18). While Tal~nymentions tested, rete~ted,testable, unlesteu' etc. as illustrations, in the emo(tion)-lexicon, the limiting attested case of conforming to the language's patterns of morphological compositionality is overemotionulizufi~~~~ (see 3.4). Analogously, the arrangement properties for "the meaning of free morphemes and words (and of idiomatic combinations of these)", making up the meaning of an expression, rearrange, by "semantic compositionality over an expression - whetlier in or out of correspondence with syntactic co~npositionality",and are only constrained by coherence of the overall conception (Talmy 2006: 18-19). Here is a small selection of tlie expression type, again from the enio(1ion)lexicon: ewzotion Iunguuge, c-'rnotioncrl Iunguuge, language of emotion, emotion in language, the expression o f elnotiow in language. the verbal expression of emo/ion, /he rerbaliza/ion c?f eniotion in Ianguuge, emotion ay exprcs.sed in Iunguage.

3.4. Some sample analyses from an evolving 'lexical network' For simple reasons of space, we will, in tlie present section of tliis exposition, fall back on few selected examples from the emo(tion)-network to

52

Murtinu Lun?perfund Gltlnther Ltrmperf

roughly sketch the consequences tliat would follow from Talmy's claims on recombinance; we have, by intention, opted for the traditionally notorious cases - those that 'sit uneasily' between morphology and syntax, predictable or 'rule-based' computation and nonpredictable or 'lexical' storage. As a first example from tlie expressions just listed that may well illitstrate the 'trans-level' nature of tlie cognitive mechanism of recombinance, and whicli cannot straightforwardly be accolnmodated to either wordformation or syntax, we revert to eniotiov~language, in contrast to (potential) referential competitors, say, emotional language and language of emolion." To focus on tlie critical argument, tlie crucial issue proves to be wlietlier enlotion lrrngzrnge (or word-formutionlword formation, for that matter) can arguably be regarded as semantically emergent (or not), which woiild Iiighlight its gestalt quality and, hence, presuppose a (quasi-)terminological usage (as is actually common practice in emotion research). Following si~cli'lexicalist' reasoning, it may be suggested that enlotion lang~ragebe conceived as a con~positethat would have to be assigned tlie status of a cognitively autonomous unit, backgrounding, by tliis analysis, the independence of the semantic components in favor of the emerging gestalt concept (cT. Talmy 2006: 17); but even under tliis composite or 'morpliological idiom' reading, recall that "awareness of the morphemes that make LIPan idioni do not greatly fade away relative to awareness of the whole idiom" (Talmy 2007a: 55). However, on account of its structural specifics (consisting of a combination of two free morpliernes), emotion Iangzrage will hake to be accorded expression status, hence cornplying, on the formal level, witli tlie constraints on arrangement of recombinational syntax. As to a possible demarcation of some domain that may these findings would have legitimately be designated as ~rord-Jorn~ution, tlie consequence that compounding would no longer be treated as an instance of word-formation but of expression formation, for all compounds consist of a combination of free morphemes. In terms of their formal arrangement, compounds would thus (rather) link up witli phrases, conform38 ing to the patterns of syntax, not of morphology. For another illustration of tlie immediate implications inherent in Talmy's compositionality view, this time focusing on the low(est) level of recombinance in a morpheme, a feasible illustration from tlie opposite end of compositionality is available just by referring to tlie evolving lexical network's base emulion. To reiterate: emolion is quite a revealing case in point to demonstrate tliat a typically entrenched, high-frequency lexeme (actitally ranging among the 1,000 most frequent words in English, accord-

Word-Formcrtion or word,fbrmation

53

ing to Longlnun Diclionury of C'onretnyorury EnglisFi 2005) that is generally presupposed as an (in one theory. stored) item i~nanalyzableby nai've language users, is readily reanalyzed/-able as a composition emo- + -tion, when such users are faced with novel compositions based on emo-, which has become tlie 'source' for myriads of compositions such as all-emo, emio(a)gc, emo-hlog, emo'er, emocrtrqi, emo-jieel, emofi, emoling, ew~or?omic.r,enlospam, enlotholic, emothon, enzoid, emornent, emosome, enlove, emofale, refro-emo, zm-el?io/unemo,~~annuhemo; or when language users, from time to time, readily produce new combinations themselves. The underlying principle may well be identified as focused attention on either level of semantic recombinance -the local or tlie global level, just as predicted by mechanisms of cross-systems cognition. Though it cannot and should not be denied tliat tlie 'need' to remember (cf. Talmy 2007a: 56), or maybe simply 'know,' a 'core' lexical item such as enlotion is considerably greater than in tlie case of elnolion language, emlotion in fact may reasonably be located between tlie former and enlo, an instance of full discontinuity as a morpheme, exhibiting a major difference in kind and associating a non-inferable emergent and cognitively autonomous higher level entity. Just as Talmy suggests, it should be emphasized tliat, as in the previous case of enlorion language, tlie difierence is seen to boil down to differences in degree of enlet-gentness, allowing or disallowing the combination to be analyzed as a predictably correlated recombination or as a noncorrelated entity with its own identity. .4nd again, for a demarcation of some domain that may legitimately be designated as word-formation, such an analysis provides a challenging counterargument in view of any proposition in defense of traditionally agreed-on linguistic levels. Here it becomes very much an issue of one's belief, as a linguist, in a specific theory to argue for rigid and precisely delineable boundaries between linguistic levels, in this case between simplex and complex words. ,411even more surprising implication appears to follow fro111the cognitive mechanism of recombinance when tlie same effects and degrees of arbitrariness, discontinuity, and lion-inferability as in the phonerne-tomorpherne transition (see emo before) beco~nesapparent in initial isms such as the alphabetism EMOED.'~ While (exclusively) capitalized EMOED represents a legal term, specifying a particular cause of manslaughter that is due to 'Extreme Mental or Emotional Distress,' hence EMOED Man.slaughter, a second meaning is instantiated in three allograplis: EMOED, ernocd, and elno 'u'. With upper and lower case spellings being attested, they serve as abbreviations of the complex plirase Every Minute of Every

54

hlurtinu Lumperr und Gunlher Lutnperl

Duy, which, according to the definition in Ihhun ~ictionary,'~) associates the pejoratively connoted meaning of 'i~nhealthyobsession.' In the same source, the meaning 'sexually molested' is recorded, though exclusively showing up in enio'd. EIMOED in the first (terminological) meaning will have to be analyzed as an extreme abbre\ iation of a complex phrase, the resultant structure being an emergent new morpheme. Even when realized in a phoneme-by-phoneme reading. as initialism, EMOED may tend toward composite status. EMOEDlc.moed in the second sense represents the typical case of an emergent non-predictable idiomatic word, which again comes in two readings: 'unhealthy obsession' and 'addicted.' While in either case the gestalt meaning is unequivocally foregrounded, the semantic components remain (as before) indeed accessible. They 'do not greatly fade away,' as a 'calculation' along the lines of a Gricean iniplicature may illustrate. If you do something 'every minute of every day,' this will probably turn into an unhealthy obsession, like being addicted to emo music, being too often in an emo mood, etc. A complementary example of a morphological idiom taken from tlie enzo(tion)-network again illi~stratesthe differences in degree of emergentness, if only for differences in formal recombinance values. Emoficon, by now a higlily current composite, in fact allows for alternative (re-)analyses, differing in degree of morphological motivation: enioti + con, enloti(c)on, and enlot + icon will act as differently valuable cues when it comes to triggering their respective associated bases emot(ion) and icon(icity). Apart from these structural differences, emoticon qualifies as a composite, representing an emergent (higher-level) entity, both formally and semantically, with its new referential identity arguably not immediately 'compi~table' from its components. This is suggested, for example, in MerrianzWeh.cter's Online Dictionary, which defines the concept as 'a combination of keyboard characters (as :-)) that typically represents a facial expression suggesting tlie writer's attitude or emotion and that is used especially in computeri~edcommunications (as e-mail)'." This is not to say, however, that most probably it would not be possible, at least for some partially trained language users, to make some educated guesses about its formal and semantic compositionality on account of the components that, as in all 'idioms,' do not greatly fade away. And just like a morpheme, the composite enloticon, an idiomatic, complex word, associates a particular form with a particular concept: Even thougl~the "overall concept is not systematically related to those of the component morphemes" (Talmy 2007a: 47), there are motivated associatio~isreadily available to recover the underlying structure, and tlie educated guesses become a regular 'generative' option.

Word-Formation or word,fi>rnlation

55

To emphasize once again, it is largely tlie constraints of morphology and syntax that apply to the arrangements making LIP such composites; that is, "the individual morphemes occur in certain structural locations that apply in general to the compositional (non-idiomatic) combination of morphemes" (Talmy 2007a: 47). Hence, there are no differences in arrangement patterns between compositions and composites, so that, as argued above, einoticon may ultimately be accessible at least to a structural analysis. As an obvious inference to be drawn from these few examples, it tnay well seem plausible that the expanded lexicon, consisting of cotnplex words and expressions, appears indeed not to be characterized by high degrees of eniergentness and cognitive autonomy. In addition, as referred to earlier, arbitrariness and noncorrelation remain marginal phenomena with exceptional status in a language's repertoire of compositionality. This assessment, then, is especially supported by the opposite extreme of multiply nested polyniorphemic compositions that represent clear instances of the recornbinational pole. As touched upon previously, overemotionalization exemplifies tlie limiting case in reco~nbinational conipositionality that is attested in the lexical network of enio(lion). This specimen of liesting (Talmy's term), brought about by ~nultipleaffixing, recombines five (or even six, when accounting for the co~npositionalityof enzo- + -tion) simplex morphemes (bound and free) to yield a transparent, hence predictable, new entity with its own identity, a complex word. In a more general perspective, this composition reveals itself as one instantiation of the cognitive mechanism of recombination, following the type of and constraints on arrangement common in morphological compositionality; hence overemotionalizarion, both in terms of structure and semantics, is a composition of morphemes, accessible by predictable and correlational analysis. To round off this sketch, basically tlie same qualification also holds for the notorious, and as yet itnresolved case of (traditional) word-formation namely the query wliether ernorion latiguage is assigned compound or phrase status; as argued above, in Talmy's system, emotion language qualifies as an expression, following the formal constraints on arrangements characteristic of syntax. It milst be considered, then, an unequivocal case of expression fortnation (and riot of word formation), with only little potential for emergentness, except for the case of bccoming a conventionalized term. Even in this reading, however, the constituent entities and component identities remain readily accessible.

56

Murtina Lunipert und Gunll~erLurnper!

4. Recombinance-gradience and integration across systems Venturing for a provisional assessment as regards Talmy's views on cornpositionality, we will focus on tlie core of his argument witli reference to tlie initial query as to wlietlier there is, first, sufficient justification of assuming a separate or unique linguistic domain of 'word-formation,' and, second, how to evaluate the distribution of reco~nbinationor emergentness against what has been reported on constritction grammar in section 2. In tliat framework, it has turned out to be almost axiomatic tliat wordformation cannot be correlated \bit11 a unique domain or level of language. In both Cognitive Gralnrnar and in Radical Construction Grammar, wordformation fragments into a Iiierarcliically organized network of constructions and constructional schemas, where one construction simply integrates into tlie next, to ultitnately become retrievable from a vast inventory of stored gestalts. I11 Construction (Cx) Grammar, this cons/ruc/icon is severely constrained, however, by tlie non-predictability condition and tlie assu~nptionof a gralnlnatical form: More bluntly put, idioms, the 'true constructions,' are segregated from non-idioms, whose form and meaning is assetnbled in terms of free combination. Word-formation, under this interpretation, is split up into constr~~ctional idioms, on the one hand, and rule-based compositions, on the other. 'What goes where' is then no longer a matter of principles but becomes an ernpirically testable decision. While Talmy obviously shares tlie conviction tliat emergent idioms must be kept apart from non-idiomatic free combinations, he offers a Inore fine-tuned view on the formation of words. On a more general level, it can be inferred from the sets of properties of arrangement and the types of constraints on these arrangements that are seen to operate on linguistic entities that a principled difference between tlie constituent meclianisms of recombinance may be identified - at least wlien it comes to the categoly's 'extreme' poles, recon~binutionand emerger?tness, respectively: For tlie former, a moderate form of digitalness is presupposed. This means tliat qualitatively distinct categories witli separate identities in their own right occur at sites relevant to those identities and combine in arrangements that, at tlie formal level, are typically contiguous and, at the semantic level, are those of conceptual coherence. Emergentncss, however, exhibits tlie strictest form of digitalness, i.e., separate identities systematically combine in alternative arrangements witli respect to each other in specific patterns that give rise to higher-level entities uitli their own identities. These arrangements are mostly sequential (though not all are necessarily contiguous),

Word-Formation or word,formution

57

while they are characterized by noncorrelated discontinuity, both formally and semantically, and by conceptual compatibility in meaning. Taltny observes 'tri~e' emergentness to be of only marginal relevance for tlie higlier levels of spoken language: significant constraints on ernergentness appear to be exhibited bq (different types ot) motivatedness across all emergent levels, from the 'most' emergent feature-to-phoneme via the phoneme-to-morpheme to the idiomatic level, all of which speak for tlie predominance of recombination in language. The highest degree of arbitrariness, resulting in discontinuity and decoupling of a higher-level novel entity witli its own idctitity from its lower-level components and their arrangement, is exhibited in "the phoneme-morpheme relationship at the heart of spoken language" (Talmy 2007a: 56). So, this form of recombinance may well "prove to be the most extreme form of recombinance to be found across cognitive systems", to the effect that tlie 'output' entities "can only be remembered, not itiferred" (Talmy 2007a: 56, emphasis added). Here, clearly, a new patterning or 'layering' emerges - but one that does not coincide at all with the classical layering of structuralist linguistics. This re-affiliation maq most clearly be seen to be instantiated in acronyms, and in tlie present context, with an additional twist of the argument at that. When, as has bee11argued in the previous section, an acronym reading is presupposed for EMO. it may readily be re-analyzable as an ernergent composite. 111 this case an acronym represents a prototypical case of a linguistic form witli an associated meaning, and the acronym will exhibit the same degree of arbitrariness as enlo, siding with the plioneme-tomorpheme type of arrangement and also requiring tliat the specific plionemic-semantic coupling be acquired, or be stored, for tliat matter. Conceived of as an initialism, wliicli will of course be a more plausible reading for most of tlie cases like Etnergenq) ikfeaturen Organizalion, Etnergenl Murkcts O~llinc,E1ectvonic.r Mectingc Orgunizulion, Ecutnenicul Mini~triesof Oregon. Emergc.licy Mun~igenienlOffice, tlie plioneme-by-phoneme constituency is higliliglited, except, perhaps, for the abbreviations referring to tlie music and youth-culture domains of European Music Office; hence, EM0 is readily expanded into a multi-constituential composition. Even if this exemplar may count as a marginal case, again, no correlation of cognitive cliaracteristics and linguistic levels call be established. Recall, however, tliat Tal~nywould estimate tlie 'amount' of such noncorrelation in compositionality as rather low: Apart from any idioms contained in thern. compositional morphology and syntax themselves do not constitute a further level of emergentness. This is

58

Martina Luny~crtond Gunlhcr L~inpcrl

because the meaning of a whole complex word or phrase is systematically related to the meanings of the component morphemes and the morphological relations or constructions there are in (and, from some theoretical perspectives, the ~norphologicalrelations and constructions available in a language are themselves morphemes of a kind). (2007a: 49-50) This is indeed a different view on the compositionality issue compared to Langacker's claim that full compositionality is the exception to the norm of emergence in morpliological constructions. At least when conceived as a straightforward generalization, the ilnequivocal inclination in Talmy's accoi~ntof recombinance in favor of the recornbination(a1) type may well recluire more empirical evidence, especially in the face of tlie small selection of sample analyses offered in tlie previous section to illustrate recornbinance in morpliological compositionality. We propose, then, that. ultimately, it must be decided for each and every individual borderline case whether there is (more) eniergentness or (more) recombination, and to complicate matters, both the structural and semantic levels would have first to be individually evaluated and then mutually accommodated; moreover, there will even be variation in the assessments of the respective results across speakers and contexts that has to be reckoned witli. In very general terms, the cognitive mechanisni of recombinance has been found operating as a cross-level integrative phenomenon, fundamentally gradient in character and, on principle, incompatible with any notion of separate levels, domains, or sub-disciplines of (tlie study of) language in the first place. From this assessment it becomes apparent that neither emergentness (in the 'strict' sense) nor typical recombination can be correlated witli traditional approaclies to word-formation. What is more, beyond such 'global' incompatibilities, more 'local' correlations with types of traditional word-formation also turn out to be unwarranted: It becomes evident that the 'partitionings' of cognitive and structural views would not coincide for principled reasons rooted in the premises that inform their basic concepts. In Cognitive Semantics the fundamental conviction prevails that language is both an integrul and an independent part of cognition. Hence, there are significant overlups as well as non-overlaps in terms of organizing principles to be acknowledged, which in turn are ultimately due to the basic design of tlie human brain. This is what Talmy's Overlapping Systems Model of Cognitive Organization (e.g., Talmy 2000: 1) and his "new neural model" (Talmy 2003: 30) in fact suggest - an interaction of a 'core' language system in tlie brain witli one or another outside substantive system such as perception, motor control, affect, reasoningtinferencing, and

Word-Fonnatiori or wor~l',fir.n~ution59

culture, allowing for accommodating properties and functions specific to language witli those pervasive in other cognitive systems. These, in turn, interact in a principled manner witli basic organizing systems or cognitive faculties sucli as attention and memory. From tliis, differences in degree of recombinance can be seen not least in differences in degree across compositionality in language, tlii~sfalling short of any a priori determination of or assignment to linguistic levels. As to tlie familiar patterns, or types, of English word-formation, tliis (non-)correlation in recombinance results in a pervasive cutting across established boundaries, which reflects long-standing categorization queries in word-formation research (cf. Plag 2003 and Sclimid 2005). While traditional word-formation ultimately disregards the problems to save the umbrella category ~ord~fornlution, and construction grammar evades them by assuming the catch-all construct of the con,siruciion, Talmy's system of recombinance yields a differentiated categorization: 1.

2.

3.

Compounds, sucli as enlotion lunguage for instance, must be categorized as expressions (in Talmy's sense) tliat will be recombinational or, less probable. rnay rather be conceived of as semantically emergent (e.g., in the case of an agreed-on term). Formally, tliey are to be considered emergent expressions, that is? -phrasal idioms,' only if tliey exhibit a specific stress pattern (or spelling conventions in many cases) tliat sets them apart from tlie free combination of tlieir constituent morphemes. Affixat ions (pre- and suffixations) like emotive or emotionul would have to be checked as to their (formal and semantic) status of emergentness; prototypically, tliougli, they qualify as compositions. Structurally, gradience, tliat is variability as to recombination and emergentness, might be the norm; to illustrate: etnofive (presumably in tlie sense 'tending to excite or capable of exciting emotion,' but less so in the sense 'pertaining to the emotions, or to emotion') will most probably tend toward morpliological idiom status (just parallel to considerable), while emoiionul (like overer~iotiot~ulizatio~i)would count as a (nonidiomatic) complex word. In tliis view, affixation, yielding complex words, would represent tlie default process of word formation (in Talmy's sense). Blends (as instantiated in enioticoli, for instance), consisting of at least two formatives, would formally qualify as morphological idioms, semantically, however, as emergent complex words, ex-

4.

5.

hibiting different degrees of emergentness according to the (dis)contini~ityof their component entities. Yet their constituents would have to be differentially analyzed in terms of the organizing principles of word or morpheme formation, with a large amount of emergentness involved in cases like the coinplex word enloticon, while, e.g., notorious smog would be accounted for by phoneme-to-morpheme recombinance, a limiting case similar to s/op/Ls/urry/Ls/~xg/s/~~A~~~/s/uice/.s/ide/.s/ick/s/ip/,s/og/s/urp. The same assessment as to formal emergentness would hold for conversion, clipping, and back-formation (cf. Plag 2003: 22-27); these 'formation' types would not be instances of word formation in tlie first place. Their specifics could be dealt with in semantic terms; otherwise they would have to be analyzed just like other It should be noted that conversion, involving a category shift, represents another basic cognitive organizing principle pervasive in language that Talmy (2000, I: 4 I) terms intracategoriul conversion. As has been argued above, an acronym, perceived as a higlierlevel entity with its own identity, would qualify as an instance of morpheme formation, in fact an extreme example of emergentness, with tlie constituent phonemes re-combining to yield a morpheme; an initialism, presupposing a phoneme-by-phoneme reading, would rather instantiate a case of expression formation. Hence an alphabetism's categorical profile may well depend both on the mode of processing by and the contexti~alizedmeaning accessible to an individual language user, and is then seen to vary on tlie level of reconibinance.

As to the suggested four structural levels of recornbinance, only the last two are of importance for any account of word formation (recalling that morphenies may be eitlier emergent or recombinational structures, semantically speaking). With Talmy distinguisliing between morphemes atid (simplex and complex) words and expressions, expressions will consist of frce morphemes only, while words involve at least one bound morplieme (together with possible root morphemes that may fuliction as 'bases'). Yet another criteria1 factor keeps idioms separate from combinations at both tlie word and expression levels, so words and expressions may eitlier qualify as compositions or composites (in the terminology adopted here). Finally, morphology and syntax, assiuned to exist in the first place, are predominantly compositional in character (as we have repeatedly emplia-

sized), but may also be emergent to some extent (when morphological or pllrasal idioms are accounted for). So, Talmy's system of recombinance eventually cuts across all traditional demarcations. In reviewing the Cognitive Linguistics approaches addressing compositionality, a crucial issue that i \ seen to reappear may be framed in terms of similarity, which is another very foundational principle of cognitiotl." At a closer look, a potential decision on the matter of 'segregating' linguistic domains (or even subdomains) ultimately boils down to the question whether the differences are conceived of as reasonably succinct to legitimate a categorial division, or the opposite view appears more plausible and similarity will 'loom large(r).' A short final illustration of this argument will remain with Talmy's account: The crucial issue would indeed rest on the decision whether, say, tlie difference in types of arrangement or in constraints on the arrangement is of such a kind that a categorial division is only to hold at the recombinance boundary (as against the lower level of discreteness, state of categoriality) or whether 'feasible' differences can be identified across the whole range of recombinance, resulting maybe from specific recurrent clustering properties. Although Talmy (2006: l5), as has been demonstrated, posits identic a l arrangement properties for "word formation" (note the deliberate choice of open spelling) and expression formation, a dflerence in constraints on types of arrangement is likely to be assumed for the opposite poles. The arrangement types of bound morphemes making up a word (whether idiomatic or not) will be constrained by morphology, and those of free morphemes yielding an expression (whetlier idiomatic or not) by syntax. So, it becomes a logical 'fact' that these two domains neither straightforwardly coincide with traditional word-formation nor do their results neatly match the set of properties of either recombination or emergentness. Thus, there is, if at all, only a weakly cognitively motivated justification for a difference in terms. Reference may be made to the limiting case of idioms: In fact, they essentially exhibit tlie very ,same type of (mostly sequential) arrangement as non-idiomatic complex words and phrases, just following the requirements on certain "affixal or lexical categories to occur in certain structural locations that apply in general to the compositional (nonidiomatic) combination of morphemes" (Talmy 2007a: 47). Idioms will not differ "in kind" from their constituents and components but "are .sirnilur in character" (Talmy 2007a: 55, emphasis added) and diferent only in degree. So far, then, Talmy's meclianism of recombinance, as we read it, will clearly favor similarity across the whole 'range' of compositionality, and similarity is seen to 'emerge' where other accounts would deny or

62

Martina Lumpert and Gunther Lanlpert

neglect it. Moreover, there is no principled difference that would exactly match the traditional linguistic domain(s) o f word-formation. Even in difference, similarity appears to predominate across 'levels.'

Notes See Lampert and Lampert (2000) for such an endeavor in view of the conceptualizations of modality, negation, and attention. Cf., e.g., Gries (2003, 2004a.b, 2006) on subtractive forms of word-formation as exemplary cases for the new statistically informed, corpus-driven approach within Cognitive Linguistics. Note that hyphenated spelling is adopted here to designate the traditional term for both the specific domain of language and the discipline of linguistics. For a similar attempt with regard to phraseology cf. Gries (2008) and, focusing on the usage-based model, Croft and Cruse (2004: 291-327). The term cot7structioncrl idiom owes its existence to Jackendoff (see, e.g., 2002: 172). But note that Jackendoff has repeatedly expressed his indebtedness to Leonard Talmy's views on "space, aspectuality, lexicalization patterns, and force dynamics" (Jackendoff 1996: 97). Thc emzotion-network has been extensively made use of in M. Lampert (2009) to reconstruct differential patterns of attentional allocation in morphological compositionality. In passing, it may be noted that most of these aspects might receive a generalized treatment in terms of attention directing in language; see Talmy (forthcoming) and M. Lampert (2009). Langacker (e.g., 2005a,b) himself has, on more than one occasion, considered Cognitive Grammar as one (radical) version of'a construction grammar. See surveys in Croft and Cruse (3004: 257-290), Fischer and Stefanowitsch (2006), Schonefeld (2006), Broccias (2006). Evans and Green (2006: 64 1 706). An interesting terminological distinction is being made in Sag, Wasow, and Render (2003) reserving the term construction for combinations of formmeaning pairings, while using the term sign to refer to simple form-meaning pairings. See http:l/www.csl.sony.frierice2007irnaterial/construction-grammar-2.pd to hold such views may be smart, but is not really helpful since it shifts responsibility to the progress of psycholinguistic research. See Croft and Cruse's (2004: 268) distinction between reductionist and nonreductionist models of constri~ctiongrammar. While in reductionist models the

Word-Formation or wovd,formatior.l

13.

14. 15.

16.

17.

18.

19. 20.

63

atomic units (say, the morpheme) are prir??itive,and the complex units are 'derived', in non-reductionist models it is just vice-versa. Along these lines, Evans and Green (2006: 70 I) consider Radical Construction Grammar and Cognitive Grammar as non-reductionist, Goldberg-type Construction Grammar as partly reductionist and partly non-reductionist, and Fillmore and Kay's Construction Grammar as reductionist. In anticipation of what will be elaborated on below, one might characterize Bybee's usage-based network theory of morphology as non-reductionist. Cf. Michaelis (2006); the term has become popular through the dissemination of Paul Kay's (1997) lectures on construction grammar, see http://www.icsi.berkeley.edu/-kay/bcg/lec02.html, but ultimately goes back to Daniel Jurafsky's unpublished 1992 dissertation. Note that terminology in the present section follows Langacker (1987). Such views lnotivate the inclusion of Langacker's account in the currently emerging forniulaic language paradigm; see Wray and Perkins (2000), Wray (2002) and the papers from the recent 2007 Milwaukee Symposium on Formulaic Language. See http:llwww.uwm.eduiDept/English!conferences/fsl/index.html. For a reading of such words in terms of gestalts cf. Panther and Thornburg (200 I: 188) who propose -'a continuun~between the lexicon as a repository for unanalyzable forms and a putative morphological component, which contains schemata for forming words our of niorphe~nicbuilding-blocks". (Note the explicit reference to the building-block metaphor again.). Another piece of evidence is found in the context of Internet censorware that is likely to (falsely) filter out "the username hilaryunne ... because it contained the word uryun" or "the name Heather because it contained the phrase eat her.! Honourable mentions were given to another enquirer who could not access a biotechnology site because its name (acces.sexcellence.org) contained the word sex. .. . Other examples of words which have been banned include cucunzhers (because it contains cum), Matsushitu (shit), ana1y.si.s (anal), class (ass), and speech (pee)" (Crystal 2006: 224). Notwithstanding such caveats, in mainstream word-formation accounts this metaphor keeps lingering on, see, c.g., Plag et al. (2007: 65, 89): "minimal building block, morphemes", and: "one common strategy of creating new words, namely adding affixes to existing bases". Consult Gundersen (2001) on building-block vs. network morphology (with applications from Norwegian). Cf. Langacker (2000: 7): "ln saying that a schema is extracted, what is necessarily being claimed is actually fairly minimal: that the comnionality inherent in multiple experiences is reinforced and attains some kind of cognitive status, so that it has the potential to influence further processing".

64

Martitlu Lamper-t and Gzrnther Lainpert

21. See Tomasello (2003); an equivalent of such principles in adult language might be the concept of the collostruction proposed in Stefanowitsch and Gries (2003). 22. Note that the functionalization of constructions as templates was ah-eady apparent in Bloomfield's conceptualization. 23. An application is Riehemann (1998). 24. But see Langacker's (2005a, b) discussion of perceived differences between his own Cognitive Grammar and CxGrammar. 25. The exposition to follow is based on "Recombinance in the Evolution of Language" (2007a) and on a handout accompanying a talk on the occasion of the 6"' International Conference on the Evolution of Language, Rome, Italy, 4112106 (Talmy 2006). Note that of the two accounts, Talmy (2007a) is actually the earlier version. In passing, one may add that Talmy's concept of recornbinance may be linked with notions of compositionality and recursion that have spawned a fierce debate on the evolution of the language faculty, especially in the wake of Hauser, Chomsky, and Fitch (2002); for reactions see Jackendoff and Pinker (2005) and Fitch, Hauser, and Chomsky (2005). The cornpositionality issue from a Cognitive pel-spective is reviewed in De Beule and Bergen (forthc.) and Verhagen (forthc.). 26. Talmy's (2007a: 26) sketch of recombinance centers around the "distinctions within the cognitive representation of information" characteristic of two linguistic systems, spoken and signed language; much of the argument in view of spoken language will, however, apply to the written modality also, and, for the focus of the present paper, such differences will be disregarded for being of minor relevance. 27. On the history and possible conceptualization of emergence see Goldstein (1999) and, with specific relevance for Talmy's account, Bonabeau, Dessalles, and Grumbach (1 995a, b). 28. Cf., e.g., Talmy (2003) for spatial schemas and Talmy (to appear) for the attention system in language. See Dodge and L,akoff (2005) for an assessment of Talmy's achievements along these lines. 29. Note that in the vocal-auditory domain of a language, a non-recombinational type is encountered in the ancient system of "vocal dynamics", encompassing the essentially analog, i.e., non-digital and non-discrete, categories such as loudness, pitch, timbre, vocal effects like nasality, tenseness, breathiness, creakiness, distinctness, rate, duration that "are realized independently without entering rearrangements relative to each other" (Talmy 2006: 13). 30. These fonnal levels of language associate four or five distinct types of arrangement properties: "phonetic features in phonemeslphonemes in morphemesimorphemes in idiomsibound morphemeslidioms in complex wordslfree morpl~emeslwordslidiomsin expressions" (Talmy 2006: 19). It may be noted in passing that actually an "exception to full discontinuity" or 'complete' emergentness, however marginal it may be, can be identified in "featural

3 1.

32.

33.

34. 35.

36. 37.

effects" (Talmy's suggested term) in phonemes, when "an English speaker may have the sense - perhaps evident in poetry - that, say, l b l is closer to l p l than to Is/"; if phonemes were "purely autonomous higher-level units", such effects should not become manifest, hence "the featural makeup o f the phonemes persists to some degree in their new identities" (Talmy 2006: 25). Jackendoff mentions the example o f scissors, which is morphonologically regular, but does not denote multiple objects, rather an object with multiple parts. The difference to Langacker's approach becomes clearly evident in a quote from Tuggy (2005: 259): "[Langacker's] model utilizes exactly the same mechanisms and definitions for syntax as for morphology. The two domains are not in separate 'modules' o f the grammar, and nothing very important hangs on whether a formation is considered to be morphological or phrasal. I t changes from one category to the other gradually over time, and speakers are unlikely to be bothered by the change at all". Cf. Talmy (2006: 19-20): "semantic components in the meanings o f morpherneslidion~slthe meanings o f bound morphemeslidioms in the meaning o f complex wordslthe meanings o f free morphemes/wordslidioms in the meaning o f expressions". This is an essential and basic distinction in Talmy's overall framework; cf., e.g. (2000, chapter 1). This general attentional priority in cognition o f the ensemble or gestalt, amply confirmed in (visual) perception, dates back to Gestalt psychologists as early as the beginning o f the 20"' century. David Navon's (1 977) groundbreaking article "Forest before trees: The Precedence o f Global Features in Visual Perception" documents the so-called Global Precedence Effect. In the visual processing o f hierarchically organized complex patterns, a general precedence, in the incongruent condition, o f the global configuration has been attested when perception o f the global pattern interferes with local features, resulting in a significantly higher speed and processing precedence o f the global to the disadvantage o f the local pattern components. It should perhaps be added that precedence o f gestalt qualities does o f course not preclude perception of local features. In selective and distributed attention, which necessarily includes both bottom-up (input-driven) and top-down (concept-driven), local components are accessible to attention, since "[iln most real situations the task o f the human perceptual processor is not just to account for given input but also to select which part o f the surrounding stimulation i s worth perceiving, attending to, and processing" (Navon 1977: 355). Note that Talmy, in this context, makes extensive cross-systems reference to motor control, vision, and music. See M. Lampert (2009: 340-356) for a closer look at the 'referential space' covered by the intersection o f 'emotion' and 'language'; here at least the following attested selection comes to the fore: emoese, emo/ioncrlese. emotion

66

38.

39. 40. 4 1. 42. 43.

Mcrr-tinu Lurnpert unu' Gutither Lurnpert languuge, enlotionrrl language, languuge of emotion, emotion in lunguage, besides more complex expressions such as cnlotion us expressed in language etc. The difference turns out to be rather due to attentional than to strictly 'linguistic' differences: Indicative of stn~cturalidioniaticity would be an alleged difference in suprasegmentals (in spoken language), such as stress pattern, and its functional equivalents in the written modality, open spelling vs. hyphen or solid spelling; cf. M. 1,ampei-t (2009). For a detailed analysis see M. Lampert (2009: 3 16-33 I); it may be added that numerous EM(>-acronyms were also attested. Cf. littp://www.urbandictionary.com. http:l/unabridged.merriam-webster.com/cgibin/unabridged?va=emoticon&x=37&y=7; cf. M . Lampert (2009: 269). Hence, zero-derivation, for instance, would most plausibly be described in terms of semantic in- or under-determinacy (Farre11 200 1). See M . Lampert (2009).

Acknowledgements

My sincere thanks go to Len Tal~nyfor generously letting me have access to substantial portions of his (then) unpublished research (including also several extended comments and individual explanations by e-mail), M. L.

References Barlow, Michael, and Suzanne Kemmer (cds.) 2000 Usuge-Bused .Mou'c.ls cfL~mguage.Stanford, CA: CSLl Publications. Benczes, Keka Creative Conipounding in E~gli.sh.The Semantics uf' Metaphorical 2006 und Meton~~micul Noun-,Vou~Combinations. AmsterdamiPhiladelphia: Be~?jamins. Bergen, Benjamin K. 2007 Construction Grammar 11: What is a Construction. http://www.csl.sony .fr/erice2007!materialiconstruction-granimar2.pdf Bloomfield, Leonard 1933 Language. New York: Holt.

Word-Forumtion or word,fbrmation

67

Bonabeau, Eric, Jean-Louis Dessalles. and Alain Grumbach 1995a Characterizing emergent phenomena ( I ) : a critical view. Revue intert7utionale de .syst&niqlre 9.3: 327- 346. 1995b Characterizing emergent phenomena (2): a conceptual framework. Revue internationu/e de .~yst6~7iq(114e 9.3: 347-37 1 . Booij, Geert 2002 Constructional Idioms, Morphology, and the Dutch Lexicon. Journal ?f Gernzunic Linguistics 1 4: 30 1 -329. 2005 Compounding and derivation: evidence for Construction Morphology. In Morpholocg;l,und its Demarcations, Wolfgang U . Dressier, Franz Rainer, Dieter Kastovsky, and Oskar Pfeiffer (eds.), 109-1 32. A~nsterdamlPhiladelphia:Benjamins. 2007a Ti7e Grammar of Wordk. An Introduction to Morphology. Second edition. Oxford: Oxford University Press. 2007b Polysemy and Construction Morphology. In Leven rnet woorden. Fons Moerdijk, Ariane van Santen, and Rob Tempelaars (eds.), 355364. Leiden: lnstituut voor Nederlandse Lexicologie. Broccias, Cristiano Cognitive approaches to grammar. In Cognitive Linguistics: Current 2006 Applications and Future Perspectives, Gitte Kristiansen, Michel Achard, Rene Dirven, and Francisco J. Ruiz de Mendoza IbaAez (eds.), 8 1-1 15. Berlinmew York: De Gruyter. Brugman, Claudia What is the Invariance Hypothesis? C'ognitive Linguistics I : 2571990 266. Bybee, Joan L. Moi.phology: A Studj~@the Reiation between Meaning and Form. 1985 Amsterdam/Philadelphia: Benjamins. 1988 Morphology as lexical organization. In Theoretical Morphology: Approac1~e.sin Modern Linguistics, Michael Hammond and Michael Noonan (eds.). 1 19-1 4 1. San Diego, CA: Academic Press. A view of phonology from a cognitive and functional perspective. 1994 Cognitive Linguistics 5.4: 285--305. Regular Morphology and the Lexicon. Language and Cognitive 1995 Processes 10.5: 4 2 5 4 5 5 . The Emergent Lexicon. In CLS 31: The Panels. Papers,fvom the 34"' 1998 Annual Meeting e f t h e Chicago Lit7guistic Society, 4 2 1 4 3 5 . Chicago: Chicago Linguistic Society. Phonolo~yand nuturul lu17guuge use. Cambridge: Cambridge Uni200 1 versity Press. From usage to grammar: the mind's response to repetition. Language 2006 82.4: 71 1-733.

Bybee, Joan L., and James McClelland 2005 Alternatives to the combinatorial paradigm of linguistic theory based on domain general principles of human cognition. The Linguistic Review 2 2 . 2 4 : 38 1 4 1 0. Croft, William 2001 Radical Constuuciion G'runlniar. Syntactic Theory it? Typological Perspective. Oxford: Oxford University Press. Crystal, David 2006 Languuge und the Internet. Second edition. Cambridge: Cambridge University Press. Croft, William, and D. Alan Cruse 2004 Cognitive Linguistics. Cambridge: Cambridge University Press. Dqbrowska, Ewa 2009 Words as constructions. In New Directions in Cognitive Linguistics, Vyvyan Evans and Stefanie Pourcel (eds.), 201--223. AmsterdaniIPhiladelphia: Benjamins. De Beule, Joacliim, and Benjamin K Bergen (forthc.) On t h e ~ m e r g e n c eof Compositionality. To appear in Proceedings of the 6"' Evolution in Luvguage C'onjkrence. Available on-line at www2.hawaii.edd-bergen. Dirven. Rene 1999 Conversion as a conceptual lnetonytny of event schemata. In Metonyn7"v in lunguuge and thought, Klaus-Uwe Panther and Giinter Radden (eds.), 273-287. AmsterdamiPhiladelphia: Benjamins. Dodge, Ellen, and George Lakoff Image Schemas: From linguistic analysis to neural grounding. In 2005 From Perception lo Meaning, Beate Hampe and Joseph E. Grady (eds.), 57-92. BerlinlNew York: De Gruyter. Evans, Vyvyan, and Melanie Green C'ognitive 1,ingui.rtic.s. An 1171roditciion.Edinburgh: Edinburgh IJni2006 versity Press. Farrell, Patrick Functional shift as category underspecification. English Lur7guage 200 1 and Lingztistics 5.1 : 1 09- 1 30. Fischer. Kerstin, and Anatol Stefanowitsch Kon.slruktion.sgrumrnatik. Yon der Anwendung zur Theorie. Tiibin2006 gen: Stauffenburg Verlag. Fitch, W. Tecumseh, Marc D. Hauser, and Noam Chomsky The evolution of the language faculty: clarifications and implications. 2005 Cognition 97: 179-2 10. Geeraerts, Dirk The interaction of metaphor and metonymy in composite expres2002 sions. In Mctuphor and Meforl)!t~?yin Comparison und C'ontrust,

Word-Formation or u~ordfbrrnution 69 Rene Dirven and Ralf Piirings (eds.), 4 3 5 4 6 5 . Berlinmew York: De Gruyter. 2006 Methodology in Cognitive Linguistics. In Cognitive Linguistics: Current Applicutions und Futzire Perspectives, Gitte Kristiansen, Michel Achard, Kene Dirven, and Francisco J. Ruiz de Mendoza Ibafiez (eds.), 2 !- 49. BerlinINew York: De Gruyter. Gibbs, Raymond W. 1990 Psycholinguistic studies on the conceptual basis of idiomaticity. C'ognilive Linguistics 1 : 4 1 7 3 5 I. Goldberg, Adele E. 1995 Con.rtruction.s: A C'onstrucfion Grurnmur Approach to Argument Structure. Chicago: University of Chicago Press. 2006 Construc/ions a/ Work. The Nature of Generalization in Lunguage. Oxford: Oxford University Press. 2009 The nature of generalization in language. C'ognitive Linguistics 20. I: 93-1 27. Goldstein, Jeffrey 1999 Emergence as a construct: history and issues. Eniergence: Conlplexifi' and Organizufioti I : 49-92. Gries, Stefan Th. 2003 Some characteristics of English niorphological blends. In C,'LS 38.2: The Pane1.r. Papers ,fronl /he 38"' Regional Meeting ofthe Chicago Linguistic Sociegl, 20 1 ---216. Chicago: Chicago Linguistic Society. 2004a Shouldn't it be breukflmch? A quantitative analysis of the structure of blends. Linguistics 42.3: 639-667. 2004b Isn't that fantabulous? How similarity motivates intentional morphological blends in English. In Langtlage, culture, and mind, Michel Achard and Suzanne Ketnmer (eds.), 4 1 5 4 2 8 . Stanford, CA: CSLl Publications. 2006 Cognitive determinants of subtractive word-formation processes: a corpus-based perspective. Cognifive Linguistics 17.4: 535-558. 2008 Phraseology and linguistic theory: a brief survey. In Phruseology: an interdisciplinuy pcrspecfive, Sylvaine Granger and Fanny Meunier (eds.), 3-25. AmsterdamIPhiladelphia: Benjamins. Gundersen, Helge Building blocks or network relations: Problems of lnorphological 2001 segmentation. In A C'ognitive Approuch lo the Verb. Morphological und Constructior~ulPer.spective.s, Hanne Gram Simonsen and Rolf Th. Endresen (eds.), 95-127. Berlin/New York: De Gruyter. Hauser, Marc D., Noaln Chonisky. and W. Teculnseh Fitch The faculty of language: What is it, who has it, and how does it 2002 evolve? Science 298: 1 5 6 9 1579.

70

hltrrfina Lunipert and Gunther Lumperf

Jackendoff. Ray 1996 Conceptual Semantics and cognitive linguistics. Cognitive Linguist i c ~7: 93-1 27. 2002 Foundutions of'lJanguuge.Oxford: Oxford University Press. Jackendoff, Ray, and Steven Pinker 2005 The nature o f the language f a c ~ ~ land t y its implications for the evolution o f language. Cognition 79: 2 1 1-225. Kay, Paul Berkeley CC'nsfruction Grunimur. Available on-line at 1997

http:llwww.icsi.berkeley.edul--~kayiI~cgllec02.html. Kemmer, Suzanne, and Michael Barlow 2000 Introduction: A Usage-Based Conception o f Language. In UsageBused Models of'lunguuge, Michael Barlow and Suzanne Kemmer (eds.), vii-xxviii. Stanford, CA: CSI,I Publications. Kemmet-. Suzanne 2003 Schemas and lexical blends. In ,Vo/ivation in lunguage; Studies in Thomas Berg, Rene honor of' C'i411ferRudden, Hubert C~~yckens, Dirven, and Klaus-Uwe Panther (eds.), 69-97. AmsterdamlPhiladelphia: Benjamins. 1,akoff. George I987 Women, Fit-e, and Dangerous Things: Whal Categories Reveal .4hout the Mind. Chicago: University o f Chicago Press. 1990 What is the invariance hypothesis? Cognitive Linguis/ic.s I: 39-74. Lakoff, George, and Mark Johnson 1980 Metaphors We Live By. Chicago: University o f Chicago Press. Lanipert, Giinther, and Martina Lanipert 2000 The conceptual .structz4rc(.s) c?f'modulity; essences and ideologies. A study in linguistic (ifietu-)ca/t.,aorizution.Frankfi~rtam Main: Lang. Lampert, Martina Attention und Recomhinunce: A Cognitive-Semantic Investigation 2009 into Morphological Conipositionality in English. Frankfurt am Main: Lang. (forthc.) Re.~pects,for similuri@. Ps~~chologicul-e.ssentiaIi.stheuristic.^ in ico-

nicity. Langacker, Ronald W. Fozrndulions of'('ogi1itive Grurnniur. Volume I: Theoreticul Prereq1 987 uisites. Stanford, CA: Stanford University Press. Found~ition.s(!f'C'ogniliveGramnlur.. Ci~lume2: De.s.scriptive Applicu199 la tion. Stanford, CA: Stanford University Press. Concept, Inluge. and k ~ ~ v n h oThe l : cognitive busis of grumnzur. Ber199 1 b IinMew York: De Gruyter. Gramrnur crnd (~'oncep/ualization.BerlinMew York: De Gruyter. 1999

Word-Formution or word,fbrniulion 2000

2005a

71

A Dynamic Usage-Based Model. In Usage-Based Models qf'Languuge, Michael Barlow and Suzanne Kemmer (eds.), 1-62. Stanford, CA: CSLl Publications. Construction Grammars: cognitive, radical, and less so. In Cognitive Linguistics: internal ~z'ynumics and interdisciplinary interaction, Francisco J . Ruiz de Mendoza Ibaiiez and M. Sandra PeAa Cervel

(eds.). 10 1-1 59. Berlinmew York: De Gruyter. Integration, grammaticization, and constructional meaning. In Grammutical ('onslructions, Back to the Roots, Mirjam Fried and Hans C. Boas (eds.), 157-1 89. A~nsterdamIPhiladelphia:Benjamins. 2007 Ten Lectures on ('ognitive Gr~mrnnicxr.Beijing: Foreign Language Teaching and Research Press. Michaelis, Laura 2006 Construction Grammar. In The Enc,vclopediu of Lungzrcrge and Linguistics. Second edition. Vol. 3. Keith Brown (ed.), 73--84.Oxford: Elsevier. Navon, David 1977 Forest before trees: the precedence o f global features in visual perception. Cognititle P g ~ c h o l o a9: 353-383. Nunberg, Geoffrey, Ivan A. Sag, and Thomas Wasow 1994 Idioms. Language 70.3: 49 1--538. Panther, Klaus-Uwe, and Linda Thornburg 200 1 A conceptual analysis o f -er nominals. In Applied Cognitive Linguistics 11: Language Pedugoa, Martin Piitz, Susanne Niemeyer, and Rene Dirven (eds.), 149-200. BerliniNew York: De Gruyter. Pinker, Steven Words and Rules: The ingredients of' language. London: Phoenix 1999 Books. Pinker, Steven, and Alan Prince Regular and irregular morphology and the psychological status of 1991 rules o f grammar. In Proceedings qf the 17'" Annual Meeting qf the Berkeley Linguistics Society, 230-25 1 . Berkeley, CA: Berkeley Linguistic Society. Pinker, Steven, and Michael T. lllllnan The past-tense debate. The past and future o f the past tense. Trends 2002 in Cognitive Sciences 6.1 I: 456463. Plag, lngo Word-Forn7ation in English. Cambridge: Cambridge University 2003 Press. Plag, Ingo, Maria Braun, Sabine Lappe, and Mareile Schramin Iritroduction f o English linguistic:^. BerlinINew York: De Gruyter. 2007 2005b

72

Murtit7u Lamper/ and Gijnther Lanlpert

Kieliemann. Suzanne %. 1998 Type-based derivational morphology. The .Journal r?f'Con7parative 2: 49-77. Gertnunic 1,ingui.sfic.~ Ryder, Mary Ellen 199 1 Mixers, mufflers and mousers: The extending of the -er suffix as a case of prototype reanalysis. In Proceedings ofthe 17"' Annual Meeling of'the Berkeley linguistic.^ Sociew, 299-3 1 1 . Berkeley, CA: Berkeley Linguistic Society. 1994 Ordered Chaos. The Interpretation qf English Noun-Noun Compounds. Berkeley, CA: The University of California Press. Bunkers arid blue-chippers: An account of -er formations in present1999 day English. English Language CIIILJ Linguistics 3 : 269-297. Sag, Ivan A,, Thomas Wasow, and Ernily Bender 2003 Syntactic Theory: A Fortt7ul I11trod7rction.Second Edition. Stanford, CA: CSLl Publications. Schmid, Hans-Jorg 2005 Englische Morphologie ur7d ni)rthildung: Eine Einjiihrung. Berlin: Erich Schmidt Verlag. Schonefeld. Doris 2006 Constructions. constructiotis-online. de. Stefanowitsch, Anatol. and Stefan Th. Gries Collostructions: investigating the interaction of words and construc2003 Linguistics 8: 209-243. tions. Inter-nationalJournal qf'(,'or~i?u.s Sweetser, Eve 1999 Compositionality and blending: semantic composition in a cognitively realistic framework. In ('ogt7itive Linguistics: Foundutions, Scope, und Me~hodology,Tlieo Janssen and Gisela Redeker (eds.). 129-162. Berlin/New York: De Gruyter. Talmy, Leonard Toward a Cognitive Semant ics. l'olutrr e I : Concept Struclz/ring Sy.y2000 terns. Cambridge. MA: MIT Press. The representation of spatial structure in spoken and signed lan2003 guage. To appeal- in Pcr:s/~ective.son classifier conslruclions in sign language, Karen Emniorey (ed.), 169-195. Mahwah, NJ: Erlbaum. Recombinance in the Evolution of Language. [unpublished handout 2006 of a lecture in Rome]. Recombinance in the Evolution of Language. In CLS 39-2: The Pan2007a els. Papers ,from the 39"' Annual iileeting qf' the Chicugo Linguistic Society. J . Cihlar, A. Franklin, D. Kaiser, and I. Kimbara (eds.), 2660. Chicago: Chicago Linguistic Society. Foreword. In Melhods in Cognilive Linguistics, Monica Gonzalez2007b Marquez, Irene Mittelberg, Seana Coulson, and Michael J. Spivey (eds.), xi-xxi. AmsterdamIPhiladelphia: Benjamins.

Word-Formution or wiord formation

73

(forthc.) The attention s.vstenz in lunguuge [to be published by MIT Press]. Taylor, John R. 2002 Cognitive Grann~~ur. Oxford: Oxford University Press. Tomasello. Michael Construclir~gu Langrloge: .A Ikclge-Bused Theory of' Lunguage 3003 Acyuisi/ion. Cambridge. MA: Harvard University Press. Tuggy, David 2005 Cognitive Approach to Word-Formation. In Handbook qf WordFormcrtion, Pavol ~ t e k a u e rand Rochelle Lieber (eds.), 233-265. Dordrecht: Springer. Ungerer, Friedrich 2002 The conceptual function of derivational word-formation in English. Atigliu 120, 534-567. 2007 Word-Formation. In Hcrndbook qf'(.'ognitive Linguistics, Dirk Geeraerts and Andre Cuyckens (eds.), 650-675. Oxford: Oxford University Press. Ungerer, Friedrich, and Hans-Jorg Schmid 2006 An Introduction to Cognitive Linguistic.s. Second edition. Harlow: Longman. Verhagen, Arie jforthc.) Syntax, recursion, productivity - a usage-based perspective on the evolution of grammar. To appear in Evidence and Counter-evidence. Fe.stschrift F. Kortland, A. Lubotsky, J. Schaeken, and J. Wiedenhof (eds.). Amsterda~nlNewYork: Rodopi. Wray, Alison, and Michael R. Perkins The functions of formulaic language: An integral model. Language 2000 and Comm~nication20: 1-28. Wray, Alison Fornzuluic Languuge lrnd the Laicon. Cambridge: Cambridge Uni2002 versity Press.

On the viability of cognitive morphology for explaining language changeA Livio Guetu

1. Introduction The dichotomy nature vs. tiurture has been defined as "[tlhe fundamental question of tlie developmental sciences" (MacNeilage 1997: 302). Nature has to do with the cognitively-grounded endowment specific of human beings, whereas nurture can be seen as the result of a learning process, which only indirectly reflects cognitively-grounded properties of the language faculty. The diachronic dimension provides a bridge between nature and nurture, in the sense that language change results, at least partially, froin the action of selective cognitive abilities associated with the single components (or dimensions) of the language faculty. On the other hand, these selective cognitive abilities may give rise to conflicts among the different dimensions of tlie language faculty, in that a certain language change, locally resulting from a natural, i.e., cognitively well-founded, process, may produce unnatural structures or configiirations along other dimensions, mhich require nurture. In particular, this paper will raise tlie question whetlier a phenomenon which is massively encountered in natural languages such as homonymy may be traced back to a large extent to welldefined "natural" patterns of diachronic evolution based on cognitivelygrounded processes. On the one hand, this is trivially the case: it is fairly well known that homonymy results from natural sound change. The fact that hear and /?ere turned out to be homophonous is due to independent sound changes, which are ipsojaclo motivated by the cognitivelq-founded articulatory and acoustic apparatus of human beings. Nevertheless, the lexical association arising as a consequence of homonymy is coirlpletely opaque to the speaker. This does not exclude that secondary motivation may arise due to folk etymold and weeds, only ogy, as in the case of weed from Old English ~ @ o 'grass' weedk$,from OE utmd 'cloth' (cf. current in the expression ~~i~d(nt''s Bloomfield 1933: 436). As commented by Ullmann (1957: 128), it is per-

fectly natural that the speaker is tempted to establisll a connection in such a case, transforming the homonymy into polysemy. On the other hand, this is trivially not the case: borrowing rnay create homonymic pairs, wliicli are not moti~atcdin any meaningfill way by the cognitive endowment, as in the case of heaver, which goes back either to OE hec?for 'castor' or to Old French buvi2re 'gag', a derivative of bave 'dribble' (cf. Ullniann 1957: 128). Again, a secondary motivation due to folk etymology is lurking here. Besides these trivial cases, the hypotliesis that is at stake is that cognitively guided semantic processes of meaning extension such as metaphor and metonymy are of paramount importance in reconstritcting semantic change. Croft and Cruse (2004) take a rather optimistic stance on the question, considerably simplifying the issue by teasing the diachronic and the synchronic perspective sharply apart. In a diachronic perspective, "homonymic units are derived from distinct lexical sources, and their orthograpliicaltphonological identity is due either to the loss of an original distinction due to language change, or to borrowing" (Croft and Cruse 2004: 1 1 1). On the other hand, "polysemic units are derived from the same lexical source, being the result of processes of extension such as metaphor and metonymy" (Croft and Cruse 2004: 1 I I). Given these premises, the following claim is made: The diachronic distinction between homonymy and polysemy is a yeslno matter, and is a question of historical fact, resolvable in principle, if not always in practice. The synchronic distinction is less firmly based, and is a matter of degree. The question is whether there is a felt semantic relationship between two interpretations of a word or not. (Croft and Cruse 2004: 1 1 1). Relying on two different case studies, this chapter shows that far from being only a yestno matter, the diachronic dimension offers much more troubles than what Croft and Cruse want 11sto believe. On tlie one hand, an apparent case of nurture will be presented, which is nicely explained as a case for nature when tlie diachronic dimension is considered. On the other hand, an apparent case of nature which has received general agreement upon a supposed "felt semantic relationship" reveals unexpected patterns of nurture, if diachrony is seriously considered. This amounts to say that if it [nay be true that a diachronic distinction between liomonymy and polyselny can in principle always be made, its bearing on the synchronic distinction is far less clear. On tlie one hand, reconstructed patterns of meaning extension may shape the onomasiological domain of certain mor-

C'ognitive morphology for euplaining language change

77

phemes (what is called layering in gralntnaticalizatio~istudies, cf. Hopper and Traugott 2003: 49), leading to uncertainty about how speakers organize the semantic space of lexical entries. On the other hand, diachronic developments may collide with synchronic assumptions about meaning extension, providing alternative interpretations in neat contrast with general views of the cognitive processes synchronically applied by speakers. This makes the diachronic perspective unescapable for a far-reaching analysis of language as a cognitively founded faculty of human beings as demonstrated in tlie following two case studies on verbal and suftixal homonymy.

