switching; speech errors; novel words; full listing; fully parsed; dual listing; mor- ... The question of morphological structure in the mental lexicon has emerged as a hot ... Francisco State University, 1600 Holloway Dr., San Francisco, CA 94132. ... nesian, Korean, Russian, Swedish, Tagalog, and Urdu. .... spark able right now.
Brain and Language 68, 68–74 (1999) Article ID brln.1999.2117, available online at http://www.idealibrary.com on
Cross-Linguistic Evidence for Morphological Representation in the Mental Lexicon Rachelle Waksler Linguistics Program, San Francisco State University Three types of naturally occurring data address the question of morphological structure in the lexical entry: code-switching, novel forms, and speech errors. The data were collected from a wide variety of languages, including Austro-Polynesian, Semitic, and Altaic languages, and a heretofore untapped set of affixation types, including infixes, circumfixes, multiple affixes, reduplication, and nonconcatenative morphology. The data are used to argue for a dual-listing model of lexical representation (e.g., Pinker, 1991; Frauenfelder & Schreuder, 1992) over full-listing (e.g., Butterworth, 1983) and fully parsed (e.g., Taft & Forster, 1975) models. The paper posits three morphological parameters which are activated language-specifically. 1999 Academic Press
Key Words: cross-linguistic; morphological structure; mental lexicon; code switching; speech errors; novel words; full listing; fully parsed; dual listing; morphological parameter.
The question of morphological structure in the mental lexicon has emerged as a hot topic of psycholinguistic debate. Two major competing models claim different units of representation. Full-listing models (e.g., Butterworth, 1983; Dell & O’Seaghdha, 1992) posit that the lexical storage/access unit is the whole word. In these models, each morphologically unique word has its own listing in the mental lexicon. Parsed models (e.g., Taft & Forster, 1975; Taft, 1994) posit single morphemes as the storage/access unit. In such models, every morpheme in the language (i.e., each root and each affix) has its own listing.
Many thanks to Johanna Andersson, Yvonne Carns, Sonja Follett, Afshan Hashimi, Suzanne Hovanasian, Kelly Hogue, Emily Jolley, Helen Kaplan, Lissa Kim, Mae Limpahan, Jovita Rose Ramos, Irit Reed, Judit Tanai, Albert Teguh, and especially Anna Payawal Scanlon for their help in data collection. I also thank two anonymous Brain and Language reviewers for their comments. Address correspondence and reprint requests to Rachelle Waksler, Linguistics Program, San Francisco State University, 1600 Holloway Dr., San Francisco, CA 94132. E-mail: rwaksler@ sfsu.edu. 68 0093-934X/99 $30.00 Copyright 1999 by Academic Press All rights of reproduction in any form reserved.
CROSS-LINGUISTIC EVIDENCE
69
To date, however, most investigations of morphological structure in the mental lexicon include only a small subset of morphological types in a handful of morphologically similar languages. Hypotheses regarding lexical representation need to be tested on a much larger inventory of morphological processes used in languages. To expand the morphological set of inquiry, as well as to identify morphological properties relevant to the question of sublexical representation, I analyze naturally occurring data from a wide variety of languages, including Armenian, Cebuano, Chinese, English, German, Hebrew, Hungarian, Indonesian, Korean, Russian, Swedish, Tagalog, and Urdu. The data are part of an ongoing collection of over 250 (previously unpublished) examples of morphological code-switching, morphological speech errors, and morphological novel forms, collected from adult native speakers. The goal of the project is to find cross-linguistic evidence from a variety of morpheme categories bearing on the morphological nature of the lexical entry. The evidence will be used to argue for/against competing theories, and to posit morphological parameters in the mental lexicon. I. CODE SWITCHING
Morphological code-switching involves morphemes from different languages combined into a single word. (Only nonborrowed morphemes were counted as code-switching examples.) The combination removes the possibility of incorrect lexical selection of whole-word forms, since the utterances are not lexical items in any language. In this section, cross-linguistic evidence supports the presence of separately listed suffixes, prefixes, infixes, and circumfixes in the mental lexicon. A. Suffixes Examples of roots in one language followed by inflectional and derivational suffixes in another language abound. In (1–4) below, code-switched roots