2. The passive auxiliary ginn in Luxembourgish In tlie Luxe~nbourgishdialect,' a rare example of surface homonymy occurs, in that tlie verb for G I V E is homo~iymicwith the auxiliary used to form the passive:'

(1)

a. ech gi gcsinn 1 give seen 'I am seen' b. ech ginn der e Buch 1 give y o u : r ~ r a book 'I give you a book'

Notice that the two homonymic verbs display different morpliolexical properties, because they select different auxiliaries, resp. BE and I1AVL, when occurring in tlie perfect construction: (2)

a. ech si gesinn ginn I am seen given 'I have been seen' b. ech hurrn der eppes gin11 I have you:DAl' something give11 'I have given you something'

Finally, the same verb is used as an inclloalive copula, and as an auxiliary for the subjunctive form:

78 (3)

Livio Guetu

a. ech gi kra~lk I give i l l 'I become i l l ' b. wann ech Z i t hatt, gkij' ech hlei~jen if I time had, ~ ~ V ~ : SI I Jstay BJ 'If I had time, I would stay'

In all these usages, the L,i~xernbourgishverb for G I V E corresponds to the Modern Standard German (= MSG) verb ~ e r d e n'become':

(4)

a. Ich werde gesehen I become seen 'I am seen' b. Ich werde krunk I become i l l 'I become i l l ' c. Wenn ich Zeit hdtte, wtjurde ich hleihen if I time had, become:sur~~ 1 stay 'If l had time, I would stay'

Except for the further usage of MSG werden as an auxiliary for the future periphrasis. which is not attested in Lu~embourgish,~ we can establish a synchronic equivalence between Luxernbourgish ginn and MSG werden. Fi~rtliermore,notice that the liomonqmy between the passive auxiliary and the verb for GIVI: is very peculiar, given that a similar pattern is only attested in some Chinese dialect (cf. Haspelmath 1990)."s this homonymy a case for nurture or nature'? Clearly, from a synclironic point of view, the homonyniy is particularly astonishing, if the selectional properties of the predicate for G I V L are considered. The latter is specified for an agentive subject and a patientive object, as opposed to the typical properties of a passive auxiliary, usually incompatible with an agentive subject. The contrast with its equivalent MSG verb u'erden is evident: for the latter, a consistent spectrum of polysemy may be reasonably assumed, in which the socalled 'fientive' verb (cf. Haspelniath 1990) has been the starting point for a grammaticalization process to start, crucially centered on the resultant state attributed by the past participle to a patientive subject, as in ich ~ ' e r d e genesen ' I am cured' which parallels ich M~erdekrank 'I become ill'. On

Cognilive morpholom,for explrrining lunguage change

79

the contrary, tlie hypothesis of a polysemic analysis looks quite improbable in the Luxembourgish case since no features appear to have been shared. Even worse, the fientive value displayed by the Luxetnbourgisli ginn appears particularly difficult to combine with the basic meaning of the verb. In the next sections, we will see that far from being an unexplainable quirky characteristic of this small dialectal area a quite natural pattern of meaning extension lies behind the gra~ntnaticalizationof G I V E as a passive auxiliary.

2.1. The cognitive representation of C;IVL A general cognitive representation of a sentence like Mory gove John o hook is given in the following picture, in which there is the profiling of tlie movement caused by a trajector, tlie giver, on a landmark, a generic thing, which goes out of her domain and enters into the domain of the second landmark, tlie recipient (cf. Newman 1996: 47):

RECIPIENT

One of the possible extensions of this image-schema involves the backgrounding of the recipient, as in the following MSG examples (but similar extensions occur in a wide spectrum of languages as well, cf. Newman 1996 for a broader picture):

(6)

a. Der Buunz gab viele Fruchte. 'The tree gave many fruits' b. Der Ofin gihf Warme. 'lit.: the stove gives heat' c. Der Geiger giht eiv~Ko~zzeri. 'The violinist gives a concert'

80

Livio G'uetu

As commented by Newman (1 996: 144): [tlhe .S~IIN(;may be viewed ... as emerging out of some physical region, and it is this way of viewing the movement of the 7111NG which motivates a large group of extensions involving emergence and manifestation of entities.

Besides the backgrounding of the recipient, this set of examples is characterized by a growing degree of abstractness for the things involved, from the concrete entities of (6a) to the event of (6c). A filrther step in this abstraction process implies the backgrounding of the causing entity as can be gathered fro111the comparison of the follom ing two image-scheinas:

The meaning extension portrayed in (7b) shows that these abstracted senses of "GIVE verbs tend to occur in constructions which lack the typical agent-patient contrast" (Newman 1996: 156). This, in turn, is responsible for the occurrence of the existential construction i n MSG, as recognized at least since Grimm ( 1 837: 266): (8)

a. Das Wetter ist .sehrgiinstig: das - e.s giht gute Ernte. 'The weather is verj favourable: this - it gives a good harvest' b. Es giht viele Kinder in &r Schule. 'There are many children in the school" c. Es giht eirwn Gott. 'There is one God'

This construction seems to have been gaining more and more terrain since the sixteenth century, especially in the dialectal area along the Rhine. The first writer to make a consistent use of the existential construction was Hans Fischart, who was active in Strasbourg in the second half of the sixteenth century. This is obviously not coincidental. We will see that this historical fact provides the key for interpreting the actual usage attested in Luxembourg.

Cognitive morphology fat* explaining language change

81

3.2. The cognitive history of the fientive extension of GIVE

It is tlie merit of Newman (1998) to have pointed out the role played by discourse i~nplicaturesin giving rise to slight meaning extensions, which finally led to the crystallizatioli of tlie ex~stentialusage of the verb for GIVE in this dialectal area, as well as in the rest of the German-speaking world.' A parallel number of similar extensions, driven by discourse implicatures, can also be assumed for the fientive extencion which constitutes the bridge for GIVE to be further gra~nrnaticalizedas a passive auxiliary. For the development of the existential constructio~l,Newrnan (1998) studied Hans Fiscliart's text Geschichtklitterurig. which is quite a free adaptation of Rabelais' Gargantua. In his analysis, he gives the following examples: (9)

a. warlr? nilr alte Weiher unnd die Hund &an ,seychten, so gehs guten Burgundi.i.\chenSaltpeter (Ge.vch. 1 25, 37-38) 'having just old women and dogs urinate on it would produce good Burgundy saltpetre' b. verzicht nzir, daJ ich euch den Suuen vergleich, sie gehen dannoch guten Speck (Gesch. 56.30-3 I) 'pardon me that I compare >ou to sons, but they do produce good bacon'

The first example is an instance of tlie existential construction, in which the premise contained in the subordinate clause allows a certain entity to collie about. In tlie second example, there is also the coming about of a certain entity; in this case, however, the musee is seen as a natural expansion of properties contained in the traiector. In other words, tlie latter is in a metonymic relation with the causee. l'hc premise for the emergence of a new cntity or condition can also be contained in tlie discourse context both for the existential and the fientive meaning, as shown by the following examples from Fiscliart and fro111a text of the sixteenth century reported in DWB, s.v. geben: (1 0) Geltet ihr Fronecken, welche nit gern spinnen, die gehen gute Wirtin? (Gesch. 135. 29f.) 'Isn't it so that your girls who don't like to spin will make (lit. give) good innkeepers/in~ikeepers'w ites?'

( 1 1 ) spcmnen sie tiicher ohen in uuf die u'tx'cher,dusz sie in der n~itteher-

ah hangen und einen sack gehen 'They hang out tlie sheets above on the roofs, so that they hang in the middle and form (lit. give) a sack' The basic difference between the two outcomes consists in the backgrounding of the trajector in the existential meaning, and in a metonymic relation between trajector and landmark in tlie fientive meaning, in that "a kind of tnovement of a new entity out of a phqsical region associated with the producing entity" is involved (cf. Neuman 1998: 317). The identity between the "producing entity" and its natural expansion, as in the case of the sow and the bacon above, is the crucial step whicli changed a typical , a fientive predicate." agentive verb like C ~ I V Finto In the following table the four steps are summarized whicli, on the basis of discourse implicatures, led to the two different outcomes, respectively the existential and the fientive meaning: Tuble I. Discourse itnplicatures leading to thc existential and to the fientive mean-

ing extension of the verb GIVI.. Tliere is a causal relationsliip between some entity X and tlie emergence of another entity Y. Y Y B Esgiht Y Tliere is some entity .Ygihl Y Tliere is a causal relationship between some Y wliich will exist entity X and the emersubsequent to the gence of another entity event described by the Y, whicli represents a antecedent clause. natural expansion of X. A Xgiht Y

+

JI

JI

JI

C E,r gih/ Y Tliere is some entity Xgzhr Y, A new entity / property Y, co~iiesabout, which Y wliich will exist is conceived as the desi~bsequent to the velopment of (natural) prior events. properties of X.

1

D E.5 giht Y

I

Y exists.

Xgiht Y,

I

X becomes Y,.

C'ogtlirive rnorphology,for explaining languuge change

83

Once that the fientive meaning came about, the conditions were met for the verb for C;lvC to cross the path of the other tientive verb werden and to share its destiny, namely of being grammaticalized as an auxiliary in the passive construction as well as in the other auxiliary functions like the subjunctive periphrasis. All of this brings us back to the initial question of tlie distinction between polysemy and liomonymy. Given that the diachronic process leading to the grammaticalization of GIVE as a passive auxiliary looks perfectly natural, how is tlie synchronic relation between the two different usages of ginn to be conceived of in Luxembourgish? I am not sure uliether a pure homonymic solutio~iis colnpletely satisfactoty, because the fientive meaning is definitely vital and allows the speakers to reconstruct a direct relation between fientive and passive usages like those mentioned above of ich werde krunk and ich werde genesen. Furthermore, the table illustrates how the linkage between the basic meaning of G I V E and its fientive extension can be reconstructed in the universe of discourse. In this light, tlie clear-cut dichotomy between synchrony and diachrony postl~latedby Croft arid Cruse (2004) is much less safe than what they claim. If it is true that this can be considered a case of diachronic polysemy, its synchronic treatment remains partially obscure and in need of further research on how the speakers of Luxe~nbourgisliperceive the lexical relation between the two verbal usages.'

3. Polysemy and homonymy in agent and instrument nouns The other case-study at issue represents in a certain way the mirror-image of what has been discussed in the previous sections. In fact, tlie polysemy of agent and instrument is given for granted in many approaches to the semantics of word formation processes."~ mention just a couple of approaches couched in different theoretical frameworks, Booij (1 986) attributes the range of meanings displayed by the Dutch suffix -er, as shown below in (12), to a universal, cognitively-founded hierarchy, wliich, from the basic meaning Personal Agent, leads to tlie other meaning extensions: (1 2) a. I~I-KSON b. OBJECT c. PI,AC'I: d. 1:VCNr

npel-er 'player' open-er 'opener' hijsluit-er 'enclosure' trejFer 'hit', 'goal'

< ,\pel-en 'to play' < open-en 'to open' < bijsluit-en 'to enclose' < treflen 'to hit'

Personal Agent > Impersonal Agent > Instrument (> Place I Event)

84

Livio Guetu

The hie!-arcliy is claimed to be supported by a universal cognitive tendency wliicli is -'presumably language-independent", so that "we expect the same polyseniy to exist for agent nouns in other languages" (Booij 1986: 5 1 1). Tlii~s,in his approach based on the autonomy of language from cognition, Booij (1986: 5 12) emphasizes that the categories responsible for the meaning extensions are "conceptual categories, not linguistic categories". Panther and Tliornburg (2002: 285) defend an opposite view, which identifies in "two high-level conceptual metapliors, personification and reification" as well as in specific "conceptual metaphors and metonymies" the "account for the polyse~nyof the -er suffix" in English. The careful investigation of the different possible meanings displayed by tlie English -er formations is sulnlnarized in the following figure, which is centered around the prototypical agent meaning, strictly connected with the instrumental extension:

\

'one w~thunexpected

Inanimate

f--

Object 'underground railroad t ~ e '

/

/

'vehicle space d e s i g ~ ~ efor d sleepmg'

Purpose-locatio n

HUMAN AGENT ---+ Instrument 'one inclincd to sleep'one sleeping'

4

'inactive spy

'

'sleeping pill'

t Quasi-Instrument -(child's) sleepwear'

E V F N T LLVI'L 'bonng event'

Figure I . Polyserny of ,sleeper (cf. Panther and Thortiburg 2002: 3 10)

From t!iis viewpoint, and abstracting away from tlie different views on the autonomy of language from cognition, Booij's and Panther and Thornburg's approaches rouglily converge in seeing a conceptual core accompanied by further meaning extensions, wliicli may be projected onto a universal conceptual level, valid for all languages. In fact, similar analyses have

Cognitive morpho/ow,for explaining language change

85

been proposed for other languages as well (cf. Dressler 1980 for a typological perspective). Quite in contrast with this received opinion, Rainer has recently suggested a radically different interpretation for the facts occurring in Spanish as well as in Italian (cf. Rainer 2004a, b). In spite of the similar pattern observed synclironically, in wliich a range of different meanings is displayed by the derivatives respectively formed with the suffixes -dor and -tore, the diachronic development undermines an explanation simply based on meaning extensions like those assumed by Booi.1 and Panther and Thornburg. The synchronic pattern occurring respectively in Spanish and Italian mirrors the Dutch and the English picture quite closely, at least for the higher positions of Booij's l~ierarchy:~ Sp.jugu-u'or 'player' It. gioca-tore 'player'

part of separable complex verb > prefix) and that this change '-implies a loss of lexical meaning" (Booi-j 2002: 218). Compare for instance the homographic voor-kom 'before-come', which can mean ( I ) 'appearloccur to', or (2) 'prevent'. In its first usage, the preposition voor- has the status of a particle since the past tense form of the verb is voor-ge-kom 'before-I>\r -come, appearedloccurred'. However, in the second instance voor- already has reached the state of a prefix in the grammaticalization process (with a meaning similar to 'pre-'): hence, the past tense is realized by a zero morpheme (i.e., 13-voor-kom 'PSI-beforecome, prevented'), as is the case in other verbs with the prefixes he- (as in be-twit 'contest'), ge- (as in ge-dru 'behave'), er- (as in er-ken 'recognize'), her- (as in her.-.tien 'revise'), ont- (as in ont-slmn 'dismiss'), and ver- (as in ver-~koon'excuse') (Taalkommissie 2002: 189). So, we can conclude that prepositions, as items from a closed-class that are generally not promiscuous, should be analyzed as either particles (i.e., as autonomous stem-like components) or prefixes, and thereby explaining their higher degree of promiscuity. All other component structures are both phonologically and semantically rather clioosy in terms of their combinatory potential in morpho-

Component structures in morphologicul constructions

1 13

logical constructions. Affixes combine on11 in certain contexts with certain stems; for example, the Afrikaans partitive genitive morpheme is highly choosy (even more so than its Dutch counterpart) since it combines only with certain ad-jectives in highly specified syntactic context (i.e., only after indefinite pronouns). Tlie same applies to all other affixes. Linking morphemes could be considered choosy since they neither occur systematically in various phonological contexts (Combrink 1990), nor do they occur systematically in all compounds. Likewise, the variety and occurrence of zero morphemes are highly limited and specified while cran morphemes couldn't be considered productive at all. One important aspect that sliould be kept in mind when considering valence of component structures in Afrikaans (as is the case in Dutch) is that Afrikaans component structures could, in broad terms, be divided in two strata: a native (Germanic) stratum and a nonnative (Romance) stratum. Examples of suffixes from tlie native stratum includes -heid as in besig-heid 'business', -ing as in hots-ing -collision', and -surd as in ryka~rrd'wealthy person'; exa~nplesfrom the nonnative stratum include -teit as in puher-teit 'puberty', -iet as in trun.sve~t-iet'transvestite', and -cur as in kontruki-ezrr 'contractor'. Witli regard to promiscuity, it can be taken as a generalizing rule that nonnative s~~ffixes combine only witli nonnative stems while native suffixes combine with both native and nonnative stems (Booij 2002: 95), thereby limiting the promiscuity of these component structures. There are of course, once again, various examples that extend from this pattern. For example, tlie nonnative adjectivalizer -ief not only combines with nonnative stems (as in kongest-ief 'congestive', and ahortiej 'abortive'), but also with native stems (as in sport-icf 'sportive', and foul-ief 'fa~~lty'); whereas, contrary to the principle stated above, the native personifier -aard only combines witli native stems (as in ~.vocst-aard'savage' and gry,s-uard 'elder'). One should note that, when applying strata considerations to promiscuity, component structures could be hybrids or could show signs of homonymily. Booi-j (2002: 96) illustrates for example that in Dutch, co~nplex units with tlie native suffix - i ~ c h(as in afgod-i,sch 'idolatrous', fantmt-isch 'fantastic', and Belg-i~ch'Belgian') have no marked status with regard to strata restrictions and -isch could therefore be considered a hybrid; the same w o ~ ~ apply l d to tlie -eer suffix in Afrikaans, in examples such as joutcer 'err' (combination with native stem) and kommunik-ecr 'communicate' (combination with nonnative stem). Tlie Afrikaans nominalizing suffix -~r.tie(as in kommunik-asie 'communication' and lekk-a.tie 'leakage') is

actually homonymous with different origins; in Dutch, this is still reflected in the orthography with two different suffixes, -atie (as in communic-atie 'communication) and -uge (as in Iekk-uge 'leakage').

3.4. Constituency Constituency is the last parameter in terms of which component str~~ctures could be characterized. It is defined as "the order in which component structures are successively combined to form progressively more elaborate composite structures" (Langacker 1987: 3 10). Since Langacker (1 987: 3 17) explicitly states that constituency is a secondary valence factor, where the "choice of a particular constituency arrangement is often not critical" (Langacker 1987: 3 10, see also Tuggy 2005: 257), it will suffice here to be only concerned with constituency on the phonological pole - where do component structures attach to each other? This parameter is necessary to make a distinction between different affixes (i.e., prefixes, suffixes, and ambifixes), including linking morphemes (as they tend to be affix-like). Since our concern here is not with composite structures, I will not elaborate on how many of which component structures are possible in composite structures; I will rather assume the simplest forms of constructions with the minimal number of component structures on a first level of constituency (i.e., lowest level of combination). Hence, for purposes of this article, I assume that n component structures are available in n combinations on n levels of constituency; this is not reflected in Figure 6.

I

- ---- -- -

~nflcctional 1

I amb~liu(L) --------

left

4

/

I- - - - - - - I del-iv;~t~onal1

I --------

I

arnbifiu (1.)

_----_ _clan I msl-pl~e~ni,j --------

I

t1

1

-------clan

-------I

--------

/ 111flecti01ia1 j j- d~ri~at1011al a~iihifir(K) I anhifix ( H i I -------

I 1~1011hcnie --------

I

I --------

right

t2

Figure 6. Relative positions of component structures in composite structures

As discussed before, stems are compulsory component structures in any morphological composite structure and are hence considered the nuclei in

any complex morphological unit. The constituency of all other component structures could therefore be defined in terms of their position relative to stems. In Figure 6 autonomous and dependent stems are thus depicted as central on the diagram and indicated with two bold arrows (Arrow 1 being the main vantage point); two sets of stems are indicated, to illustrate the possibility of compounding. Given the restricted valence of cran morphemes, they could also be considered central constituents in morphological constructions. Cran morphemes connect either to the right of an autonomous stem (such as -ves in huih-ves 'accommodate') or to the left of an autonomous stem (such as stie- in ,,tie-beuel 'stirrup'). In some (debatable) occasions, two cran morphemes can combine to form a word; compare for instance uum- and -hei in uam-hei 'hemorrhoid'. Combrink (1990: 28) also indicates that cran morphemes can partake in affixation constructions by either combining with prefixes (like -rep in onge-rep 'untouched') or with suffixes (like hug- in haglik 'precariou~').~ 1n Figure 6 cran morphemes are therefore indicated as central component structures, in line with stems. Linking morphemes occur between two stems (e.g., yerd-e-kar 'horse ca1-t') with a strong phonological affinity with the left-hand stem (i.e., perdekur is hyphenated as per-de-kur). If we take Combrink's (1990) viewpoint on linking morphemes, we should also recognize that linking morphemes attach to the right of non-variant dependent stems to allow for further derivation (e.g., the -13- in ,fils-qi-eer 'falsify', or -a/- in idiom-uties 'idiomatic'). For purposes of this article, and without going into any detail, I ascribe to the viewpoint that inflectional affixes are peripheral to derivational affixes and that derivational affixes are therefore indicated closer to the nucleus in Figure 6. Suffice it to say that prefixes attach to the left of a stem, suffixes to the right, and anibifixes simultaneously to the left and right of a stem. Since zero morphemes are always affix-like, they are positioned in line with other inflectional affixes. Note again that the representation in Figure 6 only pertains to a first level of constituency. On a next level of constituency it would be possible, for instance, to add autonomous and dependent stems to both sides of the diagram. Phenomena such as the past tense of separable colnplex verbs (e.g., uit-ge-skop ' O L I ~ - P S T -kicked ~ ~ C ~out') , or compounding on already inflected forms (e.g., k~t~jie-ko.5 'cat-1)1~-food, kitten food') could thus be accounted for.

1 16

G'erhur-dB van H~~yssteen

3.5. Summary In this section we have characterized component structures in Afrikaans niorphological constructions, using four valence factors as parameters, viz. schematicity, a~~to~iomyldependence, promiscuity, and constituency. Based on the discussion above, we can summarize our discussion in Table I , where each colnponent structure is described in terms of each of the parameters. Tuhlc I. Summary of characteristics of component structures

-

4utonomous stem Full Reduced Dependent stem Variant Nonvariant Neoa s,. s i.~ a l

I I

Affix Prclix

I 1

Suffix Anibifix I'aran~orphen~e Linking Zero Cran

I I I 1

Sche~nat~city t~~lly speci lied Fully spec~fied Fully bpecificd Partially specified Partially specified Partially specified I'artially I specified Partially schematic Partially scheniatic Partially 1 schematic l'artially schematic Fully schematic Fully I schematic 1:ully I schemat~c Fully schenint~c

I

I

I I

Phnn. Dcpcnclcncc ru11y autonom Fully a~~tononi. bully autonom. Fully dependent Fully dependent Fully depcndcnt Fully dependent Fully dependent Fully dependent t:ully dcpcndcnt Fully dependent Fully dependent I:ully depcndcnt Fully dependent r~111y dependent

I I

I I

1 I I 1

Sem. Dcpcndrncc l ullk a~~tononl. Full} autonom. Full) autotioni. Partially dependent t'arlially dcpcndcnt Partiall\ dcpcndcnt I'artially dependent Fully dependent Fully dependent Fully dependent Fully dependent I'LIIIv dcpcndcnt l:rrllv dcpcndcnt Fully dependent Fully dependent

Promiscuity r~lly promise. Fully promisc. Partially promisc.

Conc;tituency Nucleus Nucleus Nucleus

Choosy

Nucleus

Choosy

Nuclcus

I Choosy

I

Nucleus

Choosy

Nucleus

Choosy

Peripheral

I Choosy

I

(,eft

Choosy

liighi

Choosy

Both sides Nucleus1 Peripheral

I Choosy I Choosy

I I lxft

1

Choosy

LefiIrighU both ~ d e s

Choosy

Nucleus

Note that phonological and semantic dependence have been split in the table; this provides us with a more accurate description, especially of dependent stems. Also note that, as in the discussions, this table does not in

Cotliponent sfructurc.~in morphologicul constructions

1 17

any way aim to represent category membership as absolute; it is a mere visual representation and summary of the discussions above and, as in the discussions, it is based on prototypical instances of each category. Based on Table 1, we can now easily derive definitions for each of the component structures. For example, a reduced autonomous stem can be defined as a fi~llyspecified, fully autononious, and partially promiscuous syrnbolic unit tliat appears in the nucleus of colnplex symbolic units, while a linking morpheme can be defined as a fi~llyschematic, fully dependent, and choosy symbolic unit that attaches to the left of the nucleus in a complex sy~nbolicunit. Other definitions can also be derived in the same way. In the next section I will illustrate how these definitions can be made operational by applying them in tlie context of a project to develop an automatic morphological parser for Afrikaans.

4. Application: Toward a taxonomy of component structures As explained in the introduction of this article, the outco~nesof this research are being applied in a very concrete and practical field, namely the annotation of linguistic data. In a pro-ject to develop an automatic morphological parser for Afrikaans, using data-driven approaches (specifically machine learning), we need to annotate data for training the machine-learning algorithms. It is therefore quintessential (for us) to have clear-cut morphological categories for the accurate and detailed annotation of linguistic data. However, one of the very basic assulnptions in Cognitive Grammar is tliat "much in language is a lnatter of degree" (Langacker 1987: 14). In its very essence, Cognitive Grammar defies the criteria-attribute and dichotomous models of categorization and instead ascribes to models of prototypes, where category Inembers are described on continua in terms of parameters of gradation. The descriptions of component structures above serve as examples. This assumption of Cognitive Gralnlnar is therefore to a large extent at odds with the specific needs of this subpart of the project, where we have to operate with clear-cut, hard-and-fast categories for tlie annotation of data. So, the challenge is to translate the gradations and fuzzy categories of a Cognitive Gratn~nardescription into clear-cut. well-defined categories in order to make these descriptions more accessible andlor usable in an anno-

tation environment. One way of doing this is by means of taxonomies (see Taylor 2002: 128-1 39). Given the descriptions above (and specifically the representation in Table I), we are able to "draw lines tlirough" the identified characteristics in order to "translate" our descriptions in a taxonomy. Without disputing any of the assumptions of Cognitive Grammar, we are nonetheless forced to make hard-and-fast decisions about categories in order to postulate a workable categorization network (i.e., taxonomy). The result of this process is presented in Figure 7. On the basic level of categorization, we distinguish between the four main categories of component structures in terms of schematicity (as labeled in Table I), thus identifying autonomous stems, dependent stems, affixes, and paramorphemes as basic level categories (indicated in heavylined boxes in Figure 7). The alternative would have been to categorize only one category of stems (together with affixes and paramorpliemes) on the basic level (with "simplex unit" as the superordinate) and instead distinguish between autonomous and dependent stems on a subordinate level of categori~ation.However, sucli a classification would have ignored tlie fact that affixes and paramorphemes are also categorized as dependent units; hence, the distinction between autonomous and dependence (see Table 1) is made on a higher, superordinate level. On a more descriptive level (specifically categorized in terms of constituency but also taking tlie other parameters into account), we identify the various subcategories for each basic level category, sucli as linking morpheme, prefix, variant stem, reduced stem, etc. Since these are the important labels for annotation of data (it is more important to know that a constituent is a prefix, than only to know that it is an affix), one could also argue that this is the basic level of categorization; however, for purposes of this project, this '~heoretical"/"psycliogical" distinction is not that significant - it is only important to recognize these categories as finer-grained subcategories of the higher level categories. To indicate the salience of these categories, we also indicate them with a heavier-lined box in Figure 7."'

Component structures in morphological constructions

-

Figure 7. Taxonomy of Afrikaans component structures for morphological analysis

On lower levels of categorization, one can now add various subcategories as needed. For instance, we would like to distinguish between inflectional and derivatiol?al aft'ixee this is indicated in Figure 7 on tlie level of (what we call) structural andlor fi~nctionalspecification (e.g., the suffix -e fi~nctionssolnetiines as an inflectional suffix and sometimes as a nominalizer). For both these categories further levels of specification are indicated, tlie details of which are not important for the purposes of tliis article (see Groenewald and Van Huqssteen 2008). Note tliat we don't take promiscuity (and information related to strata) into consideration in the postulated network - the reason being that we want the eventual classifier to discover tliis information automatically (unsupervised), based on pure statistics. For tlie annotation of tlie data, we take an Item-and-Arrange~iientapproach to analysis (Crystal 1997: 206) since the software that we use for annotation doesn't allow for annotationlindication of morphonological processes. We therefore analyze complex units as "beads on a string": elektr-isiteit 'electricity' is analyzed as elekfr+isi+teit and perdekar 'liorse cart' as perd+e+kar. I-lowever, one runs into trouble in cases where morplionological changes occur in words. Consider, for example, katte 'cats', whose correct analysis is kut-PI (wliere PL, is the suffix -e). Should one now analyze it as kat+te, or rather as kutt+e? (Note tliat we can't just ignore the second t, since all characters in a word need to be annotated.) To solve tliis challenge, we identify on tlie level of structural/fi~nctional specification four "artificial", structural categories for purposes of tliis project. Tlie category "unchanged" is used for autonomous stems where no morplionological processes have an effect, e.g., hond-e 'dog-PI,, dogs' or venster-tjie 'window-I)IM, small window'. Tlie category "allomorph" (tlie term is here understood merely as a label and without any tlieoretical implications) is used to annotate stems with changeslalter~iationson tlie final consonants, e.g., wolw-e 'wolf-PI , wolfs' (stem is worn, or konink-ie 'kingDIM, little king' (stem is koning). In cases wliere an identical consonant is added due to morphonological processes, we use the category "allomorph plus", e.g., kalt-e 'cat-PI., cats' (stem is kul) or hall-etjie 'ball-IIIM,small ball' (stem is hal). The last category, "allomorph minus", is used in cases wliere an identical vowel has been deleted due to morplionological processes, as for example in jar-e 'year-PL, years' (stem is jaur) or strep-ie 'line-DIM,short line' (stem is ~treep). Tlie latter three subcategories actually represent mere extensions from the schema indicated on a higher level of specification and pose in so far no

Component .str.z~rur.e.sin tnorphologicul constructions

12 1

threat for the integrity of tlie theoretical basis of the categorization network. In accordance with conventions in Cognitive Grammar, these extensions of the network are indicated with dashed lines. To illustrate the applicability of this categorization network, consider the following analysis of tlie complex word elektrifik~~~ieIeiu'i~~g.sr~etwerke 'electrification pipe-line networks'. Tuhle 2.

Analysis of elektr~iku.sieleidinggsne~werke 'electrification pipe-line networks'

For annotation purposes. all nodes in the taxonomy are converted to abbreviated tags (e.g., NDS for non-vuriunt depender~tstern or PIS for plural inflec.iionu1 .tufflu), which are assigned autoii~aticallyto component structures by tlie software during manual annotation. Thus far, circa 32,000 words have been manually annotated successfully, based on the taxonomy in Figure 7. In the next stage of the project, this data will be used as training data for a classifier that will automatically analyze (parse) new data.

5. Conclusion

This article aims at solving some practical issues related to the analysis of morphological data for a human language technology pro-ject (i.e., the development of a morphological parser for Afrikaans). These issues are approached theoretically from a Cognitive Grammar perspective in order to characterize component structures in Afrikaans lnorphological constructions.

122 Gcrhurd B vun Huy.s.steer~ Various component structures are described in terms of four basic valence factors, viz. schematicity, autonorny/dependence, prorniscuity. and constituency. Based on these parameters, a theoretically unified, unambiguous categorization network (specitjcally a taxonomy) for component structures in Afrikaans morphology is postulated. It is indicated how this taxononly is extended for specific purposes of the project and how it is implemented to lnanually annotate data. Future work will include research along the same lines, but now focusing on a categorization netmork of composite morphological structures in Afrikaans. This will, of course, be a vast task but nonetheless important for a better understanding of morphological constructions in general and specifically for Afrikaans. Based on this mork, one could also adapt the current categorization network for other languages - especially languages with a niore cornplex morphology than Afrikaans. This could not only prove valuable for a better understanding of the specific languages, but also to provide new and deeper insights in Cognitive Grammar as a theoretical descriptive framework.

Notes 1.

7.

3.

Combrink's 1990 book, Afiikoansr Morfologie: Capita Exeinplaria, is the most comprehensive and also the most recent stand-alone publication on Afrikaans morphology. This work of Combrink is purely descriptive and aims to steer clear of any affiliation with any theoretical framework. As Combrink (1990: 9) states in the preface: "One way of avoiding theoretical linguistic "nearsightedness" is to analyze data by using mature theoretical linguistic notions and to adapt these notions as the data prescribes" [My translation GBVI-I]. If any theoretical presuppositions slipped into his work, it is either idiosyncratic, or Structuralist in nature. By saying this, I'm not negating the fact that various other Cognitive Granimar scholars, such as Eugene Casad, Kenneth W. Cook, Hans-Olav Enger, Laura Janda, Suzanne Kemmer, Tore Nesset, Johanna Rubba, Brygida Rudzka-Ostyn, and Arie Verhagen, have done valuable work in describing various ~norphologicalphenomena in various languages. Of course, the work by Ron Langacker, John Taylor and David Tuggy is fundamental to an understanding of morphology in Cognitive Grammar, while Vyv Evans and Melanie Green offer a valuable summary. See Tuggy (2005: 259--260) thr a discussion on the morphology-syntax boundary.

Component srructzlres in n~orphologiculconstructions

4.

5.

6.

7.

8.

9.

123

These component structures (such as the -e- in perd-e-kar 'horse cart') are also referred to as link phonemes, linking elements, interfixes, phonomorphemes, connecting morphemes, linkers, stem extenders, valence morphemes, etc. Various other valence factors, such as correspondence, profile determinacy, analyzability (Langacker, 1987), coerciveness and bondedness (laylor 2002: 330), internal and external complexity (Taylor 2002; Tuggy 1992), and entrenchment are often discussed in Cognitive Grammar literature. However, for purposes of a characterization of component structures (and not of composite structures or constructions), the valence factors mentioned in this article will suffice. Following the arguments of Tuggy (1992: 287), 1 also prefer to use the term "stem" instead of "root", and I, thus, also refrain from making a distinction between these two terms. Furthermore, this distinction is obsolete with regard to Afrikaans morphology; see also Combrink (1990). Note, however, that what I will call "non-variant dependent stems" (e.g., elektr- in electricity 'elektrisiteit') are also often referred to as roots. Under the heading "bound non-variant stems" Combrink (1990: 26) lists examples like send- in senhling 'missionary' and ren- in renperd 'race horse'. Since both send 'send' and ren 'run' are autonomous words in Afrikaans (both these words appear as lemmas in standard Afrikaans dictionaries), this analysis of his is clearly incorrect. He categorizes stems like elektr- and sekret- as 'truncated free stems', which also seems to be a misnomer for these kinds of sterns; for one, a stem like sekrelur- combines with suffixes like -is or -es.~e(both recognized by Combrink (1990) as suffixes) to form sekreturis 'secretaty' and sekre/aresse 'female secretary', thereby voiding the interpretation of these stems being truncated. To my knowledge, "paramorpheme" is not a widely used term; in fact, I found only two references to this notion. Trager (1953: 327) uses the term to name morphemes "that constitute a set of inflectional suffixes in a paradigm", while De Groot (1964: 127) uses it to denotes what I call here a linking morpheme (i.e., a morpheme that heightens the valence of two components to combine). In my terminology, I discard the notion as applied by Trager (1953), while extending the notion's meaning as used by De Groot (1964) to include other tnorphematic forms that extend far away from the prototype. Diachronically speaking, none of the examples listed could be considered cran morphemes; synchronically speaking, they are. This raises the question of the difference between morphology and etymology. For purposes of this article, I support Plag (2003) when he states that "we have to be careful not to confuse morphology with etymology. Even though a morpheme may have had a certain meaning in the past, this does not entail that it still has this (or any other) meaning and can thus be considered a morpheme in today's language" (Plag 2003: 25).

124

Gerhrrrd B vtrri tluyssteen

10. Since zero morphemes have no realization in the orthography, it is impossible to annotate zero morphemes. For the sake of con~pleteness,I include this node in grey in Figure 7 to indicate that it is not part of the practical implementation in this project. Likewise, I include "complex unit" for the sake of completeness although this is not a label used in the data annotation project.

Acknowledgments First and foremost I would like t o acknowledge and thank Sulene Pilon for many hours o f critical (and fitti) conversations, sharing o f ideas, and analysis o f data - without her, this article wouldn't have been realized. Various other colleagues and students also helped in formulating ideas presented in this article; I would especially like t o thank my colleagues Hans du Plessis and Bertus van Rooy for being soundboards. All fallacies remain miue.

References Aronoff, Mark, and Kirsten Fudeman What is Mor,phology? MaldenlOxfordiVictoria: Blackwell. 2005 Bauer, Laurie 2003 Introducing Linguistic Mol"ph01ogy. Edinburgh: Edinburgh University Press. A Glossary c?f'h.lorphology.Edinburgh: Edinburgh University Press. 2004 Booij, Geert, Christian Lchmann, and Joachim Mugdan Prcf'uce. In Morphology: .4n Interr7a/ional Hatidbook on It7flection 2000 and Word-Fornzcction C'ol. I, Geert Booij, Christian L,ehinann, and Joachim Mugdan (eds.), xxi--xxviii. BerliniNew York: De Gruyter. The Morphology of'Dutch. Oxford: Oxford University Press. 2002 T/7e Grarnrl~ar qf' Word~s: An Introu'llc~ion lo Morpholom. Ox2007 ford1New York: Oxford 'Ilniversity Press. Combrink, Johan G.H. 1990 Afrikuunse Mo~fologie:Cupita E.xenlplariu [Afrikaans Morphology: Capita Exemplaria]. Pretoria: Academica. Crystal, David 11 Dictionary of' Lingui.rtic.v and Phonetics. 4th Edition. Oxford: 1997 Blackwell. De Groot, A.W. 1964 Inleiding tot de Algenlene Tucrhvetenschap [Introduction lo General Linguis/ic.v]. 2nd Revised Print. Groningen: J.B. Wolters.

('omporient ~ t r u ~ t u r in e s morphological con.s/ructions

125

De Klerk, Ge1-t J. Die Morfologie van Afrikaans [The Morphology of Afrikaans]. In 1968 ,4fiikuun.s - Sy Aurd en Onfivikkeling [Afrikaans - Its Nature and Development], Van der Merwe, H.J.J.M. (ed.), 169-208. Pretoria: J.L. van Schaik. De Villiers, MeyetAfrikuanse Grrxmnzutiku vir Volwussenes. [Afrikaans Grammar for 1983 Adults]. Goodwood: Nasou. Evans, Vyvyan 2007 A Glossary of' C,'ognitive Linguisfic.~.Edinburgh: Edinburgh University Press. Evans, Vyvyan, and Melanie Green Cognitive Linguistics: An Introduction. Edinburgh: Edinburgh Uni2006 versity Press. Gleason, H.A. 1955 .4n In/rohction l o Descriptive Ling~istics.Revised Edition. New York: Holt, Rinehart and Winston. Groenewald, I-landrC J., and Gerhard €3 Van Huyssteen 2008 Outomatiese Lemma-identitisering vir Afrikaans [Automatic Lemmatization for Afrikaans]. Literulor. 29(1): 65-91. Haspelmath, Martin 2002 Under.sttmu'ing Morpholo~j.London: Arnold. Hudson, Richard 1984 Word G'ratnnlur. OxfordNew York: Basil Blackwell. Katamba, Francis A,Iorphology. London: Macmi llan. 1993 Langacker, Ronald W. 1987 Foundutions of' Cognitive Grummrrr Volume I: Thcoreticul Prcrequisites. Stanford: Stanford University Press. 1991 Concept, Imuge, und Symbol: The Cognitive Basis qf Grammar. BerlinINew York: De Gruyter. Matthews, P.H. 1991 hforphology. 2d ed. Cambridge: Cambridge University Press. Plag, lngo 2003 Word-firnlation in English. Cambridge: Cambridge University Press. Senekal. H.E.J., Fritz A. Ponelis, and Willem J. de Klerk 1972 Die Patroon vuri Afrikuuns [The Pattern of Afrikaans]. Johannesburg: Afrikaanse Pers-Boekhandel. Taalkotnmissie van die Suid-Afrikaanse Akademie vir Wetenskap en Kuns 2002 Afrikuunse Woordelys en Spelre~1.s[Afrikuuns Word-list and Spelling rule.^]. 9th Edition. Cape Town: Pharos.

Taylor, John R. ('ognitive Grammar. Oxford: Oxford University Press. 2002 Trager, George 1953 Russion Declensional Morphemes. Language 29, no. 3 (July-Septernber 1953): 326-338. I'uggy, David 1985 The InflectionalIDerivational Distinction. Work Papers qf the Summer lnstitzrte of Litiguistics, IJniversity qf North Dakota Session XX1,Y. (Grand Forks, N.Dak.: Summer Institute of Linguistics, University of North Dakota. 1992 The Affix-Stem Distinction: A Cognitive Grammar Analysis of Data from Orizaba Nahuatl. Cognitive Lingui~tic.~ 3 (3): 237---300. Cognitive Approaches to Word-Formation. In Handbook of Word2005 Formation, Pavol ~ t e k a u e rand Rochelle Lieber, (eds.), 233-265. Dordre~ht:Springer. Van Schoor, J.L. 1983 Die Grc~mmatikavan Standuurd-Afkikaarzs [The Grammar of Standard Afr-ikaans]. Kaapstad: L,ex Patria.

Between phonology and morphology

Hilke Elsen

1.

lntrocluction

Ilsually, gratnmatical lnodels work wit11 good, unambiguous examples or eve11 witli idealized data. l'hese data generally fit in nicely with the theory but what about doubtful exarnples and exceptions'? Should they be ignored or should the theory be refined? This paper deals witli transitional phenomena and the relevance of holistic processing in lieu of and in addition to regular morphology and presents three sets of data to support this view: (a) data from language acquisition, (b) ter~ninologyof the special langilage of chemistry, and (c) names of fantasy and science fiction novels. These case studies first of all s11ow that in addition to the analytic processing of individual sounds and morpl~elnesthe overall sound shape or Ge.~/altseems to play a role. Secondly, the borderline between phonology and morphology is not clear-cut.

2. Language acquisition The lirst research project was concerned with first language acquisition. 'The lindings reported in this paper are based on the diary data of a Germanspeaking girl, collected continuously up to the age of 2 years and 5 months (Eluen 1991). All new words, word forms, and novel pronunciations of establislied items were documented in IPA phonetic transcription. Striking facts about situation and referents, comments on frequency and mortality of individual lexical items, and notes on morphology and syntax were recorded. Imitations were distinguished froin deferred imitations and spontaneous productions. Notes were transferred to filing cards twice a day. Additional comments were added as necessary on situation, frequency of use, and changes in articulation. The entire corpils was cross-checked three times a month. Afterwards, notes were taken, at first daily, then in longer intervals.

1 28

Hilke Elsen

2.1. Phonology

In tlie following, I will present several unusual examples from children like juice (Menn 1978), [gipatul] dijjicull (Smith 1973) or [ I x d ~ d ~upple ] [wasmatu" Tustutur 'key board' (Elsen 1991). They demonstrate that segmental and prosodic features are tackled with varying degrees of precision. On the one hand, prosodic information is correctly processed, partly at the expense of segmental information. Partly, correct sounds are only found in prominent syllables. The data show varying degrees of proximity to tlie targets. Of course, there are many examples of perfect or near-perfect production: ( 1)

Munlu ['mama], 0;8,25, -mornmy' die Ruupe Nimmer,rull [di Kaupa 'n~manaz],2;3,0, book character Schlun~pf[Jlrrmpfl, 2;5,3, book and comic strip character

However, deviations are found which can be graded according to their similarity to adult words. (2)

Henry ['Iisnd~I], 2: 1, l name Kohlrabi [kol'ltabiql, 2;0,28, 'kohlrabi'

The words in (2) simply contain an additional sound (epenthesis). This can be found in language change, in synchronic variation, as well as in child language. The same holds true for the examples in (3). Here sounds change their position within the word (metathesis). In (4) the child uses reduplicated syllables to produce the words.

(3)

dreckig [ ' d ~ k u ~ g1;9,2 ] , 1, 'dirty', coll. Hilke ['ikl], 1 ;3,1; ['lika], 1 ;5,30 name Gliihbirne ['gtlbr3na], 1 ;I I 'lightbulb'

(4)

Budewur~ne['man1man~],1 ;6,3, 'bath-tub' Brombeerr~~orneode ['mamamamalada], 2;0,24, 'blackberry jam'

Reduplication is not found too often in language change, but it seems to be used systematically in some creoles (Miililhausler 1986: 123, McMahon 1994, chapt. 10.3). Related to this are consonant- and vowel harmonies, cf. (5).

Between phonology and morphology

(5)

129

Helikopter [hninknkta], 2;2,28, 'lielicopter' Lasagne [san'sanja], 2;4,2, Italian dish

Assimilation is likeuise found in language change and synchronic variation. More interesting, however, are the followi~lgexamples. The child replaced itnstressed syllables with [a], cf. (6). She also added [aja], [ala], [ala~]etc. to the first syllable of a three sjllable item with initial stress (7). (6)

Zitrone ['a'zona], 1 ;6,17, 'lemon' A4elone ['a'lona], 1 ;5 - 1 ;7, 'melon' Luterne ['ald%na], 1 ;5,6, 'lantern'

(7)

Schmetferling ['in~taja],often 1;5, 'butterfly' Hu,cten.saj/ ['hmeajai], 1 ;4, 'cough-syrup' uufiuun~en['aurfa.ja~],orten I ;4, 1 ;5. 'to clear up' Michac.1 ['mi$aja~].1;4 - 1;9, name

These fillers or master syllables were generally used as closing syllables in a word following the stressed syllable. In (8), the child reproduced the correct number of syllables and tlie position of word stress. Both child and adult words share several sounds. But this is not the case in (9). (8)

Oregano [ako'kano] 2;5,18, Italian herb Schlufunzug ['vaiiaguk], 1 ;8,16, 'pyjamas' ('on~puter[b$jluta], 1 ;10,14, 'computer' Porternom~uie['m3'tane], 1;8,27, 'purse' Tonlute [ma'lata], 1 ;3,12, 'tomato'

(9)

Scl~l.ippe['in~ta],1,4,16, 'sl~ovel' Kufer ['wiqa], 1 ;4,20, 'beetle' Suse ['tlufa], 1 ;6, name

In (9) the number of syllables and the position of main stress are correct, but the child and adult words have hardly any segments in common. To show that tlie child whose data are investigated here is not an exotic or extraordinary speaker, some forms from other children are presented (for the master syllables cf. 10, for the group in 8 cf. 1 I).

(1 0) cor?fi/t~re a:titii:t, 'jam' c o.tlun!e a:tiim, 'costume' rloisette asst, 'hazel nut' fermC ame:, 'closed, shut' (Gregoire 1937)

~rtluck[ri:tzk] disturb [rista:~] enjoy [ridmi] irlficlion [rif~ksan] conducror [ri:d~kta](Smith 1973) ( 1 1 ) C'harlenzugrze ananaii. name,

corfiture k3tiitii. 'jam' (Gregoire 1937) alj~liahethook [ z p ~ b ~ b ~ ] upple suuce [ X P A X (Menn ~] 1978) oeropltme [E:ba'ei n] /elephone [ d ~ w i:bit: n] C'opydex [d3pi:gsk] (Smith 1973) The examples in (12) show words that are only minimally similar to their target forms. ( 1 2 ) ~ipple juice [ I ~ d ~ d k ]

Lise [niUixi\](Menn 1978) C'liri,str~~a.s [gi pt it] rhir~ocerosflia~ nxatat] dtfficult [gipatul] (Smith 1973) When we look at these data, we see a very strong tendency to produce primarily tlie nulnber of target syllables and the position of target stress. The reproduction of sonie segmental features in the stressed syllables is of secondary importance. To preserve syllable number and position of stress, the children use various strategies. One is simply reduplication or using some sounds of stressed syllables for neiglibouring ones. This can be found for all children. Secondly, tlie use of inaster syllables appears. Note that the substit~ttedsyllables in all three languages are not stressed and precede or

Bel*ireen phonology and morphology

13 1

follow a stressed syllable. Obviously, children at an early age cannot cope witli long forms and instead master less salient parts of words with easily manageable templates or master syllables to maintain syllable number and position of stress (Elsen 1996). However, sometimes children's words are even less accurate. Exa~iipleslike those in (8) show that the number of syllables, the position of word stress, and nearly all stressed vowels are preserved (disregarding lip-rounding). In the case of penultimate stress both final vowels are target-like while for prepenultimate stress all three final vowels tend to be correct or tlie middle one is weakened to schwa. This is a regular process in colloqi~ialGerman and English when adults speak fast. Usually, pretonic vowels are retained. We see that tlie number of syllables, the position of word stress, and the vowel skeleton show a high degree of similarity with the target expressions. As far as consonants are concerned, tlie final or the two final consonants of a word tend to be preserved. Words with penultimate stress show similarities witli the target consonants in the final two syllables. In three syllable words with prepenultimate stress, tlie onset of the first syllable is very often retained to a high degree. In those with final stress, the last syllable is retained. Word final consonants and those of stressed syllables are relatively close to their targets. Looking at examples such as those given in (9) and (I I), it is again obvious that syllable number and word stress are target like. Vowels show nearly always target length. Consonants of child and adult stressed syllables only have some features in common, generally nasality. plosiveness, labiality, or alveolarity. There is a tendency to replace difficult sounds like Ihl, velars, and fricatives by easier ones like front plosives and nasals. This can be found for all children in all languages (Locke 1983). Thus, the similarity between children's i~tterancesand target words varies gradually from precise production to hardly any similarity. Other studies, for example comparisons between French and German clippings and acronyms (Ronneberger-Sibold 1996) or contaminations (RonnebergerSibold 2005), also showed that adult language users rely primarily on phonotactic and prosodic features. The highest amount of segtne~ital information can be found in stressed syllables whether regarding the acquisition data (Echols 1988) or the shortenings. These are obviously neither idiolectal nor language-specific peculiarities. The findings need to be treated within a framework that is able to deal with variation, transitions between correct and incorrect items, so-called exceptions, and the possibility of separate segmental and prosodic representations.

132

Hilke Elsen

Tlie data pose problelns because, first of all, tlie role of prosodic features is clearly predominant. They are very often realized at the expense of individual sounds. Secondly, there is a smooth transition from correct productions to deviant forms of the target item. These observations do not fit into general models working with abstract sy~nbolsand rules tliat neglect so-called exceptions in language production and use. Instead, an approach that considers the processing of language information in a network-like system seems preferable. Such a system imitates the neurocognitive facts - the architecture of nodes and connections and tlie mode of operation of the brain (e.g., Stemberger 1992, Elman et al. 1996, Lamb 1999). Information is processed in nodes and connections. On tlie one hand, each item such as a feature, a sound, and a word exhibits an individual pattern of activated connections. These items do not exist as entities or objects but must be understood to be a cliaracteristic pattern at a cliaracteristic position in the system. We can use names like /p/ or u'og or riozrn to refer to items or categories. But that only facilitates communication and reasoning. It does not mean that they are sounds or categories per se. They are only generalizations. A member of such a group can show a more or less prototypical structure, i.e., more or less similarity to the activation pattern of the prototype. An early and repeatedly activated area or sound is stronger than a later one. T ~ L I frequent S, sounds from the babbling stage like In/, /m/, Id/, /b/ often replace velars or fricatives in early words. Frequent word patterns of the target language are learned earlier. Infrequent patterns are more susceptible to change in younger cliildren. When too much information is processed, the most prominent part will survive, usually syllable number and stress, followed (then) by the main syllable. On the other hand, all areas in the brain, in tlie computer network, or in the hypothetical model are interconnected. Accordingly, the part that deals with prosody can process independently or not and gradually integrate information from other doniains. So, linguistic aspects grow together and complex structures emerge. The child data as well as the results on word creation indicate that prosodic information dominates over the segmental and tliat segmental information in stressed syllables dominates over information in neighbouring syllables. We might assume that the activation of connections that represent prosodic features are the strongest, followed by those whicli connect segmental information and main stress. Several examples in this and other corpora show that segmental and prosodic features are articulated with varying degrees of precision. On the one hand, prosodic information is correctly processed, partly at tlie expense of

Betweer7 phonology and nzorpllolocgy

133

segmental information. Partly, correct sounds are only found in prominent syllables. On the other hand, the similarity between children's utterances and target words varies gradually from precise production to no si~nilarity at all. We can find words exactly pronounced as the model, especially those with the prototypical German sound shape of two syllables with initial stress and a schwa in the last syllable. Especially in phonologically cornplex words, the child tackles the whole sound shape, the number of syllables, and the position of stress, thereby lositig sight of individual sounds. Network-like processing assumes transitional areas between categories, learning by pattern association, and takes into account frequency factors (Elsen 1999, 2000). The same is true for Cognitive Grammar approaches. Here, the emergence of structure from "pre-packaged" fortnations is one fundamental aspect of cognition. Analyzability is a matter of degree (Langacker 2000: 133).

2.2. Morphology To investigate the acquisition of past tense forms, computer si~nulations work with a single associative learning mechanism to answer the question whether a morphological rule is really needed to account for the formation of past tense (e.g., Rumelhart and McClelland 1986, Plunkett and Marchman 1991, 1993). In these studies models have been developed capable of lne~norizingpatterns, of generalizing regularities, and of predicting overregularizations such as "singed triggered by a sufficient number of regular verbs and irregularized verbs such as ro neul *sea11 *senlt, later in development. Astonishingly, the diary data on past participles show exactly these phenomena (for more, cf. Elsen 1997, 1998). The develop~nentfrom collecting pairs like kleckern / ge-klecker-t 'to slobber', gucken / ge-guck-1 'to look'. wickeln /ge-wickcl-t 'to swaddle' to actively using ge-t to mark the past can be understood as accumulating and analysing patterns and generalizing them to produce new verb forms. Similarly, the data on the acquisition of plurals (cf. Elsen 2002) point to the relevance of frequency, critical mass factors, and schemata. Both verb and noun inflection shows that holistic patterns or Gestalts develop into co~nplexword forms during first language acquisition. In the acquisition process, sound shapes and sounds have to be brought together and regular sound shapes develop into morphologically structured items. The diary data are not only compatible with colnpilter si~nulations

1 34

Hilke El.sen

but witli ideas about the developmental relationship between liolistic patterns and tlie origin of phonology. Viliman, for example, assumes that "plionological structure is the emergent product of tlie learning of (own and others') specific word exemplars over the course of repeated experiences witli a common pattern" (Viliman 2001 : 0). "Segments and prosody are induced from templates in the course of and as a consequence of use" (Viliman 2001 : 1 ). Frequent and simple templates show more phonetic precision. Longer. infrequent ones are targeted with less success and witli the lielp of Ersatz-strategies. Schemata are often a first step on the way to complex regularity (Elsen 2000, 2007b). Data on phonology, noun and verb niorphology, and the lexicon are consistent with predictions of network sin~ulations.The computer system and very probably a child as well are capable of abstracting and producing hierarchical and 'rule'-dependent information from surface structures witliout tlie lielp of explicit, symbolic concepts and rules. This can explain the initially separate processing of prosodic and segmental information and the trial-and-error-like variation between tokens. Atypical examples and "exceptions" find a place in this psychologically oriented model, and it may well be that "tlie basic unit is tlie whole word pattern" (Viliman 200 1 : 10). The overall sound shape of a term also appears to play a role in adult language processing.

3. Neologisms The second research project was about Gernian neologisms (for more, cf. Elsen 2004). Eight sub-corpora were compiled on the basis of, e.g., cliildren's books, tlie special lang~~ages of technology and chemistry, and advertising. Each text type was represented by approximately 500 lexemes. Striking word formation patterns were found, like various kinds of blendings and word creations which defy standard morphology. Tliese striking patterns are usually designed to trigger certain reactions from the recipients. such as attention, pleasure, or even change of opinion, as in tlie case of newspaper texts. Gestalts were found in several sub-corpora; the following section focuses on one of them. In the special language of chemistry (cf. Elsen 2004), new chemical compounds are named according to international guidelines. The results are unpronounceable systematic names. Consequently, such substances receive additional trivial names. Tliese should be easy to translate and involve in-

Berween phonology and morphology

135

formation on contents or effects. Another group of neologisms are trade natiles, which are formed for commercial purposes. They should describe tlie product, be easily pronounceablc, and arouse positive associations. Both groups should attract the interest of consumers. That is, they should sound nowGerman. Both types should also sound scientific and reliable. This aspect is satisfied with terms that resemble words of LatinIGreek origin. Tlie typical and most frequent German sound shape consists of two syllables with initial word stress and sch\\a in the second syllable. Thus, a non-German, scientific, word shape needs more than two syllables, full vowels only, many open syllables, a closed final syllable, and word final stress. Furthermore. typical German consonant clusters like IJtr-I, Ipfl-I, I-mpfl sliould be avoided. Tlie avoidance of these typical sound structures are indeed characteristic of the selection of pharmaceutical names and designations in chemistry. The alien names are created in German with the help of different word formation processes such as mal~ufacturedwords ( 1 3), blends (l4), and shortenings ( 1 5), \vI~iclishow a phonotactically coniparable pattern. (13) Acerbot?, medicine with Antihypertonicurn Lisinopril-Dihydrat Agopton, medicine with Lansoprazol against ulcers Bcryhorun,for cleaning, by Bayer Agnucuston, medicine with vitex. ( 1 4) Aflutrem, tremorgene toxin with Aspergillus flavus Bustadin, with lanthella basta and Ryanob-dependent calcium C'ulponin, protein, binding mmodulin and functioning like Troponin. ( 1 5) Aclonifin, common name for 2 - ~ r - 6 - ~ t r o - 3 - ~ o x ying i l

Bunihuterol, common name for (+-)-I -[3,5Bis(dimetl1ylcar~oyloxy)phenyl]-2-~t-~Ia1nino)ethanol. First. names like that illustrate different degrees of motivation. Secondly, many of these words show endings which sound like morphemes, but which in fact are combinations of sounds \vithout stable content, cf. -on, -01, -in (e.g., Acerbon, Bustudi'in, Bindol, Agopton; so-called pseudomorplis, cf. Elsen 2006). Word formation techniques are word manufacturing, blending, and shot-tening. The sound shape of tlie resulting words is phonotactically comparable with derivations like Adhiis-in (protein),

C'ulcit~eur-iui(protein), Alendron-at (salt), Acan~pro~s-at (salt), and Afivill-it (mineral). This stresses the third observation that various types of word formation lead to a Gestult. Tlie forms sound like Latin or Greek words and thus assume a scientific, efficient connotation. The phonological shape "steers" the line of association: substances with sucli names work properly and reliably. Taken together, we find phonotactic systematicity of sound shape and often hardly any morphological information. The goal is to coin a 1,atinlGreek sounding word and there are various ways to achieve this. Tlie transition from word creation to word formation is a gradual one. Manufactured words complement regular morphology and operate as Gestalts to transport vague stylistic and expressive meanings. Interestingly, we find such Gestalts in yet another, completely different area of the lexicon.

4. Fantasy names Tlie following data are from the research pro-ject Phuntu,slitchc Numen, Ludwigs-Maximilians-Universitat Miinchen, Germany, funded by Bau + Plan GmbH, Munich, and by the Deutsche Forschungsgenieinscliaft (EL 20112- 1, -2). Names from science fiction and fantasy stories were collected and examined according to morphological structure arid in order to investtigate whether the relationship between a name and its fictitious referent is arbitrary. All 52 books were originally written in German. Tlie texts range from classics sucli as LaBwitz (1897) to recent publications (cf. the appendix for a selection of titles). In the following, names for individuals, peoples, and species will be examined and discussed in contrast to names for substances and buildings (Roter Pultr.\l, Schwurzes Porlul, Schwurzer Turm Tempel des Toten Gottc.5, Felsenhurg, Erneritenturm) as well as bodies of water (Meer der Truuer, Sch~rrluchrotesMeer, K m a l der Ausges/c$encn, A,cphcrlt.see, Perlenn~eer, Nevernever- Wuter-FluJ, TuusendBogen-FluJ, Jamuchiniflup). Both names for bodies of water and buildings show ~norphologicallytransparent examples. Roter Pulu.~tand Schwarze.~ Portol, e.g., are Wortgruppen1exeiilc.e (cf. Elsen 2007a). 7liey equal Red Pulace and Block Portal in English. These groups of referents are mainly based on morpliologically regular for~iiswhile names for substances usually appear as in (16).

Be/iveen phonology und rnorpholog>f 13 7

(1 6) Terkonit, Ehulit, Molvedin (metals), .4rnne,sin, Ansintan, Anthygrin. C'orphorin, Energon, Er~o.\uii,Euphorit, Gl~~.~cmtiri, Somnalin, Valoro~i (drugs), S'kunit (for isolation of cables), Howalgoniurn (chemical element), Oral-Dcocul (for deodorizing the mouth). These words match with trade names from chemistry in using pseudo~iiorplislike -in, -an, -on, -it, which sound scientific but do not carry a consistent meaning. 'l'hey function as stylistic markers and may develop into morphemes. Likewise, we can find traces of sound symbolism, for example, the sound /i/ in names for small, good beings, see (1 7) ( ! 7 ) Brin [HoHe], young, good-natured prince, Elin1 [AIFla], childlike prince, Gwrgi [Pesch], small, good-natured "Sumpfling" (living in swamps). Krila [WHAGar], small, good-natured gnome, Kclwitt [EscliKel], Jombuuraner / Jonibuuran (intelligent, innocent, dolpliinlike creature), Schti [LaPla], very small winged horse.

Names for positive, magnificent, potent, and friendly characters preferably consist of more than two syllables, most of them open and many with the vowel /a/. -I lie last syllable is predominantly closed (cf. 18). (18) Sirlun!ir. [WHGar], friendly magician, kirlakcnl~nn[HoHe], friendly magician, Mandavur [HoHe], friendly elf knight, Katana Nipus [[KneiAt], friendly patriarch, Springer (species), Athanusios [Pesch], friendly scholar, Gcxldalyn [FelMa], highest druid, friendly, Racalla [HoFle] female magician. These superior, important cliaracters carry names with a lion-German but LatinIGreek-like sound shape with simple, often open syllables and fill1 vowels. Names for foreign or non-humanoid entities which are not marked as definitely vicious are striking as well. They also deviate from the typical German word pattern as they contain only full vowels. Some combinations on the phonological/graphe~niclevel are higlily ilni~sual: ( 1 9) Gorx [PR2], Gruan [WHGar], 7'e 'el [PR2], Eni 'netu [EschKel],

M o o f i ~[KneiAt].

1 38

Hilke Elsen

In contrast, names of evil alien characters show a liigh number of back phonemes, especially velar and uvular fricatives and vowels like 111,o, a/ as opposed to the "nice and small-sounding" /i/. Furthennore, many syllables contain complex onsets and codas or non-German consonant clusters, as opposed to the magnificent sounding open syllables with /a/ (cf. 20). (20) Ch 'tuon [WHGar], demon, Tcriroch [HoHe], demon, Ghuzdan [EHScha], orc, Gnooruf [FelMa], bad, animal-I i ke creature, Azrarhoth [Pescli], dark elf, Arjtrnoor [HoHe], demon, C'hrekt-Orr7 [PR2], vicious, reptilian-like creature, a Topsider from the planet Topsid, Rrul'ghargop [HoHe], orc, An-Rukhbur [FelElfl, demon. Hence, non-average characters get non-average names; that is, names which deviate from the prototypical Gennan word pattern. This contains consonant clusters SLICII a5 /jtr/. Ipfl, /mpf/ and two syllables, the first of whicli is stressed while the second contains a schwa. The names of good, important, potent people sound like Latin words and are associated with education and science. In opposition to /i/ and /el which are associated with smallness, many a-sounds appear in these names thus emphasizing tlie magnificence of the referents. Foreigners receive names with striking, thus foreign sound or grapheme combinations and contain only full vowels. Additionally, names for bad characters show preferably dark vowels and back fricatives. Jakobson explains the "ready associability of [i] with small things [...I by tlie high pitch of the vowel" (Jakobson and Waugh 1987: 187). Masuda (2007) suggests a physical basis and Ohala (1994) in his "frequency code" proposes biological grounds for some sound symbolic phenomena. One reason for the tendency to associate /i/ and /el with SMALL and thus IIARMLt>SS and GOOD as opposed to dark vowels might be found in sexual and age related ditnorphisrn of the vocal anatomy (cf. Ohala 1994). "Sounds made by a confident aggressor [...I are typically rough and have a low F< (Oliala 1994: 329). These sounds are connected with larger apparent size to give a dangerous impression. Back vowels and fricatives resemble threatening growls, roars, and snarls and will be more suitable to name dangerous and evil creatures. However, not only single sounds but also particular groups and sound patterns recur repeatedly to mark the identity of certain types of protagonists. Interestingly, for the names denoting bad, foreign characters less frequent sounds and syllable shapes are used. The most frequent sounds in the

Between phonology and movpholog~

I39

languages of tlie world are front nasals and stops and approxiniants like /I, j, W/ (cf. Maddieson 1984). The most frequent syllable structure is a consonant-vowel-combination. Coniplex clusters and velar and uvular fricatives are rare in the languages of tlie world. Most likely "rare" or "not known" is "threatening" - rare sounds, clusters, and syllable shapes niigl~t trigger preferably negative associations in addition to the fact that dark vowels and back fricatives resemble threatening growls, roars, and snarls. In combination with a non-German sound shape, they will indicate evil, foreign species Inore appropriately than a prototypical pattern such as Lene and Su.re. By contrast, to denote magnificent, good characters a Latin-like sound shape with many u-sounds seems preferable because it creates associations to "reliability. efficiency, authority, importance", arid to further positive traits. Research on sound sy~nbolislnis not very extensive. For tlie time being it will suffice to say that in naming fictitious characters, certain sound patterns are repeatedly associated with certain information patterns and that this kind of systelnaticity is neither arbitrary nor morphologically regular.

5. How to model the bridge between phonology and morphology? In sum, there is a continuum from sound shape to recurring groups of sounds to morpliemes. Not only individual sounds, but also Gestalts seem to constitute a level in language processing. The traditionally grammatical view of morphology is obviously inadequate to capture these phenomena. Instead, we need an approach which takes into account frequency factors, holistic patterns, and the fact that there is no clear-cut borderline between categories and between linguistic levels (Elsen and Michel 2007). As far as processing is concerned, network approaches provide explanations for transitions and exceptions. Frequent pattenis are processed tirst. They are more stable and easy candidates for representatives. The ability of categorization results from physiological perception, from neural feature detectors, and from filtering mechanistiis of tlie sense-organs. Neurobiological research reveals that organisms detect and react only to certain stimuli or sequences of stimuli to economize energy. In other words, certain features are selected, offering an explanatio~~ for categorization and prototypicality in cognition. Thus, from a biological standpoint, categories with representative members seem highly plausible (Muller and Weiss 2000: 59-60).

140

Hilke Elsen

From a philosophical, linguisti~,and psychological point of view Wittgenstein, Labov, Roscli, Langackcr, and others worked on similar problems, which led to (and reinforced) prototype theory. This concept was originally developed for word meanings but might be extended to linguistic objects (cf., e.g., Taylor 1995, Brdar-Szabo and Brdar 2000). Recently, Cognitive Grammar offers an integration of prototypes, network ideas, and grammatical phenomena (cf. Dirven and Versporr 1998, Taylor 2002, Ungerer and Schmid 2006, Croft and Cruse 2007), even considering aspects of iconicity and tlie problem of holistic patterns or schemas (e.g., 'Taylor 2002: 45, Ungerer and Schtnid 2006: 300). Cognitike Grammar approaches take into account frequency factors; they assume transitional areas between categories, learning by pattern association, and the grouping of entities on the basis of similarity (for a short overview, cf. Taylor 2002: 27). When contemplating on various types of games, Wittgenstein concluded: "Wir sehen ein kompliziertes Netz von fhnlichkeiten, die einander ubergreifen und kreuzen. fhnlichkeiten im GroBen und Kleinen" (Wittgenstein 1984: 276) [We see a complicated network of similarities overlapping and criss-crossing, sometimes overall similarities, sometimes similarities of detail; translation by A.E.M. Ansconibe]. Overall, similarity between words is nothing but the Gestalt, the impression of a shape, a complete whole, based on sounds and their combinations. Wittgenstein further asks: "Wie ist denn der Begriff des Spiels abgeschlossen? Was ist noch ein Spiel und was keines mehr? Kannst du die Grenzen angeben? Nein. Du kannst welche ziehen" (W ittgenstein 1984: 279) [How is the concept of a game bounded? What still counts as a game and what no longer does? Can you give the boundary? No. You can draw one; translation by A.E.M. Ansconibe]. Accordingly, we draw tlie line between phonology and morphology as follows. Looking at tlie pliantastic names we find e.g., Gorx, which is a manufactured bbord, An-Rukhbar, which is a bipartite manufactured word, Ertruser, a name with a manufactured stem and a regular affix, .JamachinzfluJ, a compound consisting of a manufactured word and the lexeme FIuss 'river', Erosun (Greek eros) and a pseudo-morph, and Terkonit, composed of a manufactured word and a pseudo-morpli. Besides, there are morphologically regular names such as Perlcnn~eer-sea of pearls' and Venusier 'humanoid living on the planet Venus.. A morphologically complex hard is composed of existing tneaningfi~l units (stems and [word fortnation] affixes). That is, it can be divided into morphemes and thus shows morpliological stri~cture.A manufactured word

Between pl~onologyand niorphology

14 1

is not morpl~ologicallystructured; it does not consist of morphemes; it is a neb stem; it operates on the level of sounds (cf. Elsen 2005). Gorx is a clear example of a manufactured word without any trace of morphological structure. It can only be described phouologically. An-Rukhhar shows the orthographic signs of a compound but without semantic support and, thus, can be called a ~na~iufacti~red word. Terkonit offers the interpretation of a derived manufactured word, but since a pseudo-morph does not carry consistent meaning, the name is manufactured. Erosun consists of a stem and a pseudo-morpli. Its morpl~ologicalstructure is that of a lexelne and a manufactured part. Ertruser, on the other hand, consists of a regular affix and a manufactured part. Similarly, .Janiuchin~flu/~ [nay count as a compound consisting of one manufactured stem and one lexeme. Tuhle I . Properties of compounds and manufactured words

Perlenmeer .JurnuchitnfluJ~ .4n-Rukhhur

morphological level Yes yes yes ?

meaningful units yes +yes no + yes no + no

T ~ ~ h2.l e Properties of derivations and manufactured words

lnorphological level Erlruser Erosn~ Terkot~it Gorx

yes ? no

meaningful units no + yes ycs + no no + no no

We [night order the names according to the degree of their lnorphological structure (cf. Table 1 and Table 2): Ve?iu,sier, Perlenrneer, Jumuchin~flu~, Ertruser, Erosan, Terkonit, An-Rukhhar, and Gorx, with a line drawn bet-

142

Hilke E1.ct.n

ween Erosun and Terkonit. The first examples are tnorphologically structured while the last three namcs only show plionological structure. However, the exa~nplesare more or less fitting. Blackbird and learner are vely good examples of composition and derivation in the same way as Perlenrneer and Vtwzfiier. Ero.\tm and .Icmltrchin?flu/3 are less clear examples, as they partly consist of non-existing sound sequences. On the other hand, Gorx is a very good example of a manufactured word. The prototypes Perlenmcer, Ventrsier. and Gorx may serve as the conceptual centre of the process of compounding, derivation, and manufacturing. They may be examples of these categories on the langue-level, to help scientists to structure the lexicon, to simplify communication, and to help the students to understand the concepts. On such a level of abstraction, marginal cases like Erosun only irritate. If we have to find a place in a morphological analysis, this is somewhere between morphology and phonology. If we have to decide, it is, per definitionern, on the side of the morphologically complex words. In fact, however, the peripheries of the categories compound and manufactured word as well as derivation and manufactured word gradually merge into one another to form degrees of membership. From a cognitive point of view, there is a continuum which is interrupted by linguists to meet Aristotelian and langue-oriented needs.

Appendix: Selection of sources for Plzantastic Names AlFla BemSt EschKel EW FelElf FelMa FraSta FraKy FraKo GlaeSt GlaeSk HoHe

Alpers, Hans Joachim. 2003. Der Flcmn~nienbund.Munchen. Bemmann, Hans. 2003. Stein z~ndFliite. MuncheniZurich. Eschbach, Andreas. 200 1. Kelwitts Slern. Bergisch Gladbach. Ewers, H. G. 1979. Die Puru-Skluven. Munchen. Felten, Monika. 2003. E@nfeuer. MuncheniZurich. Felten, Monika. 2003. Die Mucht des EIJL;nfeucrs.MuncheniZurich. Franke, Herbert W. 1962. Die Stahl~viistc.Munchen. Franke, Herbert W. 1978. Ein Kyhorg nurnens .Joe. Berlin, 196411972. Franke, Herbert W. 1964. Dcr grijne Komet. Munchen. Glaesener, Helga. 1999. Dev singende Stein. Miinchen. Glaesener, Helga. 2000. Der ,schwu~.,-eSkurubaus. Munchen. Hohlbein, Wolfgang, Hennen, Bernhard. 2001. Dus Juhr des G'reifen. Bergisch Gladbach. Klein, H. D. 2003. Phuinon~enon.Munchen. Kneifel, Hanns. 1989. Dus hrennende Lubyrinth. Zurich, 1967.

Between phonology and n~orphologv 143 KneiRau Kneifel, Hanns. 1990. Pie Ruunzfulle. Raumpatrouille Orion 6. Zurich, 1968. KneiSe Kneif'el, Hanns. 1995. Serunl des Gehorsams. Munchen. KneiAt Kneifel, Hans[!]. 1995. Atlans Todjkinde. Miinchen. LaPla LaRwitz, Kurd. 1984. Aufzwei Planeten. Berlin, 1897. MiMin Mielke, Thomas R.P. 199 1 . Mingo. Bergisch Gladbach, 198 1 . Miser Mielke, Thomas R.P. 1992. Die EnCfi4l7rung des Serails. Bergisch Gladbach. Pesch, lleltnut W., Allworden, Ilorst v. 1998. Die Ringe u'cr Mrrcht. Pesch Bergisch Gladbach. Percy Rhodun. Die dritte Mucht. 1978. W. Voltz. Gutersloh, 1961. PR I Perry Rhodan. Das Mutuntenkorp.~. 1979. W. Voltz. Giitersloh, early PR2 sixties. PR3 Perry Rhodan. Dcr (ln.sterbliche. 1979. W. Voltz. Giitersloh, early sixties. WHScha Hohlbein, Wolfgang, Hohlbein, Heike. 1996. Schatlenjugd. Eine ~~liuntu.stisc11e Geschichte. Rheda-Wiedenbriick. WHGrei Hohlbein, Wolfgang, Hohlbein, Heike. 2000. Der G r e g Miinchen, 1989. WHGar Hohlbein, Wolfgang. 1995. Die Sugu von Garth und Toriun. Miinchen.

References Brdar-Szabo, Rita, and Mario Brdar Grammaticalization and the lexicon: Core-and-periphery model vs 2000 prototype approach. In Protot)lpentheorie in u'er Linguistik, Martina Mangasser-Wahl (ed.), 139-1 59. Tiibingen: Stauffenburg. Croft, William. and D. Alan Cruse Cognitive linguistic^. 4"' Edition. Cambridge: Cambridge University 2007 Press. Dirven, Rene, and Mat-joli-inVerspoor Cognitive Explorution of Languuge und Lingu.rtics. AmsterdamIPhil1998 adelphia: John Benjamins. Echols, C. I-i. The role of stress, position and intonation in the representation and 1988 identification of early words. In Papers and Reports on Child Langziuge Developinent 27. E. Clark and Y. Matsumoto (eds.), 29-46. Stanford: Stanford University Press.

144

Hilke E/.\en

Elman. Jeffrey L., Elizabeth A. Bates, Mark H. Johnson, Annette Karmiloff-Smith, Do~nenicoParisi, and Kim Plunkett 1 996 Kclhinking Innateness. A Connectionist Perspective on Development. Cambridge, MA, London: MIT. Elsen, Hilke 1991 Evst.~pvacherwt.rh.Dev Erwerh des deutschen Laut.s~s):vtem.s.Wiesbaden: DUV. 1996 Two routes to language. Stylistic variation in one child. First Lunguuge 16: 141--158. 1997 Acquiring verb morphology: German past participles. In Proceedings of the 21st Annuul Boston liniversily Confirence on Language Developn~entVol. I, E. Hughes. M. Hughes, and A. Greenhill (eds.), 160-1 69. Sornerville: Cascadilla Press. 1998 The acquisition of past participles: One or two mechanisms? In Models of Inflection, R. Fabri, A. Ortmann, T. Parodi (eds.), 13415 1. Tubingen: Niemeyer. 1999 Ansiitze zu einer ,funktionali.s/isch-kognitiven Gram~natik.Konseqzienzen az4.s Regularituten de,s Ers/spracherwerb.s. Tubingen: Niemeyer. 2000 Das Veshaltnis zwischen segmentalen und suprasegmentalen Merkmalen in des Kindersprache. Ger~nunistyka15: 49-70. 200 1 Formen, Konzepte und Faktoren der Sprachveranderung. Zeitschriji ,Fir Gertnanistische Linguistik 29. I: 1-22. 2002 The acquisition of German plurals. In M o r p h o l o ~ 2000, ~ W. Dressler, S. Benjaballah, 0. E. Pfeiffer, and M. D. Voeikova (eds.), I 17-127. AmsterdamIPhiladelphia: Benjamins. ;3'sologismen. Formen und Funklionen neuer Wiirter in versehie2004 &nen Cirrietuten des Deur.vchev. Tiibingen: Narr 2005 Das Kunstwort. Muttersprache 1 15: 142-149. 2006 Pseudomorpheme - Fiktive Namen im ~ b e r g a n ~ s b e r e i cvon h Phonologie und Morphologie. Mutter:rprache 3: 242-248. 2007a Wortgruppenlexeme Beispiele aus Enzyklopadie, Zeitung, Baurecht und Wasserbau. Fr~chspruche.lntevnatio~alJournal cf LSP. I 212007: 44-55. Gestaltverarbeitung. Deutsch u1.v F1*en7dspruche312007: 162-1 65. 2007b Elsen, Hilke, and Sascha Micliel Wortbildung und Sprachgebrauch. Desiderate und Perspektiven einer 2007 etablierten Forschungsrichtung. Mutterspruche 112007: 1-1 6. Gregoire, Antoine L'Apl7ventis.suge du Lunguge. Liege: Librairie Droz. 1937 Jakobson, R., and Linda Waugh The SoundShupe qf'Lut7gziage. Berlin/New York: De Gruyter. 1987 -

Bet~vecwphonologv und morpliologv

145

Lamb, Sidne). f'athwuvs of the Brain: The Ncurocognitive Basis of' Language. 1999 AmsterdamIPhiladelphia: Benjamins. Langacker, Ronald W. 2000 G ~ L I N I und ~ I UCoticeptutrlizalion. Y BerlinINew York: De Gruyter. I,ocke, John L.. 1983 Phonological Acquisition and C,'hange. New York: Academic Press. Maddieson, Ian 1984 Pmtterns c f S o u n d ~Cambridge: . Cambridge University Press. Masuda, Keiko 2007 The physical basis for phonological iconicity. In Insistent Itnuges, Elzbieta Tabakowska and Christina Ljungberg (eds.), 57-71. AmsterdanilPhiladelphia: Benjamins. McMahon, April M. S. 1994 Understanding Lunguuge C'hunge. Cambridge: Cambridge University Press. Menn, Lise 1978 Ptrltem, Control, und Contrast in Beginning Speech. A Ccrse Study in /he L)evelopment qf' Word Fnrrn and Word Function. Bloomington: Indiana University Linguistics Club. Muller, Horst M., and Sabine Weiss 2000 Prototypen und Kategorisierung ails neurobiologischer Siclit. In I'rotot~pet~/heoricin der Linguistik. Anwendz~ng.sheispicle i\.tethoderirgflexion - Perspektivcn, Martina Mangasser-Wahl, (ed.), 55-7 1 . Tiibingen: Stauffenburg. Muhlhausler, Peter Pidgin R Creole linguistic.^. Oxford: Basil Blackwell. 1986 Ohala, John J . The frequency code underlies the sound-symbolic use of voice pitch. 1994 In Sound Symbolism, L. Hinton, J. Nichols, and J. J. Ohala (eds.), 325-347. Cambridge: Cambridge University Press. Plunkett, Kim, and Virginia Marchman U-shaped learning arid frequency effects in a multi-layered per199 1 ceptron: Implications for child language acquisition. C'ogniliot? 38: 43-102. From rote learning to system building: acquiring verb morphology in 1993 children and connectionist nets. C'ognition 48: 2 1-69, Ronneberger-Sibold, Elke Preferred sound shapes of new roots: On some phonotactic and 1996 prosodic properties of shortenings in German and French. In Nulurul Phonology: The Stute c!f'f'rheArl, B. Hurch and R. A. Rhodes (eds.), 263-293. BerlinJNew York: De Gruyter.

146

Hilke Elsen

2005

Zur Definition und Typologie von Wortkreuzungen. Ein Vorschlag airf der Grundlage ihrer relativen Transparenz. In Sprache zlnd Na!!lr!ichkeir. G. Fenk-Ocz!on and C. Winkler (eds.), 205-224. Tiibingen: Narr. Rurnelhart, David, and James L. McClelland 1986 On learning the past tenses of English verbs. In Parallel Distributed Processing Vol. 11. P.sychological and Biological Models, J . L. McClelland, D. Rumelhart, and the PDP Research Group (eds.), 216-270. Cambridge. MA: Bradford Books, MIT. Smith, Neilson V. 1973 The Acquisition c?J'Phonologv.A Case Study. Cambridge: Cambridge University Press. Stemberger, Joseph P. 1992 A connectionist view of child phonology. In Phonological Development, C. A. Ferguson, L. Menn. and L. Stoel-Gammon (eds.), 165189. York: Timonium. Taylor, John R. 1995 Linguistic Categorization. Prototypes in Linguistic Theory. 2"d Edition. Oxford: Clarendon Press. 2002 C'ogr~itiveGrcirnnlar. Oxford: Oxford University Press. LJngerer, Friedrich, and Hans-Jorg Schmid 2006 .An Introduction to Cognitive Linguistics. 2"" Edition. London: Pearson Longman. Vihman, Marilyn M. 2001 W-ord templates and the origins of phonology. Paper presented at the 7'" International Cognitive Linguistics Conference, July 26, 2001. [MSI Wittgenstein, Ludwig 1984 Philosophische Untersuchungen. In Tractatus logico-philosophiczrs, Ansconibe, G. E. M., G. H. von Wright, Rush Rhees. 2 2 5 4 2 1 . Frankfurt am Main: Suhrkamp. English translation by G. E. M. Anscombe. 1978. Philo.sol~hical Investigations. Oxford: Basil Blackwell.

The role of grammatical gender in noun-formation: A diachronic perspective from Norwegian

Philipp C'onzett

1. The relationship between gender and word-formation According to Corbett (1991: 1) "[glender is the most puzzling of the grammatical categories". In modern languages, however. gender is most often seen as nothing more than an abstract inherent classificatol-y feature of nouns that triggers agreement in associated words. Given this perspective of gender as a redundant category, the question arises of why it nonetheless is so persistent in a great number of languages. This question has been answered inter alia by referring to the identifying and disambiguating fi~nctiongender can have in discourse (e.g., Corbett 199 1 : 320-32 1). In this article further evidence is provided for viewing grammatical gender (henceforth gender) as an integral pat? of Cognitive Grammar, more specifically the domain of word-formation. The relationship between gra~nmaticaigender and word-formation can be approached from (at least) two different angles. In literature dealing with gender assignment, cliaracteristics of word-formation are often used as a base for assigning gender to nouns. This approach is presented in section 1. I . On the other hand, gender is described as a feature involved in the formation of new nouns. This perspective is introduced in 1.2.

1 . 1 . Gender assignment based on word-formation

Regularities between the gender of nouns and their derivational morphology can be detected in a number of languages. Gender can be tied to overt or covert derivational features. The former type is usually realized by suffixation, e.g., Norwegian klok (adj. 'wise') + skap 3 klokskap (masculine [MI 'wisdom'). The latter type w e find in nouns created by conversion, e.g., Norwegian ( u ) kmte (v. 'throw') 3 kust (M 'throw'). Furthermore, regularities between gender and derivational features can be more or less consistent. In Norwegian, some derivational suffixes are unambiguously

tied to only one gender. The suffixes -ersk, -heit, -inn, for example, are only found in feminine (F) nouns, whereas -ur, -ling, -nod are M, and -eri, -kru/i are N. Other suffixes, such as -in (FJM) or -5kap (Mlneuter [N]), are compatible with more than one gender. Naturally, these ties have led grammarians to formulate gender assigntilent rules based on derivational morpliology. For Norwegian, Trosterud (2001: 4 3 4 4 ) assumes inter alios the following assignment rules (adapted by tlie author): (1)

a. Nouns derived in -inn [...I are F. b. Nouns converted from verbal stems are N.

In case of only partially consistent regularities between gender and derivational features, other lexical properties of nouns are sometimes referred to in order to account for the assignment of gender. For Norwegian deverbal nouns derived in -ing, tlie distinction between F and M is closely related to tlie distinction between abstract and concrete. Tlius, the abstract noun kgyring (F 'driving' from krtyre v. 'drive') takes F, while the concrete noun gjelding (M 'gelded animal' from gjelde v. 'geld') is assigned M. Tlie basic idea behind tlie approach described in this section is that the derivational morpliology of a noun in some cases determines its gender. Tlie relationship between the two features is thus seen as directional, from word-formation to gender assignment.

1.2. Gender as a feature of word-formation Another perspective on the relationship between gender and wordformation is provided in literature dealing with word-formation. In some languages gender is described as a feature involved in the formation of new nouns. In particular, gender is essential in word-formation processes reflected in pairs of animate nouns like the ones in (2).' In these examples, apart from declension class, gender is tlie only formal feature that distinguishes the noun in each pair and can thus be considered as a constitutive factor in tlie word-formation process beliind these word pairs. Tlie sample set in (2) is limited to animate concepts, and tlie nouns in such pairs are sometimes called motion-nouns (cf., e.g., Corbett 199 1 : 67). Tlie wordformation process resulting in such couples is accordingly named motion (German "Movierung"), i.e., tlie formation of nouns for female animates

The role of gram~naliculgender in no~n~formation 149

from noun for male animates (or vice versa) (Bergenholtz and Mugdan 2000: 444).' M arrlic-us ('male friend') /up-us ('male w o l f ) rnaestr-o ('male teacher') b. ITA ragazz-o ('boy') c. GRE adherfox ('brother') non-os ('godfather') lieuv-i,s ('male Lithuanian') d. LIT pedagcig-a,, ('male teacher') e. OHG h2r.r-o ('master') gastgeh-o ('male host')

(2) a. LAT

F cmlic-a ('female friend') /up-a ('female w o l f ) maestr-a ('female teacher') ragazz-a ('girl') adherfi ('sister') non-a ('godmother') lieuv-c; ('female Lithuanian') pedagog-c; ('female teacher') h6rr-a ('mistress') gastgeb-a ('female host')

However, the gender pattern illustrated in (2) is not restricted to the domain of biological sex only. In Italian we find a corresponding pattern in some noun pairs denoting 'fruit' and 'tree' respectively. Examples are given in (3) below (Schwarze 1988: 14, 454, cf. also Koch 1999: 158):

(3)

F a. urur7ci-a ('orange') b. niel-a ('aple') c. per-a ('pear')

M arcxnci-o ('orange tree') me/-o (-aple tree') per-o ('pear tree')

In (3) the feminine nouns on the left denote a fruit, and the masculine nouns on the right denote the tree on which this fruit grows. The patterns illustrated in (2) and (3) can be generally labelled as gender patterns: (4) a. A genderpattern is a set of etymologically related nouns which are for~nallydistinguished solely by their gender (and optionally by their inflectional class membership). b. The assumed word-formation process resulting in such gender patterns is accordi~iglynamed gender patterning. The derivative aspect of gender outlined above is also recognized by Me17Euk, who provides the most extensive formal definition of gender known to me. "Differences in gender[...]", as he puts it, "can be exploited by the language in order to express certain derivatemes, but always in an

1 50

Philipp Con-en

irregular, unsysten~aticway - that is, in isolated cases" (Mel'Cuk 2006: 329). From this, Mel'Cuk concludes that the described property of gender "is too capricious (because it is too lexicalized) to seriously affect the essence of a gender[ ...I system" (Mel'Cuk 2006: 329). In a similar vein, Koch (1999: 158) touches upon the issue, declaring it to be "of minor importance". On the other hand, Aikhenvald (2007: 37) in her survey of typological distinctions in word-formation revalues the status of gender patterns, referring to them as a "derivational device". While these accounts provide different evaluations of tlie status of gender, they actually fail to elaborate on the phenomenon in depth but mostly rely on data presented in overview works such as reference grammars. The aim of this paper is to present a more thorougli analysis of how gender relates to the domain of lexical development. More specifically, it will be examined in how far gender constitutes a cognitively salient pattern of word-formation. In this way, the sti~dystrives to contribute to our understanding of the nature of lexical structure and organization. The analysis is limited to Norwegian data and takes its starting point in gender patterns as defined in (4). To my knowledge, this phenomenon has not been studied in any North Germanic language. The present study contributes thus to tlie tield in three ways. First, it adds to a more co~npletedescription of the Norwegian language. Second, it sheds new light on a lesser studied wordformation pattern in general. Finally, as will be argued in chapter 2, gender patterns can be accounted for in a cognitive network model. It will also be demonstrated how the model can deal with gender assignment based on word structure and, thus, cope with both aspects of gender outlined in 1 . I . and 1.2. above. In this respect, the study thirdly renders further support for a cognitive approach to tlie inquiry of language. In addition, section 2 specifies the methods and the data used in this study. The data is analysed and interpreted synchronically and diachronically in chapter 3. Finally, the article is rounded up with concluding remarks in chapter 4.

2. Background 2.1. Segmental derivation The theoretical framework for this study is founded on tlie network niodel proposed in Hybee (1985 and later). Morphologically complex words are, like root words, stored in the lexicon. But tlie lexicon in this model is not purely a list of individual items (cf. Di Sciullo and Williams 1987: 3).

The role r!f'grut~~moticuI gender in no~n~firrnutro1715 1

Rather, words are connected to each other in a structured way. Connections are found on all levels of lexical specification: phonological, semantic, and syntactic. Morphological word structure emerges from the connections between words. Essentially, these connections are bi-/multidirectional, and within a given lexical donlain directionality can vary locally. The strength of the connections is influenced, on the one hand, by the grade of similarity between the items, and, on the other hand, by type and token frequency of actual usage but also by the cognitive salience of the concept encoded in the different lexical items. 1 have argued that new nouns take their gender based on the existing patterns between gender and other lexical properties (Conzett 2006). Similarly, the formation of new words is based on patterns found in the lexicon. New words created in this way can either gain the status of permanent lexical items or they remain occasional single formations. When studying word-formation in a network model, we thus have to recognize two levels of analysis. On the one hand, there is the level of synchronic word structure, i.e., the internal structure of lexical items emerging from their connections with other items in the lexico~i.On the other hand, there is the level of the diachronic process of creating new lexical items based on existing patterns in the lexicon. In other words, we have to distinguish between word-motivation and word-derivation (cf. Dokulil 1968: 2 19, Mel'Euk 1976: 290-291). From a synchronic point of view, the relationship between the English noun freedom and the adjective.free can be described as (a) a partial segmental identity between the two, and (b) as the former being motivated by the latter, in the sense that the concept of FREEDOM is essentially based on the concept of FREE. In practice we can paraphrase the meaning of freedom with 'the state of being free'. A similar relationship is found between other pairs of lexemes like boredom - bore (n.) and bunzbleu'oni - bun~ble(n.). There are, thus, strong connections between the first elements in the nouns on the left side and the lexemes on the right side. On the other hand, there are strong relationships between the element -don1 in the nouns on the left side. A schematic illustration of the relationships between these lexical items is given in Figure 1.

Figure I. Lexical network for English nouns in -dam

Figure 1 should not be misinterpreted by reducing the connections to only being valid between segmental elements. Connections in a lexical netuork are found between all kinds of properties of words. In the exa~npleabove, there are also connections between the meaning of the involved words, their word class, gender, etc. But, of course, these connections can be of different strength. There are, for example, stronger connections between bore and bun~hleon the one hand andfree on tlie other since the two words in the former pair both share the same word class, and they are also semantically more closely related. For practical reasons, these types of relatedness are difficult to depict graphically. From this system of connections, tlie semantic word structure pattern emerges as paraphrased in 'X + h r n = the state of being X'. New lexemes can thus be created based on this established pattern, e.g., star - stardom (n.), yuppie - yuppieu'orn (n.) (cf. Plag 2003: 88). From a diachronic point of view, we can say that yuppie don^ was derived from yuppi(2 by suffixing dam. From a synchronic point of view, yzippieu'nm has an internal structure that is motivated by the noun yzippie and tlie connections emerging from the lexical network described above (cf. 'Tuggy 2005: 25 1-252).

2.2. Gender patterning

In languages with gender, the gender of noiuis is included in tlie system of lexical connections just like otlier properties of lexical items. The gender pattern described in section 1.2. is analysed in basically the same way as outlined for the formation of English nouns in -&nz above. The gender of a noun emerges from its connections to otlier nouns and to other co-

The vole of grurnmutical gender in no~n~formution 1 53

occurring lexical items that modify the noun, either in attributive or predicative position. The situation is illustrated with the Old Norse (ON) attributive modifiers in (5).

(5)

(A) a. sd grunn-i thatneighbourMASC.SG.NOM SG.NOM. b. grann-u sin-urn neigh bourhisSG.DAT MASC.SG.DArT c. grunn-ar lveir neighbourtwoPL.NOM MASC.NOM vin-r d. sh thatfriendMASC.SG.NOM SG.NOM. e. vin sin-utri I1i sfriendSG.DAT MASC.SG.DAT f. vin-ir tvcir friendtwoPL.NOM MASC.NOM

(B) su

thatFEM.SG.NOM grunn-u neighbourSG.DAT griinn-ur neighbourPL.NOM .Y u thatFEM.SG.NOM vin-u friendSG.DAT vin-ur friendPL.NOM

gra~m-a neighbourSG.NOM sin-ni herFEM.SG.DAT tvar twoFEM.NOM vin-u friendSG.NOM sin-ni herFEM.SG.DAT tv&r twoFEM.NOM

In column (A) on the left, the nouns granni (M 'neighbour') and vinr (M 'friend') are modified by a demonstrative (a, d), reflexive possessive (b, e), and a numeral (c, f). In column (B) on the right, the nouns granna (F 'female neighbour') and vinu (F 'female friend') are modified correspondingly. The gender of the nouns in (5), that is M in (A), and F in (B), is the pattern that emerges from the network of connections as depicted in Figure 2. As the illustration reveals, gender is an emergent pattern in a structured lexicon, and there is obviously no reason for why this pattern could not be involved in word-formation. In fact, the examples in Figure 2 can of course be analysed as a gender pattern. In (6) this gender pattern is described in a less cumbersome way. (6)

M '(male) person'

:

F 'female person'

Gender patterns in modern Norwegian most likely confirm that the gender pattern in (6) at one stage served as template for gender patterning.

1 54

Philipp C'onzett

Figure 2. Lexical network for gender in Old Norse"

An illustrative candidate at hand is the noun pair kokk (M 'cook') - kokke (F 'female cook'). According to Bjorvand and Lindelnan (2000: 471) kokk was borrowed from Middle Low German kok. Later, the new noun kokkc. was formed, most probabely by the process of gender patterning based on kokk, as shown in Figure 3.4 M '(male) person'

kokk (M)

:

F 'female person'

kokke (F)

Figwe 3. Gender patterning in ~ o r w e ~ i a n '

In the process of gender patterning, word-fortnation and gender assignment is carried out simultaneously. However, the network model presented above also copes with the assignment of gender to new nouns not estab-

The YO/

[

I

extension jar lid

Hogwarts headache

belly bcrtton

Potterhead

Figure 4. The elaboration-extension relationship of creative noun-noun com-

pounds (source: own construction). Accordingly, creative extension can be correlated \vitli transparency of meaning. Metaphorical and/or n~etony~nical compounds represent various levels of semantic transparency depending on which constituent is affected by metaphor or metonymy. As mentioned before, the semantic co~nplexity of a compound (jailbird or.fleahag versus godchild) does affect processing times - in other words, the creative extension of jcrilbird or Jleabog is greater than that of godchild, even thoilgli all three represent tlie same degree of elaboration (1,ibben et al. 2003). Based on the results mentioned above, it can be hypothesized that a coinpound wit11 a metaphorical or metony~nicalhead would fall higher up on tlie axis of extension than a compound with a metaphorical or metonymical modifier. However, it should be borne in mind that even further distinctions of extension (and transparency) can be established even within one single class of creative compounds if further factors, such as lexicalisation or freque~icyeffects," are also considered, not to mention the importance of context - be it textual, cultural or social.

Setting linzif.7 on creativity

23 7

4. Conclusion The paper set out to limit the notion of linguistic creativity inherent in the production and use o f metaphorical and metonymical compounds, by exploring two issues within a cognitive linguistic framework: i) what are the motivating factors in the creation of metaphorical and metonymical compounds, and ii) how can transparency of meaning and creative language use be reconciled. With regard to the first question, the paper has come to the conclusion that in most cases several motivating factors (such a s compactness, vividness, context, memorability, analogy, and remotivation) are at play in the creation of a creative compound. It has also been hypothesized that the more motikating factors there are, the greater the possibility for the expression to g o into wider use. As for the second question, in line with recent psycholinguistic research on compound processing, the paper has argued that the Langackerian notion of degree of extension is synonymous with the concept o f degree of creativity among creative compounds. The more extended a compound, the more imaginative, associative thinking is required from the listener to arrive at the compound's meaning.

Notes 1.

2.

3.

4. 5. 6.

I wish to thank Zoltan Kovecses for his useful suggestions on an earlier draft of this paper. I am especially grateful to the editors of this volume, Sascha Michel and Alexander Onysko, for their careful and meticulous editing and their highly constructive suggestions, and to Matthias Hiining and an anonymous reviewer for their cogent remarks. Needless to say, all remaining errors or inadequacies are my responsibility. According to Chomsky (1980: 322), grammar (syntax), which characterizes the properties of sentences, is an area of our linguistic knowledge that can be well ~tudiedand described. However. "the creative use of language is a mystery that eludes our intellectual grasp". The influence of context on the coinage of compounds will be discussed below, in the following subsection. The expression originates from two Australian television presenters. Source: Gruzia, UK Edition, 14 February 2005. p. 42. Source: "Potter 'gives children headaches"' (source: http:// news.bbc.co.uk/go/pr/fr/-i2ihi/ente1-taimet2645.stm [emphasis mine]). Source: http://www.wikipedia.org

7.

8.

9.

10.

France expressed strong opposition in the United Nations with regard to invading Iraq. Nevertheless, some private restaurants took on the idea. The restaurant chain Cubbies, for instance. decided to adopt the new name. The owner, Neal Rowland, explained this move as follows: "Because of Cubbie's support for our troops, we no longer serve French fries. We now serve freedom fries. [...] It's our way of showing our patriotic pride" (http:/lwww.cnn.com; 19 February 2003). This kind of linguistic activity is not peculiar to the United States: in Britain, the German Shepherd was renamed as Al.satiun during the First World War. A more recent example comes fi-on1New Zealand, where Frencli loaves became Kiwi louves in 1998, when the French government was engaged in the testing of nuclear weapons in the Pacific (Wikipedia). What is more, Danish pustrie.~ were substituted by Roses of' the Prophcjt Mz~hummud in Iran in February 2006, as a result of the Muhammad cartoons controversy (http://www.thestar.com; 16 Februa~y2006). The owner of the Cubbie's restaurant chain, Neal Rowland, in fact claimed that the inspiration forfkeedom ,fiie.s came frorn a conversation about the linguistic changes that were propagated during World War I. The close association of America with the concept of freedom can also be observed in present-day political speech. During the second presidential term, the Bush administration renamed the war against Iraq as the "war for freedom" as opposed to the "war on terrorism", which was the expression used during Bush's first presidential term. (Source: BBC News, http://news.bbc.co.uk/2!hi/a1nericas/4628377.stni; accessed 23 May, 2008) Both Downing (1977) and Warren (1978) have alluded to the location relationship as one of the compound-forming patterns of English endocentric compounds. Needless to say, Order qf !fthePhoenix heud~~che might also have been discarded due to the length of the modifying constituent. Length, however, could not have been a reason for abandoning Potter as a possible modifier it is two syllables long, just as Hogwurts. The expression comes from a sign that was spotted outside the Church of Christ in Mountlake Terrace: "God Answers Knee-mail" (Source: http:1/www.wordspy.co1n). Nevertheless, it needs to be pointed out that according to de Jong et al. (2002) there seenis to be a bias towards the family size of the modifier constituent. In an experiment participants were able to interpret compounds faster when the modifier family was large while no such affect was reported in the case of the head constituent's family, irrespective of its size. Sourcc: Macmillan English Dictiot~utyMugazine, Issue 40 (July 2006). Available online: http://www.macmillandictionary.com. -

1 I.

12.

-

13.

14.

15.

Selling 1imit.y on creativity

239

16. Source: "Proud to Be a Potterhead" by Sabaa Saleem Tahir, www.washingtonpost.com, http://www.washingtonpost.com/wp-dynlcontenti articlei2007/07/13/AR200707 I30 17 1 1 .html 17. The head is a container conceptual metaphor is extremely prevalent in English and underlies numerous idiomatic expressions. Benczes (2002) analyses idioms that contain the word head in them and shows that basic metaphors such as the head is a container and ideas are physical objects motivate idioms such as have rocks in one's head and need to have one's head examined 18. Kovecses and Radden (1 998: 64) explain that the selection of a metonymical source depends upon human experience among others. We prefer to give salience to human over non-human entities, as in "I'm reading Shakespeare". This preference can be accounted for by our anthropocentric view of and interaction with the world. 19. A similar phenomenon can be observed in metonymy: the more cognitive principles (motivations) apply to a given metonymy (such as concrete over abstract, human over non-human, specific over generic, e t ~ . )the , greater the motivation of metonymy (Kovecses and Radden 1998: 7 I). 20. The semantic characteristics of a transparent constituent correspond to the semantic characteristics of its free-standing lexical variant. 2 1. Co~npoundswith semantically non-transparent constituents that participated in the experi~iientwere mainly metaphorical or ~netonymical(see Libben et al.'s full list of the compounds, 2003: 54). 22. It is a well-known fact in psycholinguistics that a high-frequency word such as cur is recognized more quickly than a low-frequency word such as doe. This phenomenon is referred to as the word frequency effect.

References Adams, Valerie 1973 An Inlroduclion to Modern English Word Formalion. London: Longman. 2001 Complex word.^ in English. Harlow: Longman. Bauer. Laurie 1983 English Word-Formution. Cambridge: Cambridge University Press. Benczes, Reka 2002 The semantics of idioms: a cognitive linguistic approach. The Even Yearbook 5: 17-30. C'reative Compouriding in English: The Semuntic.~of Metaphorical 2006 Amsterand Metonymical Noun-Noun Combinations. dam1Philadelphia: Benjamins.

Brekle, Herbert E. 1978 Reflections on the Conditioris for the Coining, Use and Understanding of Norninal Compounds. In Proceedings of the Tu'elfih International Congre.v.s of Linguists, Wolfgang U. Dressler and Wolfgang Meid (eds.), 68-77. Innsbruck: Innsbrucker Beitrage zur Sprachenwissenschaft. BrBne, Geert, and Seana Coulson (in press) On the cognitive processing of deliberate ambiguity in newspaper headlines: the case of double grounding. To appear in Discourse Processes. BrBne, Geel-t, and Kurt Feyaerts 2003 The cognilive 1ingui.stic.s c?fincongrui~resolution: Murked refirence point strzrctllres in humor. Leuven: Katholieke Universiteit Leuven. Chomsky, Noam 1980 Rzrles und Representatior~s.Oxford: Basil Blackwell. Clark, Eve V. 1987 The principle of contrast: a constraint on acquisition. In Mechunisms oj'Langzrage ilcyuisition, Brian MacWhinney (ed.), 1-34. Hillsdale, NJ: Erlbaum. Collins Cobuild English Dictionury.for Advanced Leurners (CC'ED) 2001 3'"dition. Glasgow: Harper Collins Publishers. Croft, William, and D. Allan Cruse 2004 Cogrlilivc Linglristic.~.Cambridge: Cambridge University Press. Downing, Pamela On the creation and use of English compound nouns. Lunguuge 53 1977 (4): 8 10-842. Fauconnier, Gilles, and Mark Turner Conceptual integration networks. Cognilive Science 22 ( 1 ) : 133-1 87. 1998 Gagne, Christina L., and Thomas L.Spalding Conceptual Combination. In The Rel~resentationand Processing of 2006 Corr~poundCVorulv, Gary Libben and Gonia Jarema (eds.), 145-168. Oxford: Oxford University Press. GradeEak-Erde!jic, Tanja Euphemisms in the language of politics or how metonymy opens one 2005 door but closes the other. In Prugtilrrtics Today, Piotr Cap (ed.), 287299. Frankfurt am Main: Lang. Gibbs, Raylnond W. The Poelics ~ f ' M i n dCambridge: Cambridge University Press. 1994 Jespersen, Otto A Modern English Grornmar on Hisloricul Principles. Part VI: Mor1954 phology. London: Bradford and Dickens.

Setting lin~itsoti creativity

24 1

Kovecses, Zoltan 2000 American English: An Introduction. Peterborough, OT: Broadview Press. 2005 h'etuphor in C'ullur~:U n i v e c ~ u l and i ~ Vuriution. Cambridge: Cambridge University Press. Kovecses, Zoltan, and Giinter Radden 1998 Metonymy: Developing a cognitive linguistic view. Cognitive Lingusitics 9 (I): 37-77. Jaarsveld. Henk J. van, Riet Coolen, and Robert Schreuder 1994 The role of analogy in the interpretation of novel compounds. Journu1 of P.s~~cholirigui.s/ic Research 23 (2): 1 1 1-1 37. Jong, Nirja H. de, Laurie B. Feldman, Robert Schreuder, Matthew Pastizzo, and Harald R. Baayen 2002 The processing and representation of Dutch and English compounds: Peripheral morphological and central orthographic effects. Brain and Language 8 1 (1 -3): 555-567. Krott, Andrea n.d. The role of analogy for the production, interpretation, and acquisition of compound words. http:!!www.~pblevins.net/wordandparadig~n/~d/krott 140307.pdf (accessed: 22 August 2007). Lakoff. George, and Mark Johnson 1980 Metriphovs We Live B-v. Chicago: The University of Chicago Press. Lamb, Sydney M. Prrthway.~ofthe Brain: The Neurocognilive Basis of Lunguage. Am1998 sterdam/Philadelphia: Ben~amins. Langacker, Ronald W. Four7dutions of Cognitive Grumnim-. Volume I: Theoretical Prereq1987 uisites. Stanford: Stanford University Press. Foundations of Cognitive Grummar. Volume II: Descriptive Appli199 1 cation. Stanford: Stanford University Press. Granzmar mnd Conct.l,~zinli~ation. BerlinINew York: De Gruyter. 2000 Lees, Robe~tB. The Gra~nriiurcf English Nomina1izution.s. 5"' printing. The Hague: 1 968 Mouton. Levi, Judith N. The Syntax and Sen~anticsqfC'ornple,x Nominals. New York: Aca1978 demic Press. Li. Charles Semantics and the Structure of Compounds in Chinese. Ph.D. diss., 1971 University of California at Berkeley.

Libben, Gary 2006 Why Study Compound Processing? An overview of the issues. In The Representation and Processing of Conzpound Words, Gary Libben and Gonia Jarema (eds.), 1-22. Oxford: Oxford University Press. Libben, Gary, Martha Gibson, Yeo Bom Yoon, and Dominiek Sandra 2003 Compound fracture: The role of semantic transparency and morphological headedness. Brain and Language 84: 5 0 4 4 . Longman Dictionary of Contenlporary English. International Studenls Edition (LDOCE) 3"' edition, 6'" impression. Harlow: Longman. 1999 Oxfbrd English Dictionary, The (OED) 1989 2"d edition. Clarendon Press: Oxford. Ryder, Mary Ellen Ordered Chaos: The Interpretution of' English Nour7-Noun Corn1994 pounds. Berkeley and Los Angeles: University of California Press. Selkirk, Elizabeth 0. 1982 The Syntm of Words. Cambridge, MA: MIT Press. Talniy, Leonard The relation of grammar to cognition. In: Topics in Cognitive Lin1988 guistics, Brygida Rudzka-Ostyn (ed.), 165-205. Amsterdam1Philadelphia: Benjamins. Taylor. John R. The ecology of constructions. In Studie.~in Linguistic Motivation, 2004 Giinter Radden and Klaus-liwe Panther (eds.), 49-73. Berlinhlew York: De Gruyter. Wales, Katie A Dictionuyj~qf ,fStyli.stics.London and New York: Longman. 1989 Warren, Beatrice Semuritic Patterns 9fNoun-Noun Compounds. Gothenburg: Gothen1978 burg University Press.

Casting the conceptual spotlight: Hybrid compounding in German as an example of headframe internal specifier selection Alexander Onysko

1. Compounding as a productive process of word formation One of the most peci~liarfeatures of the German language is the process of nominalization, whicli permeates syntactic and, even more, word formational productivity in the language. The formation of nominal compounds, specifically determinative compounds, is the dominant word formational process that guides the extension of its lexicon, as generally observed by scholars investigating word formation in German (cf. Eichinger 2000: 58, Erben 2006: 58, Donalies 2005: 52, Motsch 1999: 372). According to Duden, approximately two thirds of word formations consist of nominal co~npounds(1998: 409). In line with Motsch (1999: 372), German compounds can be classified as "Koordinativkomposita" (copulative compounds), "Determinativkomposita" (endocentric compounds), and "Possessivkomposita" (exocentric compounds). This distinction is parallel to its typologically close kin English, in which the creation of compounds follows similar patterns as in German (cf. Bauer 1983, Katarnba 1993, Plag 2003). In German and English nominal compounds, the right hand element is usually considered to be the head due to the fact that it bears the ~norphological markers of the whole term (plural, case, and gender as in die Seeblunlen 'lake flowers'). In many nominal compounds, headedness is also indicated se~nanticallysince the right hand element represents the base concept that is modified by the preceding specifier/determinant. In Seehlzivl~enthe first element See denotes a specific type of the base concept 'flower'. This semantic interpretation of headedness has actually given rise to the tripartite classitication above whicli has turned into a long standing tradition i n research on word formation. On a closer perspective, however, the notion of semantic headedness merely supports a main division in copulative compounds on the one hand, atid in endocentric and exocentric compounds on the other hand. Thus, in copulative compounds, the compo-

244

Alexander O n y s k o

nents contribute equally to tlie semantic value of the construction. 7'liis is emphasized by the possibility of using tlie compound elements ambipositionally while the referent remains tlie same (e.g., Dichter-Sunger 'poet singer', Sungcr-Dichler 'singer poet'). Endocentric and exocentric cotnpounds, 011 tlie other hand, show a common semantic pattern of specifier and head, i.e., the first element determines the semantic value of the head element. This also holds for complex determinative compounds as in Ta,schengeldgesellschuft and Vern~ugensver~t~nltunggsverlrug whose head elements (Ge.sellschaji and F'ertrugl) are semantically specified by their preceding nominal compounds cum determinants. So called exocentric compounds such as Rotkehlchcn ('redbreast'), Dickkopf ('sti~bbor~i person'), or for English butterfingers and lotrdt~~outh, exhibit the same basic semantic projection from specifier to head even if tlie referent is not directly given by the head concept. As noted by several scholars, it is actually their metapliorical and metonymic reference that differentiates exocentric from straightforward sandwich) endocentric compounds sucli as Butterhrol (-butter bread'" and Feigenhlatt ('fig leaf) (cf. Motsch 1999: 372, Benczes 2006, Onysko 2007: 194-1 96, Ortner and Miiller-Bollhagen 199 1 : 1 15-1 16). The common semantic architecture of specifier and head and the fact that metaphorical and metonymic relations explain the referential range of the head element led Benczes to the conclusion to consider exo- and endocentric compounds as one phenomenon in her analysis of metaphorical and metonymic English compound nouns (2006: 183). Despite the widespread acknowledgement of the tight semantic interrelation of specitication between tlie compound elements (also cf. Erben 2006: 60-63), analyses of German nominal compounds have largely focused on the syntactic relation of the constituents, expressed in phrasal descriptions of the compound meaning, e.g., Bierlaune; Laune, die dudurch, dass jemund Bier trinkt, verur,\acht i,ct 'beer mood; mood, which is caused when a person drinks beer' (Motsch 1999: 393). In turn, syntactic extensions serve as a basis for establishing categories of specification: Referenten sind N[noun], die N1[nominal specifier] verursachen Erholungskur, 'Kur, die Erholung bewirkt' (Motsch 1999: 407) [Referents are nouns that cause the nominal specifier Recovery stay at a health resort, stay at a health resort that causes recovery] Similarly, earlier extensive research by Ortner and Ortner (1 984: 14247). has established 20 main classes and, including subgroups, a total of 58 types of compounds characterized by the specific semantic relation be-

C'usfing /he conceptuul spotlight: Hybrid cornpo~mding 245

tween specifier and head. Thus, compo~uldsof the type 1V ('constitutional compounds') are subdivided into collective-constitutional (e.g., Putien/enkollektiv 'collective of patients') and material-constitutional co~npounds (e.g., (jlvorrtzl 'oil reserves'). What cliaracterizes this attempt at classifying compounds in German is tlie fact that categories are set up to reflect common semantic functions of the specifiers as deduced from the syntactic extension of tlie compound. This process, however, runs counter to the general asymmetry of semantic projection from specifier to head, i.e., the head-centric semantic nature of compounds. By the same token, a determinant-centric classification tends to neglect the potential conceptual influence of the head element on specifier selection. In earlier research on English compounds, two studies (Downing 1977, Warren 1978) try to account for tlie relation of specifier and head elements. Both Downing and Warren classify compounds according to relational categories of determinant and detertninatum such as whole-part (duckjbot), part-wliole (urwichuir), source-result (hay jever), resemblance/co~nparison (bell hotloms), and a few others. In addition, Downing draws attention to tlie fact that the more common or liabitilal relationship between specifier and head guides the interpretation of a novel co~npound(depending on tlie amount of discourse context). Thus, a stone house will favour the decontextualized meaning of a house constructed of stones rather than a house that contains stones. Generally, both studies establish underlying categories of compounding patterns that allude to the cognitive potential inherent in the structure of compound nouns. However, Downing and Warren remain focused on tlie relation of specifier and head without touching on the aspect of selectional contiguity of the specifier and the head noun. More recent allusions to underlying cognitive mechanisms for the creation of compounds in German are given in Eicliinger's introduction to German word formation (Deutsche Wortbildung: Eine Einfulirung, 2000: 10): Klassematische Kategorien, die aus der lexikalischen Bedeutung der Elemente abgelesen werden konnen, erleichtern die genauere Identifikation der Beziehung. Bei SchieJhude wird man, wegen des Elements { b u d } , sicher eher auf eine lokale Relation ,wo etwas stattfindet' - komtnen als bei detn Zweitelement von SchieJeisen oder ,5'ch~eJ~~uffe, wo die instrumentale Relation - ,womit man etmas tut' z~eifellosdie nahe liegendste Wahl ist. [Cla~semiccategories, which can be established from the lexical meaning of the elements, facilitate the more precise identification of the relation. Because of the element {hoo/h/stull} in shootrng boothhtall, the locative relation - 'where something takes place' will be close at hand while the sec-

-

-

ond elements in shooting iron or shooting weupon, certainly emphasize the instrumental relation - 'what it takes to do something'.] Eichinger's observation that the element -hz~& evokes a locative sense since these types of stalls are typicall) found in particular locations such as markets and amusement parks is but one of the inherent conceptual associations of -hude. Compound-internally, the specifier &hie$ in SchieJhun'c stresses another associative potential of the head element. That is the notion that a -bt{le has a particular function, purpose, and commercial content. Accordingly, the specifier explains the purpose of the stall as a place for amusement shooting and, as sucli, appears associatively grounded in the head concept -bude. l h i s associative relation between the specifier and the head element in compound nouns is the cornerstone of the followir~gtheoretical discussion and analysis of hybrid nominal compounds in German. Before going into more theoretical and exemplary details, it is necessary to outline the set of working hypotheses which frame the ensuing analytical approach to compound formation in German: (a) In determinative nominal compounds semantic information is projected in the direction from the specifier (determinant) to the head (determinatum). (b) The direction of projection indicates that the specifier is compound-i~~ternally dependent on tlie head, which fi~nctionsas a conceptual base of the construction. (c) In line with this direction of projection, specifier selection in determinative nominal compounds is head-centric, i.e., the conceptual/semantic frame of the base concept guides the selection of the determinant. (d) Tlie amount of contiguity among the different associative potentials (i.e., frame internal meaning potentials) varies according to tlie dynamic prototypicality of the associative relations in the semantic frame of the head. (e) Tlie degree of prototypicality (i.e., frame-internal associative strength) influences specifier selection on a cline from less context dependent (most prototypical) to more context dependent (least prototypical). (0 Type and token frequencies of frame-internal associations in compounds provide an indirect measure of tlie associative strength (i.e., the prototypicality of the contiguity) of frameinternal meaning potentials.

Casting /he concep/uul s/~otlight:Hybrid comnpou17ding

347

These hypotheses are couched in cognitive linguistic theories which provide support for the relevance of these claims. Section 2 will discuss the main theoretical background for the above hypotheses of compound formation.

2. The cognitive base of head-frame internal specifier selection in nominal compounding3 "Framing is pervasive in language: [. . .] all linguistic units evoke a semantic frame" (Croft and Cruse 2004: 40) This statement is indicative of how the mind of a speaker constructs (and categorizes) tlie conceptual world. In contrast to truth-conditional and structural semantics, Charles Fillmore emphasizes that experientially-based associative links exist between concepts so that the activation of one concept will prime related concepts which can be metaphorized as constructing the semantic frame of the base concept (cf. 1982, 1985). For example, the conceptualization of RISK involves a variety of related sense elements constituting its semantic frame: chance (uncertainty about the result), potential harm, potential gain, actor, valued object (at risk), situation (context of the risk), deed (cause of the situation), purpose and motivation (of actor), and the beneficiary or victim (cf. Fillmore and Atkins 1992). It is a further feature of semantic frames that, in natural discourse, participants can focus on selected elements of tlie frame while in the background tlie whole semantic frame is conceptually primed. As concept~~al spaces of interrelated knowledge, semantic frames exhibit a culturally specific degree of conventionalization; however, semantic frames are not purely static entities consisting of a closed set of associations, but language users can construct semantic frames dynamically and online according to discourse context (Croft and Cruse 2004: 96). In a similar vein, Givon (2005: 70) describes semantic frames from the perspective of network theory when he states that: Consider the conceptual node 'house', which in our culture most com~nonly thus automatically activates tlie closely-linked nodes 'root', 'walls', 'floor', 'living room', 'bedroom(s)', 'bathroom(s)', 'kitchen', 'basement' etc. Presumably because of such automatic activation, when house is introduced into the discourse for the first time, each of its closely-linked nodes are automatically activated or 'primed', and are thus mentally accessible to the interlocutors even without explicit mention.

-

-

111order to account for context dependent priming of a word in less liabituated contexts, Givon postulates that tlie activation of the core or prototypical nodes of the word's semantic frame shift to neighboring less prototypical nodes. Such context-induced shifting of activation patterns is characteristic for tlie creation of metaphorical meaning, multiple senses, and for semantic change in general (2005: 7 1). Tlie understanding of tlie semantics of linguistic expressions in terms of frames is an essential tenet that has given rise to further developments in cognitive linguistics. Thus, aspects of se~nantic frame structure (e.g., frame-evoked roles, cf. Sweetser 1999: 13 1) appear in the notions of active zones and profiling in Langacker's influential theory of Cognitive Grammar (cf. 1999, 2008). Semantic frames also underlie the construction of mental spaces (Fauconnier 1985), wl~icliin turn influenced tlie creation of conceptual blending theory (Fauconnier and Turner 2002). The notion of selected, context dependent activation of meaning from the larger frame of semantic potentials is crystallized in Allwood's ~lnderstandingof meaning as contextually instantiated meaning potentials (2003: 29-66). Apart from the foundational contribution of semantic frame theory to cognitive linguistics at large, semantic frames also bear applied explanatory potential for the process of compound formation. By implication of previous research, the internal semantic structure of non-copulative nounnoun compounds appears ~~nderlyingly guided by a pro-jection of semantic information from the specifier to the head element.' This general relation between the compound constituents is reflected in tlie descriptive terms of specifier and head and the synonymous determinant and detern~inutuni(cf. Marchand 1969, Katamba 1993), or, L)eternlinans atid Deferniinrrtuin in German (cf. Eichinger 2000: 1 17). (Cognitive-)Fi~~ictionally speaking, the specifier modulates the referential range of tlie head element. As such, the internal projection of semantic information emphasizes the (conceptual) dependency of the specifier on the head (as expressed in postulates [a] and [b] in section 1). Since, compound internally, specification occurs in the semantic frame of the head, tlie general direction of meaning projection in noun-noun compounds puts the semantic frame of the head at the center of attention. Accordingly, meaning potentials (i.e., sub-frames) in tlie head-frame appear as possible candidates of specification. The frame of 'button', for example (cf. note 4), primes meaning potentials related to the type of garment (its superordinate fi~nctionalentity), to the position of tlie button (on the garment), and to the size, shape, material and colour of the button.' In the metaphorical compound belly button tlie specifier 'belly' indeed instantiates tlie positional

Cosling rhe conceptuul spotlight: Hybrid compounding

249

aspect in the semantic frame of 'button' and, thus, complements the visual analogy that inspires the metapliorical reference to a 'navel7. (cf. Benczes' analysis of belly button in n conceptual blending framework, this volume). A few more, randomly chosen German nominal compounds ( K U ~ V rrluschine 'coffee macliine', Fensterbrc>tt'window sill', Worlhilu'ung 'word formation', and Tischbein 'leg of a table') can serve as further examples to illustrate the pattern of head-frame dependent activation of tlie specifier. ('coffee') extends tlie PIJKPOSE of the machine as a primary T ~ L I Kujfee S, extension of its semantic frame. Fenster ('window') specifies the LOCATIONAL, FIINC I ION within the conceptual frame of Brett ('board'). In Wortbildung, tlie specifier describes the OBJLCT of the formation, and Tischbein is based on a metonymic relationsliip between specifier and head. In tliis case, the semantic frame of the head ('leg7) forms part of the larger entity expressed in the specitier 'l'iscli' ('table'). The overall meaning of the compound Tischbein emerges by alialogical metaphorical projection of the sliape and function of the human leg onto the supporting parts of a table. Such metapliorical and metonymic semantic qualities of compound constructions have more recently been explained by Reka Benczes (2006, and partly this volume) based on conceptual blending theory (Fauconnier and Turner 2002). While conceptual blending indeed provides a powerful framework that shows how individual semantic traits of the compound constituents (i.e., the input spaces) feed into the blended space (i.e., emergent meaning of the compound), two observations delimiting its application have to be made: First of all, conceptual blending explains the process of metapliorical projections between tlie compound elements that shape tlie overall emergent meaning of the construction. However, conceptual blending appears inactive in non-metaphorical compounds which are based on a straightforward interpretation of the primary sense relations of specifier and liead as guided by the semantic frame of the head. Secondly, in metaphorical compounds, conceptual blending operates on top of tlie frame based relation of head and specifier. As such, the instantiation of the specifier in the semantic frame of the liead appears as the underlying conceptual process of compound formation. To illustrate the basic application of head-frame internal specifier selection, it is necessary to reconsider the analysis of a few nominal compounds previously disc~~ssed in the framework of conceptual blending theory. In the compound gun wound, for example, Seana Coulson asserts that the first input might contain a kame in which 'gun' participates, and the second input might contain a frame for 'wound'. 1.. .] From the 'gun' frame,

the blend inherits information about the cause. From the 'wound' frame, the blend inherits information about the result. The two inputs share (among other things) a common associated event sequence, although they profile different aspects of it. In the resultant blend, the wound has been profiled just as it was in the input space for the head noun, 'wound'. (200 1 : 130-1 3 I)

This description of 'gun wound' alludes to the underlying semantic relation between the specifier 'gun' and the head noun 'wound' even if conceptual blending fails to highlight the contiguity of the specifier and the head and thus the potential cause for the creation of the compound. In detail, the explanation starts out with a given juxtaposition of two input frames ('gun' and 'wound') from which tlie elements are derived that build the meaning of tlie blend. This is essentially a cause-result chain happening on the background of an associated event sequence. Finally, the blend is characterized as inheriting the same profiling of 'wound' from its input space. The last observation is actually indicative of the fact that the head 'wound' retains its semantic frame (no-metapliorical projections) in the compound and is merely profiled according to a specific aspect inherent in the semantic frame of wound. Thus, a wound evokes the image of an inflicted area that is potentially dangerous for its bearer and is accompanied by sensations of pain and suffering. Since a wound symbolizes a disruptive, exceptional state, the possible C A U S I , % ~ this potentially dangerous disruption is primed in the semantic frame of tlie concept. Specification with 'gun' instantiates this meaning potential inherent in the head-frame. Quoting Hampton (1991) earlier in the same publication, Seana Coulson contrasts 'gun wound' with 'hand wound' in an attempt to show the difticulties to draw general conclusions on tlie meaning of the specifier in compounds: While 'hand' functions as the patient in 'hand wound', it serves as tlie instrument in 'hand repair' (2001: 126). Again, if the formation of the compound 'hand wound' is approached from the perspective of the head, the frame-induced activation of the specifier lies close at hand. Apart from the reason of the wound, the location of tlie woutid on the (human)body, wliich is implicitly related to tlie degree of danger, plays a further primary role in the semantic frame of .wound'. Among other body parts, 'hand' is thus a prime candidate of specification in the head-frame of 'wound'. The discussion of these examples emphasizes that the semantic frame of the head offers a basic conceptual map from which specifiers can emerge via the instantiation of inherently contiguous sub-frames or meaning potentials. By contrast, conceptual blending theoty seeins to only vaguely relate to this potential and its application appears to blur the semantic relations of

Ca.\tit.rg the ~.onceptuulpotl light: Hybrid conrpounding

25 1

lion-metaphorical nominal compounds. This observation leads to the question of how the conceptual relation between the specifier and the head is expressed in ~netaphoricalcompounds that involve blended sense projections. Fauconnier and Turner discuss a set of counterfactual compounds: cuffkine henduche, tnoncy problem, and rlicotine Jit (2002: 227-31). In a condensed form their analysis reads: These straightforward phrases - referring to a headache that comes from lack of coffee, a problem that involves lack of money, a fit brought on by a lack of nicotine [...I - all set up an integrat~oninvolving a counterfactual link between spaces. [...I In the linguistic construction shared by 'caffeine headache', 'money problem', and 'nicotine fit', the first noun picks out the element in the desired input whose absence in the blend is causal for the unwanted state, and the second noun picks out the bad state that obtains in one of tlic inputs and in the blend.

[...I For example, we could read 'caffeine headache' as referring to a headache caused by the caffeine. For both networks [factual and counterfactual], there is a cause-effect relationship in the blend - in the first case, between ahtencr of cuffelne and herrdachc, and in the second case, between presence o f cafleinc and heudache.

[...I The notion of ahsetice is not explicitly indicated by any part of the expression 'caffeine headache'. It emerges from tlie entire network, as prompted by this grammatical construction. Essentially, their description stresses the fact that there is a cause-effect relation between the compound elements and that the cause is projected counterfactually. At the same time, the head-frame 'headache' is held to be obtaining in one of the input spaces and in the blended space. This is reminiscent of Coulson's analysis of gull wound and implies along the same lines the central role of the head-frame in tlie compound. In analogy to wound, a headache, a problem, and a fit describe unwanted, bad states and, as such, prime tlie possible cause of this condition in their semantic frame. On top of this basic process of head-frame internal priming of the cause, tlie actual instantiation of the specifier is guided by a counterfactual projection. The understanding of the counterfactual mapping is dependent on encyclopaedic and experiential knowledge. Thus, the possession of 'money' prototypically evokes positive connotations so that in default contexts proble~nswill only arise for the lack of it. 'Caffeine' and 'nicotine' are known as two stimulant substances whose lack can cause unwanted physiological reactions. In these cases, however, the counterfactual map-

ping is less transparent since overuse of stimulants can lead to similar states so that tlie actual interpretation depends on the usage context. The actual counterfactual blending thus occurs between the activated meaning potentials in the head-frame and tlie instantiations of its contiguous causal sub-frame. As sucli, blending operates on top of the basic process of headframe internal specifier selection. To move a step deeper into the realm of metaphorical compounds, Fauconnier and Turner discuss the construction of con~pufervirus (2002: 27475) from tlie perspective of co~iceptualblending. They cogently show the cross-domain mapping of the basic metaphor 1 tCllNOLOGICAI, EVOISITION IS UIOI,OC;IC~\L EVOI,IJI ION as having developed from the initial blend of conipuler virus. The internal semantic relation between tlie compo~lnd components, however, is again steered by the semantic frame of the head. As a virus is conceptually tied to illness, the host and the effect of a virus appear as prominent roles in its semantic frame. In coniputer virus the role of the host is metaphorically instantiated by projecting tlie aspect of anirnacy inherent in the virus onto an inanimate host. Another interesting example is tlie cornpound land yachl, which is analyzed by Fauconnier and Turner in the following way: I-low do we go fiorn the linguistic unit to the conceptual elements or from the conceptual elements to the linguistic unit? In the case of nominal compounds, the formal unit names two elements in two different spaces, and directs the understander to find the rest. We will call these conceptual elements the ncrn~edelenlents. Consider 'land yacht' as a reference to a large, luxurious automobile. Clearly, 'land' and 'yacht' come from different domains: Yachts are associated with water as opposed to land. -Land yacht' gives us land from one space and yacht fiom another, and asks us to perfonn a mapping between these spaces. In this mapping, yacht corresponds to luxury car, land corresponds to water, driver corresponds to skipper, and the road for the car corresponds to the course for the boat. (2002: 356) This passage indicates that blending theory tends to gloss over the internal semantic relations between the named elements of the compound. In fact, the creation of compounds is described as a naming of elements from different spaces, thus creating the impression that the specifier and the head element are initially selected on an unrelated basis. As before, a closer focus on the semantic frames of the specifier and the head, however, emphasizes their semantic contiguity. Thus, the concept of 'yacht' is by default associatively connected to the fi~nctionaldomain of water. Cancelling this prototypical association immediately primes specification by the ex-

C'asting the cot1c.cq7tuc1l.~pofligh/: Hybrid coinpoundirig

253

ceptional functional domain of yacht. As such, the choice of 'land' as a specifier emerges from an antonymic mapping in the functional domain grounded in the semantic frame of yacht. Depending on tlie discourse intention and context, a speaker can use the antonymic mapping to construct the reference to a luxurious car.' This presupposes the projection of the relevant properties of the head ('luxurious, big vehicle') onto the exceptional environment ('land'), in which cars appear as prototypical vehicles. As Fauconnier and Turner demonstrate, this metaphorical mapping process can be explained in a conceptual blending framework. However, the discussions of the various compounds show that conceptual blending in metaphorical nominal compounds is based on underlying semantic relations between the head and its specifier, i.e., tlie specifier is grounded in tlie semantic frame of the head. Thus, a process of head-frame internal specifier selection appears as the basic characteristic of the formation of nounnoun compounds while co~iceptualblending accounts for cross-frame mappings in metaphorical/tnetonylnic compounds. Recognizing the conceptual grounding of the specifier in tlie semantic frame of the head noun evokes the question on which principles such associative links are formed. In his article on "17rame arid Contiguity" Koch provides insight into this issue ( 1999: 139-67). He defines conceptual relations in a semantic frame by rneans of contiguity. Thus, contiguity is defined as "a salient relation that exists between the elements (or sub-frames) of a conceptual frame or between the fratne as a whole and its elements" (1999: 154). As one of its instantiations, Koch exemplifies the influence of contiguity for the formation of English compounds, "in which the concept FKIIIT) and the concept designated by N I + NZare designated by N I (I,EMON contiguous (whereas the relation between the concept designated by N l (I'KEE) and the concept designated by N I + N2 (LEMON TREE)is that of taxonomic subordination" (1999: 158). To expand this analysis in terms of head-frame internal specifier selection, the head-frame TREE evokes contiguous associations to its gestalt, in particular to its leaves, branches, and fruits which play a salient conceptual role as distinctive characteristics among different types of trees. Specification according to this distinctive and, in the case of l.F,MON, also functionally salient characteristic is thus contiguously primed in the semantic fratne of I R E E . To emphasize the conceptual quality of contiguity and frames, Koch points out that it is "our knowledge of the world that determines contiguities" so that "contiguity has to be considered as constituting a conceptual, extralinguistic and not intralinguistic relationship" (1999: 145). This experiental, extralinguistic relationship is realized in language by guiding the

formation of compound nouns according to head-frame internal specifier selection as described in the previous analysis of compound nouns and more extensively so in section 4. Another essential cliaracteristic of contiguity is its dependency on prototypical salience of tlie frame-internal associative links. In other words, concepts sucli as 'tree', 'yacht', headache, 'fireplace', and so on evoke prototypical associations to salient components such as leaves, water, pain (as an unwanted state), and fire. Such prototypical associations apply in most contexts of conceptualization and are thus vital for tlie mental image of the concept. It is important to mention that prototypicality is not a static notion but varies dynamically according to discourse focus (cf. Geeraerts 2006: 2 7 4 7 , Givon 2005: 71 ). This means that frame internal contiguity can fluctuate according to discourse context. While unmarked, default contexts evoke prototypical associations proper, rare discourse contexts can Iiighliglit different conceptual aspects of the frame which become contiguous for tliis specific discourse context. If we consider the connection between usage frequency and prototypicalitys in head-frame internal specifier selection, the observation can be made that frequent onomasiological extensions of the same conceptual domain in the semantic frame of the head will be indicative of the degree of contiguity between this conceptual domain and the liead-frame. The semantic frame of tlie anglicism Team, for example, primes tlie members of a team as one of its contiguous subSrames. When it comes to the formation of hybrid compound nouns, the head Tearrz is indeed most frequently specified by reference to team members (cf. section 4). This emphasizes tlie dominant conceptual contiguity of the MEMBERS in the semantic frame of Team. To sum up, the discussion in tliis section has gathered and provided evidence that tlie word formational process of nominal compounding (as realized in English and German) is steered by contiguous conceptual relations in the semantic frame of the head. Thus, associatively linked concepts in the frame of the liead noun offer basic meaning potentials for specification in non-copulative co~npoundnouns. Based on these contiguous associations, the actual selection of the specifier is guided by context dependent associative strength, i.e., dynamic prototypicality. The emergent meaning of the compound is constructed following tliis basic principle of headframe internal specifier selection and can involve various degrees of metaphorical and metonymic mappings as demonstrated in the framework of conceptual blending theory. Taking tliis theoretical framework as a basic analytical tool, the remainder of tlie chapter will focus on hybrid com-

C'usfing the con~ept~rnl spotlight. Hybrid conipounding

255

pound nouns as a specific segment of nominal coinpo~~nds in German, which provide a fairly recent perspective on word formational productivity.

3. The concept of hybrid compounding: previous findings and preliminary results Hybrid colnpounds emerge as tlie result of language contact involving lexical borrowing in languages that use compounding as a word formational strategy such as English and German. Constructio~iallyspeaking, a hybrid compound can be defined as tlie combination of native and borrowed free morphemes that form a lexical and conceptual entity. Since German exhibits a predilection for nominal compounding, borrowings from English also participate in this process. As previous research has shown, liybrid compounds actually account for tlie largest number of anglicism types (70% of a total of 16,663 types) in the German newsmagazine Dcr Spiegel, year 2000 (cf. Onysko 2007). The analysis of hybrid compounds such as Erjblgs~fory('success story'), Borsericrash ('stock market crash'), Krisenwiun~rgement('crisis management'), PcxrtygrcVJe ('famous party person'), Conpufer-Muus ('computer mouse'), Internet-Nutzer ('Internet user'), and so on has established a few general observations of hybrid compounds in German (i) Hybrid compounds follow the main specifier and head schema of non-copulative conipounding in German. nouns occilr both in specifier (ii) Anglicisms in hybrid compo~~nd and liead position albeit there is a general tendency for individual anglicisms to prefer the specifier position (e.g., Internet). A functional switch between specifier and head is accompanied by a change in the co~icept~~al quality of the anglicism. While in Partyk~rrizler ('party chancellor') tlie specifier pro-jects the qualities associated with a party ('fun', 'happiness', 'carelessness') onto the agentive liead Kunzler, in Hau,spart,v ('house party') the anglicism is conceptualized as an event which is specified by its location. (iii) There is a correlation between the overall use of anglicisms as single words and their occurrence in hybrid compound nouns. This is in line with general observations in language contact scenari \vIiicli show that as part of their integration process, bor-

r o ings ~ assume lexically productive functions in the receptor language (cf., among others, Thomason 200 1 ) The last observation also raises another issue which relates to the understanding of productivity of a borrowed term in the receptor language. Apart from the copying of form and meaning pairs from a source language (SL) to a receptor language (RL), language contact can also remain on the conceptual level, stimulating the expression of an SL-concept by RL-means. This is the case, for example, in calques which represent conceptual influence in the RL by various degrees of translational equivalence from fairly literal to free renderings (e.g., ~ I c y ~ ~ c r ~in l p eEnglish r as tlie conceptuallexical stimulus for the creation of Wolkenkratzer 'cloud scraper' in German). According to this process, some hybrid compounds could represent partial translations from an English model (e.g., ,Ypitzenmanuger from 'Top-Manager', Hqjencity from 'port city', and FuJbullfun from 'soccer fan'). 'These examples might inspire the conclusion that the anglicism has actually not been used creatively in German but appears as a remainder of a partially borrowed compound expression from English. This assumption raises the question of why the original compound has not been fillly translated into German. In fact, the answer higliliglits the productive potential of the anglicism element in the hybrid construction despite the existence of a partially translated compound model in English. Thus, the English term in potentially modelled hybrid compounds most often represents an established borrowing in tlie German lexicon and, cruciallq, it also frequently appears in other compounds which are not directly related to an English model (e.g., Kunsf-City 'art city', Slrommunuger 'electricity manager', Milch.stru/jen-Munuger 'milky way manager', Mcirchenfun 'fairy tale fan', Wulflei.tchj~m'whale meat fan', and many others). This underlines that these anglicisms basically function as productive elements in German regardless of individual possible proximity to an English model. Furthermore, translation inherently designates a productive process in the RL since tlie realization of a concept by lexical means of the RL is essentially a process of linguistic appropriation of form-meaning units that build their separate conceptual and formal domains in the RL. As such, whether by partial translation or by free combination, hybrid compound formation in German is based on the language intrinsic schema of combining elements into a conceptual whole. Since there is evidence that the basic principle of internal semantic projection from specifier to head seems to apply similarly in English and German, translational equivalence does not

C'ustlng t/?cc onccptual spotlight: Hybrid compounding

25 7

interfere with the formation of compounds in German. Accordingly, assessing the level of producti\ ity of hybrid compounds in German by degrees of productivity (from literal translation to free combination) is irrelevant for tlie analysis of the conceptual relation between specifier and head element. In fact, nominal hybrid compounds are particularly interesting for investigating the underlying conceptual relations between tlie head and the specifier for several reasons. First of all, liybrid compounds are mostly recent creations wliich allow an analysis of tlie process of compounding from a current perspective. They are less likely to be lexicalized and stored as whole elements in a speaker's mental lexicon (cf. note 9). Secondly, borrowings tend to bear a more restricted semantic scope in the receptor language as compared to long-extant native elements. In terms of analyzing compound nouns, the semantic frames of anglicisms as nominal heads thus provide a clear view on possible associative contiguities that account for specitier selection. For this reason, tlie approach of head-Game internal specifier selection is tested on a sample of hybrid compound nouns extracted from Der Spiegel 2000 (cf. Onysko 2007). The data consists of hybrid compounds formed by the one hundred most frequent anglicisms as head elements. The frequency of the anglicisms in head position is determined by type frequency of hybrid compounds. Selecting the most productive anglicisms in hybrid compounds is necessary in order to acquire a large enough pool of hybrid compounds per anglicism head as a basis for establishing contiguous conceptual domains in the head-frame. Appendix 1 provides the complete list of tlie one hundred most frequent anglicisms as heads in hybrid compound nouns. For the purpose of this study, the five most frequent and the five least frequent anglicism heads from the list in the appendix are analyzed in detail and categorized according to their conceptual implications. These investigations are complemented by an analysis of ten randomly selected hybrid compounds whose anglicisms are only used once as a head of hybrid compounds in the corpus. The following table provides an overview of the ~iiostand least productive anglicism heads, whicli will be analyzed in detail in section 4.

25 8

Alexander Onysko

Talde I. The five most and the five least frequent anglicisms as heads in hybrid compound nouns among the one hundred most frequent in the corpus (cf. Appendix 1 for the complete list)

I

I Total type fre-

I Manager

quency 1211

Rank Anglicism head 1

5 96 97 98 99

1

Show Kids Bestseller Sewer Power

I Total token fre-

101

177

7

9

16

111 11

6 6

I

quency 623

10

Table 1 shows the overall filtered type frequency, i.e., rnorphological and orthographical variants of the same cornpound are counted as tokens of the same type (e.g., Expcrtenteunl and Eq3erten-Tcums are two tokens of the same type). In Appendix 1, type frequencies represent unfiltered data, which means that variants are counted as separate types (according to Wordsmith Tools). These raw figures have served as the basis for data selection. In general, type frecluency represents the number of different hybrid compounds with the same anglicism head (e.g., C;e.rl~en.stersho~~, 'ghost show', Li/erutuushow>,'literature show', and Musikshow 'music show') while the overall token frequency is calculated by the number of occurrences of each individual type in the corpus (e.g., the type Fern,teh,show 'TV-Show' occurs eleven times in the Deu Spiegel 2000).~Finally, type and token frequencies are applied in the analysis as indicators of contiguity strength (i.e., prototypicality of frame internal associative links) for tlie five most frequent anglicism heads.

4. Head-frame internal specifier selection: evidence from hybrid compounds In the following analysis of the five most productive anglicisms, the five least productive anglicisms, and ten sporadic creations, particular focus is

Custitzg the conceplzrul spotlight: Hybrid conipounding

259

attributed to tlie contiguous meaning components in the head-frame which are realized by various clusters of specifiers in tlie liybrid compounds. Type and token frequencies of the specifiers are taken as an indication of the most prototypical semantic contiguities of specification. In case of potential ambiguity, the meaning of a hybrid compound was disambiguated in tlie usage context of Der Sl?iegel articles. The complete data of hybrid compounds with the five most frequent anglicism heads are provided in tables and graphs visualizing frame internal contiguities in the appendix.

4.1. Highly productive anglicism heads and their specifiers in liybrid compounds According to Table I, the five most productive anglicisms as heads in hybrid compounds are Manager, Film, Test, Team, and Show. Their productive patterns of liybrid compound formation will be concisely demonstrated below drawing from the pool of their individual compounds in the corpus. Murruger: The anglicism appears as a head in 2 1 1 different types of liybrid compound nouns. The denotative core of its semantic frame refers to a person who is in charge of a structure (social, institutional, material) to be controlled and managed. Tliis function of a manager implies a certain complexity and inhomogeneity of the governed structure whose performance is improved under tlie guidance of a manager. The connotative base of the frame evokes associations to business (capitalism, profit), social status, professionalism, wealth, power, and emphasizes tlie modern nature of the profession. These basic meaning conlponents of manager set up a semantic franie that implies the institution as tlie action space of the manager, the object of the managing activity, and, in line with its agentive reference, the qualities and characteristics of a manager in exercising Iier/liis profession. An analysis of the hybrid compounds indeed crystallizes these meaning potentials as the essential contiguous associations in the semantic frame of manager since, apart from two examples, all the specifiers in the corpus revolve around these three aspects of its semantic frame. In sum, the following distribution is observed: (a) Most of tlie types (13 1) and tokens (346) specify the AFFI1,IA7 r ~ o ~ /of othe ~manager, ~ ~ is., ~ the ~ specifier ~ ~ ~determines o ~ the institutional affiliation and the type of occupation of the manager. Tliis function particularly involves proper nouns as

names of companies and institutions (overall, 99 types and 261 tokens, e.g., Elf-Munuger, Te/ekon~-Manager,Kirch-Manager, and T/iy.s.ten-Munuger).I0 (b) Specification by the S U I ~ I L CofI the managing appears as the second most frequent process (68 types and 2 12 tokens, e.g., Fondsmanager, Krisenn~unugcr, Finunzrnunuger, and Projebnianuger) . (c) Specification relating to the AGLNTIVL QlJALITICS of a manager is the least frequent regularly employed frame internal contiguity ( 1 0 types and 6 1 tokens, e.g., Spitzcnmunager, Erfilg.snzanuger, arid Zupuck-Manager). The only two examples that do not immediately emerge from these inlierent semantic components are Polit-Manager and Ternliniiiunuger. The latter is based on a nietonymic projection from AGENr (animate) to OHJECI (inanimate). Thus, the agentive fi~nctionof a manager (as scheduling meetings and events) is mapped onto an object that is designed to record ones appointments. This metonymic prolection emerges from the close associative bond between a manager and herlhis prototypical concerns and activities (holding and participating in business meetings); so, the sense extension relies on the specifier expressing the obvious in the se~iianticframe of the head which facilitates the projection of the prototypical function from an agentive to an objective referent. At the same time, the instrumental reference of Terminr~ianugerfeeds back into the basic agentive frame of 'manager' as a typical instrument of managers. Polit-Manager (3 tokens) appears as a near copulative compound that describes a manager cirni politician viz. a politician with managerial attitude? and qualities. The emerging meaning describes the persona of a politician and manager evoking connotative associations to being modern, dynam ic, and economically oriented: ( I ) lm Zeitalter von lnternet und Globalisierung gehe es, so bekennt ein ehemaliges Kabinettsmitglied der Union, nicht um den Wettstreit der wirtschaftspolitischen Konzepte der sei im Sinne eines pragmatischen Reformkurses entschieden. Entscheidend sei, welche Partei die iiberzeugenderen Polit-Manager prasentiere. (Der Spicgel2000: 913 1 ) [As a former cabinet member of the Union [Christian Democratic Union] admits, in the age of the Internet and globalization. the competition for concepts of economic policies is not important. 'This is decided along a general line of pragmatic refor~iis.The decisive fact is which party is able to present the more convincing political managers.] -

An essential structural feature of the compound is the clipping of the suffix(es) -ik/-er (Politik/er 'politician/politics') which creates a certain vagueness of the first element in the compound and leads to the activation of the nested input frames of politics and politician. As a result, the compound evokes the notion of a manager as a politician and of a manager of politics. Film: The semantic fralne of Filni extends from a variety of denotative and connotative core meanings of the term. The following denotative components are active: -

film as medium (audio-visual component); unreal but projection of reality - tiction, tied to a display medium (TV, movie theatre, Internet) film as material substance (film strip; thin layer of substance) - filni as story (characteristic themes and topics of films) - film as genre (genres are abstracted from recurrent usage of labels relating to topics, audiovisual components, and audience reactions) A few Inore vital semantic traits tint the meaning of Film with a connotational hue: film as action passively consumed - film between entertainment and infor~nation film as a bounded work of art According to these semantic fibres, the basic concept creates associations to a bundle of contiguous meaning potentials. The conceptualization of film as artwork primes its creative source, its audience and its artistic valne/quality. The aspect of film as story promotes associations to its topic, setting, personae, and audience. Film in its concrete, material sense (film strip) cognitively highlights its gestalt (shape and constitution) and can metonymically project its property of being a thin layer onto other materials (see below). The general audience-centric functions of films from entertainment to information imply purpose, quality, and temporal boundedness, and the aspect of film as ~nediurnanticipates its embedding in an organ of projection and broadcasting. As regards the corpus, these contiguous meaning potentials in the semantic fralne of 'film' indeed account for the semantic relation between specifier and head in 145 out of 147 types of hybrid compounds (499 out of 50 1 tokens). The two not irn~nediatelycontiguoils hybrid colnpounds are

262

Alexander Onvsko

due to metaphorical projection explainable in the framework of conceptual blending theory: Aniei~enjies.serfi11~1 ('anteater film') and Masalu-Film. The emergent meaning of the iriput spaces of 'film' and 'anteater' is a movie of very little financial success so that one would have to produce many such films in order to gain sufficient profit. Since the blend is built on an associatively distant space of a movie as a silbstance to live off, i.e., a director trying to earn one's living by producing movies and ants that only in bulks provide adequate amount of food, the blending is made explicit to the reader in the context of the Spiegel-interview: (2) Das sind fur rnich alles Aineisenfresserfilme. Katzenberg bezeichnet je-

den Film, der nicht wahnsinnig vie1 Geld verspricht, als Ameisenfresserfilm weil Ameisenfresser sich nun ~nalsehr miihsam ernahren miissen, von sehr kleinen Portionen. (Der Spiegel2000: 291198) [For me, these are all anteater movies. Katzenberg uses the term anteater movie for every film that does not promise large financial success - because anteaters have to strive for their food and feed on small portions] -

The blending Mcrsulu-Film draws on the cultural specific reference of Masulu (a blend of spices used for Indian cooking or for the preparation of tea) to establish the image of a kitschy Bollywood movie. As such, the specifier is a metaphorical extension of the underlying contiguous association to the quality of a movie. As far as tlie bulk of hybrid compounds built on the liead-frame of 'film' is concerned, the contiguous meaning potentials grounded in the prinlary denotative and connotative traits of the concept account for a spread distribution of specifiers. In total 1 1 frame bound meaning potentials are realized in the hybrid compounds. They are briefly discussed below in the order of type over token frequency: TOPICof a film (34 types, 87 tokens) appears on a general level as tlie most productive association, i.e., a film is most likely to be specified by its I'OPIC/TI~E,ME. In the corpus, topic particularly relates to events and actions (e.g., Antikriegsfilni 'anti war film', Holoctr~~stfiln~ 'holocaust film', Inze~t.filni 'incest film', and kirtu,rtropher~filr?i'catastrophies film') and some of the topics have acquired the quality of genre terms (e.g., Dokurnentarfilnz 'documentary film', Krinzinalfilrn 'crime film', and Heirnutfilm 'homeland movie - sentimental film in a regional setting'). Associatively closely related to LOPICare the PE,KSONAC (14 types, 20 tokens) and the SETTING (3 types, 4 tokens) of a tilm: Fr~mkensteinfilni, Terroristenfilm 'terrorist movie', and Van~pirfiln~ 'vampire movie' (specification by PERSONAE) and

('ustirzg the conceptzlul spollrght: ffyhrld compozmding

263

BerKfilnl 'mountain tilm', tfaup.studCfilrn 'capital city film', and Wasser,film 'water movie' (specification by SE~'I'ING). Another frequent pattern of specification relates to the criE.4TIVe SOIIRC:F: of the filrn (director/producer). In this case, determinants as proper nouns (25 types, 29 tokens, e.g., Diirrie-Filn~,Dresen-Filrn, and FelliniFilm) by far outnumber instances of categorial nominal specification with only two examples in the corpus: h'achwuchsfilnz 'offspring movie' and Stu&ntenJilm 'student movie'. The PURPOSE/FUNC I'ION of a filrn appears as a further productive pattern of specification (19 types, 121 tokens). This contigi~ityarises from tlie inherent meaning of a film as serving entertaining and/or informative functions. Thus, a film takes the form of an instrument that is capable of influencing tlie audience's state of mind. This fi~nctionis spelled out in a couple of hybrid compounds that qualify tlie manipulative effect of a film on its audience, e.g., Aufklarung.~filrr~'educational film', Fr~temi~sierungsfilm 'fraternization movie', Propagrmdujilnl 'propaganda film', Werhejiln? 'advertising film', and Unterhaltung.c.Jilm 'entertainment movie'. In addition, the I'IJRPOSE/FUNCTION of a movie can relate to general purposes and (so'closing cial) events, as in Eri$fnungsfiln~ 'opening movie', AbschluLsL~.Jilnz movie', ('unnes-Film, Dehiitfilm 'debut movie', Pilotfiln~'pilot movie', 'competition movie'. The high token frequency in the and Weftbe~~erh,sfiln determining domain of P ~ J R P O S E / F I J N C ~ I O Nis due to a particularly common hybrid conlpound noun Spieljilt11('feature film', 65 tokens) that functions as a hypernym referring to entertainment movies in general. Further contiguous clusters of specification describe tlie QUALITY of a film (14 types, 30 tokens) in relation to tlie production and tlie response of the audience, e.g., for production, Billigfiln~'cheap movie', Montag.~jlnz 'Monday movie', and Qualit~tsfilni 'quality movie', and, for response, E~fi,lgsJ;lni 'successful movie', Gzite-Laune-Filnz 'good mood movie', Liehling.giln~'favorite movie', and Skunublfilnz 'scandalous movie'. Specification by the MATERIAL C~MI'ONI:NT of film creates hybrids such as fitofiltrl 'camera film', Kleinhildf~lwl 'miniature film', and Schn~alJilrn 'substandard (in size) film'. By metonymic extension of its prototypical shape as a thin layer, tlie head Filni is specified by other materials, creating the image of a thin layer of the determinant material, e.g., Eiwe$J;lnz 'protein fil~ii'and 6@lrr1 'oil film'. Finally, the head Filnz is specified by the following less productive frame internal contiguities: -

MEDIUM OF BROADCAS I'ING (6 types, 1 19 tokens, e.g., Fernsehfiln~'TV film, Kinc!filni 'cinema movie')

-

-

-

27 tokens, e.g., Schwars~lei/ljilm'black and uhite film', Stumnifilm ', Zezchen/rickfiIr~i 'animated tilm') TEMPORAL B O ~ ~ N D L ' I ) N55 I (5 types, 15 tokens, e.g., DreiSfunclen-Filrri 'three hours movie', Fort~etzung~sfiIn7'sequel movie', KurzJilrn 'sl~orttilm') Target A U D l r N C E ( 5 types, 13 tokens, e.g., Fanzilienfilni 'family movie', Fruuenfiln~'film for women', Jugendfilm 'youth film') A r J v 1 o - v r s u i z l I F A I ~ I I (7 ~ ~ types, S

Despite the various conceptual domains that account for the selection of specifiers with the head Filnz, an altogether unified picture of specification emerges since the individual domains arise within tlie semantic frame of the anglicism head. In other words, the specification is guided by contiguous associations to primary denotative and connotative meaning components of the term. The data indicate that 'IYIPlC' and CRE.ATIV1: SOURCE (directorlproducer) are particularly salient for specifier selection while P U R I ' O S E / F I J N C ~ ' I ~ N , QUAI,ITY, MIZTEKIAL COMI'ONENT, MEDlllM O F 13KOADCASTIN(;, AUDIO-VISUAL FEATIIKCS, TEMPORAL, ROCJNI>EI>NESS, and ,AIJDIL;NCI: account for the rest of head-frame internal specifiers in the hy-

brid compounds. Teun~: The anglicism Team occurs as a head in 104 types of hybrid compounds (281 tokens). Its semantic frame is construable upon the denotative tneaning components of 'a group of people, typically professionals, that interact cooperatively to achieve a common purposelgoal' and the connotative aspects of 'cooperation', 'beneficial social interaction', 'modernity', and 'common activity'. As an anglicism, tlie term came originally into use with reference to sports teams but has extended its meaning today to the field of business and it can be applied more generally to instances when a group of professionals interacts cooperatively to achieve a goal. As far as the data of hybrid compounds is concerned, specification of the head element Team follows its contiguous meaning potentials arising from its core traits of meaning. Most prominently, a team is specified by its MLMBLRS (40 types, 95 tokens) as in An7uteurtean1, A r z t e f e ~ ~'doctors' i team', Astronuutenteanl 'astronai~ts' team', Beraterteanz 'advisors' team', and Betreuertearn 'coach's team'. Partly, this type of specification is individualized to one specific member, who acts as the leader or tlie salient figure of a team. In this case proper nouns occur as determinants, e.g., Alex-Teanl, Fetz-Team, and Goddio-Teann. Closely related to members, specification can also em-

Ccrstini,. the conceptual spotlight: Hybrid compoz~nding 265

phasize the LlIAII4CTLRIST1(' Q U A I 1TY of the members and, by metonymy, of the whole team (4 types, 4 tokens), as in Innovation.steom 'innovation team', Kartlpferteam 'fighter team', Trauwlteam 'dream team', and Spitzenteam 'top team'. of the team emerges as anNext to MEMBERS, the DISCll'l INE,/CONCI;KN other contiguous component triggering specifier selection in 35 instances ( 1 34 tokens). Part of these hybrids denote sports disciplines and events as 'national league team', FuJjhallteam 'soccer team', and in E~~sfligu-Team Tricrthlontean~'triathlon team'. Others refer to professional concerns such as Au.s,stellung,ctearri 'ex11 i bit ion team', Drehtean~ 'film shooting team', Fern.selzteam 'TV team', and K~wlerateam'camera team'. Again, proper nouns can function as specifiers in both domains, e.g., DFB-Team 'German national soccer team', RTL-Team 'German television station team', and SPD-Teurr~'social democratic party team'. The remaining specifiers in the corpus address the implied I'UIiPOCE/(;OALof a team (1 6 types, 35 tokens, e.g., Entwicklungsteam 'development team', Expeditionsfeuni 'expedition team', and Fo~schungsteam 'research team'), the SOCIAJ s PA I IJS of tlie team (7 types, 2 1 tokens, e.g., Hoch.scl~ultean~'university team', Konkurrei~zteam 'competitor team', Leitmngsteani 'team that is in charge', and Ldnderteanl 'national tearn'), and the SIZE of the team (2 types, 2 tokens, Runlpfteanz 'shrink team' and Vierer-Team 'team of four people'). In sum, MLMBEK and D~SC~PL~NE/CONCCRN appear as the most salient domains of spccit'ication since they are immanent elements of a basic conceplualization of 'team'. The frame internal contiguous meaning components outlined above comprise all the hybrid compounds of Team in tlie corpus. Test: Test functions as a head in 1 15 types of hybrid compound nouns (254 tokens). Its denotative range involves the semantic traits of 'examination', 'a~ialysis', 'measurement', and -trial' (not the real thing). On the connotative side, a test evokes a range of emotions on a cline from uncertainty to reassurance and tends to be percei\led as a stressful situation. Furthermore, a test owns a scientific flavour and is frequently held in a formal setting. As regards the data, the majority of specifiers determine the 51JBIkCl of the test as entity or activity (45 types, 116 tokens, e.g., A b g a ~ t e ~'exhaust t gases test', Ahwehrtest 'defense test', Aton~homhente.st'nuclear bomb test', and Bugelte,st 'ironing test'). Besides the crucial contiguous aspect of the sub.ject of testing (what is tested), a test also implies a result. Thus, not

266

Alexander Onysko

surprisingly, the head noun Test is also frequently specified according to 58 tokens). Specification by RESULT involves general concepts referring to achievements and mentallphysical states and phenomena as in Abschlusstest 'final exam', Aujkahmetest 'entrance exam', Ausdauertest 'endurance test', Belastungstest 'fitness test', and Eignungstest 'screening test'. In addition, the result can also be expressed in the determinant with reference to substances tested for as evident in Aidstest (note the metonymy of RESULT FOR CAUSE, i.e., sickness for virus), Drogentest 'drug test', Hepatitis-Test, and Kokaintest 'cocaine test'. In this sense, the unwanted result from the perspective of the person tested is determined in the specifier. Further determinants arising from contiguous sense relations in the head-frame of Test relate to test DESIGN~METHOD(21 types, 29 tokens, e.g., EEG-Test, Inparottest 'infrared test', Kaltekammertest 'cooling chamber test', Liigendetektortest 'lie detector test') and to CHARACTERISTIC QUALITY of a test such as duration, frequency, and significance (1 1 types, 30 tokens, e.g., Alltagstest 'every day test', Dauertest 'continuous test', Friihtest 'early test', Kurztest 'short test'). This leaves 5 types of hybrid compounds in need of further explanation as they involve sense projections that build upon but go beyond basic head-frame internal contiguities: Idiotentest 'idiot test', Hartetest 'resistance test', Holland-Test 'Netherlands test', Lackmustest 'litmus test', and Realtest 'real test'. Out of context, the combination of real and test seems contradictory since test and reality display antonymic traits. In the original article, however, Realtest refers to the role of a political party that has made promises during the election campaign to effectuate certain policies but then struggles to actually implement them in the test situation of the political reality of governance. RESULT of the test (33 types,

(3) Dabei ist der linke Kietz-Vorsteher Profi genug, um die Zwange der Realpolitik zu kennen. [.. .] Doch der Glaube, es besser zu konnen als die

Altparteien, envies sich im Realtest als grandiose Selbstiiberschatzung. (Der Spiegel2000: 1 1141) [In fact, the leftwing Kietz-director has sufficient professional experience to know the restrictions of realpolitik. . .. However, in the test situation of real politics, the belief of being able to do better than the old parties turned out to be a huge overestimation of one's own capabilities.] The excerpt underlines that the hybrid compound Realtest is linked with Realpolitik, which is introduced earlier in the same passage. Thus, the hy-

Costing the conceptual spotlight: Hybrid conlporrnu'ing

267

brid appears as a clipped version of *Realpolitikte,st and, conceptually speaking, the determinant describes the SE'ITIN(; of tlie test. The example of Holland-Tesf is based on a metonymy ofthe type country name for national team. As such, the compound denotes a test soccer game between Germany and the Netherlands while compound internally tlie specifier determines tlie criicial PAR'TICIPANT in tlie test. Similarly, Idiolentest involves a metonymic personification of a psychological condition by derogatorily labelling people tliat suffer from mental deficiencies. In context, tlie hybrid compound stands for a psychological test for traffic delinquents to regain their driver's licence. Compound internally, the specifier realizes the contiguous sense relation of the (negative) RESULT of the test (cf. above). The hybrid Hurtetesl is another example of head-frame internal realization of tlie RESIJ1,T in tlie specifier. In this case, the compound is used in an abstract sense derived from the pro.jectioli of concrete, material reference onto abstract psychological resistance. Finally, Lacknzustest is an example of a compound tliat has shifted from its original meaning so that tlie compound internal sense relation of head and specifier are no loriger transparent and relevant for its meaning. Originally, Lucktnuslest referred to a specific chemical test which ascertains the ph-value of a substance under the addition of the colour substance LackThis shows that tlie conipound originated from the contiguous sense WIUS. relation between the head-frame Test and its specific DESIGN/MEI'HOD (substance used for testing). In general usage, the technical meaning of the compound has shifted to denote a situation wliicli demands a clear, usually eitherlor decision by the people involved. This seems conceptually related to the German nietaphorical idiom Farbe hekennen, 'to clearly decide for one option, i.e., to show one's colors'. The following two excerpts from Der Spiegel illustrate this common meaning of the hybrid compound: (4) ,,Putin steht vor einem Lackmustest", sagt die U S Expertin Amy Smithson. ,.Entweder er entscheidet sich fir die alte Garde [...I oder fur die Wissenschaftler. die ihre Kenntnisse friedlichen Zwecken zufuhren wollen.'. (Der Spiegel2000: 4 11192) ["Putin sees himself before a litmus test", says US expert Amy Smithson. "Either he decides for the old establishment or for the scientists, who would like to use their knowledge for peaceful reasons."] (5) Der Lackmustest, der zweifelsfrei Wesentliches von Unwesentlichem scheiden konnte, ist noch nicht erfunden. (Dev Spiegel2000: 49124)

[The litmus test that can clearly distinguish between the essential and the unessential has not been detected yet.]

In general, the analysis of hybrid compounds with the anglicism head Test show, as in the previous cases, that the specifiers arise from contiguous sense relations within the basic semantic frame of the head, which is construed by implication of its basic meaning components. This also holds true for the individual cases which inkolve figurative pro-jections that can change the overall meaning of the compound and, from the perspective of current use, veil internal semantic relations of the compound constituents. As such, figurative processes in compounds operate on top of basic headframe internal contiguous sense relations. ,5'h0~?: The anglicism ,5'how occurs as a head in 101 types of hybrid compound nouns in the corpus (177 tokens). While its denotative scope can essentially be captured by the notion of a 'staged performance/exhibition/event', show radiates a rich set of connotative associations among which the following seem prevalent: 'entertainment', 'passive reception and intangibility from outside' (both aspects foregrounding the audience), 'remoteness from reality' (sets up an own bounded world), and the creation of 'tension, surprise and inlpressiveness'. Along the same lines as in the previous analyses, the formation of hybrid compound nouns with the head Show is generally guided by contiguous associations couched in the semantic frame of the head. One of the primary contiguous meaning components relates to the question of what the show is about, that is, its topic or theme. In the hybrid E show is indeed the most compounds, specification by T O ~ ~ I C ~ H EofMthe prevalent process (32 types, 48 tokens) as in Auto-S'how 'car show', Barsenshow 'stock market show', Erolikshow 'erotism show', Tanzshow 'dancing show', and Wi,s.sensshow 'knowledge (quiz) show'. Further conceptual contiguities of show that surface frequently in specifiers relate to SCIIGI:NINC; (19 types, 56 tokens, e.g., Ahendshow 'evening show', Fernsehshow 'TV show', and Sonnt~rgsshow'Sunday show'), SAISENT ACTIVI.~Y/QUALITY/M/\.I'EUIAL ELEMENT' (22 types, 3 1 tokens, e.g., Ausziehshow 'strip show', Nebel.~ho~) 'fog show', and Verleumdungsshow 'slander show'), arid to PAK7'1CIPANl'S (19 types, 25 tokens, e.g., Alligulor-Show, Expertenshow 'expert show', Haraid-Schnliu'l-Show). While these conceptual contiguities of show feature equally prominent in the corpus (and thus allude to their equal contiguity strength in the frame of show), associations related to the AUDIENCE of a show only account for a few determinants: Frauenshow 'women's show', Mus.sen.sho~~ 'show for the masses', ,Spunnerslzow 'voyeur show', and Vojieursshow.

C'asting the conceptual spotlighl: Hybrid compounding

269

To complete the analysis of the hybrid compounds with the head Show, five forms remain which involve metonymic and metaphorical projections from underlying contiguous sense relations: Briillshow 'howling show', FlieJhuwd-Show 'conveyor belt show', KaJ;g,show 'cage show', Testo.sterorz-Show, and Unterleib,~shw'womb show'. The latter two are examples of metonymies that expand the contiguities of Show in terms of its PARTICIPANI'S and AUDIENCE. In Testosferon-Show the prototypical male hormone stands for a male target audience and tlie major participants of the sliow which indulge in typical "manly" activities. Unterleihsshow is based on a (body) PART FOR WHOLE metonymy. The body part determines the salient element of the Show, and, by implication of its spectatorship, tlie meaning of 'peep sliow' emerges from the hybrid compound. Rather than metaphorical imagery, Briillslzow involves a double entendre which, on the one hand, refers to the PAR'I'ICIPAN'I'S of a talk show vociferously accusing each other and, on the other hand, to the AUDIENCE who, ideally, burst out into similarly loud laughter. The metapliorical reference of Kiifigshow and Fliepb'band-Show can be construed in a single scope conceptual blending network (cf. Fauconnier and Turner 2002: 126-1 3 1 ). The input frame of Kufig specifies the basic SEI'TING of the show as an extension of its contiguous sense relation in the head-frame. On tlie level of generic space, both elements of tlie compound foreground the existence of an audience as observing animals in a cage and as being spectators of a show. While the input frame of Show represents the organizing frame of tlie blending, the emergent meaning in the blending draws from both input spaces. It is based on the common implicational sense of 'audience', on the head-frame sense of 'a staged performance for entertainment reasons', and on the sense of 'enclosure and captivity' from the determinant input (Kufig). This inspires the emergent blended meaning of .a sliow that is characterized by observing its captivated participants'. The tight compression of semantic information from the two input frames, however, creates a certain semantic vagueness that allows room for referential indeterminacy. Therefore, tlie actual reference of the metaphorical hybrid compound needs to be (and actually is) determined in the usage context: weniger junge Zuschauer zwischen 14 und 49 Jahren als noch zu Anfang wollen der Kafigshow .,Big Brother" [...I, beiwohnen. (Der

(6) [...I

-

Spiegel2000: 5 1/ 104) [... - fewer young viewers between fourteen and forty-nine than at the beginning still want to \catch the cage-show "Big Brother". . .]

The immediate association between the hybrid compound and tlie name of tlie TV show is a token for the creational dependence of Kaf;ggshowon the concept of "Big Brother", which provides tlie vital referential grounding of the blend. The constituting frame of the blend (Sl~ow)emerges from the basic classification of "Big Brother" as a type of reality TV show while tlie determinant input Kujig relates to the participants' choice to live as "captives" in an artificial community. In addition, the image of tlie cage feeds allusions to a zoo into tlie blend highlighting the (anonymous) safety of tlie observers and the disparity between human observers and dependent animals. This dependence is materialized in the TV show with the role of the audience in deciding the fate of the inmates (who is allowed to stay, who has to leave). In contrast to Kdfigsl?ow, the metaphorical projection in FlieflbundS h o ~ iis based on a more direct mapping between the two input frames which arises from the prototypical function of the determinant ('conveyor belt'). This is underlined by the implication that other conceptual associations of FlieJband (e.g., typical shape and location of use) are highly incompatible with the semantic frame of the head. Tlius, the emergent meaning in the blend is not dependent on a contextual referent but draws from the functional aspects of automatized production. This image involves a conceptual clash with the head-frame Show which implies the semantic components of 'entertainment, diversity, ingenuity. and wit'. In the blended space, the features of a machine-like, non no to no us production prevail characterizing the meaning of tlie hybrid co~npoundas 'a show that is a reproduction of similar shows, rapidly produced for quick profit, and generally lacking ingenuity and artistic wit'. In terms of underlying frameinternal contiguities, the determinant specifies the head-bound sense relation of tlie PRonucTloN of a show. The discussion of the hybrid cornpounds involving Show as a head element reverberates the common pattern of head-frame internal specifier selection as the basic productive criterion for the formation of (hybrid) co~npoundsin Gertuan. Even the comparably rare cases of ~netaphorical compounds that allow for a description along the lines of a single scope conceptual blending network are underlyingly framed in the contiguous sense relations of the head. In ~netaphorical compounds, however, the emergent meaning of the blended input spaces creates rich semantic imagery reaching beyond basic head-frame internal contiguities. Before testing these observations on instances of less productive patterns of hybrid compound nouns, another observation has to be made, which will be picked up in more detail in the concluding section of tlie chapter. This is

('usling the concep!ztul spollrghf. 11)brid compounding

27 1

the notion of graded fuzziness of contiguous sense relations in headframes. In other words, contiguous sense relations do not appear as clear cut categories within the franie of the head but allow for overlap between contiguities that can be cognitively metaphorized in the sense of Giv6n's depiction of graded membership and fi~zzyboundaries of prototypes (2005: ~ ~al show interacts with PAII'I'ICIPAN'I'S 4 6 4 7 ) . For instance, r ~ r l c / ' l ' ~ l tof of a show (e.g., Alligator-,Yhow, 'a show featuring alligators and a show contiguity focussing on alligators') and with elements of the compo~~nded A 1II.I:MI:N~r (e.g., in Au,rzieh.show of SA1,It:N'I' A C ' T I V I T Y / Q I J A I ~ I ~ ' Y / MI'I:RIAI, the specifier describes a topical activity while the determinant in T ~ z ~ ~ z s h o ~ ) can be concepti~alizedeither as a static nominal theme or as an activity). Bearing this in mind without further elaboration, it is time to move on with the discussion and analyze selected examples of hybrid compounds whose anglicism heads have been far less productively involved in the formation of hybrid compounds.

4.2. Less productive anglicism heads and their specitiers in hybrid compounds In order to provide a mole coniprehensive picture of hybrid compound formation, this ~ectionexpands the scope of analysis to less productive anglicism heads and to sporadic creations. In detail. the investigation focuses on the five least productive anglicism heads in hybrid compounds among the list given in Table 1 and on ten compounds each of which represent the single instance of an anglicism cum head in the corpus. 7 he latter were drawn from the result list of 111brid conlpounds in Der ,piegel 2000 according to their occurrence in alphabetical order of reverse word sort (cf. note 1 1 ) In Table 1 , kid.\ (7 types, 9 tokens), hc~tscller(6 types, 1 I tokens), cclrver (6 types, I I tokens), ~ o ~ l (6 c rtypes 10 tokens), and surfir (5 types, 6 tokens) occur at the bottom of the list of the one hundred most frequent anglicisms as heads in hybrid compound nouns. Their individual patterns of specification are briefly analyzed below. Kids: 'l'he hybrid compounds with the head kid.^ are: GroJstadt-Kids 'big city kids', Kon.so1cnkids 'gamepad kids', Konsum-Kids 'consumption kids', Llr-~u~skidv 'luxury kids', Reichc-Lcule-Kir1.s 'rich people kids'. MotorrodKid.7 'motorcycle kids', and Wun&rkid.s 'miracle kids'. The anglicism Kids

272

Alextrnu'er Onysko

basically denotes children and adolescents and is connotatively loaded witli tlie meaning of being modern, trendy, and slightly spoiled. These connotative traits are generally reflected in tlie contiguous specifiers which determine tlie ACTIVITY of tlie kids (Konsolen-, Konsuni-, and Motorrad-), their (GroJstadt-, Reiche-Leute-), and their attribution to a SOC:IAL,NE.I'WOIIK ARII,ITII:S (Wunder-) and Dr:srnris (Luxus-). Be.st.seller: Tlie hybrid compounds formed witli tlie head Best.seller appear in the corpus as Buchbe,st.seller 'book bestseller', .Juhre.sbe,s/.seller 'yearly bestseller', Krinzi-Be.stseller 'detective story bestseller', Literatur-Bestseller 'literature bestseller', iihertr.schtmg.chcstseller 'surprise bestseller', and Wclthe,stseller 'world bestsel ler' . Related to the meaning of the head ('something that sells very well'), tlie determinants specify Puonuc'r (Buch-, Krimi-, Literalur-), MAUKE'I' (Well-), DUKA'rION (.lulires-), and C:ONSIJMF,R (more specifically their buying beliaviour wliicli is expressed counter expectations in ~berrmchungs-). By metonymic sense extension, M A R K I T size and DURATION emphasize the extraordinary impact of the bestseller. The sole occurrence of books as PRODUCT specification reflects the primary semantic scope of the anglicism, wliicli was initially borrowed in the context of bestselling books. Continued use, however, has expanded its range to other artistic media and genres (e.g., CDs and movies) and, by fiirtlier extension, to any type of product that sells well. Server: Tlie anglicism Server ilicorporates a specific technical meaning wliicli can be described as 'a computer in a network used to provide services (as access to files, shared peripherals, or the routing of e-mail) to other computers in the network (cf. Merrialn Webster Online). The name and the definition imply the fi~nctionalsalience of this type of computer. When it comes to the formation of hybrid compound nouns, FuNC [ION indeed emerges as tlie most prevalent contiguity for specifier selection in the corpus: Filterserver 'filter server', Nelzserver 'network server', Progranzm.server 'program server', Sprach-Server 'language server', and Tau.sch.server 'exchange server'. The only remaining hybrid specifies tlie MATEKIAI, component of the server: Kartcgyel-Server 'potato server'. This form was coined in reference to a hoax invention by a team of British cornputer tcctinicians, who claimed that they were able to use potatoes as an electrical source for running a processor chip.

('urlzng /he conceptual ~pollight Hybrid conlpoundlng

273

Power: The anglicism Power denotes force and strength in a general sense combined with connotations to modernity, young age, and dynamic force. As a head in hybrid compound nouns, Power evokes the AGr NT as holders of the power in Burgerpower 'citizen power'. Frauenpower 'women power', and Madcher~pocver'girl power'. In addition, the rvrJl;:of power is specified in Hal-und-Stech-Pmr 'hewing and stabbing power', and Rechenpower 'calculation power'. Finally, the rwlwo\c of power is made explicit in ~ikopo~lier 'ecological power'. The creation of the last example is doubly contextually motivated. First of all. i)kopower is used to label the political move of tlie then German minister of economics to include an ecological perspective in policy making. Secondly, the selection of the head power plays on a close association to Powerfiau 'powerful woman', which is a lexicalized concept in German as emphasized by its recurrent use in the corpus (TF 9). This implicatioli becomes particillarly evident in view of the textilal position of the hybrid compound as part of the headline and because of the general focus of the article, which profiles the woman selected to f i l l tlie void of ecological policy making. (7) Okopower fur Muller Mit einer ungewohnlichen Personalie will Wirtschaftsminister Werner

Muller das blass-grune Image seines Hauses aufpolieren. Als Leiterin des Referats ,,Okologische Grundsatzfragen" sol1 Kristina Steenbock, 46, bisher Chefin des Berliner Greenpeace-Buros, ab Mitte Januar den Umweltschutz in Mullers lndustrieabteilung verankern. (Der Spiegel 2000: 5212 1 )

[Ecological power for Muller Minister of economics, Werner Muller, wants to brush up the pale-green image of his cabinet with an unusual person. Kristina Steenbock, 46 and until now head of Greenpeace in Berlin, is supposed to take care of environmental protection in Muller's industrial cabinet.]

Surfer: The anglicism Surj>r has followed the semantic development of its English source term so that today Surfer can refer to a person practicing the sports of (wind-)surt?ng and snowboarding, and it can denote someone surfing in the Internet. The basic meaning for both uses can be described as a 'person who is indulging in moving in a mediuln (e.g., water, snow, the World Wide Web) by choosing the direction of tlie movement in a creative and spontaneous manner while relying on the medium's intrinsic thrust.' The concept also bears essential associations to a joyful and playful leisure time activity. When SurfiYr fi-111ctiotisas a head in hybrid compound nouns in the

corpus, specification reflects contiguous sense relations extending from its 'snow core meaning. TIILIS,in il'etxurfir 'net surfer' and Schnee.su~~jhr surfer' the MI:[>IlJM of the actorlactivity is specified while Gesundheits'fi~nsurfer' specify a PURPOSE, tlie sulfiv" 'health surfer' and S~?uJ!~.surf~r latter by extension of its connotative implication. The last example, Dauer.surfe~'constant surfer' refers to a person who is constantly surfing in the internet, i.e., the DURATION of the agent's involvement in the activity is specified. Thus far, the analysis of the less productive anglicisms has confirmed the role of contiguous sense relations between tlie head and tlie specifier in liybrid compound nouns. The process by wliich the determinant instantiates a head-frame internal conceptual contiguity emerges as a basic cognitive explanation of hybrid compound formation. While drawing conclusioiis on more or less prototypical contiguities in individual semantic head-frames is bound to indirect evidence of typeltoken frequency wit11 highly productive anglicism heads, even the less productive anglicisms above hint at the existence of different contiguity strengths arising from the basic meanings of the heads. TIILIS,Kids prinie tlieir crucial field of A C ~ I VI IY, Bestseller closely implies tlie bestselling PRODUCI, a Server is designed for a specific I lINC'lION, Power is intrinsically dependent on an AGENT as source, and for a LSuy-rthe MFI)IIIM is vital for performing the activity. To complete tlie conceptual analysis of hybrid compound nouns across different layers of productivity, tlie final data set to be discussed consists of ten forms that are characterized by the sirigillar occurrence of an anglicism as a head in liybrid compounds. The examples are taken from the corpus in the order of tlieir occurrence" to avoid a bias in data sampling. This led to the extraction of the following compounds: Aupair-Boy 'au pair boy', Schlachtfild-Dun~nzy'battle field dummy', DNA-.Jockey, Mu.\terbody 'perfect body', Surrealisten-Dan 'surrealist dandy', EfJizienz-Output 'efficiency output', Gumntie-Agreement 'guarentee agreement', PronzinenlenPlacement 'celebrity placement', Le~k~r-C'al.suit 'leather catsuit', and Bonsenclipper 'big shot clipper'. From this selection, Schluchtfeld-Dumny, Mu~terbody, EffizienzOu/put. Gurantie-Agreen~erzt.I'rominel~ten-Plucc~t~ent, and Ledel*-C'atsuit straightforwardly emerge from the principle of head-frame internal specifier selection. The semantic contiguities realized by the specifiers are PURPOSF (of the dummy), APPFARANCE (of the body), RESlJL I (of the output), K1,9111 P (of the agreeruent), \rrUI tc I (of the placemetlt), and M A F t K l A L (of the type of clotliing).

('crstirrg the conceptual syotlighf: Hybrid compotma'ing

275

The remaining hybrid compounds are created according to diverse mechanisms. DNA-Jocke,v is coined in analogy to disc jockey, which, as a well established anglicism (preferably used in the abbreviated form DJ), provides the basic pattern for the formation. Compound internally, the ION ~ of i specifier is evoked in the head-frame of .Jockey as the main o ~ c u l > the head. The allusion to disc jockey is emphasized in context which introduces the verbal anglicism samplen (typically used with DJs) to describe the activity of a DNA Jockey: "oder es vom DNA-Jockey samplen lassen" [...or have it sampled by a DNA Jockey ... 1 (Der S'yiegel 2000: 481190). While in Bonzenclipper the compound internal specification realizes the OWNER in the liead-frame of a 'luxurious boat', the actual formation of the compound invokes a metaphorical projection reminiscent of Fauconnier and Turner's analysis of 'land yacht' (cf. section 2). In this case, tlie metaphorical blend [naps the source domain of a luxurious boat onto the target domain of a luxurious airplane used by big shot com~nunistparty members in the former GDR. The compound Surreuli,sten-Dandy is applied in reference to the painter Salvadore Dali. As such, it is an exatnple of a copulative co~npound equally ascribing both characteristics to his personality. 'The perhaps most interesting compound from this sample is the term Aupuir-Boy. From a head-centric perspective 'boy' is specified frame internally by his OCCIII~ATION.However, in this case, this explanation fails to i~nveilthe intricate semantic relation between the elements of the conipound. Since au pair tends to create prototypical associations to a feminine agent rather than to a masculine, the mention of 'boy' emerges from the semantic frame of au pair, specifying the unusual gender of tlie person. In addition, the compound Aupcrir-Boy contextilally attributes to the description of a deranged family situation in which the mother, her lover, and her two daughters with their male ail pair have moved to a hotel while the father takes care of their sons. (8) Wahrend die von einem Aupair-Boy betreuten Franzi und Fritzi bei der Mama und dem Engelbert im oberbayerischen Hotel wohnen, versorgt der diipierte Doktor Heidenreich arn Rhein die Jungs. (Der Spiegel 2000: 17197) [While Franzi and Fritzi, taken care of by a male au pair, live with their mum and Engelbert in a hotel in upper Bavaria, the duped doctor Hei-

denreich provides for their sons at the Rhein.] From the context, the selection of 'boy' is thus inspired by its function as a marker of unusual male gender of tlie au pair and by its connotative value

which alludes to tlie young age and attractiveness of the caregiver. This underlines the exceptional semantic structure of the compound in which tlie first element functions as the frame-setting head and what appears to be the struct~~ral head, i.e., tlie right hand elenlent, semantically fi~nctionsas a postmodifying determinant." Tlie sniall random sample of hybrid compound nouns formed by a one time only occurrence of an anglicism in head position basically confirms the general conceptual drift in hybrid compounds fro111the head element to the determinant, i.e., the specifier tends to instantiate associatively contiguous facets in the semantic franie of tlie head. As in the examples of more productive head-frames, head-frame internal specifier selection appears as underlyingly present in metaphorical compounds and, as tlie last example sl~ows,it can also hold regardless of tlie position of tlie conceptual head element in tlie compound. Based on these observations, the conclusion will summarize the main findings and address a few general issues in order to synthesize the approach of head-frame internal specifier selection as a general cognitive principle of compound formation.

5. Summary of the main findings and some concluding thoughts on prototypical contiguities Tlie evidence drawn from the analysis of the selected hybrid colnpou~id nouns allows for a few generalizations as far as word formational productivity of liybrid compounds is concerned. First of all, it has become evident that in headed compound nouns the selection of the specifier is guided by head-frame internal contiguous sense relations that arise from the core traits of the meaning of the head. This contiguous sense relation between specifier and head has been referred to as head-frame internal specifier selection througliout the chapter. Secondly, even in cases where tlie meaning of the compound emerges from metaphorical and metonymic projections or becomes lexicalized by abstraction from its original sense as in Lackriiuste,st ( ' l i t m ~ ~test'), s tlie underlying contiguous sense relations between head and specifier can be unveiled by close analysis. As such, headframe internal specifier selection provides a baseline for a conceptual blending franiework, particularly for single scope networks. The data of metaphorical hybrid compounds show that the selection of input spaces (or frames) in the blending follows from contiguous sense relations of the head. Thus, conceptual blending operates on the basis of head-frame internal specifier selection in metaphorical compounds while in non-

metaphorical compounds head-frame internal specifier selection alone can account for the conceptual motivation of compound formation. In general, the data indicate that metaphorical and metonymic sense projections can occur freely across hybrid compounds but that tlie actual amount of metaphorical compounds is minimal compared with non-metaphorical constructions (e.g., 2 out of 2 1 1 types of Manager compounds, 2 out of 147 types of Filrn compounds, and 5 out of 104 of Test compounds). While tlie principle of head-frame internal specifier selection applies across the data of compounds, generalizations on contiguities are only valid for the specific head and its semantic frame, since the contiguities arise from the individual core semantic traits of the head. This is evident when comparing the extensions of contiguous sense relations among the most productive anglicism heads. The specifiers of the head Munuger, for instance, emerge from three primary contiguities determining the AFFILIA'I'ION/OCC:IIPAI'ION of the manager, the suoJEc~rof the manager's occupation, and the AGFNTIVI: QUAI,ITIF,S of the persona of the manager. Film, on tlie other hand, evokes a rich array of contiguities extending from its complex semantic architecture: ~I'ol~lc, I'EKSONAE, SETTING, QIJALITY, MATERIAL, MEDIUM OF UROAL~C:ASfINCi,ALIDIO-VISUAI, FIA'I'UKES, I'FMPORAI, I3OUNI)EI)NESS, and ALII>IENCE. The graphs in the appendix illustrate the semantic frame and the contiguities of specitication for the most productive anglicism heads. The visualizations draw attention to the crucial issue of conceptual fluidity that characterizes semantic frames and their inherent contiguous sense relations. Even though different contiguities are represented as separate spaces in the semantic frame of the head, contiguous senses can interrelate and partially overlap. As mentioned earlier, for example, 'l'oplc of a film interacts with characteristic PIRSONAI': and Sk:l'-I'ING. The specifier in Massentest 'mass test' could be evoked by PARTICIPANTS or could elnerge from test DESIGN. Similarly, the specifier in Hfirtetest 'hardness test' can realize the SIlBJ6C.T or the RESULT of the test depending on the usage intention. To indicate conceptual fluidity among contiguous sense relations in the illustrations, associated contiguities are connected with dashed arrows. In general, tlie contiguous sense relations in the semantic frame can be metaphorically visualized as poles in a continuum of contiguity. Despite the fluid nature of contiguities, individual contiguous poles in the semantic frame of a head tend to cluster Inore specifiers than others. Manager, for example, is most frequently specified by AFFILIATION/OCCIJPAI'ION (13 1 types which is 62% of all its hybrids in the corpus) followed by the sul3.rt:cT of the ~nanagelnerlt(68 types which is 32% of all

27 8

Alexander O ~ y s k o

its liybrids). The head Team draws the majority of its specifiers from MEMBERS of a teain (40 types wliich is 39% of all its liybrids) and from 1)rsclI'I.INlr: of'a team (35 types which is 34% of all its hybrids). In the conceptually close lieads Film and Show (both share tlie semantic traits of ~ c as the 'entertainment', 'creative source', and 'audience') tlie - r o ~appears most productive contiguity for specifier selection (34 types for Film which is 23% of all its hybrids and 32 types for Show wliich is 32% of all its hybrids). Due to their conceptual relatedness, Film and Show share a similar pattern of specifier recruitment which tends to spread out Inore evenly alnong its contiguous poles. Finally, the liead Test is specified by three contiguous sense relations for the vast majority of hybrid compounds: SUBJECT (45 types wliich is 39% of all its hybrids), RESULT (33 types which is 29% of all its hybrids), and DCSI(;N/METH~D (2 1 types wliich is 18% of all its hybrids). The small number of occurrences among the less productive anglicism heads does not allow establishing distributional patterns of frame internal contiguities with tlie exception of Server, which is regularly specifed by its FUNCTION in all but the exceptional, playful creation of Kurtqffel-Server 'potato server'. These distributional tendencies of specifier selection reflect tlie different strengths of frame internal contiguities. For example, a Team immediately implies its MEMBEKS and its DISCIPI,INE, and a Test primes most of all SU13JECT and to lesser extent RESIJIX and MFI'HOI)/I)ESIGN. The gradience in type frequency of contiguous specifiers thus reflects prototypical frame internal associations that guide tlie selection of the determinant element. The strongest. i.e., prototypical, contiguities emerge from the most conceptirally salient contiguous senses. Of course, this does not suggest that other specifications of a head are ruled out. It is merely indicative of the notion that a default context would prefer the activation of the most prominent contiguous senses whereas special usage contexts can demand more exceptional determinants as discussed above in a couple of instances, e.g., Ameisenfresserfilm 'anteater film', Reulres! 'reality test', and Kafigshow 'cage show'. The graphs in the appendix try to incorporate varying contiguity strengths, i.e., degree of prototypicality of specifier selection, by thickness of tlie arrows branching out of the core meaning. Overall, the study has tried to account for the word forinational productivity of compound formation from a cognitive perspective by focussing on anglicisms as a productive segment of the German language. The analyses of the semantic relations between specifier and head have shown that the specifier generally elnerges froin contiguous sense relations in the head. As such, the specifier in hybrid compounds is activated within tlie semantic

Custing the conceptual spotlight: H,vbrid compounding

279

frame of the head concept. This process has been established as the principle of head-frame internal specifier selection. The investigations of hybrid co~npoundshave indicated that head-frame internal specifier selection provides a foundation for a conceptual blending analysis of metaphorical compounds and that it unveils the existence of prototypical head-frame internal sense contiguities. While the data of this study prokides evidence for the cognitive relevance of head-frame internal specifier selection, it is the task of further research to probe into other segments of compound formation and to expand its scope to other languages sharing similar word formational processes. In view of the generally blurry boundaries between conipounding and syntactic formations, a further potentially insightful endeavour could be the analysis of syntactic structures of ~nodification along similar conceptual lines.

280

Alexander Onysko

Appendix 1: The one hundred most frequent anglicisms as heads in hybrid compound nouns in Der "+iegel2000 Rank

2s

I londy/ Handys

36

62

Munagers l"i1111i F~ln~ei

1 t: 1 36 37

1

&.s~gn D.sign.s Code: Codes Desrgner,' Desrgnerrril Designers

I Burl Bars feri

Look' Looks

27

1 !i 1 37

I 82 X3

85 86 .-

" 88

.

.Shop/' Shol~s

11

13

1l~srnes.s

11

12

/*.lop/t,,lop.~

10

16

Krcher

10

16

fister

10

15

(./I,,/ (-11,l.S

10

I?

Oll~f/t/ 0utf;ls

10

10

10

10

Pool/ Pools Rroher Rrokern //omepuge

93

'" ')'

C)8

"' O0

13

9

9

Junkie/'./unkres

'

9

Krck

9

9

8

19

SI'OI

92

9

Rocker; Hochees

8

17 II

Lady/ Ludys

8

.Irrnrho/Jttnzbos

8

!Monopoly

8

8

ti,&

7

9

~eslse~er

6

11

.s~l7,er

6

II

f'o~oer

6

10

i

6

'S?,l:ti?r

8

-

Appendix 2: 'Tables of hybrid compounds for the five most frequent anglicism heads

Manager:

Rcrtelsmann1-hyssen-Manager

( a ) ,,IFFII,IA-

1)aimlcr-C'hryslcr-

6

Manager Yello-Manager

6

ABB-Manager

5

1:lich-Manager

RMW-Manager

5

Siemens-Manager

I lolzmann-

5

~ r o n ; / o c v ' c ;I ~7/0.\ ~

Managcr Rayer-Manager

3

Metro-Manager

1

Vodafone-

4

BouyguzsManager

1' I

Comnion cut~.gorial nouns

I

ager

I

Oppositionsnianager

(cop.) Riistungsnianager

Korridor-Manager

I

Ersatzkassennianager

1

Biirsenmanager

Musikmanager

3

Kantincnmanager

I 1

Atonimanager

-?

Wohungsrnanager

1

I

Forschungsmanager

2

Ladenmanager Arkadenmanager

1

Kreditmanagcr

Krenil-Manager

1

Kulturmanager

liennstallmana~er

I

Markenmanager

2 2 2

Stahlrnanager Programmman-

Sum 32

(b) S( :H./ECl

85

Kadiornanager Rennn~anager Spirituosenmanager

agcr

(

Uisikotiianager

1

Familienmanager Imniobilienman-

I

1

-7

Lebensmittclmanagcr

2

Tabakmanagcr

I I

Ruchmanager

ager Fernschnianager

2

I'lug'cugn~anager

%ei(schriftenman-

2

Karrierenianager

9 Werbemananer

I I

BundesligaManager

I

Sicherlieitsmanager

1

Tertnhmunuger* (insir.)

I

Ehrenamlsman-

I

Ilrlaubsmanager iionsmanager

Sum. 70

Fusionsmanager Wisscnsmanagcr Ausstellungsman-

(c) AGI:,.L7Il't ()(:.I/ /TIES

Wcst-Manager

Weihnachtsfilm

Lupach-Manager

Ingrid-Noll-F~lm

Sum: 10

Sum: 34

Jarmusch-l'ilni Kiarostami-Film Kicslowski-l,'ilrn

Film:

Land-Film

(a) Tow('

Sam~iraiiilni Sandalenfilm Scliauspielerlilm

Nachwuchslilni

1

Ncshat F,'ilm Otto-F~lm Palandt-Film

Schindler-Film

Schelme~ifilm

Wenders-Ftlm

Serienkiller-l'ilm

Wong-Film

-l'erroristenfilm Vamnirlil~n Sum: 14

Setting

Haupstadtlilm

Kricgstilm Kriminaltil~n

120

1

(c) P~IRPOSE / FUW1701V

Absclilussfilm AufklBrungslilm Cannes-Film

Kuha-k ilm

Exper~mentalfilm Dehiitfilm

Mafia-Film Musikerlilm

1

NS-Film Olynipia-Film

I I

I'ornofiliii

6

Casting [he corlceprzrul ,spotlight: Ffybr-id cor~~pour?ding 285 I!nterIialtungsfilln

1

Werbctilm

8

Wcttbcwerhsliln~ Sum 19

6 121

(d) ( ~ [ : . ~ I . I T(prodzdcY tion, audier?ceresponse)

Sum: 5

Team : (a) ME';LIBER:,

'l'rickfilln %cichentricktilm Sum: 7 Gruseltilm

27

(g) .A U L I I L A.r:

Kultfilm Lieblingstilln Montagsfilm lugendfilm lungsfilm Ktndcrtilm Skandaltilm Sp~tzenfilni Sum: 14

(e) M,,f T E R ~ ~ I ,

Sum: 5

13

( h ) hilELUli~(qfhroadcasting) Type I'crl~schfilni Kinotilm

52

II'1.L-Film Videotilni Mikrotilni

Videokamera-Film Sum: 6

I'hospholip~dFilm Schmaltilm

(i)

TEIL1POK,41,,401 :.ill-

EDILE.SS

Sum- X

Film

2 86

Alexander On-vsko 1-crnsclitcam

20

Fluminense-Team

I

FuDballteam

2

tliirfirnktearn

I

Inditex-Tc:un

I

Spcrialistcntcam

I

Kameratearn

28

Stolte-Team

1

LKA-Team

I

Studcntcntcam

1

Mapei-Team

I

'Tai~clierlea~n

I

Olyn~piatearn

II

Technikerteam

3

Oranje-Team

1

i)berset~erteam

I

I'rojektteani

4

Wawra-Team

I

Kedaktionsteam

I

Wisscnschalllerteam Sum: 40

0

Renntcam

0

Itl'L-l'eam

1

95

E l Verhandlungsteam Verkaufsteam Vorhut-Team

(d) SOcl.4 1, STATUS' I lochschulteam

Schauhiihnenteam 12

Nationalteam SPIEGEL-'I earn Reserveteam

Sportteam

I Triatlilonteam 1 TV-I eam

Innovationstea~n tiB11lpfcrteam Spitrentcam

II 1'1

I I

Sum: 7

(e) S/zk

Werksteam

Sum: 4

ZD1:-Team 124

Sum. 35

(b) Dlsc'l-

(c) P~JKrosE

Test:

I

kM P-Team Frstliga-Team I'allschirmteam

3

( a ) A~OB,JE('l(en-

Expedit~onsteam

3

1ityductivil~g

I:orschungstea~~~ 5

CCF- I earn Drehteam

Entwichlungsteam

11

I~'uhrungstearn

h

(;rabungstcarn

2

Inlegrat~onstea~n 2 Operationsteam

I

OP-Team

2

Produktionsteam

2

Aroma-Test

I

21

C'usting the conceptutrl spotlight. Hybrid con~pounding 287 Sum: 45

I I6

(b) RESGLI

I Drogentest I Epo-Test

15

11 I

Hepatitis-Test

Ger7eral concepts (uchievemenl, mental and pl7y.~icul states/ phenonemu)

1 1

Kohaintest

I

Nandrolon-Test

I

Prionen-Test

I

Troponin-Test

2

CJK-Test

2

Zuckertest

1

Type sum: I I

Tk 25

Sum: 33

58

(c) DESIG~VMETHOD

Farbcindringtest F-Show

3

Sum: 19

56

FlieBhand-Show Kafigshow Tcstostcron-Show

Morgenshon

(e) A 1!nlcxt .E

Ilnterleibsshow Sum. 5

NIIR-Show Nctr-Show

Appendix 3: Graphs of head-frame internal specifier selection of the five most frequent anglicism heads Graph (Munugeu): Semantic frame of Manuger highlighting contiguous meaning components as sources of specifiers in hybrid compound nouns - Solid arrows indicate contiguities - Thickness of arrows sy~nbolizescontiguity strength - Dashed arrows mark relations between contiguities

' I

'

:

'

(preoccupation of agent): ; subject of managing

I

I I

'

1----------------------:

;

,---------------: metonymic mapping - frame I bound

I

I I

I I

I I

I

I

I I

1 - - - - - - - _ _ - _ _ _ _ _ - - 1

A \

..........................

denotative components: an agent in (instrument for)

I control, power, ordering; I contra chaotic complexity ! and inhornogeneity of socialf institutional, and material structures

f

I I I I I

I

I connotative components:

I

1 I

;

I

modernity, profession,

I achievement I _ _ _ _ - _ _ - - - - - - - - - - - - - - - - - - a

i

------------------

.....................

I

I

: I

(type of agent): affiliationloccupation

I

I

6

...................... I I I

I I I

agentive qualities I I

('r~sfingthe concep/zlcrlspotliph/: Hybrid cornpozrnding

29 1

Graph (Film): S e ~ n a ~ i lframe ic of Film highligllting contiguous meaning co~nponentsas sources of specifiers in hybrid compound nouns - Solid arrows indicate contiguities - Tliickncss of arrows sy~nbolizescontiguity strength - Dashed arrows mark relations between contiguities r - - - - - - - - - - - - - - - - - - - - - -

i

f metaphoricallfigurative

; pro-jection

I

-I - - - - - - - - - - - - - ---------------I

;

,

I - - - - - - - - - - - - - - - -

:

I

;

personae

I

purposeifunction

;

topic

I I

denotative components:

I

8

:

I characteristic ; : It-: I setting I

I I _ _ _ _ _ _ _ _ _ - _ _ _ ,

t

L

'

I

: 8

1

_

_

_

_

-

-

_

-

_

_

_

:

_

connotative compon e t : a w o k enter-

,

I

I

:

I

I

I

I

;

tainment - information, audience centered , passively consu~ned

_

,--------------f creative source

I

I

I

_

I

I

I

I

features

audience

I

:

I

'

'

I

---------------

I

I

:

I

I

I

I

I

I

stance, audio-visual imagery, story, bounded creation

I

mediilmof broadcasting

L - - - - - - - - - - - - - - -

,

; I

I

------------------I

j material substance

:

pr~v~ar?. sense nleforlynly

(temporal) bounI dedness I I I- - - - - - - - - - - - - - - - - ; 1

; I

I I

> %

I

, '

:

Graph (Tecxni): Semantic frame of Team highlighting contiguous meaning components a s sources of specifiers in hybrid compound nouns

- Solid arrows indicate contiguities - Thickness o f arrows symbolizes contiguity strength - Dashed arrows mark relations between contiguities

: : I

/

members (common characteristics)

;

'

I I

I

I _ _ _ _ _ _ _ _ _ _ _ _ - _ _ _ _ _ _

;

I

- - - - - - - - - - - - - - - - - - -I~

I disciplinelconcern

( I

-------------------

I

: I

I

r _ _ _ - - - _ - _ - - _ _ - - - - - - - - -

; group of people. com! mon purpose, occupation I

: :

I

;

\

connotative components:

a

I

cooperation, achieve; ment of goal, modernity,

\

f _ _ _ _ _ _ _ _ _ _ _ - _ _ _ _ _ _ - - - - I

v

v

I ,_____________---____------_--; defining quality I purpose : : ' I (size) , _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ I-------: , I social status ; '

I

I

I

I

I

'

I

___--------------I

I

I

I I I

I

I I

I

I

Casting the conceptual spotlight: Hybrid co~npounding 293

Graph (Tc..s/): Semantic frame of Tes/ highlighting contiguous meaning components as sources of specifiers in hybrid colnpound nouns

- Solid arrows indicate contiguities -

Thickness of arrows symbolizes contiguity strength Dashed arrows mark relations between contiguities

I metaphorical!figu!-ative

;

projection

,. . . . . . . . . . . . . . . . . . . . . . .

\ I I I

Graph (Show): Semantic franie of Show highlighting contiguous meaning components as sources of specifiers in hybrid compound nouns Solid arrows indicate contiguities Thickness of arrows symbolizes contiguity strength - Dashed arrows mark relations between contiguities

-

I

.........................

I metaphoricallfigurative pro- : I jection ------------------I

I

I

I

I

I

I------?

I

I

I

I

staged performance

I

:

connotative:

I

sive reception, impressive, removed, intangible

I entertainment, pas,,I r

I

' ' I

I

---- ---- - ----- --

r _ _ _ _ _ _ _ - _ - _ _ _ _ _ - _ _ _ _ _ _

I 1 I

I I I

screening

-.- -

I

audience

I

Castirzg he cot7ceptuul spotligl11: Hybrid com/~ounding 295

Notes 1.

2.

3.

4.

5.

6.

The multi-part compound I'ermiiget~sver.wultungsver~~ugallows for two possible interpretations depending on the degree of lexicalization of the internal + Vevtrug and (b) Vermijgen -1 Venualcompounds: (a) Vt.vnziigensverwultz~t~g and slightly tzitigsvertrag. Collocational preference for I~'errniigensverw~uItutig differing semantic scope of I~erwul~ungsvertrug (implies a more general sense of an institutionally bound administrative contract) favour analysis (a). For further general discussion on the segmentation of multi-word compounds cf. Donalies 2005: 52-54. Literal translations of compounds are given to emphasize the German conceptualization. Due to the constructional similarity of German and English compound nouns, the theoretical approach appears basically valid for both languages as emphasized by the literature on compound nouns in English and German (cf. Eichinger 2000, Donalies 2005, Motsch 1999, for German, and Bauer 1983, Katamba 1993, Plag 2003, Benczes 2006 for English). However, there is still a need for an extensive analysis of English compounds along the principle of head-frame internal specifier selection in order to substantiate this claim. This compound internal projection of semantic information is not on par with a building block approach to the overall meaning of the compound but allows for emergent meaning of the construction as a whole. Thus, information is projected from the specifier to the head in niicrowave oven (referent of the compound is a specific type of oven) and in belly button (cf. Benczes, 2006: 98, and this volurne) despite the fact that in the latter case the actual referent ('navel') emerges by metaphorical prqjection. At this stage, it is important to briefly refer to the interrelation of context on the one hand and prototypicality on the other hand as factors guiding the activation of conceptual elements in a semantic frame. Thus, 'button' could also be specified by the country of origin, e.g., a Chinese button. However, arguing along the same line as Downing (1 977), prototypicality in default contexts will evoke associations to most common frame elements pertaining to the garment, function, position, and material aspects of a button. Coulson expresses an ambivalent attitude towards the notion of cause. While she explicitly mentions cause as a feature of the blend derived from the input space of gun (see quotation in text), she continues with the following claim: "Given the philosophical quandary surrounding the proper definition ol'a 'cause'. it's tunlikely that this concept adds anything to a cognitive account. For example, in 'gun wound'. is it really the gun that causes the mound? One might argue that the bullet cnzrses the wound. Alternatively, one might argue that it's the shooter who causes the \\ound. On the cognitive account, one need not appeal to an ob.jectivc definition of causality. but irlstcad appeal to an idealized model of thc event sequence surrounding getting shot. Regardless of the actual cause. we undcrstand the compound 'gun wound'

296

Alexander Onysko because the cognitive rnodel makcs -gun' cognitively accessible" (200 1 : 13 1 ).

In my view, this statement is slightly misleading since an objective detinition of causality is not necessary to account for basic cognitive processes guiding the formation of compounds such as conceptilal profiling, sense projection, and analogical framing. In this sense 'gun' as the instrumental source of the wound, 'bullet' as the object source of the wound, and 'shooting' as the action source of the wound are all conceptually possible metonymically related instantiations of the prominent role of CAUSE in the semantic frame of 'wound'. In contrast to English, German shows a predilection to highlight the type of action as a causal specifier of wound (cf. Schusswunde 'shooting wound', S/ich~~zrnu% 'stabbing wound', Schii~;fin/unde'scraping wound', Branntwzdnde 'burning wound', and Schnitlwunde 'cutting wound'). The actual selection of the causal specifier of wound is most likely a question of categorization by hypernymical designation. Since gun denotes the class of 'shooting irons' in English, its appearance as causal specifier is not conceptually far-fetched. 7. While this metaphorical meaning of 'land yacht' has become the conventionalized denotation listed in dictionaries (cf. Merriam Webster Online), discourse context could also license its reference to a luxurious boat that for special reasons is not put into use in water but on land (c.g., as an exhibit, as a pub) 8. As Geeraerts notes, the degree of prototypicality or salience is indirectly related to usage frequency. Accordingly, "the explanation of prototypicality should not restrict itself to the senzu.siologica1perspective (in which each category is considered on its own), but that the onomusiologicul point of view (in which it is studied how several items may express similar or identical concepts) should be taken into account as well" (2006: 75). Similarly, in her network model approach to morphology and phonology, Bybee stresses the importance of type and token frequency to create prototypical referential schemas (cf. 1985,2001). 9. The distribution of token frequency per hybrid compound type indicates the productive use of anglicisms in German. From a total of 4,939 types of nominal hybrid compounds with an anglicism head, 3,837 (78%) occur merely once in Der Spiegel2000. 572 types (12%) occur two times, 193 (4%) three times, and only 3 17 hybrid compounds (6%) are used four times and more often. Apart from the fact that filtered data would slightly increase individual token frequencies, the general tendency of distribution emphasizes that there is a high number of "Augenblickskomposita" (non-lexicalized spontaneously created compounds), which provide an extensive base for the investigation of conceptual processes in compound formation. 10. The appendix contains a complete list per anglicism head of all the hybrid compound nouns found in the corpus and the frame-based visualizations of the contiguities between anglicism heads and their specifiers.

Cmling the concepl~rulspotlight: Hybrid compounding

297

I I . The order of occurrence was determined in Wordsmith Tools according to alphabetization by reverse word sort of the result list of "hybrid cornpound nouns with anglicism heads" in Der Spiegt.12000. 12. This is one of the rare examples where the first element assumes conceptual, head-like functions in the compound. As an anonymous reviewer points out, particularly partony~niccompounds as discussed earlier in the article (e.g., Tlschhein 'leg of a table' and Fensterbrett 'window sill') could also be conceptually interpreted as left-headed compounds, i.e. the leg forms part of the frame of table. In such cases, a frame-based interpretation of the semantic relations can be motivated bidirectionally. For further research, an investigation of the same element in structural head and specifier position will shed more light on this issue and, to quote the reviewer's words, "would actually extend and enhance" the current theory.

Acknowledgments

I would like t o express my gratitude t o Marta Degani for invaluable discussions on an earlier draft o f this article. Furthermore, I would like t o thank Reka Benczes and an anonymous reviewer for insightful remarks on the paper. All remaining errors and inconsistencies are my responsibility.

References Allwood, Jens 2003 Meaning potentials and context: Some consequences for the analysis of variation in meaning. In Cognitive Approaches to Le-xical Semantics, Hubert Cuyckens, Rene Dirven, and John R. Taylor (eds.), 2966. BerliniNew York: De Gr~iyter. Bailer. Laurie English Word-Formutior?.Cambridge: Cambridge University Press. 1983 Benczes, Reka Creutive cornpounding in Engli.sh: The sen~unticsof' metaphorical 2006 and metonyn?ical nozm-nour? combinations. AmsterdamiPhiladelphia: Ben-jamins. Bybee, Joan L. 1985 Mor*pl7ology.AmsterdamiPhiladelphia: Benjamins. Phonoloa and Lungztage Use. Cambridge: Cambridge University 2001 Press.

298

Alexander Onj.sko

Coulson, Seana 2001 Seninntic Leups: Frame-Sl~$ing and Conceptual Blending in ,tileaning Construction. Cambridge: Catlibridge University Press. Coulson, Seana, and Todd Oakley 2003 Metonymy and conceptual blending. In Metonymy und Pragnlatic !r~fi.renc:ing,Klaus-Uwe Panther and Linda L. Thornburg (eds.), 5 179. AmsterdamIPhiladelphia: Benjamins. Croft, William, and Alan D. Cruse 2004 Cognitive Linguistics. Cambridge: Cambridge University Press. Donalies, Elke 2005 Die Wortbildung des Deulschen: ein Ijherhlick. 2. Auflage. Tiibingen: Narr. Duden 1998 Gru,nma/ik der deutschen Gegenwcrrts.spruche. 6"' edition. Mannheim: Dudenverlag. Downing, Pamela 1977 On the creation and use of English compound nouns. Language 53 (4): 8 1 0 4 2 . Eichinger, Ludwig M. 2000 Dezrtsche W(~rthi!dung:cine Einf2hruny. Tiibingen: Narr. Erben, Johannes 2006 Einfuhrung in die deulsche Wortbildungslehrt.. 6. Auflage. Berlin: Schmidt. Fauconnier, G illes 1985 Mental Spuces. Cambridge: Cambridge University Press. 1999 Methods and generalizations. In Cognitive Linguistics: Foundations, Scope, and Methodologv, Theo Janssen and Gisela Redeker (eds.), 95-128. BerlinINew York: De Gruyter. Fauconnier, Gilles, and Mark Turner The Wcry We Think: C'oizcep/zral Blending and the Mind's Hidden 2002 Coniplexities. New York: Basic Books. Fillmore, Charles J. Frame semantics. In Lingui.stic.s in the morning calm, The Linguistic 1982 Society of Korea (ed.), 1 1 1-137. Seoul: Hanshin. Frames and the semantics of understanding. Quuderni di semanticu 1985 6,222-254. Fillmore, Charles J, and Beryl T. Atkins 1992 Toward a frame-based lexicon: The semantics of RISK and its neighbors. In Fr.urnes, ,fields and contru.s/s: new es.~uysin semantics and lexical orgunizution, Adrienne Lehrer and Eva Feder Kittay (eds.), 75-102. Hillsdale, NJ: Lawrence Erlbaum.

Casting the conc.eptuu1 spotlight: Hybrid coi?~pounding 299 Geeraet-ts, Dirk 2006 Words u t ~ dOther wonder.^: Papers on Lexicul and Scti~unticTopics. BerlinlNew York: De Gruyter. Givon, Taltny 1986 Prototypes: Between Plato and Wittgenstein. In Noun C'1as.se.s and C~ategorizution,Colette Craig (ed.), 77-102. Amsterdam1 Philadelphia: Benjamins. 2005 Context as Other Minds: The Prugrnatic.~qf Sociality, Cognition and Benjamins. Comniuriicution. A~nsterda~nlPhiladelphia: I-lanipton, James A. 1991 The combination of prototype concepts. In The P.sychology of Word Meanings, Paula J. Schwanenflugel (etl.), 91-1 16. Hillsdale, NJ: Lawrence Erlbaum. Janssen, Theo, and Gisela Redeker (eds.) 1999 Cognitive Linguistics: Foundutions, Scope, und Me/hodolop. Berlin/New York: De Gruyter. Katamba, Francis 1993 Morphology. Basingstoke: Palgrave Macmillan. Koch, Peter 1999 Frame and contiguity: on the cognitive bases of metonymy and certain types of word formation. In Metonymy in Language und Thozlght, Klaus-Uwe Panther and Giinter Radden (eds.), 139-167. AmsterdamIPhiladelphia: Benjamins. Langacker, Ronald W. 1999 Granztnur and Conceptuali~ation.Berlinmew York: De Gruyter. 2005 Construction Grammars: cognitive, radical, and less so. In Cognitive Linguisti~:~: Internal Dj1~umic.s und Interdisciplinary Interuclion, Francisco J. Ruiz de Mendoza Ibafiez and M. Sandra PeAa Cervel (eds.), 10 1-162. Berlinmew York: De Gri~yter. Cognitive Grumn~ur:A basic infroduction. Oxford: Oxford Univer2008 sity Press. Marchand, Hans The Cutegories o t ~ dTypes qf'fresenf-DuyEnglish Word-Fornmtion. 1969 Munchen: C.H. Beck. Motsch, Wolfgang Deutsche Wortbildung in Grundzugen Berlinmew York: De Gruyter. 1999 Onysko, Alexander Ang1icisni.s in German: Borrowing, Le,uical Pr*odz~ctivity, and Written 2007 C'ode.switching. Berlin/New York: De Gruyter. Ortner. Hanspeter, and Lorelies Ortner Zzrr Theorie und Praxis der Kon1~7ositufbt-schzmg. Tubingen: Narr. 1984

Ortner, Lorelies, Elgin Miiller-Bollhagen, et.al. 199 1 Deutsche CVortbildung Tvpen und Tendenren in der Gegenwurtsspruche: Vierlcr Haupttcil -- Substuntivkomposita. BerlinMew York: De Gruyter. Panther, Klaus-Uwe, and Giinter Radden (eds.) 1999 Melonyny in Language und Thought. AmsterdamIPhiladelphia: Belljamins. Plag, lngo 2003 Word-Fornzation in English. Cambridge: Cambridge University Press. Ryder, Mary Ellen 1994 Ordered Chaos: The interpretation of' English noun-noun compounds. Berkely, Los Angeles: University of California Press. Sweetser, Eve 1999 Compositionality and blending: semantic composition in a cognitively realistic framework. In Cognitive Linguistics: Foundations, Scope, and Methodology, Theo Janssen and Gisela Redeker (eds.), 129-1 62. BerlinMew York: De Gruyter. Thomason, Sarah Grey 200 1 Languuge Conlac!. Edinburgh: Edinburgh University Press. Warren, Beatrice 1978 Seniuntic Patlerrls of'Notm - Noun Conipounds. Gothenburg: Gothenburg University Press. -

Does love come from to love or to love from love? W h y lexical motivation has to be regarded as bidirectional

Birgit Umbreit

1. Introduction The direction of derivational relations, especially of conversion pairs, is a subject that has been tackled by many linguists but has not yet been satisfyingly resolved. From a motivational point of view, the same proble~n arises: Even if it is likely that a word a is motivated by a word h, can the opposite direction really be excluded'? Tlie present contribution argues in favour of a bidirectional ~uiderstandingof lexical motivation and proposes that motivational relations can nevertheless display a certain tendency towards one direction of motivation. The paper is organized as follows: The first section presents the definition of lexical niotivation as the basis of fi~rtlierinvestigations. The second section continues with the directionality probleni while the usefulness of various criteria meant to determine the direction of derivation is disci~ssedin section 3. Section 4 provides evidence for bidirectionality from a cognitive point of view, and section 5 proposes a continuum in order to accoi~ntfor different degrees of directionality.

2. Lexical motivation: old and new approaches Tlie question whether there is a natural connection between a form of a word and its meaning is a vely old one. It is extensively discussed as early as in Plato's Cratylos dialogue, which, however, does not take a clear position on the subject. Our recent understanding of synchronic lexical motivation is essentially based on Ferdinand de Saussure's distinction between motivation and arbitrariness of words in his C'our.5 dc li~igui~tique ge'lze'rale. While, according to de Saussure, linguistic signs in general have to be considered as arbi-

trary (e.g., Fr. wzrr 'sister', vingt 'twenty'), some signs in every language show a so-called "relative motivation": Le principe fondamental de l'arbitraire du signe n'emp6che pas de distinguer dans chaque langue ce qui est radicalement arbitraire, c'est-a-dire immotive, de ce qui ne I'est que relativement. Une partie seulelnent des sig~ies est absolument arbitraire; chez d'autres intervient un phenomene qi~ipermet de reconnaitre des degres dans I'arbitraire sans le supprimer: le ~ i g n epeui Btre velativement motivi. ( 1966a: 1 8 0 1 8 I , emphasis in the orginal) [The fundamental principle of the arbitrariness of the sign does not prevent our singling out in each language what is radically arbitraly, i.e. unmotivated, and what is only relatively arbitrary. Some signs are absolutely arbitrary; in others we note, not its complete absence, but the presence of degrees of arbitrariness: the sign ma?? he relntively molivated. lranslation according to de Saussure 1966b: 13 1 ; emphasis in the orginal] -

A word that de Saussure considers as relatively motivated is, e.g., Fr. dixneuf 'nineteen' because the complex sign evokes its components u'ix 'ten' and ncz~f'nine': (I)

Fr. dix-ner!fLnineteen'

~ l i x'ten', rlruf 'nine'

t'

The reason for the fact that de Saussure speaks of "relative motivation" instead of "absolute motivation" is that the components of motivated signs are themselves arbitrary (1 966a: 182). While this possibility of decomposing complex signs could forlnally be described as "morphological motivation", de Saussure (1 966a: 10 1-1 02) also talks about another type of relative motivation which concerns onomatopoeia (e.g., Fr. tic-tuc 'lick-tock', ouaouu 'woof-woof). This phenomenon, however, is not part of a lexical approach to motivation and will not concern u s here any filrtlier. Another important theoretician of lexical motivation is Stephen UIImann (1957 and 1962), who, apart from onomatopoetic words, distinguishes two kinds of lexical motivation. The first type corresponds more or less to what de Saussure calls relative motivation and is based on wordformation, e.g., suffixation, as in En. Icuder, wliicli is motivated by to lead, and co~npositionas in En. blackbird. motivated by black and bird (UIImann 1957: 87). (2)

En. leader +- lo lead

(3)

En. blackbird +- black, bird

Lexical niotivation as bidirectional

303

Going beyond de Saussure, Ullmann proposes a second type of motivation called "semantic motivation". This refers to metaphorical and metonymic meanings of words, i.e., to relations between different senses of a polysemous linguistic sign. For example. En. bonnet in the sense of 'cover of a motor-car engine' is motivated by the sense 'woman's hat tied under the chin' (Ullmann 1962: 91). (4)

En. bonnet 'cover of a motor-car engine' the chin'

t

'woman's hat tied under

The definition of synchronic lexical motivation taken as a basis in this paper systematizes a point that Ullmann mentions only en pussant: morphological and semantic motivation may occur simultaneously.' A reconsideration of Ullmann's examples shows, however, that this combination is not an exception, but the norin. Leader is a suffixation 011 tlie basis of to lead i.e., it is ~norphologicallymotivated. At the same time, there is a semantic connection between tlie two words. Since a leader is a person who leads someone or something, a semantic contiguity or a metonymical relation exists between the lexical items. Thus, synchronic lexical motivation has two dimensions: a morphological or forinal one and a semantic or cognitive one (Koch 2001, Koch and Marzo 2007).' Both dimensions can be cross-classified. This holds for all results of word-formation processes, as there are no forinal motivations that do not display a semantic motivation at the same time, and vice versa.4 Consequently, lexical motivation is always two-fold and can be grapliically represented as a motivational square:

Figure 1. Motivational square (Koch 200 1 : 1 156). C stands for concept, L for lexical item.

304

Birgit Umbreit

At this point, it is important to point out that the phenomenon of lexical motivation in this paper does not refer to words or lexemes, but to "lexical units". This term is based 011 Cruse (1986: 77), who defines a lexical unit as "the union of a lexical form and a single sense" whereas a lexeme is a "family of lexical units" (1986: 76). The units belonging to one lexerne then have to be semantically related. This very useful definition allows us to describe the motivational relations based on overt word-formation processes and those based on polysemy on the same level, as will be explained below. On the vertical axes in Figure 1, the two lexical units in question (in what follows referred to as niotivational pair) are represented with their concept on the higher level and their form on tlie lower one. On tlie horizontal axes, the two relations are indicated: contiguity on the level of the concepts and suffixation on the formal level (leader t lead). The two senses of the polysemous bonnet, apart from being metaphorically connected, are also related on the formal side: their word forms are identical, which can be seen as a maximal formal similarity. Making use of Cruse's term "lexical unit", we can consider the two senses of bonnet as two semantically related lexical units, which manifest the special characteristic of having identical word forms. Thus, in the case of polysemy, tlie motivational square can be reduced to a triangle:j

COVER OF CARENGINE

metaphorical similarity

I

bonnet

WOMAN'S

HAT

I

L1=z

Figlire 2. Motivational triangle (Koch and Marzo 2007: 265)

On tlie vertical axes in Figure 2, both concepts are linked to the same form, whereas on the horizontal axis, the semantic connection of metaphorical similarity between the concepts is indicated." With this two-dimensional definition of lexical motivation. we have set the background for examining a central proble~nof motivational relations: the "directionality problem" (Katamba and Stonham 2006: 1 19).

3. The directionality problem

In the majority of the works on lexical motivation that have appeared so far, motivation is conceived as unidirectional. According to Gauger (I 97 1 : 12-1 3), wlio bases liis account of transparent, motivated words mainly on de Saussure, Fr. po111niit.r'apple tree', for example, is dependent on ponlnie 'apple'. It is motivated by this base word. However, the motivation does not work the other way round. Pontn~c.is considered as opaque (not motivated) because, according to Gauger, it is totally independent from pomtnier or other Inore complex words of this word-family:

(5)

Fr. pommier 'apple tree'

ponir~ie'apple'

t7

De Saussure and Ullmann also understand their examples as unidirectionally motivated [cf. ( I ) and (2)]. At a first glance, the unidirectionality of motivation seems to make sense since, as de Saussure (1966a: 18 1-1 82) i~nplicitlystates, only derived wordsXare motivated. Similarly, lacobini (2000: 866) points out that derivation corresponds to "a semantic atid morphophonological grouth wliich acts upon a base to form a derived word". In tlie examples, we can indeed clearly define a base unit and a derived unit: Fr. ponrnlier differs e only in the additional suffix that leads to morphophofrom p o n ~ n ~not nological growth, but also has -'extra meaning", viz. a "greater semantic specificity" (lacobini 2000: 866). This means that an apple-tree is not an apple, but the tree where apples grow. Examples ( I ) and (2) can be explained similarly. lacobini's definition is widely used as a general rule to determine which words are motivated and wliich are opaque. Hobever, this observation has a few sliortcomings as the following examples demonstrate: (6)

Germ. Schiinheit 'beauty'

+-

schRn 'beautiful'

(7)

En. mouse 'computer device'

+ 'small

grey rodent'

The motivational pair in (6) clearly shows formal growth of SchBnheit in comparison to schiin. The derived word bears an additional suffix. Semantically, however, tlie two units are absolutely identical except from tlie different word categories they belong to. Consequently, we cannot speak of any semantic growth here. Example (7), on tlie contrary, does not show any "morplioplio~iological", i.e., formal growth, as it displays polysemy and, thus, obviously relies on the identity of word forms. On tlie semantic level, it is diFficult to ascertain semantic growth: From our extra-linguistic knowledge we are able to deter~ninethe meaning 'computer device' as derived, because the computer tool was invented only some decades ago. Certainly, tlie computer mouse bears some visual resemblances to the animal mouse, and it is this resemblance that creates tlie metaphor. But where exactly is the "extra meaning" or the "greater semantic specificity" claimed by lacobini? The proble~nsapplying lacobini's definition lead us to three interim conclusions concerning his rule:

i. ii. ...

111.

In tlie derived unit, morphoplionological/formal growtli can occur without semantic growtli in comparison to tlie base unit. If the extra meaning is not obvious, it is not quite clear what is meant by semantic growtli. Apparently, it is rather extra-linguistic knowledge tliat influences decisions about the direction. In the derived unit, semantic discrepancy (to avoid the term "growth") in comparison to tlie base word can occur without morplioplionological/for~nalgrowtli.

One solution might be to modify lacobini's criterion to the effect that either formal growth or semantic discrepancy are sufficient in order to determine which unit is the motivated one. IJnfortunately, however, the following cases show tliat even tlie refined version fails, as there is neither formal growtli nor semantic discrepancy:

(8)

Germ. f i g .period of 24 hours' H ? ~Tug 'time of daylight'

(9)

En. love H? to lo1~e'~'

(1 0) Fr. iternel 'eternal' H? iternilk 'eternity'

Lexical rnolivation us hidirectionul

307

Example (8) represents another case of formal identity. Again, there is no semantic growth, and extra-linguistic knowledge does not help either. The meaning 'period of 24 hours' is motivatable by 'time of daylight', but the opposite direction of motivation is just as good. The same holds for the motivational pairs in (9) and ( 1 O), botli of whose meanings show semantic identity. The only formal difference between the units in (9) is their word category while the motivational partners in (1 O), displaying truncation, are equal in formal complexity too. All three formal processes exemplified in (8) - (lo) are based on common phenomena in the creation of new words in many languages: a form acquires an additional meaning (which leads to polysemy), converts or is truncated. However, tlie direction of motivation cannot be establislied in these examples. Does love come from lo love or lo love from love? Eitlicr way would make sense, similarly as for (8) and (lo). The unidirectional conceptions of lexical niotivation presuppose that only one unit is motivated on the basis of the other one, i.e., that the direction is identifiable. Consequently, the cases discussed here represent a problem for this approacli." I-lowever, it would be counterintuitive to deny a motivational connection between the two meanings of Tug, between love and 10 love, and between kterntl and kternitk, because the members of the pairs are clearly related botli formally and semantically. As linguists have to deal witli the same problem in other linguistic areas, such as morpliology and word-formation, various criteria have been developed to determine the direction of derivation, especially in tlie case of conversion pairs. If these criteria could also be applied to liiotivational pairs, the directionality problem would be avoided. Unfortunately, despite the vast amount of literature on this sub.ject," most of the criteria are not completely satisfying, as the next section will show.

4. Criteria for the direction of derivation

In the following subsections, tlie most interesting criteria for deciding on the direction of derivation are presented, together witli tlie criticism they can be subjected to.

4.1. Historical data 'The simplest solution to tlie directionality question would be to consult historical information, i.e., "dates of first records" (Balteiro 2007). Tlie earlier attested unit would then be considered as the base, the later attested one as the derived form (cf., for example, Biese 1941, Tournier 1985, ~ t e kauer 1996, and Balteiro 2007). Unfortunately, in the example of the pair love - lo love, both words are first attested in the year 825 according to the Oxford English Dictionary (OED), so that tlie criterion is of no use here. In any case, it is generally problematic to base synclironic relations on diachronic evidence, especially, "because complex semantic changes may overwrite the original direction of conversion" (Plag 2003: 108; cf. also Adanis 1973: 40-41). That means tliat the direction of derivation suggested by diachronic information may contradict tlie intuitions we liave today about a certain derivational pair, and, in a synclironic analysis, the latter liave to be regarded as more decisive tlian historical indicators.'" Plag illustrates his critique with tlie English verb to crowd tliat seems to be derived from the noun crowd although, according to tlie OED, the verb emerged first. Tlie inversion of tlie direction is due to various semantic changes wliicli the verb has undergone in the course of the centuries. Thus, dates of first records cannot generally account for synclironic lnotivational analyses.

4.2. Frequency Another common criterion is frequency of occurrence. It is generally assumed that the less frequent or marked unit is derived from tlie more frequent or unliiarked unit (cf. Tiersma 1982, Bybee 1985, Plag 2003). At first sight, this rule seems appropriate because, as Plag (2003: 1 1 I) argues, derived units are often (but not always, cf. section 3) semantically more complex and consequently liave a snialles extension than their base units, with the consequence that they fit into a fewer number of contexts. Altliough she does not give precise figures, Eaton (1967) confirms that people use much more often the concepts black and bird tlian tlie quite specific complex word blackbird. If we apply the frequency criterion to one of our unclear cases, French iternel and iternitk, the latter unit would be the derived one according to Eaton (1967) since iternel is part of the thousand most frequent words in French while iternitk ranges only among the fourth thousand most frequent words. Note, lioweves, that frequency dictionaries are

iunreliable sources in that they normally only provide information on tlie occurrences of forms but seldoln on the occurrences of meanings." So, we cannot subject polyse~nyto the frequency criterion. This is especially problematic, as most words have more than one meaning. Apart from this shorlcoming of frequency dictionaries, there are other proble~nsconcerning the frequency criterion: First of all, the criterion only works well if there is a considerable difference in frequency between the units in question. If both words belong more or less to tlie same frequency range, the divergence is not strong enough to be convincing. Eaton (1967), for example, is of no help for deciding on love and to love, because both concepts are part of the 662 most frequent words of English. Secondly, frequency always depends on the type of sources used to count it. Even if a tnajor number of different text types is used, the representativeness is still only relative (~tekauer1996: 129). In texts that are older than fifteen years, for example, the lexical wnit n?ou.ce 'small grey rodent' is certainly more frequent than the lexical unit m o u , ~ in tlie sense of a cornputer tool. Tliis, liowever, might not be the case in more recent texts. Thirdly, frequency can change over time. While Plag's observation concerning the narrower semantic range of the derived word might be true shortly after having been coined, it is possible that the derived unit obtains a higher frequency because of semantic or extra-linguistic, cultural change (such as tlie increasing importance of computers that leads to a higher frequency in the use of words from the corresponding semantic fields). We thus have to conclude that the frequency criterion is only of limited reliability.

4.3. Semantic criteria The most important criterion based on semantic arguments is "semantic dependency", as first established by Marchand (1 964).Ii Accordingly, "the word that for its analysis is dependent on the content of the other pair member is necessarily the derivative" (1964: 12). Marchand explains this rule with the help of the English conversion pair S U M , and to sow: The verb saw must be derived from the substantive suw. SUM] sb is satisfactorily defined as a 'cutting instrument with a blade, having a continuous series of teeth on the edge'. That the instrument may be used for the action of .sawing need not be included in the definition. On the other hand, the content analysis of the verb must necessarily include the semantic features of the substantive SUW: suw vb 'use a saw, cut with a saw'. (1 964: 12).

The semantic dependency criterion has the advantage of corresponding to the intuitive opinions that speakers - \vhether linguists or not - often have about the direction of derivation or motivation (cf., for example, Lieber 1981: 129). Unfortunately, however, it lacks objectivity and risks to be circular because the result of the analysis depends directly on the formulation of tlie paraphrases used. Or, as ~tekauer(1996: 128) puts it, "we can adjust the definition of semantically related words in accordance with our intentions". ~tekauerdemonstrates that one could perfectly well argue in favour of the opposite direction of derivation for saw and to saw, defining the verb as 'to cut with a toothed instrument' and the noun as 'instrument for sawing'. The same point can be illustrated with the conversion pair love and to love: The noun is derived from the verb if the former is paraphrased as 'the feeling of loving somebody' while the verb is dependent on the noun if defined as 'to feel ~ o v e ' . ' ~ Consequently, tlie semantic dependency criterion does not seem to be completely wrong as everyone might intuitively prefer one paraphrase over the other. Since it is too subjective in its interpretation, however, it cannot serve as a valuable indicator for the direction of derivation. Another semantically based conception of conversion is Dirven's cognitive approach to conversion, further developed by Schrnid (2005). Dirven (1999: 277) regards conversions as metonymically induced, in that one component of a certain action schema stands metonyniically for another component of the same schema. For example, the English verb to clean has a case frame that contains the semantic roles of "an agent, a patient, and possibly an instrument, a manner and a result" (Dirven 1999: 277). The corresponding adjective clean denotes the role 'result', which seems to be tlie most salient element in the action schema. If the adjective converts to the verb to clean, the result is regarded as standing metonymically for the whole action schema. Note that this approach involves directionality, as the adjective is seen here as the basis for the verb conversion. Ungerer (2002: 560) and Schmid (2005: 198) discuss various metonymy-based groups of conversions that display different degrees of entrenchment of one or both conversion partner(s). There are especially two groups which are also distinguishable in terms of directionality: The first group is characterized by deeper entrenchment of "the source item of the conversion" in comparison to the "target item" (Ungerer 2002: 561). According to Schmid (2005: 200), this is the case when the shared conceptual substance or basic concept can be attributed more or less uneq~iivocallyto a certain concept type. En. to juther, for example, is seen as a denominal verb because the concept b~ I t IER is a typically nominal concept and thus is

Lexicul n~otivationas bidirectional

31 1

better entrenched than the corresponding verb to ,father, where the agent tnetonymically stands for the whole action. Other concept types belonging to this category are, e.g., deverbal nouns where an action stands for the result of the action (cut 'result of cutting') or deadjectival verbs where the resulting quality stands for an action (to e m p y 'to make empty'). Thus, for these cases, the direction of derivation, viz. their motivation does not seem to represent any problem. The second group, on the other hand, consists of conversion pairs where both the source and the target item are equally well-entrenched. This pastitularly holds for emotions (101.e - /() love), mental states (doubt - to h u h / ) , and modal concepts (try - to /r,v), hence relatively abstract notions. Here, the respective concepts are not so easily attributable to certain types because they are more open to different ones. Conseqi~ently,the direction of derivation cannot be determined with certainty. Schmid (2005: 202) sees the reason for this difficulty in the fact that the corresponding abstract schemas are not structured enough to be suitable as frames for metonymic processes. To sum up, Dirven's approach as well as Schmid's first group of conversions support a unidirectional point of view. While for Dirven questions of directionality do not seem to play a role at all, Schmid argues that some concepts are more typical for certain bases than others. However, Schmid's argumentation is no less sul7jective than Marchand's criterion of semantic dependency. Although Schmid's decisions about source and target concepts probably correspond in most cases to peoples' intuitions, they do not rely on objective criteria. For example, there is no reason why the noun C I A I , in Schmid's eyes derived from the corresponding verb to cut, could not be conceived either as the basis for the verb. We could perfectly well consider the noun cut as typically nominal and define the verb to cut as 'to produce a cut'. The same criticism holds for the notions of entrenchment and salience. It is not yet clear how the notions of salience and entrenchment can be objectified and in how far they differ between cultures and individuals. Another reason why the metonymy-based approach to conversion has to be considered with caution is the fact that even if some kinds of metonymic conversion processes are rather impsobable," we cannot foresee which ones might one day take place between the constituents of a schema. Besides, many metonymic processes have proved to work bidirectionally, as in PART-WI-IOLE and WIIO1,tI-PAR7 metonymies.

Apart from these critical considerations, however, Schmid's approach is convincing as lie recognizes that tliere are different degrees of how tlie direction of conversion can be determined.

4.4. Formal criteria As far as formal criteria are concerned, the two important ones by Sanders (1988) and Aronoff (1981) will be discussed here.18 Sanders' indicator of the direction of conversions (or, in his terminology, of zero-derivation) is called tlie "Overt Analogue Criterion" (OAC). Its "restricted" version is described as One word can be derived from another word of the same form in a language in the language where the same deriva(only) if there is a precise analog~~e tional function is niarked in the derived word by an overt (nonzero) form. ( 1 988: 160).

For the conversion pair unswer and to unswer, this would mean tliat tlie noun has to be considered as derived since there is an overt analogy, viz. to unnounce +19 unnouncement, where the noun is marked by a suffix and is thus forlnally more complex (Sanders 1988: 156). For love and to love, the decision remains difficult because there are not Inany overt processes of suffixation in the domain of feelings. However, in line with pairs such as lo au'n?ire au'lnir-ution and to uppreciate --t uppreciution one could consider tlie verb as basic. As far as the OAC is concerned, tlie author himself takes a critical stance. Tlie first point Sanders mentions is tliat, if applying the criterion, one has to accept that conversionlzero-derivation is dependent on overt derivation, i.e., as a marked form of regular suffixation. Tlie author considers this as "seriously questionable" due to tlie generality and productivity of the process at least in English (Sanders 1988: 165). Apart from various quantitative and qualitative problems with reasoning based on analogy in general, Sanders admits tliat the litnits of the applicability of tlie OAC are reached when there are analogous overt derivational processes for both directions, as in En. cover and to cover." So, even if the OAC might be helpfill for some conversion pairs - those which parallel overt processes it cannot apply to all of them and is thus not a sufficient solution to the directionality problem. Another forrnal criterion is tlie "distributional" argument forniulated by Arotioff (1 98 I) in his generative work on word-formation:

Lexical motivution as hidirectionul

3 13

If ... a class of words ,l' is derived from another class of words Y, then for every word x, in .Y there shoirld be listed a correspondingy, in Y, but not vice versa. ( 1 1 5 )

For pairs such as En. ornamenf and to ol-~~unient, this rule means that the verb has to be tlie derived unit since we find many nouns ending in -ri~ent which do not have a corresponding verb, e.g., element - *to elenzent but hardly any verb in -merit without a corresponding noun." There is no doubt that tlie distributional argument may be a valuable indicator of the direction of derivation in cases where tlie number of occurrences of one word category is significantly higher than the number of words belonging to the other word category. However, just as with the frequency criterion, the question arises how big tlie difference between the numbers of occurrences must be in order to be meaningful. The criterion cannot work whenever tlie number of derivational partners of both word categories is more or less equal. In any case, tlie distributional criterion can provide us only with a relative probability of direction but neither with an absolute certainty nor with a proof of its psychological reality (Rainer 1993: 52). Taking into account all four kinds of criteria discussed in the preceding subsections, all of them have the status of "bloBe Hilfskonstruktionen" [mere auxiliary constructions] (Rettig 198 1 : 169) to the effect that none seems to be absolutely suited to solve the directionality problem: The historical criterion cannot be used in synchronic analyses at all, frequency and distributional criteria are only relative indicators under specific circunistances, semantic criteria are too subjective, and the overt analogue criterion can account only for a small number of cases. As far as our example of love arid to love is concerned, only formal criteria might hint at the derivational origin, but Sanders's criterion alone does not provide enough evidence as to be sure of the direction of derivation. Besides, most criteria have so far been applied to conversion only since it is the most common process which results in directional ambiguity. It is doubtful whetlier the criteria, especially the formal ones, can be usefill for instances of polysemy at all. The question of hierarchy is another problem. As one can normally apply more than one criterion to an alnbiguous case, and as the different rules can lead to contrary decisions, it is not clear which criterion has to be regarded as most decisive. Formal criteria might be more important because of their greater objectivity whereas semantic factors better reflect intuitive preferences for one direction over the other.

TIILIS,we can only conclude tliat even if we use the above-mentioned criteria for some unclear derivational pairs, tlie directionality problem of derivation and motivation as a whole still remains unsolved. In tlie next section an alternative solution for tlie problem is suggested from a cognitive point of view: tliat is to account for lexical motivation as a bidirectional phenomenon.

5. A bidirectional approach to lexical motivation Bidirectional conceptions of either lexical motivation or derivational relations are of course less common than unidirectional ones. Within tlie generative framework, several authors have proposed non-directional or bidirectional?' redundancy rules for tlie areas of phonology and morphology (e.g., Leed 1970, de Cliene 1975, .lackendoff 1975). Tiersma (1978) proposes non-directional relational rules that apply to allomorpliy and lead to "bidirectional leveling" (1978: 67). Lieber adopts a non-directional position on conversion altliougli she considers directional analyses not as "impossible" but as "~~ndesirable"(1981: 125). In her own morphological framework she proposes redundancy rules instead of affixational approaches to account for conversion. However, as she considers a semantic analysis to be independent of her own purely syntactic one, she does not reject Marchand's semantic dependency criterion, arguing "that the sernantic analysis can involve directionality without arguing in any way against the non-directionality of the syntactic analysis" (1981 : 129). She, thus, adopts an intermediate position with respect to directionality. Becker (1990) also represents an intermediate view. On the one hand, he accepts non- or bidirectional relations between words, e.g., Germ. pjlegen 'to care for' tt Pj'ege 'care'. On the other hand, lie proposes a markedness theory tliat accounts for tlie so-called "asymmetrical relations", e.g., Germ. glucklich 'happy' t Gluck 'happiness' (Becker 1990: 5 1-52). Ford and Singh (1984) claim bidirectionality of word-formation processes, which is in their eyes justified by language acquisitioll and diachrony. Leech (1 974) defines lexical derivation as bidirectional because of the phenomenon of back-formation, where the word-fornlation process takes place from the more complex to the simpler form, i.e., into the unexpected direction: Since historically the derivation can move in either direction. it is arguable that lexical rules should be formulated (for the purpose of representing lin-

Lexical rnolivulio~us hidireclio~ul 3 15

guistic competence) in a bi-directional form, the predominance of derivations from the silnpler to the more complex form being regarded simply as a matter of historic probability. (1974: 224). As far as the phenomenon of motivation itself, or, as he calls it, of "motivatability"" is concerned, Rettig ( 1 98 1 ) strongly defends a bidirectional point of view, treating both lnotivational partners as equal: Sobald beide Einheiten gleichbel.echtigt nebeneinander in der Sprachkenntnis vorkommen, kann man keine allgemeine Aussage iiber die Richtung der blotivierbarkeit meht- machen. Man muR vielmehr annehmen, dal3 eine Motivierung sowohl von der einfacheren Einheit zur komplexeren als auch umgekehrt verlaufen kann. Motivierbar ist sowohl poniniier durch pomme als auch ponzme durch poninzier ( 1981 : 167, my emphasis) [As soon as both units coexist equally in linguistic knowledge, it is no longer possible to make a general statement about the direction of motivatability. Rather, one has to assume that a motivation can run from the simpler to the more complex unit as well as the other way round. Ponimiev is motivatable by pomme as well as potnme by pommier my translation and emphasis.] -

From a Cognitive Grammar perspective on conversion in English (mainly based on Langacker), Twardzisz (1997: 60) regards motivation as multidirectional: Motivation ilnderstood in this way [i.e., as semantic sanctioning from many sources at the same time] does not only go along one line from a phonologically simpler form to a more con~plexone but instead is viewed as a complex multi-directional sanctioning phenomenon.

Similarly, as already described above, Schmid (2005) implicitly accepts bidirectionality for conversion pairs expressing emotions and mental states as directional metonymic processes cannot be established in these cases (cf. 4.3). From a bidirectional point of view, the hitherto proble~naticexamples in (8) to (10) would be motivated reciprocally:

(8a) Germ. Tug 'period of 24 hours' (9a)

En. love

+-+

t ,

Tug 'time of dayliglit'

to love

(10a) Fr. kternel 'eternal'

t ,

kcernilk 'eternity'

3 16

Rirgit Urnhreit

As tlie double-sided arrows indicate, both directions of lnotivation are possible: Tug in tlie meaning 'period of 24 hours' can be explained on the basis of tlie meaning 'time of daylight' as an extension of the temporal period of reference. The latter meaning can be motivated by 'period of 24 hours' undergoing a restriction of the period of reference. Similarly, as already indicated in subsection 4.3, love can be derived from the verb if paraphrased as 'the feeling of loving sb.', or tlie verb can be motivated as 'to feel love', i.e., on the basis of the noun. Basically, both directions are fine. The same reciprocal motivation holds for (lOa). In addition, the bidirectional understanding of motivation is intended to account for the cases hitherto analyzed as unidirectional: (5a) Fr. pommier 'apple tree'

++

(6a) Germ. Schiinheit 'beauty'

++

por~zwze'apple' schiin 'beautiful'

Altliougli most speakers would probably prefer to motivate pornr~~ier by pon~nze,it sounds perfectly acceptable to motivate the word pair the other way round, i.e., ponzme as 'fruit that grows on apple-trees' (cf. the quotation from Rettig 1981 above)." As far as (6a) is concerned, again both directions of motivation make sense since semantic identity seems especially apt to be analyzed bidirectionally: schiin can be explained as 'sho\ving beauty' as well as Schunheit can be paraphrased as 'the state of being beaut i fu 1'. A bidirectional conception of lexical motivation consequently has tlie advantage of allowing a uniform analysis of all types of motivational relations, independently of tlie difference in formal complexity or in semantics of their members. This approach does not intend to say that tlie concrete processes of lexical innovation are bidirectional but rather claims that speakers can motivate words in both possible directions. Evidence for bidirectionality is provided by cognitive processes belonging to different linguistic areas, such as the mental organi~ationof words. second language acquisition, and specific experiments on lexical motivation carried out within tlie framework outlined in section 2. These three kinds of evidence are presented in tlie following subsections.

5.1 . Evidence from cognitive worcl-family organization According to Augst, ~ i a t ~ vspeakers e possess a so-called "synchronic etynol logical competence" (1975: 176-177). This means that all language users are able to analyze co~nplexwords and to establish synchronic relations between words both for~nallyand semantically because they have an implicit or even explicit knowledge of word-family organization. A wordfamily is understood as a "set of words derived from a given stem by means of either compounding ... or derivation ..." (Schreuder and Baayen 1997: I ~ I ) . "Thus, the English word-family-filh consists of words such as lo f i h, ~jisher, .fishery,,fi.5 hemcxrr, and pearl fithery (Schreuder and Baayen 1997: 133). Augst's assertion is corroborated by cognitive revelations, which show that the notion of word-family is a "wesentliches Strukturmerkmal des Sprachbew~~sstseins"[an essential struct~~ral characteristic of linguistic consciousness] (Blank 2001). TIILIS,it plays an important role for the mental organization of lexical ~lnits. A word-family is by definition hierarchically organized since there is always a -'core word" (cf. "Kernwort" in Augst 1975) or "stem" (Baayen and Schroder 1997) that constitutes the base in relation to the organization of all other family members, e.g., j k h in the above-mentioned example. Psycholinguistic experiments, however, show a diversified picture about tlie cognitive access and representation of derived words. Studies conducted by Tyler et al. (1993) and Marslen-Wilson et al. (1994) have resulted (with some restrictions) in significant effects for different constellations of simple and derived words that were morphologically and selnantically related: Suffixed and prefixed words prime their stems and vice versa; prefixed words prime suffixed words and vice versa; finally, prefixed words prime each other, but suffixed words do not prime each other. According to the authors, this is due to a so-called "inhibitory l i n k between suffixes attaching to the same stem (Marslen-Wilson et al. 1994: 18-1 9). Thus, we can say that

i.

There are not only links between each of the word-family members and their base word but also between the derived members, at least between prefixed words and between prefixed and suffixed words if tlie results obtained by Tyler et al. and Marslen-Wilson et al. are generalizable.

318 ..

11.

Birgit Unibreit

The links between the base of a word-family and each of the derived members as well as between the derived members themselves are bidirectional.

In addition, word-family size influences reaction time in lexical decision tasks and frequency ratings (Schreuder and Baayen 1997): the larger the word-family, the shorter the reaction time in lexical decision tasks. Subjects recognize more easily that a nionomorphemic target is a well-formed word of the respective language." In frequency ratings, simple nouns are assigned a higher frequency if the size of the word-family is bigger. Thus, word-families, organized bidirectionally or even multidirectionally, are a central device for the cognitive representation and access of words.

5.2. Evidence from L2 acquisition

Learners of a foreign language often encounter the problem of unknown vocabulary, both in written texts and in oral communication. But the unknown word does not always disturb the comprehension of the entire text or conversation. Sometimes, the meaning can be inferred from the context. At other times - and this is the case that interests us here - it can be inferred from familiar words belonging to the s a n e word-family. Denninghaus calls these words "potentieller Wortschatz" [potential vocabulary] and argues: Im

Gegensatz zur linguistischen Betrachtungsweise, die in zahlreichen der

[...I Paare Wortbildungsmodelle sieht, wobei der eine Partner originar und der andere Partner abgeleitet ist, sind bei der Betrachtung der Paare als ~ortbezugsmodelle"beide Partner gleichwertig. Der Aufbau des potentiellen Wortschatzes vollzieht sich nach beiden Richtungen, vom originaren Wort zum abgeleiteten i~ndumgekehrt. (1976: 7, my emphasis) [Contrary to the linguistic approach that considers marly of [. . .] the pairs as word-formation patterns where one of the pair members is the original and the other one is the derived form (from a perspective that regards the pairs as word reference pattenis), both pair members are equal. The formation of potential vocabulary takes place in both directions, from the original word to the derived one and vice versa. my translation and emphasis.] -

Denninghaus thus states that the inferences of potential vocabulary runs in both directions, from the derived to the "original" word and vice versa, as both of the derivational partners are equal. MeiDner (1989: 377) agrees

with this opinion in that Iic distinguishes a "zuriickfuhrender Typ" [backleading type] of formation as in (I I), from a "erweiternder Typ" [expanding type], as in (12).:' (1 1 ) Sp. tranquiliu%u"calmness'

t

frcxnyuilo 'calm'

( 12) En. conscious +- con,~ciousness

These two directions of inferences, wliicli probably the ma-jority of L2 learners makes use of, speak in favour of bidirectionality.

5.3. Evidence from speaker judgements on motivation In systematic studies on lexical motivation realized in a research project at liibingen ~n iversity,'" ordinary speakers have to decide whether and by wliicli other unit a lexical unit is motivated. This experimental approach is justified by the synchronic etymological competence of native speakers claimed by Augst (1975). In addition, it provides us with evidence on the average linguistic consciousness of the language users. Introspection of linguists, by contrast, would certainly reveal different motivations to those found by non-linguists. Another advantage is that by asking a certain amount of informants for each stimulus (nornially between 20 and 30), the problem of conflicting motivations is avoided, as, in most cases, one motivatiolial unit is more frequently chosen than the other possible ones. On these grounds, the most salient motivational partner for tlie corresponding sti~nuluscan be classified (Marzo and Rube 2006: 154). The studies on motivation reveal that speakers motivate given stimuli mainly by forrnally and selnantically simpler but also by more or equally co~nplexwords. Selected data from two experiments, a pilot study with German stimuli and a study with Italian stimuli will illustrate this. In the German experiment "Alpha", the infhrmants had to answer the following question for fifteen randomly selected German words: "Woher kommt das angegebene Wort?" [Where does tlie given word come from?]. To answer this question, tlie respondents were able to choose one out of four possibilities and provide the respective source terms of the derivation.

1. von einem anderen Wort derselben Wortfamilie 2. von zwei anderen Wortern 3. von einer anderen Bedeutung desselben Worts

320

Birgit Onibreit

4. [I. 2. 3. 4.

von keinem anderen Wort oder keiner anderen Bedeutung from another word of tlie same word-family from two otlier words from another meaning of tlie same word from no otlier word or meaning]

Answer (1 .) accounts for all kinds of formally and semantically related words; answer ( 2 . ) accounts for compounds; answer (3.) indicates polysemy; and informants who clioose (4.) consider tlie stimulus to be not inotivated at all. Note that all kinds of informant's motivations, be they simpler or more complex than the stimulus, were accepted provided that a formal and a semantic relation in accordance with the definition of motivation outlined in section 2 could be found. Table I . Percentage of answers not representing a ~notivationalbase in experiment "Alpha"

stimulus + meaning Krone

'dental crown'

answers representing a motivational basis Krone 'crown of a king or queen'

answers not representing a motivational basis Kronung kronen Krone 'crown of a

others

opaque

tree' unmijglich

'impossible'

unferfor(ier11 'to ask too little

from sb.'

kijclt eln

8.8% vermcigen Moglichkeit 8.90/0 Unterforderung zmter/Forderung unten/Forderung uherfordern ,fordern 8.8% Koch

'to simmer' 1.8% average of the stimuli in tab. I average of all 15 stimuli in Alpha

7.1 %

3.0%

1.7%)

8.8%

1.8%

14.3%

Lexicrrl tnotivution u.s bidirectional

32 1

Table 1 presents the results for four of the fifteen stimuli. The first column shows the sti~nulustogether with the meaning in wliicli it was given. The second column contains the percentage of answers wtiicli represent what is intuitively supposed to be a motivational base lor formal andlor semantic reasons, while tlie third colun~nshows tlie percentage of' answers that are sitpposed to be Inore complex than the stimulus. The column named "others" subsurnes all kinds of answers tliat could not be attributed to column two or three, ~nostlybecause they were not interpretable. The last column named "opaque" contains the percentage of answers where 110 motivation was found, including the last answer possibilitj presented above [Tlie word comes from no other word or meaning.]. These four stimuli show an average deviation from the expected motivational base of about 7%. This number is even lower (about 3%) with regard to the deviation from the niotivational base of all of the fifteen stimu l i from Alpha, which cannot be presented here due to limits of space. In a later study on 25 randomly selected Italian words, experiment "Gamma", tlie percentage of non-directional answers was a bit higher (cf. Table 2). This time, the question that the (native) participants had to answer was formulated as "Perchd, secondo voi, la parola puo essere uiilizzata nel senso spiegato?" [Why, in 4011r opinion, can the given word be used in the given meaning'?]. 'This formillation differs from tlie previous one given in Alpha ("Where does the given word come from"?), which suggests a diachronic answer. This is of course undesirable in a synchronic study. Tlie answer possibilities were the same as in Alpha, formulated this time as "La parola e connessa ad un'altra parola della stessa farniglia di parole." [The word is related to another word of the same word-family] etc. in correspondence to tlie modified question. Resides, an additional answer type was given, namely "La parola e connessa ad una parola di un'altra lingua." [The word is related to a word of another language]. This allows the identification of loan words. Since this aspect does not interest us in the context of non-basic motivations, they are subsumed under "others" in Table 2." Apart from this small increase, the subdivision of the colu~nnsin Table 2 corresponds exactly to the one made in Table 1. For the stimuli listed in Table 2, the percentage of non-basic motivations is about 21 %. If all of the 25 Italian ctimuli of Gamma are taken into account, the average number of answers tliat do not represent a motivational base is 11.9%. This is still significantly higher than the average deviation from the motivational base in Alpha. How can the higher number of non-basic answers in Gamnia be explained? One reason might be the modified question "Woher komrnt das angegebene Wort?"Where does the

given word come from?], used in Alpha, which suggests a diachronic answer or at least a motivational partner from which the stimulus can be derived for~nallyor semantically. The question used in Gamma "Perche, secondo voi, la parola puo essere i~tilizzatanel senso spiegato?" [Why, in your opinion, can the given word be used in the given meaning?] is less suggestive of a certain derivational source. So, respondents might feel less Table 2. Percentage of answers not representing a ~notivationalbase in experiment

"Gamma" stimulus + meaning grande 'important'

answers repre- answers not rep- others senting a moti- resenting a motivational basis vational basis grande grande 'excellent' 'big, tall' grande 'wellknown' il grande 'adult'

opaque

grandezza 55.9'Yo

buono 'delicio~~s'

huono

14.7%

29.4%

'good, bontir

pleasant'

huono 'good char-

acter' 47.1%

nuovn 'newly arrived'

nuovo 'new'

23.5%

5.9%

23.5%

20.5%

26.5%

novitu nuovo 'unknown,

unfamiliar' vero 'original, authentic'

26.S0/o

26.5%

vero 'true' vero 'real'

veritii

32.4%

20.6%

2.9%

44.1%

40.5%

21.3%

7.3%

30.9%

43.0%

11.9%

5.5%

39.6%

average of the stimuli in tab. 2 average of all 25 stimuli in Gamma

restricted in giving their answers. I-lowever, another reason, which I consider Inore important, is the nature of the stimuli. In Alpha, over 53% of the stimuli were formally co~nplexones while a bit more than 33% were formally simple but selnantically no ti vat able (i.e., cases of polysemy). The

Lexical n~otivationas bidirectional

323

rest of the stimuli are supposedly opaque. In Gamma, 68% of the stiniuli belong to the category of formally simple but semantically rnotivatable stimuli; 4% were supposedly opaque, and only 28% are fortnally complex. While formally co~nplexstimuli can be easily linked to a motivational partner belonging to tlie same word-family (mostly the base word), this is not the case for the formally simple but semantically derived ones." Here, the informants either give a formally different (and more complex) form that is still more salient than another meaning of the same stimulus, or they provide another meaning that is not necessarily more basic. This is evident when coinparing Table 1 with Table 2, the latter of which showing a smaller numerical difference between column 2 and 3 (motivational partners representing a base and ~notivationalpartners not representing a base). Another symptom of the reduced salience of the stimuli presented in Table 2 is their high score of opacity. What this subsection has shown is that there is still a smaller or even an eqilal number of answers that deviate from the expected motivational bases even if more basic motivational partners are the norm. Thus, both directions of motivation seem to be part of a language user's linguistic consciousness. Consequently, a bidirectional conception of motivation is much Inore suitable in the given framework since it can account for all kinds of no ti vat ions found by the speakers: motivational bases as well as equally or Inore cornplex units. However, for some stimuli, there is a high uniformity in what is chosen as a motivational base whereas in other cases it is not always clear if a possible motivational unit is a base or not. Section 6 constitiltes a first attempt to conceptualize these different degrees of directionality as a continuum.

6. Towards a continuum of directionality

As was demonstrated in section 5 , a bidirectional approach to lexical motivation not only allows a uniform analysis of all types of motivational relatio~is,but it is also cognitively more adequate. Yet, there is probably widespread agreement that in many cases an intuitive preference for a certain direction of motivation remains. Even if both ways are theoretically possible, it still seems to be tilore appropriate to motivate pommier by pornnle than vice versa. How can this preference be explained? One solution might be to represent the strength of this tendency towards unidirectionality as a continuuln with a bidirectional centre between two unidirectional poles:"

324

Birgit C'n~breit

pole 1




pole 2

Figzrre 3. Directionality continilu~n

The more certain a direction of motivation appears, tlie closer the motivational pair has to be situated at the respective pole. If the direction of motivation cannot be deterniined, the motivational pair in question is situated in the middle of the continuiun. Coming back to lacobini's cliaracteristics of the motivated unit refined in section 3 , we suppose that either a formal or a semantic discrepancy between tlie motivational partners is sufficient for a directional tendency. Accordingly, the German pair a. Schiinheit - h. schiin, manifesting a formal difference only, should have its place near pole 1 ." Tlie same holds for the pair a. n~ouse'computer device' and h. mouse 'small grey rodent', wliicli is only semantically different but could be assigned to pole 1 on extra-linguistic grounds. If neither formal nor semantic reasons can be found in order to decide about tlie direction of motivation, as in the case of love - to love, the motivational pair can be portrayed as occupying the bidirectional area between the two poles of tlie continuum. It is quite obvious, Iiowever, that such an intuitive placement of motivational pairs on tlie continuuni remains vague and sub-jective. In order to objectify these decisions, speaker judgements could be helpful once more. i 4 As described before (cf. 5.3), the consultation of a fair number of informants leads to lnore representative results than a decision based on introspection. If tlie continuum illustrated in Figure 3 is changed into a scale, the answers of the informants can be given more ob-jective~~:'~

pole 11

I

I

I

I

I

I

4 pole 2

Figzrre I.Directionality scale

Tlie Inore informants choose one direction of motivation, tlie more clearly the corresponding motivational pair can be attributed to the respective

pole. If the informants do not agree on the direction of motivation or if they are not able to make any statement about it, tlie motivational pair will be situated in the middle. 'The speaker-based investigation of motivation will also allow us to verify the modified "growth"-criterion first formulated by lacobini: Is fortnal growth between tlie motivational partners enougli in order to decide tlie direction of motivation or not? Does tlie same result hold for purely semantic discrepancies? Are there clearly bidirectional pairs, or do individual speakers always prefer one direction of motivation over the other? Speaker judgements will also provide infor~nationon the relation between formal and semantic devices and directionality: Which formal relations, which semantic relations, and whicli combinations of both of them allow for a determination of the motivational direction and which fail to do so? All these questions will Iiopefully be answered soon - after having asked ordinaly language users. In addition, speaker judgements can also provide evidence for Schmid's (2005) two groups of conversions (cf. 4.3), which differ in their determinability of direction for semantic reasons and would thus seen1 to fit well into a scale of directionality. The results will show if Schmid's cognitively based distinction can be confirmed from the point of view of language users. Furthermore, these metalinguistic judgements on directionality will be compared with an indirect method of testing directionality effects, namely with a priming task similar to the ones by Marslen-Wilson et al. (1994; cf. 5.1). The comparison of these two different methods for investigating the directionality of lexical motivation will show whether the results from conscious and unconscious tests on directionality converge or diverge - or, to put it differently, whether the metalinguistic speaker judgments on the direction of motivation are contirmed by priming tasks tliat reveal the cognitive organization and activation of motivational relations.

7. Conclusion The preceding sections have higlilighted tliat tlie traditional unidirectional conception of lexical motivation is problematic when defined as a relation between lexical units that always manifests itself both formally and semantically. Thus, certain lexical devices, e.g., conversion, truncation, and polysemy, are ambiguous as far as their direction of motivation is concerned. Even if different criteria for determining tlie direction of derivation are applied, there are no means to decide about all unclear cases of motiva-

tion. Thus, a bidirectional understanding o f motivation is better suited t o account for all kinds of motivational pairs. Evidence for bidirectionality comes from cognitive approaches, such a s the mental organization o f wordfamilies, second language acquisition, and from speaker judgements on motivation. Yet, nlotivational pairs tend in different degrees towards unidirectionality, which, in fact, can be visualized on a continuum of bidirectiotiality. It can be hypothesized that the preference for unidirectionality correlates with formal o r semantic differences between the members of motivational pairs and that the preference for bidirectionality, on the other hand, correlates with little formal and semantic differences between the pair members. Specific speaker judgements on the direction o f motivation will liopefi~llysoon allow Inore precise statements about the directionality continuum.

Notes I. 2.

3.

4.

5.

6.

7.

The arrowhead running from right to left stands for "is motivated by". Ullmann (1962: 92) cites the case of En. blue hell, which is "a transparent colnpound and at the same time a metaphor based on the bell-like shape of the flower". C1.. Rettig (1981) for a similar conception of lexical motivation. For Panther and Radden (2004) and for Lehmann (2007) the combination of morphological and semantic motivation is one of the possible types of linguistic motivation. Purely formal si~nilariticsbetween words without any semantic connection, e.g., Fr. porter 'to carry' - porte 'door', and purely semantic relations which do not bear any formal similarity, e.g., Fr. vigne 'vine' - raisin 'grape' (Koch and Marzo 2007: 267), are not considered as motivated. Note that this definition distinguishes polysemy from homonomy: While two polysemous lexical units are linked semantically and can be attributed to the same lexeme, holnony~no~~s lexical units do not manifest any semantic relation and constitute two autonomous lexemes. A systematic cross-classification of all possible formal and semantic relations leads to a two-dimensional grid for the description of lexical motivation which is presented in Koch (200 I : I 160) and Koch and Marzo (2007: 268). At this point, the one-sided arrow fi-om right to left can be specified as standing for unidirectional motivation.

Lexicc~lmo/ivation as bidirectional

8.

9. 10.

l I.

12. 13. 14. 15.

16. 17.

18.

19. 20.

2 1.

22.

327

The term "derived word" is used in a general sense here, namely as a word that is created by a process of lexical innovation. All motivated nits can consequently be regarded as derived. The bidirectional arrow followed by a question mark indicates that the direction of motivation is unclear. Contrary to e.g., Vennemann (1972) and Bergenholtz and Mugdan (1979a, 1979b), conversion is here regarded as a derivational process and is consequently based on motivational relations. Cf. also Scheidegger (I 98 1 : 54-56). For a first overview cf. lacobini (2000). Cf. Bergenholtz and Mugdan (1979a), Scheidegger (1981: 54), and Rainer ( 1993: 5 I) for the same point of view. Eaton is a laudable exception as she mostly specifies the meanings of the given word forms. Cf. also Lieber (I98 1 : 129), Cruse (1986: 132-133), Olsen (1986: 122-123), Fleischer and Barz (1 995: 2 10-2 1 I), Rainer (1993: 5 1-52), and, despite some critical comments. Plag (2003: 109). For a critical view of Marchand's criterion cf. also Bergenholtz and Mugdan ( 1979: 160), Sanders (1 988: 173- 174), and Becker (I 990: 49-50). Dirven states that conversion processes "are mainly applied to non-human participants such as patient (as object), instrument, manner, locative and attribute", whereas human participants that typically occupy agentive and dative roles are rarely the basis for conversion processes. He explains this observation with the cognitive principle of anthropocentrism: "Since human beings are already the focus of attention in most linguistic structures, they cannot be focused upon again in the conversion process, at least not in the agent or dative roles. which are prototypically I?uman roles" (1999: 285). Other formal criteria. e.g., the ones formulated by Marchand (1 964) and Kiparksy (1982) are of an even more limited applicability as the ones mentioned here and are therefore omitted in this article. The arrow running from left to right means "is the basis for". Following Sanders (1988: 167). lo cover can be considered as derived from cover in analogy to chain -t to enchain or, vice versa, in analogy to to cleave -+ cleaver. According to Aronoff ( I98 1 : 1 17) except for two cases: to,foment and to den7ent. Bi- and non-directionality are not always distinguished as neatly as in Balteiro (2007). who uses the term "non-directionality" where a derivational relation between conversion partners is denied, and "bidirectionality" for nonunidirectional approaches that are based on derivation. In de Chene (1975), Tiersma (1978), and Becker (1990) for example, it is not always clear if biand non-directionality are used synonymously or stand for different concepts. This is why non-directional accounts are also discussed in the present paper.

Rettig (198 1: 75) creates this term in order to underline the metalinguistic aspect of speaker-based approaches on lexical motivation. Section 6 will show how this implicit contradiction can be resolved. Schreuder and Baayen use the term "morphological family" for what is traditionally referred to as "word-family". Note that the definition for word families cited above includes compounds formed on the basis o f t h e base word of the respective word-family, but not inflectional relations. Similar effects are obtained by de Jong et al. (2002) for certain types of compounds, while for others they find a so-called "position family frequency effect". De Jong et al. (2003) show that word-family size influences various types of adjectives to different degrees. "Wortbezugsniodelle" [word reference patterns] are detined by Denninghaus (1976: 5) as regular relations existing between two types of words as to their formal and semantic structure, for example with respect to the position of the root morpheme or the use of certain affixes. Word reference patterns often appear in whole series of two or more words related among each other. Cf. also Liibke (1977) who implicitly shares this view. Project B6: "Lexical Motivation in French, Italian, and German (L,exiTypes,,,)", Collaborative Research Centre 441: Linguistic Data Structures: On the Relation between Data and Theory in Linguistics: http:/lwww.sfb441.unituebingen.delb6. This is the reason why the average number of "other" answers in Table 2 is higher than for Alpha, which did not include this question. For more details cf. Marzo (2007). I owe this suggestion to Verena Rube. The numbering of the poles is insignificant. The important point is that the two poles are directly opposed with respect to the direction of motivation. Cf. Bergenholtz and Mugdan (1979a) and (1979b) fbr a first speaker-based test on directionality. A similar proposition of a directionality scale for conversions by CI-occo Cialeas (1997) is not useful for the approach presented in this paper. Firstly, being rooted in the framework of Natural Morphology, it is typologically oriented. Secondly, it aims to determine the direction of derivation u priori, whereas in the speaker-based perspective developed in section 6, the assignment of a motivational pair on the scale is only made a posteriori according to the average answers of the informants.

References Adams, Valerie 1973 An lntroduclion to Moden7 English WordTfbrniation. London: Longman. Aronoff, Marc 198 1 Word Fornwtion in Generative Gramnlar. Cambridge, MAlLondon: MIT. Augst. Gerhard 1975 Untc~vuchungenzun1 ,2.1orpherninver7turder deulschen Gegenwartssprrrche. Tubingen: Nan. Balteiro, Isabel The Directionulity of' Conversion in English. A Dia-Synchronic 2007 Stugv. Frankfurt am Main: Lang. Hecker, Thomas 1990 Anulogie utld tnorphologische Theorie. Munchen: Fink. Bergenholtz, Henning, and Joachim Mugdan 1979a Eir7j2hrung in die Morphologie. Stuttgart: Kohlhammer. r und Wortarten. In Deulsche Ge1979b 1st Liebe primar? ~ b e Ableitung ger~wurt.sspruche.Ent+vicklmr~gen,Enlwiirfi, Di.skussion, Peter Braun (ed.), 339-354. Munchen: Fink. Biese, Yrjii. M. 194 1 Origin arid develop~nentof conversions in English. Ar/nule.v Acaderniue Scientiurum fit/t~icneB 4512, 1--495. Blank, Andreas 200 1 Einfuhrung in die lexikalische Semantik fur Romanislen. Tiibingen: Niemeyer. Bybee, Joan L. Morphologv: a stud) of'the relution hetween meaning und,form. Am1985 sterdam1Philadelphia: Benjamins. Crocco Galeas, Grazia 1997 Me/qjiora Mo~:fblogicii. Saggio di niorfblogiu naturule. Padova: Unipress. Cruse, D. Alan Lexical Semantics. Cambridge: Cambridge University Press. 1986 De Chene, Brent The Treatment of Analogy in Formal Grammar. In Pupeos,fron? lhe 1975 eleventh regioncrl meeting. Chicago Lir7guistic Society, April 18-20, 1975, Robin E. Grossmann, L. James San, and Timothy J. Vance (eds.), 1 5 2 1 6 4 . Chicago: Chicago 1,inguistic Society. -

De Song, Nivja H., Laurie B. Feldman, Robert Schreuder, Matthew Pastizzo, and R. Harald Baayen 2002 The Processing and Representation of Dutch and English Compounds: Peripheral Morphological and Central Orthographic Effects. Brain and Languugc 8 1, 555-567. De Song, Niv-ja H., Robert Schreuder, and R. Harald Baayen 2003 Morphological resonance in the mental lexicon. In: Morphological Structzire in Language Processing, R. Harald Baayen, and Robert Schreuder (eds.), 65-88. BerlinINew York: De Gruyter. Denninghaus, Friedrich 1976 Der kontrollierte Erwerb eines potentiellen Wortschatzes im Fremdsprachenunterricht. Praxis d m neusprachlichen Unlerrichts 23, 314. Dirven, Rene 1999 Conversion as a Conceptual Metonymy of Event Schemata. In: Metonyniy in Language and Thozrghl, Klaus-Uwe Panther, and Giinter Radden (eds.), 275-287. AmsterdamIPhiladelphia: Benjamins. Eaton, Ilelen S. 1967 An English-French-Gert~~an-~Cpa~~i~rh word .frequency dictionary. A correlation qf the ,fir-st six thousatid words in four single-language Jrequency 1i.rt.s. New York: Dover. Fleischer, Wolfgang, and Irmhild Barz 1995 Worthildung dcr c l ~ ~ ~ ~ t sGegen~~art.s.sprache. chen Unter Mitarheit von Maricrnne SclirBder. 2., durchgesehene und erganzte Auflage. Tiibingen: Niemeyer. Ford, Alan, and Rajendra Singh 1984 Remarks on the Directionality of Word Formation Processes. In Proceedings o f the First Eastern Stute Confirence on Linguistics, 205213. Columbus, Ohio: The Ohio State University. Gauger, Hans-Martin 197 1 Llurchsichtige Wiirlel-. Zlrr Tl~eorieder Wortbildung. Heidelberg: Winter. lacobini, Claudio 2000 Base and direction of derivation. In hlorpholo~/Morpholo&. An Internutionul Hunu'hook on Inf7ec/ion and Derivution/Ein interncr/ionales Handhuch sur Fle.~ionund Worthildung, Geert Booij, Christian Lehmann, and Joachiln Mugdan (eds.), 865-876. Berlin/New York: De Gruyter. Sackendoff, Ray S. Morphological and Semantic Regularities in the Lexicon. Langztuge 1975 51,639-671. Katamba, Francis, and John Stonham 2006 Morp/i(>log)i.Second Edition. Basingstoke: Palgrave Macmillan.

Lexical rnofivcrtiot~a.s bidirectional

33 1

Kiparsky, Paul 1982 Froin Cyclic Phonology to Lexical Phonology. In The structure qj' phonological repre.sen/ution.s. Part I, Harry van der Hulst, and Norval Smith (eds.), 13 1- 175. Dordrecht: Foris. Koch, Peter 200 1 Lexical typology from a cognitive and linguistic point of view. In Language Typology and Lunguage Universals/Sprachtypologie und sprachliche Universalien/Ln tjyologie &s lungues eet Ies universazrx lin,uui.stique.s. An Internatioriul HaridbooUEin internuiiona1e.s Handhuch/Munuel international, Martin Haspelmath, Ekkehard Konig, Wulf Oesterreicher, and Wolfgang Raible (eds.), 1142-1 178. Berlin/New York: De Gruyter. Icoch, Peter, and Daniela Marzo 2007 A two-dimensional approach to the study of motivation in lexical typology and its first application to French high-frequency vocabulary. Studies in Lungzruge 3 1, 759-29 1 . Leech, Geoffrey 1974 Sernuntics. Harmondsworth: Penguin. Leed, Richard L. 1970 Distinctive Features and Analogy. Lingua 26, 1-24. Lehmann, Christian 2007 Motivution in Ianguugc. Atlen~ptat a systemati-7ation. In: Spruchliche Motivatiotz: Zur 1nterdqenu'c.n~von lnhalt und ,4us&uck, Peter Gallmann, Christian Lehmann, and Rosemarie Luhr (eds.), 105-140. Tubingen: Narr. Lieber, Rochelle 1981 On the organization c?J'tlielexicon. Bloomington: Indiana University Linguistics Club. Liibke, Diethard 1976 Der potentielle Wo~tschatzim Franzijsischen. P ~ L M&S~ Stieusprachlichen Unterrichls 3 1, 372-379. Marchand, Hans 1964 A Set of Criteria for the establishing of derivational relationship between words unmarked by derivational morphemes. Indogermunische Forschlrngen 69, 10- 19. Marslen-Wilson, William D., Lorraine K. Tyler, Rachelle Waksler, and Lianne Older 1994 Morphology and meaning in the English mental lexicon. Psychological Review 1 0 1 , 3-33. Marzo, Daniela 2007 What metaphor, metonymy and polysemy can tell us about scales of diagrammatic transparency. Talk, 10th International Cognitive Linguistics Conference, Krakow, 15.-20.07.2007.

Marzo, Daniela, and Verena Rube 2006 What do you think where words come from? Investigating lexical motivation empirically. In The VIII-th International Clonfkrence Cognitive Modeling in Linguistics Proceedings. Vol.1, Valery Solovyev, Vera Goldberg, and Vladimir Polyakov (eds.), 152-161. Kazan: Kazan State University. MeiRner, Franz-Joseph 1989 Grundwortschatz und Sprachenfolge.,franziisisch heute 4, 377-387. Olsen, Susan Wortbildztng im Deutschen. Eine Einfuhrztng in die Theorie der 1986 Wortstruktur. Stuttgart: Kriiner. Otto, Walter F. (ed.) Platon. Surntliche CVerkc. Vol. 2. In der Ubersetzung von Friedrich 1964 Schleiermacher mit der Stephunzrs-Numerierung. Hamburg: Rowohlt. Panther, Klaus-Uwe, and Giinter Radden (eds.) 2004 Studies in Linguistic ,bfoti~,alion.Berlin/New York: De Gruyter. Plag, lngo 2003 Word-Formation in English. Cambridge: Cambridge University Press. Rainer, Franz Spunische Wortbildungslehre. Tubingen: Niemeyer. 1993 Rettig, Wolfgang Sprachliche Motivution. Zeichenrelu/ionen von Laulform und Bedeu1981 lung urn Beispiel,fr.anzci,~i~scher Lexikoneinheiten. Frankfurt am Main: Lang. Sanders, Gerald 1988 Zero Derivation and the Overt Analogue Criterion. In: Theorelical Michael I-lamniond, and Michael Noonan (eds.), 155il4(>rpholo~)~, 175. San DiegoiNew York: Academic Press. Saussure, Ferdinand de Cours u'c lingz/istiqlre gc;nkrule. Public! par Charles B u l [ ~et Albert 1966a Sechehuye uvec la collaboration lie Albert Riedlinger. Paris: Payot. Cozrrse in Genertrl Lingui.stic.s. Edited by Charles Bully and Albert 1966b Sechehuye. In colluborution with Albert Riedlinger. Tran.sluted with an introduction and notes by Wade Buskin. New YorkIToronto: McCraw-Hill. Scheidegger, Jean Arbitruire el niotivation en jrun~uiset en alleniund. Examen critique 198 1 des thBses de Churle~Bully. Bern: Francke. Schmid, Hans-Jorg Englische Morphologie und Wortbildung. Eine Einfuhrung, Berlin: 2005 Erich Schmidt.

Lexicul i~zotivutionus bidirectional

333

Schreuder, Robert, and R. Harald Raayen How Complex Simplex Words Can Be. Journal of Memory and Lun1997 guuge 37, 1 18-1 39. Simpson, Sohn A., and Edmund S. C. Weimer 1989 The OxjOr-d English Dictionary. Second Edition. Oxford: Clarendon. [ = 0EDl ~tekauer,Pavol 1996 A Theory of Conversion in English. Frankfurt am Main: Lang. Tiersma, Peter M. 1978 Bidirectional leveling as evidence for relational rules. L,ingua 45, 6577. 1982 Local and general markedness. Language 58, 832-849. Tournier, Jean Introduction de,scrip/ive u lu 1exicogL;nc;tiquede I'angluis contempo1985 rain. ParisiGeneva: Champion - Slatkine. Twardzisz, Piotr 1997 Zero Derivation in English. A Cognitive Grammar Approach. LAblin: Wydawnictwo IJMCS. Tyler. Lorraine K., Rachelle Waksler, and William D. Marslen-Wilson 1993 Representation and Access of Derived Words in English. In C'ognirive M~~u'c.1.~ of Si~eecl?Processing: The Second Sperlonga Meeling, Gerry T . M . illtniann, and Richard Shillcock (eds.), 125-140. Hove: Lawrence Erlbaum. Ullmann, Stephen The Principles qfSernuntics. Second Edition. Oxford: Blackwell. 1957 Scmuntic.~. An introduc/iot~to the science qf meaning. Oxford: 1962 Blackwell. Ungerer. Friedrich 2002 The conceptual function of derivational word-formation in English. ,4nglia 120, 534-567. Vennemann, Theo 1972 Rule inversion. Lingzcir 29, 209-242.

Doubler-upper nouns: A challenge for usage-based models of language?' Bert Cappelle

1. Double -er coinages: rare but regular visitors of the English language In English, there is a rather odd way to form agentive nouns based on phrasal verbs: besides well-established cases where the -er suffix attaches to the verb, like passer-hv and runner-up, one occasionally finds cases in which the -er suffix is attached to both the verb and the adverb, like dropper-inner and wmher-upper. One's immediate reaction to such forms might be that they are intentional errors. For example, a respondent to an internet post about such forins writes: "I always thought 'picker upper' is (or at least originated as) a joke form that deliberately breaks rules".' Indeed, the suffix -er is not meant to be doubled, nor is it meant to be attached to an adverb. That they are deliherate creations is supported by a quotation from the animation series The Sinysons, in which the incompetent lawyer and charlatan Lionel Hutz prornotes himself as follows:

(1)

Lionel Hutz, your new agent, bodyguard, unauthorized biogrupher, and drug dealer... er, keeper-an~ayer. (www.snpp.com/guides/hepped.up.l~tml, accessed 1 May 2008)

?[he . triple occurrence of -er (not even counting the pause

... er) in tlie last

noun i~ndoubtedlyserves to enhance tlie humorous effect created by the all too conspicuous denial of a criminal activity. In this paper, I want to show that such reduplication forms do not just deliberately break rules but that they also obe-y a number of rules. In fact, in the example just given, adding an -er to tlie adverb makes the agentive nominalization more acceptable than a nominalization without such a second -er (cp. ?*~lrugh a l e r ... er, keeper-away). In other words, doubling the -er may be the most felicitous option when deriving nouns from phrasal verbs. Moreover, although a fi~rtherplural affix could potentially be added to either -er affix, we see a clear difference in acceptability between, e.g.,

336

Bert C ' a ~ ~ e l l e

"wa.5her.s-upper and +vasl~er-upl~ecs (Blevins 2006: 514). If double -er forms were plain wrong to start with, we probably should not have such a clear intuition about which -er affix has to serve as tlie affixation site for plural inflection. So, even thougli doitble -er forms are not considered as Standard English. these two observations suggest that they are not just performance errors either. Put differently, if this morpliological structure is an error of some sort, then I agree with Mclntyre (2004) that "it would have to be admitted that it is a fairly natural kind of 'error' for many speakers". But herein lies a curious paradox: on the one hand, double -er forms are rather rare and sound as if they are made LIPfor tlie occasion; on the other hand, they seem to be a quite sensible choice in some usage situations and we do not find it hard to tell how these forms can be further inflected. But how can we find tliis form (relatively) 'natural' and how can we make granimaticality judgments about alternative plural forms and other issues if me hardly ever encounter this structure? To quote again from Mclntyre (2004): I don't know if it's a good idea to dismiss the construction as some sort of idiosyncratic constructional template perpetuated by imitation. Its actual text frequency is extremely low; even very advanced L2 English learners are mostly unaware of its existence. I have known about the problem since at least 1995, and have since then come across only two examples in normal interactions with English speakers (i.e., excluding corpus work, etc.). One wonders if this is enough for us to learn 'picker upper(-er)' by direct evidence, or whether it is somehow a natural response to the problem of how to affix left-headed structures given the less-than-wonderful status of nonreduplicative solutions ('washer-up', 'wash-upper') for a sizable set of speakers. (Mclntyre 2004)

So, Mclntyre's question is: How does a speaker's grammatical knowledge about tliis marginal structure arise if not as the result of frequent exposure to double -er instances? Clearly, this is a question which transcends the morphological phenomenon at hand, since it poses a challenge for usagebased accounts of language and language learning, like the ones proposed in, for example, Langacker (I 988), Barlow and Kemmer (2000), Tomasello (2000, 2003), and Bybee (2006). 111 SLICII accounts, linguistic competence has less to do with any innate general rules than with linguistic experience acquired via increasing familiarity with concrete usage forms. Double -er forms may be too infrequent to give rise to a routinized pattern, and hence would seem to cast doubt on the validity of usage-based accounts. 1 will

Doubler-upper nouns

33 7

claim tliat, even though double -er instances may not be sanctioned directly by a double -er constr~ictionalschema and should instead more likely be seen as resulting from the application of various higher-level rules, these rules tnay themselves have come about as generalizations over m~~ltiple usage events. Therefore, a rule-based account of a linguistic phenomenon should not necessarily be considered as radically opposed to and incompatible with a usage-based account. This paper is organized as follows. In Section 2, 1 will discuss the question whether double -er forms are really so infrequent and whether they are all made for the nonce. It will be shown tliat they are indeed quite rare and that most, if apparently not all, double -er nouns are created on the spur of the moment. In Section 3, I will look at various attempts by other linguists to account for these nonce-usages. Most of these attempts are very tentative and sketchy, but there is at least one exception, by Acke~naand Neeleman (2001, 2004), which deserves closer investigation. In Section 4, I will offer an account in terms of interacting rules which conspire to bring speakers to produce double -er forms. In section 5 , l will consider a closely related structure in wliich a second -er is added to a preposition rather than an adverb (e.g., videogunie writer-abouter). In Section 6, 1 will reflect briefly on what tny analysis implies for usage-based theories. Section 7 succinctly summarizes the main views.

2. The extent of use of double -er forms In this section, I discuss how common (or not) double -er forms are in English. I first give a survey of the literature, at least with respect to this question and then formulate a hypothesis about their first appearance. Next, I report on original corpus research into the frequency and productivity of the double -er structure in present-day (British) English, showing that the structure is indeed quite infrequent, altliough one fortn Gxer-upper) has gained some currency and some other forms tnay be coined Inany times over, thus boostilig their frequency. In a final subsection, I sutn~narizemy main findings and explore the role of imitation in the use of double -er forms.

33 8

Bert Cuppelle

2.1. Usage comments in the literature All authors who have signaled the existence of the double -er type have given an assessment of its frequency or productivity, often accompanied by a stylistic appreciation. The prevailing comment in the literature is that double -er forms are attested colloquialisms, as will appear from the following brief overview.' Ackerna and Neelernan merely state that such forms "are attested in adult English" (2001: 24). This formulation suggests that the authors realize that the reader may not have come across double -er forms and therefore needs to be convinced of their occasional occurrence. Bauer writes that double -er forms "tend to feel very clumsy, and as a result tend to be used mainly in colloquial speech" (1983: 289). Together with Huddleston, he later describes them as "non-standard nonce-words" (Bauer and Huddleston 2002: 1655). Bolinger (197 1 : I 16) calls them "popular coinages", which "began to multiply in the 1920's and 1930's", a point that I will take LIP below. Blevins (2006: 513) calls thein "colloquial", implying that they're not part of the standard language. Busuttil (2001 : 38), discussing a form like doer-upper of houses, comments that "we're not dealing with an isolated example" (my translation). Farrell (2005: 103) writes that "it is customary to double -er" but adds that "[this structure] may not yet be fully conventional" (2005: 103) although it "appears to be at least somewhat productive" (2005: 103, fn. 1 I). Miller (1 993: 132) speaks of a "new productive pattern" and remarks that affixation which both precedes and follows the non-verbal part is not only one of the "logical possibilities" but also that all the possibilities "are found". This latter addition implies Miller's realization that the reader might be unfamiliar with the pattern and could have supposed that the double -er structure was just a logical possibility without evidence in actual use. Mclntyre (2004) goes further by stating that there is "good attestation" of such forms, but, as mentioned before, he considers it very unlikely that individual speakers have encountered a sufficient number of them to be able to extract a re-usable pattern from them. We will come back to this point later. 2.2. Earliest written occurrence: not before 1900 Perhaps the most interesting discussion about the usage of the structure (hilt not of its deeper motivation) is provided as early as 1936, in the journal Anzericun Speech. The author, Wentworth, describes it as "[tlhe current, popular, grotesque way of forming new low-colloquial and slang

Doubler-uper nouns

339

compound nouns" (Wentuorth 1936: 369). Amid rather sub-jective judgments - Wentwortli also wonders why "the evil genius of slang permits -er [. ..] to be so misused" (1 936: 370) - we find valuable information about wliat must be the original distribution of this structure: it is "national [i.e., nation-wide] in occurrence, journalistic in origin, collegiate in vogue, and economical in expression of ideas" ( 1936: 369). While we will discuss the latter point (semantic economy) in Section 3.2, it is worthwhile to dwell on the question whether Wentworth really commented on a phenomenon that was still very recent when he wrote his article. The earliest occurrence of a double -er word that I could find in the OED dates from just a few decades earlier: a new draft entry (March 2006) of the OED mentions an instance dating back to 1913, taken from the C'hicago Sunday Tribune: "For every fling-arounder..tliere is a busy little picker-upper". This also seems to confirm what Wentworth writes, namely that its origins are journalistic. Wentwortli lists 18 attested specimens of double -er nouns. For four of these, I could find attestations in the OED, namely Jixer-upper, waker-upper, builder-upper, and tearer-downer, and all of these are indeed from around the time of Wentworth's article: 1932, 1935, 1936, and 1942, respectively. Note also that one of Laurel and Hardy's sI1ol-t films, titled The Fixer Uppers, dates from 1935, the year before the publication of Wentworth's article. It is possible that this title picks up on a then-trendy usage. Remember, finally, that Bolinger (I 971) also considered the structure to have bloomed in the 1920s and 1930s. We have no idea wliat he based this claim on, but it seems to tie in with what we read in Wentworth's article. Recent research by Denison (2008: [4]), drawing on the Google News Archive, strongly suggests that "[i]t does look as if the pattern enters American newspapers in some force in the 1930s". So, double -er forms appear in writing only in the twentieth century, and it is only in the nineteen twenties and especially nineteen thirties that its popularity gains momentum. Of course, given the more formal and conservatively edited nature of writing compared to spontaneous speech - a difference which might have been starker in the nineteenth century than in the twentieth and twenty-first century - the absence of evidence that double -er nouns occurred before 1900 should not be taken as evidence of their absence in the language.

2.3. Freqi~ericyand productivity in present-day English There is no reason to believe that double -er forms have spectacularly boo~nedsince the 1930s. Otherwise. the authors cited in Section 2.1 would not have felt the need to i~nderlinethe fact that they do occur at all. So, evactly Iiow frequent are such nouns? To get an idea, I searched for instances in the British National C'orj7u.s (BNC'), a 100 million word corpus of fairly recent ( 1980s- 1993) British English, containing both written and spoken language, and in the Cbrpus of Conten?porary An~ericanEngli~h (C'OC'A), an even larger and still growing corpus (at the time of research, 360+ million words) of current (1990-present) spoken and written American English. Since double -er forms should be seen as competitors of at least two other derivational patterns (kerb-er-adverb and verb-adverb-er) 1 have looked for instances of these patterns as well. The results are given in Table 1 and Table 2, respectively. The tables give the results for the number of tokens (number of word hits), types (ni~mberof distinct words, i.e., lexemes) and hapaxes (number of words with only one occurrence). The relevance of this last parameter will be explained shortly. The adverbs searched for were about, ucros,s, ahead, along, apart, around, crside, avtuy, buck, b y , down, home, in, ofl on, out, over, round, through, logether, and UP. Tuhle I

Number of tokens, types, und hupuxes for three competrrig k m i s o j -er nomrnalnntrons n'c.rtvedfrom phrasal verhv rn the BNC

verb-er-adverb-er Tokens

4

TYpes

3

Hapaxes

2

verb-er-adverb

verb-adverb-er

Table 2 Number of tokens, 1-vpes, and hopc~xe.cfor three conlpeting kinds nonzrnnlrzut~on~ devrved fiorlz phrusul verbs in COCA

verb-er-adverb-er Tokens

150

Types

37

Hapaxes

30

verb-er-adverb

of

-er

verb-adverb-er

When we look at Table I, we are immediately struck by tlie liigh relative frequency of the verb-er-adverb structure. The BNC data apparently prove wrong a passing statement to the contrary in Ackema and Neeleman (2001 : 24), where it is said, rather surprisingly. that the double -er structure is the most frequent of the three structures. 0 1 1 the other hand, for American English (cf. Table 2), the double -er pattern does seem to have equal Qpe frequency as the verb-er-adverb pattern (or perhaps even ever so slightly higher frequency tlian it), although tlie latter patter11 still far outweiglis the former in terms of token frequency. In any case, to tlie extent that results from BNC' and from COC'A can be compared, we can infer from Table 1 and Table 2 that there is a clear association between derivational structure (the choice of morphological pattern) and variety of English (British or American). Chi-square tests reveal that this association is highly significant, both when the token numbers are compared (i.e., when we compare the first row in Table 1 with the first row in Table 2; 2 = 39.4; p < .001, DF = 2) and when the type numbers are compared. (i.e., when we compare tlie second row in Table 1 with the second row in Table 2; 'X = 64.5; p < .001, DF = 2). More specifically, there is a positive association between the verb-er-adverb pattern and British English and between the two other patterns and American English. The finding that the double -er structure is more colnlnon in A~nericanEnglisl than in British English is confirmed when we compare the proportions of double -er tokens per total number of words in the differently sized corpora: tlie difference between 4 tokens per 100 ~nillionwords and 150 tokens per 360 lnillion words is significant at the ,001 level. The difference in the proportions of types per corpus (3 types per 100 million words vs. 37 types per 360 million words) is still significant at the .05 level.

342

Bert Cr~ppelle

As for the pattern with a single -er attached word-finally, tliis is clearly tlie least frequent structure, both in number of types and in number of tokens. Not counted as true instances of this derivational type were cases whicli are derived from a complete verb phrase with an object, e.g., crunch-you- upper, .figure-it-outer, gro.,s-rne- outer^, wake-rne-upper, etc. These forms are perhaps better analyzed as the output of an altogether different morpl~ologicalprocess, namely -er affixation to (uninflected) verb phrases. Nominalized verb phrases can only have word-final suffixation: just like you cannot have *banner-the-bomb or *banner-the-bomber alongside bari-the-bonzber (derived from bun the bomb), you cannot have "cruncher-you-up or *cruncher-you-upper as alternatives to crunch-youupper. Such alternatives do exist for what I considered to be real instances of the verb-adverb-er pattern (e.g., try-outer-,\ and give-upper). What I also counted as instances were come(-)bucker and touchdowner (8 and 5 tokens, respectively, in COCA), even tlioi~ghthese are probably not directly derived from to conie buck and to touch down but 'indirectly' from the conversions conieback and touchdown. Since there is just one type with one token for this pattern in the BNC', I ran an extra search on Google.co.uk to find out whether instances of tliis structure are really so infrequent, even compared to double -er nouns. This was indeed confirmed: only 7 out of 78 phrasal verb derivations ending in -uppers were of tlie verb-adverb-er pattern."hese instances were colter-uppers, dial uppers (3 hits), hurryupper., (2 hits) and tidy-uppers. We will see in Section 3.1 that the bisyllabic structure of the verb may play a role in the selection of this derivational pattern. However, at least one phrasal verb consistently selects the verb-adverb-er pattern: someone who 'comes out' is typically referred to as a come-outer - a word attested in the OED - rather than as a corner-out or a coiner-outer. Also, the word hung-urounder, which can also be found in the OED, seems to occur at least as frequently as its alternatives, hangeraround and hunger-urounder, probably also for metrical reasons. Such individual preferences cannot deny tlie larger picture, which is that the verb-adverb-er pattern is by far the least productive of the three patterns. Another measure of productivity proposed by Baayen and his coworkers (see, e.g., Baayen and Lieber 1991) is based on the proportion of lexemes that occur only once in a corpus, tlie so-called hapax Iegonzenu, or hapaxes for short (see also Chapman [2008] for a discussion of this method with respect to tlie pattern under discussion). The idea behind counting hapaxes as a means of measuring productivity is that single-token types in a large corpus may very well correspond to neologisms in real use. Obviously, a necessary requirement for a pattern to be called productive is that

it allows novel instances to be based on it. So, using hapaxes as a measure of productivity seems to be a more valid way tlian just looking at the raw oi~tputof a pattern: for all we know, a pattern with many instantiations might have been very productive in tlie past while being unproductive in tlie present. In other words, we might prefer to look at a pattern's "potential productivity" rather tlian its *'realized productivity" (Baayen 2009). The numbers of Iiapaxes for the three patterns are given in the third row of Tables 1 and 2. The relationship between hapaxes and coinages is in fact an indirect one. The more productive a morpliological pattern, the larger the set of possible types based on it. However, even a decent-sized corpus may not contain all of the Inany possible types, especially not all of the still infrequent neologisms. Yet, it is to be expected that at least some of a pattern's new offspring will nolietheless occur in the corpus, usually just once, and occasionally even twice or more often. Therefore, if there are many hapaxes instantiating a morphological pattern in a corpus, we can infer that language users 'outside' the corpus can create many new instances of that pattern, even though some or all of these liapaxes in the corpus may not actually be newly-coined and may already have been around in the language community for a long time. Hapaxes are thus an indication of tlie potential to form neologisms but should not be identified with neologisms themselves. One of Baayens' hapax-based measures of productivity is calculated by taking the quotient of the ~iuniberrz, of hapaxes of a given pattern in the corpus and the total number N of tokens of all the words displayi~igthat pattern in the corpus:

Applying this formula to tlie data for American English, we obtain the results shown in Table 3. Table 3. P-values (lfq?ax/Tnken r-utios) ,fbr three competing kin& of -er nominalizations derived,fi.om ,~~hrilsal vohs in COCA

verb-er-adverb-er

verb-er-adverb

verb-adverb-er

344

Bert C'uppelle

This time, the verb-adverb-er pattern comes up as most productive. However, the P-value for this pattern might not be very reliable because the sample of instances is too low. It has been proposed that Baayen's productivity measure should not be used with sample numbers lower than 50 (cf. Bauer 2001 : 15 1). Indeed, this is the reason why we did not calculate the Pvalue for the BNC data, where the verb-adverb-er pattern has only one type with one token. In other words, 'all' of its occurrences in the BNC are hapaxes. This would mean that this pattern has total productivity (P = I), a co~iclusionwhich runs counter to our intuitions. The P-value for tlie double -er pattern in BYU C'orpus of Anierican English can be calculated more reliably. It is exactly 0.2 (30/150), whicli is quite high: one in five of the double -er tokens collected in this corpus are hapaxes. Given that this carpus is a very large one indeed, capturing for all intents and purposes the language as used by a speech conimunity in the real world, it is also safe to assume that many or most of these hapaxes are genuinely newly-coined words; that is, words that were not existing members of the vocabulary prior to their usage. Sure enough, the fact that they occur in a corpus does not turn them into well-established members of the English lexicon; they are therefore better termed 'nonce-words' rather than 'neologisn~s' (see Bauer [2001: 381 for the distinction, which is based on the degree to which a coinage is familiar to a large enough set of the speech community. An objection to the measure of productivity given in (2) has been made by Van Marle (1 992: 156), who sees no reason why the denominator in the fraction should be the token frequency of the morphological pattern. Van Marle rightly wonders why the frequency of words already formed on tlie basis of a patter11 should play a role in gauging the degree to which that pattern can be used to coin new instances of it. Let us take the verb-eradverb pattern lo illustrate the problem. This pattern contains a small number of very frequent types. In the BNC, there were 547 occurrences of the type runner-up, 2 12 occurrences of the type passer-by, a ~ i d68 occurrences of the type hanger-on. Together, these 3 types represent up to nearly three quarters of all the tokens of that pattern, with the remaining 52 types only accounting for just over the remaining quarter of tokens. In C'OC'A, these three types even take up 98% of all the tokens (248012526). In other words, if it were not for these three highly frequent lexemes, the verb-er-adverb would score much higher on the hapax-based productivity measure given in (2). Accordingly, if we adopt a hapadtype ratio instead of a hapadtoken ratio, then we see that the productivity scores for the three patterns are I I I L I C ~more similar (cf. Table 4).

Doubler-upper nouns 345 Ttrhle I. illternutive P-vulues (Hupux/Type rutios),for three competing kinds of -er notninalizatiot~.~ &rived,fi-omphrasal verbs in COCA verb-er-adverb-el

verb-er-adverb

verb-adverb-er

Again, even though the numbers of types and hapaxes might be a bit too low to allow accurate calculations and comparisons, these results suggest that the double -er pattern, which we are especially interested in, is far from an unproductive pattern. It is perhaps even the most productive of all three competing patterns on tliis alternative hapax-based measure. Let us now have a closer look at double -er types with liigh tokenfrequency. One would expect that types with many occurrences must be well-established in the language community or must at least be familiar to a large sub-portion of it. Tliis seeins certainly true for a noun like$xerzq?per, for wliicli there are more than hundred occurrences in BYU Corpus c!f American English. The estimated number of search results for $xerupper on Google even exceeds one million. Tliis very high number can partly be explained by its semantically lexicalized status, in other words, by the fact that it has a specialized meaning. Fixer-upper originally referred to someone who fixes things up (as in the title of the Lai~reland Hardy short film already mentioned). However, in a second meaning which has been in use since at least as early as 1948 (judging from the earliest attestation of this meaning in the OED),jixer-upper refers to a cheap building which needs to be fixed up by the prospective buyer.' Not all forms with high token-frequency are lexicalized, however. As an example, consider pulter-onner. Google estimates that tliis derivation comes up almost a thousand times in its indexed pages. However, this form usi~allyoccurs in so-called synthetic compounds, like make-up putter-onner, weight putteronner, c.o,,tunle-putter-ont~er,bumper sticker putler-onner, etc., all of uliich sound as if they are made up for the occasion. There is no reason to assume that language users creating such conlpounds make use of a noun (putter-onner) which is already a firm part of the English vocabulary. Put on as a phrasal verb has liigh token-frequency itself, due to its semantic non-specificity and hence flexibility. Google estimates that there are about 78,400,000 results for "put on", and it is therefore only to be expected that many speakers, independently of one another, may suddenly find them-

346

Bert C'q?pelle

selves in a circi~~nstance in wliicli tliey need to norninalize tliis phrasal verb. If this view is correct, tliis means that high token-frequency of an item does not necessarily correlate with degree of conventionalization in the language.

2.4. Interim sumlnary and transition Linguists who have remarked on tlie existence of double -er derivations have generally called them substandard nonce-nouns. One doiible -er noun, fixer-upper, milst undoubtedly be familiar to many speakers, especially those working in real estate and people having dealings wit11 them. This term can therefore not be called a coinage whenever it is used in the sense of 'dwelling sold cheaply but in need of repair'. It is a stored entity in the mental lexicon of a significant group of speakers. The majority of double -er nouns, however, are probably created just for the occasion. The first known occurrence of a double -er noun dates from the second decade of tlie twentieth century, soon after which such nouns seem to have gained some popularity, especially in the 1930s. Since then, their use may not liave dramatically expanded so as to become part of the standard language. If they had, linguists would not still call them colloquial nonce formations or find it necessary to point out tlie mere fact tliat tliey are really attested. Attested tliey are, but with only four occurrences in the entire BNC' (100 million words) and even with a somewhat more respectable 150 occurrences in COCA (360+ million words at tlie time of research), it does not seem unwarranted to claim tliat tliey remain too infrequent to be licensed by a readily available double -er co~istruction. On tlie other hand, double -er forms need not be used very frequently to get entrenched in the mind of the language user. The fact tliat these forrns can be used for comic effect means that when one such noun is uttered, it will not go by unnoticed. Hearers who have been exposed to such a reduplicative form just once may repeat it in their minds, where it 'gets stuck' as a result. If a single occasion can lead to mental storage, it is reasonable to assume that many double -er nouns are coined by analogy of such an entrenclied form. In other words, language users may not liave a readymade double -er construction, but they may well have access to one or two exemplars of a 'virtual' double -er construction, that is, a schema which does not liave mental unit status as sucli but wliicli can potentially be extracted from the exemplars so as to be applied to new instances.

Doubler-upper nouns

347

What, then. is the role of imitation in tlie use of double -er forms? The right answer seems to be that most double -er forms are coined (that is, invented), but this coinage may be facilitated by tlie availability of some similar forms tliat fi~nctionas models. So, the coiners may make use of some example cases altliough these are probably not plentiful enough to lead to the establishment of a real construction in tlie mind. There is some support for the view that imitation plays a role in the coinage of these special derivational forms. One piece of evidence is tliat, as we have seen, sucli forms seem to liave become more popular in the nineteen-thirties. The only explanation for such an increase, I believe, is that speakers started hearing or reading sucli forms in tlie language community, and that some users picked up on this trend and helped multiply this derivational type by coining some similar forms of tlieir own. Another piece of evidence is tliat an even rarer form with triple -er (e.g., washer-upperer) has somewhat higher frequency in Australian English than in other varieties of English (cf. Cappelle 2003). This fact can most naturally be explained if we accord some recognition to tlie role played by imitation of morphological precedents. This said, my central claim is tliat we do not need to posit a routinized double -er construction, not even a 'virtual' double -er construction, in the sense given above, to account for the occurrence of double -er forms. This is because such forms make sense by themselves, complying as tliey do with a number of general principles (or 'rules') for whose existence speakers tnust definitely liave ample evidence. It is tlie interplay of these general principles which must have enabled tlie creation of the very first double -er creation, which could not possibly have been coined by imitation. Before I give an overview of these principles in Section 4, it is liigli time we looked at various explanations given in the literature for the occasional occurrence of double -er nouns.

3. Explaining the occurrence of double -er nouns In tlie previous section, we liave seen that double -er nouns are rather too infrequent to be simply sanctioned by an idiosyncratic construction which is ready for retrieval from the language user's 'construct-i-con' - a point already made by Mclntyre (2004) - but tliat tliey might still be formed by analogy with some rare examples tliat are mentally entrenched because of tlieir relatively high token-frequency and/or high conspicuousness. However, suppose, just for tlie sake of tlie argument, tliat there were no forms

available whatsoever which functioned as examples to the speaker. This may not even be a far-fetched thought experiment, because we cannot exclude the possibility that some language users come up with a double -er form without ever having heard or seen one before. I claim that, even in the likely unavailability of a double -er construction and in the theoretical unavailability of any model cases, there is no need to throw one's hands in the air and simply state, as Farrell (2005: 103) does, that -er is doubled "for whatever reason". In what follows, I will review various kinds of explanations that have been offered: plionological ones, semantic ones, and morpho-syntactic ones.

3.1 . Phonological expressiveness Miller (1993: 133), quoting Sproat (1985), states that ihe second -er in ~ncrrcher-thmz4ghc~r is an "expressive P(honological] F[orm] copy". At first sight, this is a rather unsatisfactory analysis, which seems to simply translate into generative jargon tlie pre-theoretical idea that -er is added for reasons of word-internal rhyme. Such a linguistically naive view can be found in another respondent's comment to the internet post mentioned in Section 1 above: Those words or the construction of those words probably originated from a desire to rhyme; intentional or not. Think of it in terms of poetry. "Picker upper" simply sounds better than "picker up" or "pick upper". (ablauttime.blogspot.co1ni20041091passers-by-be-damned.html accessed 23 August 2007).

A desire to rhyme does not play a role in many other morphological processes. However, English does have a number of rhyming words like arlyfurty, easy yeasy, heller-skelter, hurdy-gurdy, itsy-bitsy, nlumho jumbo, etc., so an explanation of double -er nouns in t e r m of rhyme can perhaps not be rejected altogether as cluptrup." On the other hand, in such reduplicative~,not just the final unstressed syllable of each part rhymes but also the preceding stressed syllable (and, if present, an intervening nonstressed syllable, as in tlie case of higgle&-pigledy, hickory dickory, Herkinzer ./c>rkinler, jiggledy-wigglcdy, joukery-cookery, nipperty-tipperty, etc.). Rhyme must therefore be deemed to be of limited importance to the creation of double -er nouns, but its role cannot be ignored. Maaike Belien, in personal communication, gives me the example of the slogan "Bounty, the quicker picker-upper", used in an American com~nercialfor a brand of

Doubler-zlpper nouns

349

kitchen towel (cf. also Chapman 2008). In this slogan, tlie verbal part of the derivation rhymes with the preceding adjective, and the two agentive -er cuftixes rhyme with the comparative -er suffix. Again, however, the bisyllabic rhyme of the adjective and the verbal part of the derivation does not extend to the non-verbal part. Considerations which more certainly seem to play a role are metrical ones. For instance, as mentioned above, the type with a single final -er is very rare but sounds quite natural with bisyllabic verb bases, e.g., coverupper, diul-upper, .figure-outer, hurry upper, tidy-upper. Similarly, when the adverb starts with an unstressed syllable, the double stressed-unstressed sequence is satistied by the final -er type. Consider again the sentence with what may be the earliest double-er occurrence in the OED: "For every fling-arounder..there is a busy little picker-upper", and observe how the first nominalization does not double the -er suffix. One possible reason is that such a doubling would disrupt the alternation of stressed and unas stressed syllables (cp. fl2l,lc.u.ROUND.er and P~l~/~2ic;.er.a.ROUND.er), noted also by Denison (2008). Haplology (i.e., the avoidance of similarsounding syllables), may also explain the selection of tlie verb-adverb-er pattern here: an -er added to the verb would sound exactly like the first part of crround. We can also invoke stress as a partial explanation for why someone who gets picked up can be called apicker-uppee (apart from apick-uppee) but not a *pickee-upper.' For some similar examples, consider (3a-b): (3)

a. I always said it's harder to he the breaker-upper than the brenker-uppee hecau.se the duniped person alwayksgets all /he sytnputhy. (kvetcli.indiebride.com/index.php?t=117sg&goto=53276&rid=O&S =98aa32ea9af9f63e75a2f48faf4d5cb7, accessed 22 May 2008) b. ikfy duy hegun with Drives Us All Crazy pulling sowie girl's pig tail out? Hone,rtly, I never got the whole slory. The pig tail puller-nutee w w sobbing .so lozrdly I couldn't understund u dung thing she said (hOkieerin.bIogspot.co1n/2006~0 1-01 -arcliive.html, accessed 22 May 2008)

Miller (1993: 132-133) takes such forms in -ee as evidence that the first -er is "essentially vacuous", which seerns correct. He then goes on to explain tlie fact that the -ec suffix necessarily attaches to the adverb by appealing to case-theoretical principles. Basically, his argument is that the

3 50

Bert Cuppelle

deep-structural thematic object of tlie adverb must combine witli tliat adverb before they are both syntactically incorporated with tlie verb, which possibly has already received an affix in the lexical component before this incorporation. Not only is it hard, if not impossible, to translate his account in theory-neutral terms, but it also seems to me that a simpler explanation can be given here. The productive patieiitive -ee suffix always carries the nuclear stress in the derivation, unlike agentive -er. Words ending in -ee therefore have a characteristic word-final stress. The form PIrK.er.up.EE respects this typical word-final stress pattern tliat words in -ee have, while tlie anomalous form *pick-EE.up.c.rdeviates from it. Restoring the familiar stress pattern of -ee nouns by stressing the word-final -er (pick.~l.:up.ER) would only make matters worse, of course, because it would wrongly underline that the derivation is agentive. In Section 4, 1 will give an extra reason why the form picker-uppec makes more sense than *pickee-upper to denote someone who gets picked up.

3.2. Semantic expressiveness We have already quoted Wentworth's (1936: 369) view that the double -er structure is "economical in expression of ideas". Wentworth clarifies this claim by stating tliat . .. two-thirds of the handful of specimens in question cannot be so concisely worded in standard English. The flippancy of waker-upper[,] holder-outer, dropper-inner, putter-off'er, and lohhj~-dr(fter-throz1gheris almost counterbalanced by the circuitousness of [hilt which wakes one up, one who holds out, one who drops in, one who pzrts soriiething off; and one who d r $ s through a lobby, and by the formality of awakener, intractable, chance visitor, procrastinator, and casual inspector (?f'hotclpatronage. ( 1 936: 369)

So, according to Wentworth, tlie double -er structure allows a complex concept to be succinctly encoded witli the help of every-day English words and morphemes. With two -er suffixes, ho\vever, tlie doi~ble-cr derivational structure is not as concise as it could be. Bauer (1983: 289) finds tliat all tlie derivational structures considered so far, and a couple of others not mentioned here, are rather awkward. This, together with the fact that there are so many competing types, is considered by Bauer to lead to the frequent avoidance of adding an -er suffix at all. The process wliicli allows users to avoid the suftix attachment problem altogether and, hence, which produces

Doubler-upper nouns

35 1

even more concise derivations is conversion. Examples are drop-out, runrmiuy,~how-(!fl, and sland-in. Yet, when it comes to creating a noun which refers to the agent of the event expressed by a phrasal verb, conversions may not be the most obvious choice. A brief browse througli derivations of this type revealed that agentive interpretations represent the minority. Among the BNC's twenty ~iiostfrequently-used conversions ending in up, there was just one lexeme wliicl~can be interpreted as referring to a person, namely back(-)up. The other derivations (e.g., build-up, cover-up, line-up, etc.) typically refer to an objectified event. Back-up, too, can refer to an event, and this may indeed be the more basic meaning. This is clear from the fact that when hack-up refers to a person, it glosses more naturally as 'someone who can provide back-up' than that hack-up in the sense of 'support' can be glossed as 'what is provided by a back-up'. Since conversions do not code agentivity directly, they are semantically less expressive (or less informative) than -er derivations. However, the double -er structure is not more expressive than its rival derivational structures with -er. Accordingly, the need for conciseness in expression certainly cannot be the reason why some speakers prefer washer-upper over ~1a.tlzer-upor wush-upper.

3.3. Lack of morphological compounds of tlie adverb-verb type Ackema and Neelernan (200 1 : 23-26; repeated in 2004: 1 59-1 62) attempt to provide an explanation for the fact that there is such a variety of derivational processes which turn plirasal verbs into agentive nouns. They contrast this variety with the lack of options for agentive synthetic compounds like truck driller, for which there are no grammatical alternatives like *driver-/ruck, *hive-trucker or "driver-trucker. Their explanation is that agentive derivations from phrasal verbs, ulilihe agentive synthetic compounds, are not formed directly from a verbal compound. That is, in English, there are no such adverb-verb compounds as *to away-throw, *to instand or *to u'owr7-let alongside syntactic combinations like to throw away, to stund in or to let down (200 1 : 23-24). Because of the lack of such morphological forms, speakers are forced to take the verb-adverb combination, which they see as a syntactic rather than morphological combination, as the host of tlie -er affix. Each possible phonological output of this addition is then fraught with a different violation. A form like cutter-up contains a violation of "linear correspondence", by which a structurally external element (like an affix) should stay plionologically external to its host. A form

3 52

Bert Cappelle

like cut-upper violates tlie "input correspondence" principle, wliich says that an affix whose struct~tralinput is a category or a phrase headed by that categorq sliould phonologically attach to that category. And as to cuttery?per, finally, this form violates the "quantitative correspondence" principle, according to which affixes cannot be phonologically realized more than once. If a morphological co~npoundlike *to up-cut were available, then "upcutter would be formed, by analogy with truck driver, and it would block all tlie forms directly derived from cut up. The main merits of Ackema and Neeleman's account is that it attempts to explain the ~nultiplicityof forms and that it spells out reasons for what Bauer (1983: 289) calls their "awkwardness". However, there are also a number of proble~nswith their account, whicli I will discuss in the following three subsections.

3.3.1. There are no N- V cony~oundsbur there ore Adv- V conipounds Ackelna and Neeleman's (2001, 2004) niorphological analysis of a synthetic compound like truck driver is shown in (4): (4)

[N [V

truck driveler]

Sucli an analysis is proble~naticbecause truck driver cannot be glossed as 'someone wlio *truck-drives'; rather, it Inore naturally glosses as 'someone wlio drives trucks'. A more satisfactory constituent structure of truck driver is therefore tlie one given in (5): (5)

[N [N

truck]

[N

[" d~*ive]er]]

So, the fact that we do not have forms like "driver-truck, *drive-trucker or

"driver-trucker cannot be based on the presumed availability of verbal compounds of tlie noun-verb type, wliich is a non-existent morphological type in ~ n g l i s l iBy , ~ contrast, verbal compounds of tlie adverb-verb type, whicli Ackerna and Neeleman clairn do not exist, can actually be found in English. True enough, prefixed verbs (e.g., overact, underfeed etc.), wliich they do give some exarnples of, are not co~iipo~ilids: they all have stress on tlie verb and the prefix has a meaning related to degree, wliicli is rather different from its meaning as an adverb. However, English also has cases SLICII as backtrack, download, clown.scale, u'o+tlnsize, inbreed, &set, aut-

wurce, and upgrade, all of which have or can have stress on the adverb and which therefore (also) qualify as compounds and not (just) as prefixed verbs.

3.3.2. Not all verb-adverb conzbilwtions are syiztactic, nor are all their derivations morphological

Acke~naand Neelenian's (200 I , 2004) account suggests that a verb-adverb combination, when not yet nominalized, is a syntactic entity, and that the diverse nominalizations types are all ~norphological entities. There are reasons to object to this view. As regards non-derived verb-adverb combinations, I more or less follow Farrell (2005) in assuming that most of these can be realized either as word-level verb-adverb compounds, making up a colnplex verb, or as phrase-level structures, with the verb heading a VP and the adverb potentially heading an adverb phrase within that VP. For example, blurt out is a compound in (6a) and part of a verb phrase in (6b): (6)

a. "Out with it, " blurted out the Capfain brusquely (www.poetrymagazines.org.uk/magazine/record.asp?id=1 1405, accessed 22 May 2008) b. He just blurted the word., right out.

(www.boston~iiagazine.com/articles/the~sins~of~the~father/, accessed 22 May 2008) In (6a), blurted out cannot be a phrase, since the quotative inversion construction only allows verbs (without complements or adjuncts) to occur before the postposed subject (Toivonen 2003: 175-176). In (6b), blurted and ottt make up a V P which also (among other things) contains the N P the ~iord.s.Claiming otherwise (i.e., .sh(~utout is a phrase in (6a) or a cotnplex word in (6b)) would require the application of operations in order to turn phrasal constituents into a word in (6a) or to separate the parts of a word in (6b). Such operations can easily be avoided by assuming dual realization of phrasal verbs as words or phrases. As to nominalizations from phrasal verbs, I also believe that some are word-like in nature while others are phrasal. The verb-er-adverb structure is a phrasal kind of derivation. To see this, note first that there is clear evidence that nominalized verbs in -er can still take complenients to the right of them, as in (7a-f):

(7)

a. fighter.fir,freedonl b. giaopers in /he durk (www.time.com/tirne/magazi11e/article/0,917 1,l 10195 1023133289,00.htmI, accessed 28 May 2008) c. giver qf weallh lo all peoyle ( w w w . a n c i e ~ i t i i ~ d i a . c o . u k / h i n d u i p ~ s l i a s . l i t ~accessed nl, 28 May 2008) r moneyfor good causes d. r a i ~ e of (www.rhodesfami ly.org.uk/blog/2003/ 1 1/to-be-frank-this-isthird-in-my.ht~iiI,accessed 28 May 2008) e. reader of stories to grandchildren (www.dundurn.com/ .../category-id, 15 1 /manufacturer-id,O/ opI/, accessed 28 May tion, co1n-~irtue1nart/Ite1iiid,28/~1iiccIih, 2008) f. niuker ofpeace wijh /he Palestinians (www.almeghari-palestine.bIogspot.com/,accessed 28 May 2008)

Unless these sequences are to be analyzed as word-level entities - an analysis which is absurd and untenable -there may be something to be said for a parallel analysis for, e.g., giver-in, namely as a verb-derived noun accompanied by a complement. together forming a nominal phrasal structure. Another argument for such a phrasal analysis is that an inflectional affix (e.g., the plural morpheme) typically attaches to tlie -er affix (e.g., lookers-on) rather than to the following adverb (e.g., ?*looker-ons). If the adverb in tlie verb-er-adverb pattern were felt to be fully incorporated into the morpliological derivation, we would expect to find the word-final plural more often. Such a plural form can often be found forpasser-by, but the standard plural form (pu.sser,s-by) still reigns supreme. The only difference with phrasal nolni~ialderivations like those in (7c-f) is that the adverb obligatorily precedes rather than follows the nominal complement which corresponds to the verb's direct ob.ject: (8)

a. b. c. d.

keeper {uway) of drug deu1er.s { *uwa-y) picker {up) of trash /*up) cleaner {out) oj'sewers /*out) (BNC') cl7utter {up,' ofgirls /*up: (BNC)

By contrast, double -er derivations are more morphological. While they also occur in phrasal nominal structi~reslike those in (8), yielding con-

Doubler-upper nouns

3 55

structs like keeper-awayer oj h u g dcalerc, their use in such nominals is considerably less frequent than tlie use of tlie verb-er-adverb type (but inore frequent than conversions in si~chstructures: compare the ~tujferupper c?f the conccw and the sttifJup [*of the concert] [Mclntyre 20071). A~nongthe 25 different double -c.r lexemes with upper gathered via a search on Google (see note 4), only 3 had a following colnplelnent (cleaner-upper5 of the natural ~.vorld,washer-uppers of dirty cup., and plate,, and owner-upper\ of modern I&). By contrast, among the 5 5 verbcr-adverb lexemes in the BNC', we found 17 lexernes which occurred with a phrasal complement to their right (e.g., knocker-about of buildings and ca\tlc\, pziller-down of ~tutme\of the Shah crnd his futher, brznger-in cf new czi.ttom,\, cohhler-up qf ,ztconis, etc.). Doubling the -er is not necessary in such nominals since there is no need to prevent the adverb from being analyzed as a complement. Such a need does exist, apparently, when the affixed verb is the head of a synthetic compound. Consider the following examples:

(9)

a. b. c. d.

drug dealer keeper awayer trash picker uppelsewer clecrner oziler girl chatter q p c r

Without the second -er morpheme, these constructs would not be acceptable, as we have noted in the introduction. Compare: (1 0) a. *drug dealer keeper awuy b. ? *trmh picker up c. ?*sewer clecrner out d. ? *girl chutter up The reason is that the presence of a noun preceding the head of the derivation forces us to consider the derivation as morphological, while a nonaffixed adverb, as we have just argued, is considered to belong to a phrasal kind of derivation (i.e., a nominal and not a noun). Observe that bonaJide phrasal coniplernents are also often (though not always)9 hard to retain to the right of synthetic compounds:

(1 1) a. ?storyteller to chilu'l.er7 and udu1t.s (attested but of dubious acceptability) (www.stolytnagic.co.uk, accessed 2 May 2008)

b. :)present giver to niy brother (idem) (jaimescli~~~arz.blogspot.co1n12005/ 1 Olive-finally-beenreplaced.litml, accessed 2 May 2008) c. "[email protected]. d. "bricklayer drinker under the table Much more natural than (I la) is ~torytellerfor children, which has 259 hits on the U K pages of Google (October 9, 2007), as opposed to just one hit for storyteller to children. By contrast, the VP tell stories.for children has only 4 hits as opposed to 646 hits for the VP tell stories to children. 'This distribution is not at all surprising if we assume that the to-PP is a verbal complement (i.e., an argument) and thefir-PP a postmodifier to a noun. On the other hand, ( I la) and (1 I b) sound somewllat more acceptable than (I Oa-d). This may be because the verb and the adverb have a strong collocational bond which prevents the adverb from remaining affix-less when the verb is turned into the head of a synthetic compound. A strong collocational bond would then explain why (I lc) is also ungrammatical. However, there is no collocational link between drink, bricklayer and table in (I Id). Here, the explanation might be that bricklayer in syntactic structure is not the object of drink but rather the notional subject of the understood predicate nlove/go under the table. As a result, *brickluyer drinker in itself is a faulty compound (unless when intended as a dvandva), whether or not it is followed by a nominal complement. Some speakers apparently find intriguing ways of incorporating this complement into the synthetic compound as well. Since it cannot legitimately bear an affix (*light tripper fanta.t/ic-er), it can only be synthetically integrated as a non-head, following the other non-head. And indeed, both light funtustic tripper and brickIqyer under the tuhle drinker call be found on the web, with one occurrence each. It is clear that these constructs are not to be considered as Standard English. 0 1 1 the other hand, their occurrence reveals a strongly felt unease with agentive synthetic N+N compounds which are followed by a complement (here:,fantus/ic; under the table) of the nominalized verb. What is important to remember is that we have shown that washer-up and warher-upper are different kinds of nominalizations: the former can be analyzed as phrasal in nature (i.e., as a nominal), whilst the second -er in the latter forces us to analyze the adverb as an integral part of a single colnplex word (a noun), wliich may further take part in synthetic compounding.

Doubler-upper nouns 35 7 3.3.3. Not one hut three correspondence viol~rtion.~ ,jOr the double -er slructure

A third problenl with Ackelna and Neeleman's (2001,2004) account is that it considers each of the three competing nominalization types with -er to be stig~natizedwith just one violation - Inore specifically, each type violates a different principle. However, if we take Ackema and Neeleman's account to its logical conclusions, it becomes clear that the double -er structure actually violates three principles: not only does it infringe the "quantity principle", which dictates that only one affix be added, but just like the verb-er-adverb pattern, it violates what they call the "linear correspondence'' principle since (the in this case first) -er follows only a part of its morphological host, and, just like the verb-adverb-er pattern, it violates the "input correspondence principle" since the adverb to which (the in this case second) -er is attached is not the head of the morphological input. 'This triple violation should in principle make it worse than the verbadverb-er pattern, which is taxed with only one violation. Yet, remember from Section 2.3 that the double-er type is more frequent than the verbadverb-er type. To summarize section 3.3, an account of the double -er structure is not satisfactory if it hinges on any of the following wrong assumptions: (i) adverb-verb co~npoundsare i~navailable(as opposed to noun-verb compounds) so that we cannot have, say, *upfixer (as opposed to, say, truck driver); (ii) all verb-adverb combinations are syntactic phrases, irrespective of the ordering of the particle with respect to an ob-ject N P (e.g., wash the dishes up and wash up the dishe,,), and all their derivations are single morphological words (e.g., wmher-up and washer-upper); (iii) there is a simple set of mapping principles between morphosyntactic stritcture and phonological output, such that each derivational type flouts just one principle. We have shown each of these assutnptions to be untenable. As a consequence. we need to look for an account which does not rely on them.

3.4. Overcoming problems with left-headed compounds Finally, let us explore Mclntyre's (2004) hypothesis that the double -er type might be "somehow a natural response to the problem of how to affix

left-headed structures given the less-than-wonderful status of nonred~~plicative soli~tions(-washer-up', '\\ash-upper')". 111 fact, this seems to me to be tlie key to an understanding of the double -er structure. The main problem witli tlie verb-er-adverb structure is that the adverb keeps 'dangling' at the right-hand side of the verb-derived noun. This in itself is not problematic, unless the nominalization undergoes synthetic compounding. In this case, there can be no word-external material to the right of the compound's liead, as we have seen in Section 3.3.2 (examples (IOa-d)). The main problem witli the verb-adverb-er structure is, as Ackema and Neeleman (2001, 2004) point out, that tlie suffix is riot attached to tlie head, as it ought to be - or, at least, as it usually is. The double -er type then overcomes both problems by more clearly integrating the adverb into the nominalization while retaining tlie verb as (co-)bearer of the derivational suffix. According to Busuttil (2001) and Chapman (2008), the double -er structure is not so much a synchronic compromise between the verb-er-adverb structure and the verb-adverb-clr structure, as a diachronic transition between the former and tlie latter. Such a develop~nentwould be somewhat reminiscent of changes in inflectional morphology whereby the inflectional affix does no longer always attach to tlie head in a left-headed complex word, as in si~ters-in-lawor Atlorney,,c-General, and is increasingly attached to the periphery, as in .,cis/er-in-1aw.r (cf. also the term in-laws) and Attorney-Generals, tlie latter form having practically become the norm. I am not so sure whether the double -er structure can be dismissed as merely a transitional form. Such a view would be supported by the emergence and firm establisliment of double -.\ plurals like si~ters-in-luwsbefore the appearance of a peripheral plural like sister-in-laws. I do not think such a development can be substantiated by data from diachronic corpora, altliougli this still needs to be checkcd in sufficient detail (see Chapman [2008] for a first attempt). More plausibly, I believe, sisters-in-luws and washer-upper are to be seen as synchronic attempts to do two things right: add an affix at the morphological head and add an affix at tlie end of a word. One might point out that there is still the "input correspondence" violation caused by tlie -er attached to the adverb (cf. Section 3.3.3). That is, since the adverb is not tlie category or the head of the phrase which serves as the morphological host of the -er suffix, it sliould not be the plionological host either. However, an alternative way to look at the double -er structure is not as a combination of a suffixed verb and a suffixed adverb, but as a blend of a norninalized verb and a nominalized phrasal verb (cf. Blevins

Doubler-lrppcr nouns

3 59

2006: 5 14). That is, we might argue that the second -er does not attach to the adverb per se but to tlie entire phrasal verb:

c. [N wmher-upper] From this perspective, the double -er structure is no longer in violation of Ackema and Neeleman's (2001, 2004) "input correspondence" principle. After all, tlie phrasal verb (i.e., the left-most V-node in [12b]) is the category which is the morphological and phonological host of the second -er. According to Elenbaas, tlie fact that tlie -er affix can appear on both the verb and the adverb is illustrative of the ambiguous nature of English VPCs [i.e. verb-particle constructions B. C.], which straddle the boundary between syntax and morphology. Suffixation on the verbal part follows syntactic principles, whereas suffix placement on the predicate part signals an interpretation of the VPC as a word. Reduplication of the suffix acconimodates the principles of both syntax and morphology. (2007: 18) -

This tics in with what we noted at the end of Section 3.3.2: When -er is added only to the verb, a phrasal kind of nominalization is produced; in addition, when -er is added to the adverb as well, this invites an analysis of the verb and tlie adverb as forming a ~norphologicalunit. However, there is a difference in perspective between Elenbaas' (2007) discussion and mine. In the quoted passage, Elenbaas explicitly links tlie two affixation sites (head and word boundary) to the ambiguous status of the construction from which the nominalization is derived, while I have compared verb-er-adverb and verb(-er)-adverb-er nominalizations in terms of tlie status of the derivational product. It might be tempting to assume that there is perfect correspondence between tlie status of the input and the status of the output of tlie derivation. However, it remains an issue of f~~rtlier research to find out whether a combination which serves as input of the morphological process and which is perceived as 'syntactic' will produce a phrase-like entity as output, and whether an input combination which is analyzed as a 'word' will produce a word-like derivation as output.

360

Bert Cuppelle

4. Interacting rules

We are now at a point where we can explain how the double -er structure comes about. In the prebious section (Section 3.4), we argued that this structure might arise as a way of complying with two principles: adding an aftix to the morphological head and adding an affix to the boundary of a word. Language users have ample evidence that both of these principles are at work in the language. Usually, these principles converge. English is generally right-headed, and, therefore, if a suffix is added to the morphological head, it is usually added to the right periphery of a word. In cases where tlie head of a morphologically complex word is located on the left, we see a tension between these two principles (cf. also Chapman 2008) should tlie suffix appear on the head or should it appear at the end of the word? This tension can be resolved by giving the affix two phonological appearances. Or, as Denison (2008: [9]) puts it, "it is plausible to regard double marking with -er as a compromise between opposing demands". This conflict resolution differs greatly from the way conflicting gram~naticalprinciples are ilsually considered to interact. In Optimality Theory, for example, principles are ranked according to their importance, and in tlie case of a conflict between competing principles, more important principles take precedence over less important ones. In the case of double -er nouns, however, there are two conflicting principles ('suffix should go on the verb' versus 'suffix should go on the right word-boundary') without one principle winning out over the other - rather, they are both winners. 1 cannot think of other structures in which this is the case, although morphological blends might be an obvious place to start looking for that. In fact, there are several niorphological and phonological principles whose workings could be argued for in double -er nominalizations but which exist independently of such nominalizations. A quick overview is given in (1 3a-g). The individual principles will then be brietly commented on further below. (13) a. A derivational affix attaches to the head of a morphologically complex entity. b. A derivational suffix attaches to the right boundary of its base. c. The head in a synthetic compound is not (normally) followed by any further elements which correspond to syntactic complements of the base of the head.

d. A morpliologically co~iiplexentity which consists of two rhyming parts produces a jingling effect in (actual or imagined) pronunciation. e. A ~norpliologicallycomplex entity which consists of two parts with identical stress patterns produces a jingling effect in (actual or imagined) pronunciation. f. Tlie final derivational suffix of a word, if there is one present, determines tliat word's categoriality and its global semantics. g. A plural andlor genitive affix attaches right after the noun's category-determining derivational suffix, if there is one present, and by default word-final ly. Tlie morpliological principles stated in (13a) and (13b) have already been mentioned in this section and will not be further discussed here. The principle given in (13c) has also been discussed already: in Section 3.3.2 it was invoked to explain why there is something awkward about nominal derivations like tlie one in (14): (1 4) I hereby hequeuth to you, iZ.fury, the title c?f'bestpresent giver to r ~ y brother. Cjaimescliwarz.blogspot.co1~il20O5l1 Olive-finally-been-replaced.litm1, accessed 28 May 2008)

Such structures do occur, but the presence of a syntactic co~nplementfollowing a synthetic compound yields a somewhat unco~nfortablemixture of syntax and ~norpliology.When nominalizing transitive phrasal verbs, one can avoid mixing morpliology and syntax by adding an -er affix to tlie adverb in such a synthetic compound (compare again ?*girl chatter up and girl chutter upper). It is for a similar reason that an extra -ed suffix is often added to participial ad-jectives which are derived from phrasal verbs and which function as tlie ad-jectikal base of nouns ending in -ness. Thus, in addition to fucked-upne.c.c, one can encounter fucked-up@)edness (cf. Mclntyre 2004; Peters 2006). The reason is presumably that -ness requires its base to be a word-level adjective and that fucked up may be felt to be too much like an ad-jectival phrase to meet tliis requirement. Adding an -ed suffix to the end of tlie base gives tlie derived adjective the desired form for further -ness suffixation. Some support for tliis claim emerges from the fact that a form like,ficked-upped without -ne,ss hardly ever occurs - its rare occurrences can be explained as backformations from tlie noun ending in -ne.\s.1 0 Tlie more general rule thus seems to be that the input of a mor-

362

Bert Cuppelle

phological operation (whether it be synthetic compounding or suffixation) should look like a word-level (Xo) category, if at all possible. Since phrasal verbs often function as words syntactically (cf. for example [6a] above), the adverb may be analyzable as incorporated into the verb and, hence, as eligible for carrying a suffix. When it receives a suffix, the verb-adverb combination acquires de facto word-level status. The principles in (13d) and (13e) are phonological. The avoidance of 'jingling' effects has been discussed extensively with respect to consecutive ing-forms (see, e.g., Ross 1972, Milsark 1972, Berman 1973 and Pulli~m1974 for some early references). Doubling -er also seems to have a jingling effect, and this effect is probably intended - in any case, the jingling is in all likelihood the main factor why double -er constructs sound comical. Remember from Section 2.4 that there are also instances with triple -er, e.g., bringc.r-backerer, sorter-outerer, washer-upperer, etc. Note that a third -er is almost always added to the second, hardly ever to the first. This could be due to the fact that the first affix is less important than the second, as we have seen in connection with forms like picker-uppee. I can see two different reasons for the occurrence of these 'retriplications' (as Mclntyre 2004 calls them). One is that the producer of a double -er form realizes that the result is (or will be) finny and, in an attempt to underscore this self-awareness, makes it even funnier by adding a third, wholly superfluous -er. For the same reason, perhaps, some users produce even quadruple -er forms like hunkrer-ouferer, pmserer-onnerer, pickerer-upperer, etc., which are attested on the web. Another reason is that language users may try to niitigule the jingling effect by adding a third -er, which usually results in the combination of a trochee and a dactyl instead of two consecutive trochees. Some further speculations are briefly provided by Denison (2008). Principle (13f) is the principle which makes us conclude that a word of the form X-able is an ad-jective meaning 'fit for X-ing or worth being X-ed' or that a word of the form X-it-y is a noun referring to the 'state or quality of being X', etc. Note that it is only derikational suffixes and not prefixes that have a category-determining effect in line with the Righthand Head Rule (Williams 1981). For instance, while we know that de- can be used to form verbs (e.g., derange, dejorps/), a word which starts in cr'e- is not necessarily a verb - it can be a noun deriked from such a verb (e.g., derangeinen/, &forestation). Similarly, while the inf'ixal element -0- is used to modify or form adjectives (e.g., socio-economic, verhocen/ric), it is the right-peripheral affix which ultimately indicates the word class of the derivation (cp. socio-econon?icn, verhocentricity). So, what matters to the in-

terpretation of a word in terms of its categoriality and schematic meaning is the final derivational morpheme. It is probably because of principle ( I 3 9 that a form like breaker-uppee is not to be interpreted as agentive, despite tlie presence of an -er morpheme. The -ee morplierne is right-peripheral, so it determines the categoriality (noun) and meaning (patientive) of the derivation. Finally, principle (13g) states that when a marked inflectional form is needed, tlie inflectional ending is attached to the (last) derivational suffix, if there is one, and as a general rule word-finally. Thus, the plural of computer user is not, of course, *con~pu/ersuser but coniputer users. After all, a conlpu~eruser is a kind of user and not a kind of computer, so the plural is marked on the part (including tlie affix) which determines the general meaning and categoriality of tlie derivation. Usually, this affixation site coincides with the end of tlie word, which explains why the plural of dvandvas like singer-songwriter and hunler-gatherer is usually only marked word-finally (singer-songwriters and hunter-gatherers), although some speakers might mark it on both parts (singer,s-songwriters, huntersgatherer-s). I believe that principle ( 1 3g) also accounts for a fact mentioned in note 4, namely that the plural (andlor the genitive) of a double -er form is generally marked only 011 the second -er, not (also) on the first. In other words, we can say two washer-uppri., but not " t w o ~lmhers-uppersand definitely not *two ~)a,rhers-upper.This again ties in with tlie observation that the first -er is vacuous. It is overshadowed by the second -er, which attracts any further inflectional morphemes. If the category-determining derivational affix is not word-final, as it usually is, tlie inflectioiial affix generally won't be word-final either. This is why tlie competing verb-eradverb pattern normally receives a plural andlor genitive -s on the -er (e.g., runners-up). However, since a category-defining affix, if any, is typically word-final (as per principle 130, and since tlie default inflectional affixation site is at the right word boundary anyway, one can witness the occasional occurrence of forms where an inflectional affix appears word-finally without being directly attached to a mid-word category-determining derivational affix (e.g., :'*runner-ups, ? *passer-bys).

5. Prepositions rather than adverbs taking -er An interesting case concerns similar-looking constructs in which the second -er attaches to a preposition rather than an adverb. Here are some attested examples:

364

Berr ('uppellr

( 1 5) a. lobl~y-drifier-/hr*ougI~cr (Wentwortli 1936: 369)

b. videogame writer-abouter (ps2.ign.co1n/articles/797/797664p I .litml, accessed 28 May 2008) c. hand~lagonjuniper onner (w~v.ocre1nix.org/foru1ns/~Iiowtliread.plip?p=57338, accessed 28 May 2008) d. window-looker-outer (boosIiay.blogspot.c01n/2005~10~0 I-arcliive.litml, accessed 28 May 2008) Most adverbs tliat occur in phrasal herbs double as prepositions: about, LIL.TOJ.\, long, arozmd, aside, by, down, in, o f , on, out, over, round, lhrough, and zq) are words that can occur with or without a complement NP. It is mainly for this reason tliat adverbial particles in phrasal verbs have often been called "intransitive prepositions" (see Cappelle 2004 for references and extensive discussion). Tlie examples in (1 5) seem to invalidate Miller's claim tliat the double -er structure only occurs witli "intransitive prepositions" (1993: 132). Of course, Miller is right to the extent tliat we cannot have, for instance, *&ifter-6hroughe1~-lobbie.s.However, 1 maintain that the host of the second -er in each of tlie examples in (1 5) is to be considered as a transitive preposition, not as an intransitive one (which I would call an adverb). Its N P complement is realized as the non-head of a synthetic compound. Tlie non-head in a synthetic compoulid normally functions as a complement of the verb from which the head is derived. In these cases, though, tlie initial noun is not a colnplenielit of the verb but a complement of tlie preposition. Tlie question arises whether sucli instances are created by analogy witli forms like Jewer cleaner outer and other such forms involving derived verb-adverb combinations (cf. tlie examples in [9]). If so, it would mean that a liypotliesized construction with tlie form 'noun-verb-er-adverb-er' is extended to encompass prepositions as well as adverbs. In tlie light of what we have come to learn about tlie double -er structure, an account based on tlie existence of sucli a low-frequency construction is not only implausible but also unnecessary. We can explain tlie exa~nplesin (15) by taking recourse to tlie principles stated in Section 4, plus the independent principle that prepositions in English can be stranded (e.g., this is the lobby which he used to driji through). In (16), however, the stranded prepositions are interpreted as syntactic elements in otherwise fully inorpliological structures. Just like an i~n-affixedadverb following the head of a synthetic compound

Doubler-upper nouns

365

(e.g., *,sewer clcuner out), such a stranded preposition violates tlie principle given in ( 13c): ( 16) a. "lobby-drifier-through

b. *vi&ogurne writer-about c. *ban~lwugon jun~peron d. *window-looker-out It is by adding an extra -er to the preposition that the derivation becomes morphological. If tlie affixed verb is not t l ~ ehead of a synthetic compound but the head of a nominal structure, there is no reason to add a suffix so as to integrate the word-external stranded preposition into the compound. The following examples are all nominals (rather than nouns) whose head is a noun derived from a verb. ( 1 7 ) a. drijier fhrough the burs and honky-tonks of South Loz~isiana

(liomepages.paradise.net.nz~triffid/wotired/liolidays58.litm,accessed 28 May 2008) b. wriler about gunie5 (www.gamesetwatch.com/2007/05/garne jour~ialist-todeveloper -o.php, accessed 28 May 2008) c. junlper on the b a n d ~ ~ ~ r g o n (www.youtube.com/watch?v=vfiqzFhHhd8,accessed 28 May 2008) d. looker oul the window (webble.orangecrayon.co1n/archives/000023.html,accessed 28 May 2008) In short, much tile same general principles apply for the derivation of verbPP sequences as for the derivation of verb-adverb sequences (known as

phrasal verbs). There is no need to explain instances like lobby-a%iJierfhrougher as being tlie result of an extension of a special low-level constructional template.

6. Can a usage-based view be salvaged'! This study started out by raising tlie following question: Given the rarity of double -er forms, how come speakers use them, and how come they know how 1 0 use them'? Before answering that question, it was necessary to make

sure that double -er forms are actually as infrequent as they have usually been made out to be. A systematic search in the BNC' (on *abouter, "ucro~.\er,*aheuder, etc., where '*' is a wildcard) returned only four occurrences. COCA contained significantly more occurrences, but the overall number was still small. This confirms that the frequency of double -er is too low for speakers to derive all of'the fine gramnlatical details of this structure from its very instantiations. For example, it seems very unlikely that speakers know merely on the basis of their rare encounters with double -er forms that the plural is formed on the second -er or that the double -er structure is preferred over its main competitor when it combines with another noun as head of a synthetic compound (compare, e.g., trash pickerupper and *trash picker-up). We therefore have been forced to invoke the existence of several more general principles which together explain the occurrence of double -er forms and their gratntnatical properties. This conclusion might be thought to seriously undermine usage-based accounts of language. In such accounts, learners acquire knowledge of constructional patterns through frequent exposure to instances of these patterns. Clearly, if there is hardly any direct evidence for a double -er structure available in learners' linguistic input, they must acquire their knowledge of this structure via more general rules rather than from familiarity with tlie pattern itself. Yet, I believe that just because a linguistic structure requires a rule-based account, this does not necessarily mean that usage-based lnodels of language learning have been proved untenable. First of all, there can still be some item-based learning. While double -er coinages can best be explained as resulting from a variety of general rules rather than as being licensed by an idiosyncratic double -er construction, there are also a few double -er forms which may be entrenched in the language. This is probably the case for fixer-upper ('house which needs to be repaired and which is sold at a bargain price') and rnaybe also for picker-upper or even washer-upper. These forms can still be derived from morphological rules, but their relatively frequent occurrence could easily lead to their redundant storage in tlie lexicon. This is why Langacker (1987) re-jects the traditional distinction between rules and lexically stored exceptions to rules (his so-called 'rule/list fallacy'), claiming that certain phenomena might perfectly conform to rules and still be listed among the memorired units (that is, witho~ttbeing exceptions to tlie rules). Expanding this point, it has recently been shown that rules of very high generality, such as those governing tlie formation of questions with long-distance dependencies, may not only co-exist with lower-level templates (e.g., WH do you think ... Y or WH did you .say... ?) but may also be less salient than these

Doubler-upper nouns

367

lexically more specific schemata (cf. Dqbrowska 2006). It is not very likely that this is also tlie case for the linguistic rules which underlie tlie formation of double -er forms. Yet, we have suggested that a speaker's familiarity uith reasonably well-entrenched forms like Jixer-upper and pickerupper might lower tlie threshold to create a similar, novel double -er form. Such exemplar-triggered facilitation would involve the creation of an ad hoe constructional template - what I have called a 'virtual' construction in this chapter and what might correspond to what Kay (2002) has called a 'pattern of coining'. Of course, it is possible that not every language user is equally good at extracting such a template from concrete instances. To use an example 1 have given elsewhere (Cappelle 2008), internet searches re(?ff, wit11 such substiveal a creative use of the idiom verb one *.( {u.s.s/~u~I) tutions for the object head as behind, booty, hotty, hunt, buns. huttock~, funny, hiney, jacksie, rear, tail, tzuh, etc. - in fact, any possible alternative for u.ts tliat could be found in a good thesaurus (cf. also Culicover 2007). Yet, most people, if they were to use this expression in the first place, would only use tlie version wit11 either UA,S or butt, failing to create a virtual pattern in which the object's head noun is replaced by 'any word referring to one's behind'. In the same way, some language users may be slower than others to seize the opportunity to create analogical instances to fixerupper and/or wanher-upper by means of a low-level generalization. Whether such a generalization is extracted at all by some speakers is an open question, but the possibility exists in principle. 'There is another reason why the (predominantly) rule-based account proposed in this study can be compatible with usage-based theory. The thrust of my account has been to show tliat speakers could create a novel (or, for that matter, existing) double -er noun even without ever having heard such a noun before. 1 would like to stress tliat this does not mean that these general rules tliemselves cannot have been learned in a way that is advocated by usage-based theories. In other words, there is no reason to assume that these rules should be part of our innate linguistic endowment: they can very well be learned on the basis of direct positive evidence, that is, from frequent exposure to concrete instances which satisfy the rule and hence, in effect, collectively embody the rule. Admittedly, there can be 110 direct positive evidence for the rule given in (13c), which states that a synthetic compound is not naturally combined with syntactic complements of the nominalized verb. Even so, this rule could perhaps be learned on the basis of indirect negative evidence, that is, on the basis of a conspicuously

3 68

Bert Cuppelle

low number of structures wliicli might reasonably be expected to occur much Inore often. Whetlier or not these interacting rules are indeed learned via positive or indirect negative evidence is beyond the scope of this paper. All that is being argued for here is the fact that while speakers manage to master the tine grammatical details of the double -er pattern in spite of its rarity, this does not ,force us to conclude that speakers rely on grammatical rules which must be innate. The double -er pattern is a kind of grammatical crossbreed whose general structures giving rise to it (in particular, head + -er and word + -er) are so common that we certainly cannot exclude that their properties are learned on tlie basis of linguistic evidence.

7. Conclusion In sum, rare but regular double -er nouns call for an analysis wliicli sees them as tlie output of various interacting grammar rules. These rules euplain not only their occasional occurrence but also their surprisingly stable grammatical properties, wliich can hardly be learned on tlie basis of these infrequent occurrences alone. However, this apparent victory for rulebased language knowledge need not be interpreted as a victory for the linguistic innateness view or as a defeat for usage-based models of language knowledge. For instance, just because speakers who have never encountered the plural of the patientive form breaker-uppee know it must be hreuker-uppees and not *breuker.c-uppee does not mean that speakers' knowledge has to be described in terms of innate principles of universal grammar. On the contrary, this knowledge could be ascribed to speakers' familiarity with a wealth of examples in wliicli tlie plural is marked right afier the meaning-determining derivational suffix in words (e.g., drivers, inhabitanls, divorcees, etc.) and an even greater wealth of examples in wliicli tlie plural is marked at the end of words (e.g. cats, dogs, houses, trees, etc.). So, the rules which account for tlie grammatical properties of the double -er pattern, and for the very use of the pattern itself, can very well be learned by imitation (i.e., via positive evidence) or by avoidance of what is expected but conspicuously absent (i.e., via indirect negative evidence). Certainly, I liave not demonstrated that this is how these rules are actually acquired, so I liave not fumislied any evidence for a kind of 'indirect' usage-based acquisition of the doubler-upper construction, that is, for acquisition based on the observable behaviour of other patterns. Yet, to tlie ex-

Doubler-upper nouns

369

tent that I have shown that u'otlhler-upper. nouns share their grammatical properties with other, much more familiar linguistic phenomena, I hope t o have at least argued convincingly that doubler-upper nouns pose less o f a threat t o a usage-based morphological nod el than their rarity would initially lead us t o assume.

Notes In shaping my ideas about the morphological phenomenon discussed in this paper, I have benefited fiom a lengthy e-mail exchange with Andrew McIntyre, who I am confident will eventually come up with a more penetrating analysis than I can offer here. I also thank the audience attending a presentation of an earlier version of this paper at the Second Conference of the UK Cognitive Linguistics Association for various helpful suggestions. This paper also benefited from helpful comments by the editors of this volume and by Judith Hcidc. I gratefully acknowledge financial support fiom the Research Foundation F1unu't.r~(FWO-Vl~randeren)for a postdoctoral fellowship. 2. ablauttime.blogspot.com/2004/09/passers-by-be-damned.html,accessed 23 August 2007. 3. For some additional references, see McIntyre (2004). Some of his references are to non-published studies and are thus not easy to come by. The publication by Chapman (2008) and the discussion of it by Denison (2008) only came to my attention after the present paper had undergone reviewing. Therefore, no attempt has been made to fully integrate the analyses and findings of these recent publications into the literature overview. Nevertheless, I have added occasional references to these studies in niy text when they provide significant complementary information or when they fonn interesting points of convergence or differcnce with my paper. For Denison's paper, page number references are to the version that is fieely available online. 4. 1 searched on Google.co.uk for matches of the following string: "the~these~those~some~manyjfew(several * uppers" This yielded 37 1 matches in the viewable pages. The reason for using the plural in my search query was to see whether the verbal part was also inflected for plural which it turned out not to be, although McIntyre has noted the webattested example debris pickers uppers and I have also come across passersh.y.7 (without hyphen!) in COCA and messers-arounders and movers-uppers on the Internet. Using the plural also excluded large numbers of unwanted occurrences of upper as an adjective. In addition, the reason for using a determiner followed by a wildcard was that I would otherwise retrieve too many occurrences in which the free morpheme upper is used - the upper of a shoe is the 1.

-

-

3 70

5.

6.

7.

8.

Bert Cuppelle part above the sole (and apart from this use, zq~/rersis also used to refer to stimulating drugs, as opposed to doiowner.~,which are sedatives or tranquilizers). With the search command shown, I minimized the number o f noise occurrences mainly to cases in which uppers was preceded by an adjective or qualifying noun, e.g., leuther uppens. .satiri uppers, sturdy uppers, suede upper.~,14'0ven uppers, etc. Bauer and Huddleston's example o f blower-upper has a similarly patientive reading: "'a shell which should be blown up"' (2002: 1655). Likewise, a singer-alotiger is not 'someone who sings along' but a 'song with which you can sing along' and a sopper-upper (as it is used in COCA) is not 'something which sops up' but 'something to be sopped up'. The existence o f such nonagentive readings, while rare, is not unpredictable, since they also occur with non-reduplicative -er nouns. Examples are keeper ('something which should be kept'), must {ln~yer/haver,'('something that you must buy/have'),,/i~lder ('a sheet that is folded and used to hold documents'), sipper ('drink that is meant to be sipped'), scrutcher ('lottery ticket that has to be scratched to reveal the potentially winning combination'), etc. For a list o f 177 such words in English, see the category "head (consonant) shift" on this website: www.tr~rssel.com/flipflop.htm. An even longer list o f reduplicatives can be found on this site:

people.scs.fsu.edu/-burkardt/fun/wordplay/itty~bitty.html. Double -ee seems to be reserved for combinations which sound deliberately childlike or which are used to (self-)mock the English as spoken by Chinese. Some web-attested cases are: ( i ) a. We've been using one o f t h e Exergen tenzporal artery tliermometers. Works very well, und no stickee 11pee buttee. Wal-Mart has them ,for under $30. (in a self-help site for fathers who stay at home to take care o f their children) b. he has durk brown, ulniost hluck, stick-ee up-ee baby hair c. .4ccording to the Observer: 2/2/03, Ronaldo9.s wife Milene, holds the WorId,foothall keepee uppee record, rrlthough not an olytnpic sport, certainlj~an achievement. ( i i ) a. Me jump-ee buck-ee b. Ying 's Cllinee Tukee-Outee in .Jack.sonville, FL c. "Me lun backee tukee him sufee,"the younger of'the Emtern adventurers went on, pointing to his ,father. "Then me mukee walkee ull ulonk you, takce you hack same p l ~ c e eyou comee,froni. ... " (note also the reverse order o f verb and adverb) The -ee suffixes in these usages do not carry a patientive meaning. In the examples in ( i ) , they are in fact typographic variants o f the more frequently adopted -.y ending (.stick-upjy, keepy-uppy; etc.) Seeming exceptions like to air-conciition, to bruit~wu.sh,to breast-fived, to glohr-trot, to headhutit, to purty-crush, to stage-nzanage, to window-shop, etc.

are probably all back-formations (in these cases from uir-conditioning, brecrs/,fieding, globe-trotter, headhunter, party-crashing, stage manager, windowshopping); cf. Marchand ( 1969: 58-65), Selkirk (I 982: 16-1 7 ) , Bauer (1 983: 208), Bauer and Huddleston (2002: 1654). 9. Attested cases like peacemaker wi/h /he Palestiniuns and Iifigiver to our plum/ (as said of the sun) seem fine. 10. Evidence that the first -ed is vacuous, just like the first -er in the double -er structure, can be found in the fact that this -ed can be omitted, as in the following examples from Peters (2006): beat-upedtiess, h1ol.v-upedness, cutchupedness, cr-uck-upedness, mix-upedness, screw-upedness, sex-upedness, etc. In Peters's list, we also encounter forms like blowing-upedness, hookingupedness, mashing-upedness, suckin '-upedness, not-waking-upedness, which are in some respect similar to attested cases like wushing upper, ,fbllowingupps-, and the forms in (i): (i) the going-uppers, he corning-u'o+~)ner*s, the /?overing-about-in-themiddle-u-hi/-e~s(www) The last forin and the forin not-~vuking-upedness(rather than un-wukingupedne.s.s) suggests that we are dealing with a phrasal base that undergoes conversion or derivation: (ii) a. [N [V [VI' [hovering about in the middle a bitller] b. [, [,?, ,[ not waking up]ed]ness] A similar structure as (iia) can be given for crunch-you-upper and ban-thebonlber. mentioned in Section 2.3.

References Ackema, Peter, and Ad Neeleinan 2001 Morphological selection and representational modularity. In Yearbook of Morphology 2001, Geert Booij and Jaap van Marle (eds.), 15 1. DordrechtIBostonILondon: Kluwer. Beyond h.lor/,holog~~:Interfuce Conditions on Word Formation. 2004 Oxford: Oxford University Press. Baayen, R. Harald 2009 Corpus linguistics in morphology: Morphological productivity. In Handbook qf Corpus Linguislics, Anke Iiideling, Merja Kyto, and Tony McEnery (eds.), 899-9 19. (HSK I Handbucher zur Sprach- und Kommunikationswissenschaft I Handbooks of Linguistics and Communication Science 29.), BerlinlNew York: De Gruyter.

3 72

Bert ('uppelle

Baayen, R. Harald, and Rochelle Lieber Productivity and English derivation: A corpus-based study. Linguis199 1 iics 35: 861-877. Barlow, Michael, and Susanne Kemmer 2000 Usage-Based Models oj'Lau7guuge. Stanford, CA: CSLl Publications. Bauer, Laurie 1983 Etiglish Woru'fi,rnzation. Cambridge: Cambridge University Press. 200 1 Morphological Productivity. Cambridge: Cambridge University Press. Bauer, Laurie, and Rodney Huddleston 2002 Lexical word-formation. In The Cumbridge Grammar efthe English Language, Rodney H~lddleston, and Geoffrey K. Pullum (eds.), 162 1-1 72 1. Cambridge: Cambridge University Press. Berman, Arlene 1973 "Triple -ingl'. Linguistic Inquiry 1:4 0 1 4 0 3 . Blevins, James P. 2006 English inflection and derivation. In The Handbook of English Linguistics, Bas Aarts and April McMahon (eds.), 507-536. Malden, MAiOxford: Blackwell. Bolinger, Dwight The Phrusul Verb in English. Cambridge, MA: Harvard University 197 1 Press. Busuttil, Pierre 2001 Quelques elements de retlexion sur des conversions et derivations de verbes complexes. In M6lange.s en I'honneur de Gkrurd Dele'chelle. Fabienne 'Toupin (ed.), 27-40. (GRAAT 25.) Tours: Presses Universitaires Fran~ois-Rabelais. Hybee, Joan 2006 From usage to grammar: the mind's response to repetition. Languuge 82: 529--55 1. Cappelle, Bert 2003 Meervoudig -er bij Engelse partikelwerkwoorden ['Multiple -er on English particle verbs']. Talk given at the "Morfologiedagen 2003, University of Ghent, Belgium, December 4-5, 2003. 2004 The particularity of particles. or why they are riot just 'intransitive prepositions'. In Adposition.s of Movement, Hubert Cuyckens, Walter de Mulder, and Tanja Mortelmans (eds.), 29-57. (Belgian Journal of Linguistics 18.) AmsterdamIPhiladelphia: Benjamins. 2008 What should 'stockings' look like? On the storage of linguistic information. In Du ,firit grcrmmaticrrl uu ,firit cognitif/ From Gram to Mind: Grammar as Cognition, Jean-Remi Lapaire, Guillaume Desagulier, Jean-Baptiste Guignard (eds.), 178-1 89. Bordeaux: Presses Universitaires de Bordeaux.

Dozibler-zlpper nouns

3 73

Chapinan, Don Fher-zrppers and pu.s.ser:s-by: Norninalization of verb-particle 2008 constructions. In Studies in rhc History ofthe English Language IV: Enipirical and Analytical Advances in he Slu4: of English Language Change, Susan M. Fitzmaurice and Donka Minkova (eds.), 265-300. BerlinMe~!York: De Gruyter. Culicover, Peter 2007 Syntactic nuts, core and periphery, and universal grammar. Talk given in the series RCEAL Tuesday Colloquia, Cambridge, UK, May 15,2007. Dabrowska, Eva 2006 The mean lean grammar machine meets the human mind: Empirical investigations of the mental status of linguistic rules. Talk given at the 2nd International Conference of the German Cognitive Linguistics Association, Munich, Germany, October 5-7, 2006. Denison, David 2008 Patterns and productivity. In Studie.~in the History qf the English Language IV: Enipirical rrnd Ar?a(yricul Advances in the Study of English Langzcirge Change, Susarl M. Fitzmaurice, and Donka Minkova (eds.). 207-230. HerlinINew York: De Gruyter. [prefinal version available online: ]

Elenbaas, Marion 2007 The ,Synchrotiic and Diachronic Syntax ofthe English Verb-Particle C'on~bination.(LOT 119). Utrecht: Netherlands Graduate School of Linguistics. Farrell, Patrick 2005 English verb-preposition constructions: Constituency and order. Language 8 1 : 96- 1 37. Kay, Paul 2002 Patterns of coining. Talk given at the Second International Conference on Construction Grammar, Helsinki, Finland, September 6-8,2002. Langacker, Ronald W. E'oz~nclaufion.~ of Cognitive Grammar. Volurne I : Theoretical Prereq1987 uisites. Stanford, CA: Stanford University Press. A usage-based model. In Topics in Cognitive Linguistics, Brygida 1988 Rudzka-Ostyn (ed.), 1 2 7 16 1. An~sterdamiPhiladelphia:Benjamins. Marchand, Hans 1969 The Categories and Tipe.s ofPre.vent-Day English M'ord-Formation. Second Edition. Miinich: C. H. Beck. Mclntyre, Andrew 2004 Sum: English Affix Redup!ication. The Linguist List 15.1929.

3 74

Bert Cappelle i :http:iilistserv.linguistlist.orgicgi-

biniwa?A2=ind0406d&L=linguist&P=6698 > Particle verbs and argument structure. Language and Linguistics Cornpuss 1 : 3 50-3 67. Miller, Gary D. 1993 Con~plexVerb Fori~~trtior~. AmsterdamiPhiladelphia: Benjamins. Milsark, Gary 1972 Re: Double -;rig. Linguistic Inquiry 3 : 542-549. Peters, Mark 2006 Sex-upedness. Wordlustituu'L.,Friday, February 24, 2006. < http:iiwordlust.blogspot.com/2006602~0 I_archive.html > Pullum, Geoffi-ey K. 1974 Restating Doubl -ing. G'1os.s~8: 109-120. Ross, John Robert 1972 Doubl-ing. Linguistic Inqzlil?; 3: 6 1-86. Selkirk, Elisabeth 0 . 1982 The ,?vntax of Words. Cambridge. Mass./London, England: The MIT Press. Sproat, Richard 1985 On Deriving the Lexicon. Ph.D. Dissertation, Massachusetts Institute of Technology. Toivonen, Ida 2003 A'on-projecting word>s.A cirse study o f ,Y~veu'ishpurticles. Dordrecht: Kluwer. Tomasello, Michael 2000 First steps towards a usage-based theory of language acquisition. Cognitive Linguistics 1 I: 6 1-82. 2003 Constructing u Lungucrge: A C~suge-Bused Theory qf Lunnguge Acquisition. Cambridge, MA.: Harvard llniversity Press. Van Marle, Jaap 1992 The relationship between morphological productivity and frequency: A comment on Baayen's performance-oriented conception of morphological productivity. In Yearbook qf Morphology 1991, Geert Booij, and Jaap van Marle (eds.), 15 1-163. Dordrecht: Kluwer. Wentworth, Harold 1936 On adding the suffix of agency, -er. to adverbs. Anicricun Speech 1 1: 369-370. Williams, Edwin 1981 On the notions 'lexically related' and 'head of a word'. Linguistic Inq~ir);12: 245-274. 2007

The influence of morphological structure on the processing of German prefixed verbs Judith Heide, Antje Lorenz, AndrL; Meinunge~: and Frank Burchert

1.

Introduction

I. 1. Polymorphemic words in tlie mental lexicon One of the central issues in psycholinguistic research is how words are represented, accessed, and processed in a speaker's mind. The mental lexicon, LC., the dynamic organization of words in long-term memory, is a vast and complex network of mental rcprescntations, associations, and processes which ensures the great speed and automaticity of word recognition and production in everyday language use. One field of research on the mental lexicon centres around tlie question how morpllologically complex words are represented and accessed. Taft and Forster (1975) were the first to show that cornplex words are decornposed into their constituent morphemes in language recognition and that aftixes and stems may have distinct mental representations (see also Taft 1986, 1988, 2004, Taft, I-lambly, and Kinoshita 1986). However, other authors argue for a holistic representation of polymorphemic words (Manelis and Tharp 1977, Butterworth 1983, Bybee 1995, Rumelhart and McClelland 1986, Sereno and Jongman 1997, Plaut and Gonnertnan 2000), and there is no unequivocal evidence for either n~odelsof decomposition or full-listing approaches. Current psycholinguistic lnodels therefore postulate dual routes or dual mechanisms to allow for both lnorphological decomposition and full form processing (e.g., Caramazza, Laudanna, and Rolnani 1988, Pinker and Prince 1994, Baayen and Schreuder 1999, Clalisen, Sonnenstuhl, and Rlevins 2003). Various psycholinguistic parameters such as regularity, semantic transparency, and frequency are assumed to influence the processing of polymorphemic words, and there is evidence that irregular, opaque, and high-frequent words are listed as fiill forms in the mental lexicon while regular, transparent, and low-frequent words undergo mor-

phological decomposition (Bertram, Schreuder, and Baayen 2000, Baayeti, McQueen, Dij kstra, and Sclireuder 2003). Tlie dual mechanism account has been applied to tlie German mental lexicon by Clalisen and colleagues for inflectional (Clahsen, Eisenbeiss, and Sonnenstuhl 1997, Sonnenstuhl, Eisenbeiss, and Clahsen 1999, Sonnenstuhl and I-1~1th2002) and derivational processes ( S o ~ ~ n e n s t ~and ~ l iHutli l 2002, Clahsen et al. 2003). To account for the different experimental findings on German inflection and derivation, Clahsen et al. (2003) suggest a refined version of tlie dual mechanism approach. Tlie authors argue that regular inflected words are decomposable into stem and affix (and are, thus, sensitive to root pririiing) while irregular inflected words match liolistic entries (and, thus, show an effect of full form frequency). 111the case of derived words (no~ninalizationsand diminutives), both an effect of root priming and of full form frequency was observed (Clahsen et al. 2003). This led the authors to the conclusion that derived words are represented as full for~iisin tlie mental lexicon, but that they are linked to the underived stem form. However, as Clahsen et al. (2003) only used suffixed words in their experiments, it remains an open question whether tlie assumption of a "linked full form representation" for derived words is specific to suffixation or whether it holds for prefixed items, too. Literature on (German) word formation takes different views on tlie relationship between prefixation and suffixation. I11 contrast to Olsen (1986) and Erben (2000), for example, who argue that both suffixation and prefixation are derivational processes, Fleischer (1982) liighlights tlie differences between suffixation and prefixation. He concludes that only suffixation is a derivational word formation process, but that prefixation has to be regarded as a third word fonnation process aside from compounding and derivation. Thus, from a tlieoretical perspective, it cannot be easily decided whether tlie results of Clahsen et al. (2003) for suffixed words are extendable to prefixation. Therefore, the aim of tlie present study is to investigate how prefixed words are represented and processed in the mental lexicon. Specifically, we focus on German verbs which are prefixed with ver- (e.g., verlaufen 'to get lost'). The linguistic properties of these verbs are described in the following section.

1.2. Linguistic properties of Gertiian ver-prefixed verbs Together with he-, ent-, er-, and zer-, the prefix ver- is one of the central German prefixes which form niorpliologically complex verbs by derivation

from different roots. Unlike most other prefixes in German, these prefixes are usually unstressed (except for contrastive contexts) and immobile, i.e., they cannot be separated from their root in a sentence context as shown in ( 1 ) and (2). Thus, it is not trivial to ask whether these prefixes may have separate mental representations. (1)

vergeben a. Jlcli vergehe Dir b. *Icli gebe Dir ver I.forgive .you

(2)

abgeben a. *Ich uhgebe etwas b. Jlch gehe etwas ah I h~rndsoniething in

Among tlie central prefixes, ver- represents the biggest group with 693 verbs (CELEX database, Baayen, Piepenbrock, and Gulikers 1995). Pseudo-prefixed verbs are very few in number. These are verbs beginning with the letter triplet ver- and where initial ver- is not a pretix (e.g., verhalwieren 'to verbalize'). Because of this numerical salience and the limited risk of misparsings, ver-prefixed verbs should have prominent mental representations which can be parsed efficiently (Scliirmeier 2004). Initially, all ver-verbs seem to form a homogenous group. They consist of tlie prefix ver-, a root, and the infinitival suftix (e)n,' resulting in a ver+ root+(e)n surface structure (e.g., verlaufen - ver+lauf+en). However, the internal structure of these items may differ considerably. First, the root type varies, as tlie prefix ver- can be added to nouns (Schlzi,t.sel 'key' ver.schlun.\eln 'to encode'), verbs (Iuufen 'to walk' verlaufen 'to get verhlassen 'to fade'), and bound morlost'), ad-jectives (blass 'pale' phemes (*ge,\s(et~J vergessen 'to forget').' For most words, tlie root can be identified unambiguously. 1-lowever, in some instances it remains unclear whether the prefixed verb (e.g., verliehen 'to fall in love') was derived from a noun (Liehe 'love') or from a verb (liehen 'to love*) (for a further discussion of this issue see Umbreit, this volume). More generally, one can distinguish between ver-prefixed verbs derived from a lexical root (verb, noun, and adjective) and verbs containing a non-lexical root, i.e., a bound morpheme. Second, some ver-prefixed verbs do not seem to follow the Righthand Head Rule (RHR, Williams 1981, cf. Donalies 2007: 21-22). The RHR

+

+

+

+

states tliat the riglitliand constituent of a morphologically complex word is the liead of tliat word and determines the lexical category of the full form. By definition, prefixes are added to tlie left side of a root, and they do not change its lexical category but alter or specify the meaning (Fleischer, 1982). This would mean that the prefix ver- should be added to a (simple) verb, wliicli then constitutes tlie riglitliand constituent of tlie morpliologically complex word and determines tliat the prefixed item, too, belongs to tlie lexical category of verbs. This analysis is uncontroversial for those verprefixed verbs which were derived from a simple verb (e.g., verluufen). However, verbs like ver.schliisseln, verblu.ssen, and verpssen are analyzed as being derived from a noun, an adjective, and a bound morpheme because tlie riglitmost element does not form an existing German verb (*schliisseln, *blu.ssen, *gessen). In these cases, it seems to be tlie prefix, i.e., tlie leftliand element, which determines tlie syntactic properties of the fill1 form. Olsen (1991) suggests that tlie prefix is the head of the complex verb and provides syntactic information ofthe full form (e.g., ver,schliisseln begins with ver- and is therefore a verb althougli it is derived from a noun). However, violating the RHR might not be necessary: One possibility to account for a verb like ver.schliiLsselnis to assume a virtually existing verb .sch/iiLsselnto which the prefix ver- is attached. This, however, implies a two-stage derivation process involving a non existing intermediate form (Plank 198 1). Alternatively, following Fleischer and Barz (1 995), two derivational processes may apply to a root like Schliissel simi~ltanouslyby adding the prefix and changing tlie lexical category at the same time. In such a case, a special word formation type "prefix conversion" must be postulated. Finally, it is possible to derive verbs like verschliisseln and vcrbl~s~sen by circumfixation. Here, prefix and suffix form a discontinuous affix and are added to tlie root simultaneously. Circumfixation is well described for tlie for~nationof the German past participle, and Drijkoningen (1999) shows how the RHR is capable of explaining circumfixation. For tlie case of ver-prefixed verbs, though, one would need to specify whether all verbs are derived by circumfixation or whether some verbs are prefixed and others are circumfixed. Irrespective of the tlieoretical explanation, it is obvious that for some ver-prefixed verbs tlie combination of root+(e)n corresponds to an existing German verb (e.g., luufin in verluujen), while in other cases tlie root+(e)n combination is non-lexical (e.g., *schlii,s.selnin ver.schlii.s.seln)." Finally, the prefix ver- can bear different semantic properties and can change tlie meaning of the root accordingly. Fleischer and Barz (1995) distinguish three main types of ver-prefixed verbs: perJc;ctive verbs indi-

3 79

Processing of German prefixed verbs

cate the gradual completion of a process (e.g., verheilen 'to heal'). A nzodal reading implies that the action is performed incorrectly (verluufen 'to get lost') or too excessively (versulzen 'to oversa~t').~ In an intensive reading, the prefixation with ver- results in an intensification of root meaning (verbleiben 'to remain'). In addition, the semantic transparency of the prefixed verb varies, i.e., the extent to which the meaning of the root is still recognizable in the meaning of the full form. While the meaning of a (perfective) verb like verurmen ('to impoverish') is accessible via the combined meaning of the prefix and the root ( a m 'poor'), this is not true for a verb like vertrugen ('to tolerate'), which is derived from trugen ('to carry'). The concept of semantic transparency does not apply to verbs derived from a bound pseudo-morpheme as the latter do not have any meaning in modern German. As shown above, although ver-prefixed verbs have the common surface structure ver+root+(e)n and seem to form a 1iomogenous group, subsets differ considerably with respect to (a) the lexical category of the root (root type), (b) the lexicality of the root, (c) the lexicality of the root+(e)n combination, and (d) the semantic transparency of the full form. Table 1 summarizes the linguistic properties of German ver-prefixed verbs and gives an overview of the existing subsets. Table I . Overvie~jqfthe linguistic properties of German verbs prefiued with ver(n. a,: not applicable)

s

to sketch

E %

opaque

verstehen to undevstand steh stand stehen to stand

schleudem to sling ve~aten to betmy Rat advzce raten to advzse

vermuten to assume Mut couvage *muten

verelteln tofoil e ~ t evazn l *eiteln

n. n.

380

Heide, Loren-, Meinunger, and Burchert

2. Morphological priming o f ver-prefixed verbs: The study o f Schirmeier, Derwing, and Libben (2004) In a recent study, Scliirmeier. Derwing, and Libben (2004) investigated whether tlie mental lexicon contains any morphological links between a ver-prefixed verb (the full form) and its constituents (prefix, root, and suffix). They used a lexical decision paradigm and applied visual morphological priming to 72 ver-prefixed verbs wliich were either derived from a verb, a noun, an adjective, or a bound morpheme. The target items (e.g., verhittern 'to embitter') were preceded by either a related or neutral root prime (H117EK / \AUBI,II) or a related or a neutral root+(e)n prime (BlITERN / SAUBERN). For all root types, response latencies were shorter in tlie related than in tlie neutral root priming condition. However, the amount of the priming effect was bigger for ver-verbs derived from adjectives and nouns than for full forms containing a verbal or a bound root. Root+(e)n priming showed to be equally effective for all four subsets. Thus, for the verb and bou~idgroup, root+(e)n priming was more effective tlian root priming, while the opposite was the case for ver-verbs derived from adjectives, and there was no difference for the noun subset. In addition, a colnparison of the different root types in both neutral priming conditions showed elevated response latencies for the adjective subset although the subsets had been matched carefully for factors known to affect lexical decision latencies. Schirmeier et al. (2004) explain tlieir results in terms of the psycholinguistic parameters "lexicality" and "morphological structure". According to the authors, lexicality accounts for the fact that the adjective subset benefits more from root tlian from root+(e)n priming, whereas tlie verb subset shows bigger effects for root+(e)n tlian for root priming. While this explanation is convincing for ver-verbs derived from an ad-jective, as tlieir root in contrast to the root+(e)n combination represents a lexical ent~y,it is less plaiisible for items derived from a verb. In the latter case, both root and root+(e)n are lexical (tlie root being the familiar imperative and root+(e)n tlie infinitive). So, either prime frequency (as the infinitive form root+(c)n is likely to be more frequent tlian tlie imperative root) or the extent of visual overlap of prime and target, which is bigger in the case of root+(e)n, may account for the difference found. It also remained unclear whether the (non)-lexicality of a full form's root has an influence on the processing of tlie whole word. On the orie hand, stimuli with a bound root showed only a small root priming effect, which, according to Schirmeier et al., "confirms [their] prediction of an effect of root lexicality in the priming study" (2004: 78). On the other hand, in the neutrally primed conditions,

Processing

c?f

German prefixed verbs

38 1

stimuli with a bound root were not recognized more slowly than those with a lexical root and the authors conclude that "whether or not the root is a real word does not seem to affect whole-word processing" (2004: 85). 'I'lii~s, the results of Schirmeier et al. (2004) cannot entirely answer the question whether lexicality has an impact on the processing of the prefixed verb. Furthermore, if lexicality alone influenced the processing of the full form, one would, according to Schirmeier et al. (2004), expect equal effects of root and root+(e)n priming for items witli a bound root because witli these items both root and root+(e)n are non-lexical. By contrast, the noun subset should have been primed better by the (lexical) root than by the (noti-lexical) root+(e)n combination. As this was not the case, Schirmeier et al. (2004) introduce ~norphologicalstructure as a second parameter supposedly influencing the processing of the full form. Schirmeier et al. (2004) argue that the default morphological structure for ver-prefixed verbs is right-branching [ver+root+(e)n]. Tliis applies to verbs derived from verbs, bound roots, and nouns, as for tliesc subsets root+(e)n priming showed to be more effective than root priming. For tlie ad-jectival subset, however, priming with a root was more effective than priming with root+(e)n, which Schirmeier et at. explain witli the "structural anonialousness" of these items (2004: 85). Althougli tlie idea that the internal ~norphologicalstructure may influence the processing of a morphologically complex full form is promising, tlie interpretation and conclusion of Schirmeier et al. (2004) are somewhat problematic. The authors state that ver-prefixed verbs derived from an adjective (e.g., verurmen) are ,,structurally anomalous", presumably because they contain a lexical root (ur~n)but also a non-lexical morphological substring (*urnzen). I-lowcver, this is also true for ver-verbs derived from a noun (e.g., vcr.schluLs.seln'to encode') with Schliissel ('key') as the lexical root, while the substring *.~chlu.tselndoes not exist.5 Furthermore, the items used in Schirmeier et al. (2004) in the adjective and noun condition contained lexical root+(e)n combinations (e.g., verlzeilen 'to heal' with heilen 'to Ileal' as an existing German verb) as well as lion-lexical ones (e.g., vereiteln -to foil' witli tlie non-existing substring "eiteln). Thus, items within one subset did not necessarily have a common internal morpliological structure making a coherent interpretation of the results difficult. Finally, the assumption of morpliological structure and lexicality being independent parameters is not straightforward as the internal structure of a ver-prefixed verb (right-branching \ s. flat) may depend on the lexicality of

382

Herde, Loren=, Meinunger, and Bzrrchert

its substrings (root and root+(e)n). If this is the case, ver-prefixed verbs with a lexical root+(e)n combination should have a right-branching structure, while verbs with non-lexical root+(e)n have a flat structure.

3. Aim of the present study and research questions The present study investigates in how far the processing of polytnorphemic Gernian verbs prefixed with ver- is influenced by their internal niorphological structure. In the previous section, we argued that there are some flaws in the ~naterialused by Schirmeier et al. (2004). This is why we aimed at controlling our itern sets even more rigorously. Accordingly, we refined the classification systeni and posti~lateda fifth subgroup for items with an a~nbigousroot, which may be either derived from a noun or frotn a verb (cf. 1.2). Secondly, we only used items with non-lexical root+(e)n combinations in the noun and ad-jective subset so that, within a subset, all items were consistent with respect to the lexicality of root+(e)n. Our experiment is designed to test whether the processing of verprefixed verbs is influenced by their root type and their internal morphological structure. The overall question is whether ver-prefixed verbs are represented and processed holistically or in a decotnposed form and whether this depends on the specific linguistic properties of the verb.

4. Empirical investigation 4.1. Methods

62 students (58 women, 4 men, mean age 23.6 yrs) at the University of Potsdam (Germany) participated in the experiment. They are all native speakers of German, right-handed and have normal or correct-to-normalvision and no reported reading disorder. For participation, they either received course credits or were paid 6 Euros.

Processing of German prefixed verbs

383

134 German verbs prefixed with vcr- served as target words. They are listed in the appendix. The prefixed verbs were derived from either an adjective (verbluks.sen'to fade' / h/uLssAdi 'pale'), a noun (verschliisseln 'to encode' 1 LYch/iis.selN'key'), a verb (vcrnrischen 'to mix' I mischen" 'to mix'), or a bound morpheme (verletzen 'to hurt' / "letzten)). For a fifth group of items, the root could not be unan~biguouslyidentified and was either a noun or a verb (verliehen to fall in love I LieheN love1 lieben" to love). All ver-verbs were taken from the CELEX database (Baayen et al., 1995), and subsets were matched for fitll form frequency, root frequency, word length, and number of orthographic neighbours. Prior to the experiment, 60 native speakers of German had classified all verbs as being semantical ly transparent." To counterbalance for yes and no responses in the lexical decision task, we int,ented 134 non-existing but morphologically complex verbs. They were matched in pairs with the targets words according to root type, root frequency, and word length. 25 si~nplenouns, 25 simple verbs, and 25 simple adjectives, as well as 75 simple non-words served as monolnorphemic fillers. Thus, the material contained 41 8 items which were either morphologically complex (n=268) or simple (11=150) with an equal number (n=209) of existing words and non-words. Each target item was associated with four different primes, i.e., a related and an unrelated root prime and a related and an i~nrelatedroot+(e)n prime. Root primes were free standing monomorphemic adjectives (BLASS) and nouns (SCHLUSSEL, LIEBE) (for the subsets of adjective, noun, and nounlverb) as well as the monomorphernic familiar imperative (MISCH) for the verb subset and a nonlexical but root-like element (LETZ) for the bound root subset. For the root+(e)n primes, the suffix -(e)n was added to the root, resulting in infinitival verbs (MISCHEN, LIEBEN) for the subsets of verb and nounlverb and leading to infinitival-like but nonlexical combinations for the subsets of adjective, noun, and bound root (BLASSEN, S C H L ~ ~ S S E LLETZEN). N, All prime words were used twice, once in the related and once in the unrelated condition. To keep the priming procedure constant throughout the experiment, polymorphemic fillers were also asscociated with a related and an unrelated root prime and a related and an unrelated root+(e)n prime, respectively. For the monomorphemic words (e.g., Apfil 'apple') and non-words (e.g., "Upjel) we used identity (APFELIUPFEL) or unrelated (HITZEIPITZE) primes.

1.1.3. Procedure

We created four lists, each containing all items in the same order. Targets and fillers appeared only once per list, and all four priming conditions (related root, related root+(e)n, unrelated root, and unrelated root+(e)n) were equally represented in each list. Each prime appeared in only one list so that across lists each target was preceded by a different prime. Thus, all participants were confronted with all items but in one priming condition only. Tlie experiment was run on a Maxdata notebook using UDAP 3.34 (Universal Data Acquisition Program, Zierdt 1998-2007). Items and primes were presented with black letters (Aria1 24) in the centre of a white screen. In tlie masked priming paradigm (cf. Forster and Davis 1984), each trial consisted of tlie following sequence: A 450 ms fixation cross (+) was followed by a blank screen (5001ns). A forward masking pattern (####), matched in length to each prime word, appeared for 500 ms and was followed by the prime which was presented for 66ms. Tlie prime immediately changed into tlie target string (word or non-word), whicli remained on the screen until the participant responded. Primes were presented in upper case and targets in lower case.' Participants were instructed to respond as rapidly and accurately as possible whether the target string was a German word or not. Participants made the lexical decision by pressing tlie left key of a computer mouse for words ('yes') and the right key for non-words ('no'). Response latencies as well as accuracy of response were measured. Participants were not informed about tlie presence of prime-words, and no reaction time or accuracy feedback was given. The experiment was subdivided into 4 blocks with breaks in between. For each participant, the experiment lasted about 40 minutes.

4.2. Results 8308 responses were recorded for the target items. Data from three participants were excluded from the analyses because their overall response accuracy was below 80%. Nine targets (n=4 from the adjective subset, n=3 from the noun subset, n=l form the nounlverb subset, n=l from the bound morpheme ~ u b s e t )were ~ excluded from the analyses as less than 50% of the participants had identified them as an existing German word. In addi-

Processing of German prefuced verbs

385

tion, we removed from our item set three items which had been classified incorrectly with respect to the part-of-speech of their The overall accuracy for the remaining targets was 92% (range 58loo%), and did not differ significantly between the subsets. However, it should be noted that eight of the nine items that we did not analyse due to enhanced error rates contain a non-lexical root+(e)n combination. We will come back to this observation in the discussion section. To analyze response latencies and priming effects, we only used correct responses with latencies differing not more than 2 standard deviations (SD) from the mean reaction time of the group. Thus, 6396 RTs (i.e., 90% of analyzable and 97% of correct responses) went into the RT analysis, ranging from 262-1541 ms. Figure 1 shows the overall response latencies for the different subsets of ver-verbs collapsed over the four priming conditions. Figure 2 depicts the priming effects measured for root and root+(e)n priming, respectively. Priming effects were calculated by subtracting the mean response latency in related priming conditions from the mean latency obtained for unrelated primes.

Figure I. Response latencies (in ms) for different subsets of ver-prefixed verbs collapsed over all priming conditions.

386

Heide, Lorenz, Meinunger, and Burchert

r root prlwiuru Ircrnt+icl~rW H ~ I ~ P

Figure 2. Priming effects (in ms) for different subsets of ver-prefixed verbs meas-

ured for root priming and root+(e)n priming. We performed a 5 x 2 x 2 analysis of variance (ANOVA) by participants (F1) and by items (F2) with root type (verb vs. nounlverb vs. noun vs. adjective vs. bound morpheme), prime type (root vs. root+(e)n), and prime relation (related vs. unrelated) as independent factors. Both analyses showed a main effect of prime relation (F, [1,58] = 63.45, p < .001; F2 [ l , 1171 = 56.98, p < .OO 1 ) indicating that targets were responded to faster after related than after unrelated primes (686 ms vs. 734 ms). In addition, there was a root type effect (F, [4,2 121 = 32.33, p < .001; F2 [4,117] = 3.59, p < .01). We did not find a main effect for prime type (F, [1,58] < 1; F2 [1,117] < 1) and no interactions. Thus, root+(e)n priming does not seem to be more effective than root priming (or vice versa), and there is no evidence that the effectiveness of a specific prime type depends on the root type of the items to be primed. Pairwise comparisons (Bonferroni correction for multiple comparisons applied, all p < .05) revealed that both the verb and the nounlverb subset differed significantly from the adjective, noun, and bound morpheme sub-

Processing qf German prefixed verbs

3 87

set. Neither tlie verb and nounlverb subset nor tlie adjective, noun, and bound rnorphelne subsets differed significantly. The observation that response latencies were significantly shorter for items derived from a verb or a nounlverb tlian for items derived from an adjective, a noun or a bound morpheme led to a second ANOVA where we replaced root type by nlorphological .\tructure. This factor merged the verb and nounlverb items into a new subset "right-branching structure" wliile the adjective and noun items formed the "flat structure" group. As items with a bound morpheme are not deco~nposableinto lexical units they constituted a third subset. We liypotliesized that the root type effect observed would now be reflected in differences between groups differing in terms of morpliological structure. Furthermore, we expected that for right-branching itelns root+(e)n priming is more effective than root priming, wliile the reverse pattern should be found for items with a flat structure. We again ran analyses by participants ( F , ) and by items (F2), now with morphological structure, prinze type, and prinle relation as independent factors." Both analyses showed main effects of morphological structure (Fl [2,1 161 = 48.2 1, p < .001; Fz [2,12 I] = 5.69, p < .O I) and prime relation (FI [1,58] = 65.02, p < .001; F1 [I ,1211 = 47.66, p < .001) but no main effect of prime type (F, [I ,581 < I ; Fz [ I ,1211 < 1). There were no interactions between tlie three factors. In particular, there was no interaction of morphological structure and prime type (F, [2,1 161 < 1 ; F2 [2,12 I] < 1) and no interaction of ~norphologicalstructure, prime type, and prime relation (Fl [2,101] = 1.51, p = .23; F2 [2,121] = 1.75, p = .IS). Pairwise coinparisons showed that response latencies were significantly shorter (p < .001, Bonferroni correction applied) for itelns with a right-branching structure than for all other items. Items with a flat structure and items wliich are not deco~nposableinto lexical units did not differ significantly.

4.3. Discussion We analyzed response latencies and ~norphological priming effects in a visual lexica1 decision task for different subsets of German ver-prefixed verbs. Response latencies differed significantly between tlie root type subsets. Shorter response latencies were obtained for the verb and nounlverb subset tlian for the noun, adjective, and bound root subset. Based on these findings the following latency hierarchy depending on the root type emerges (with '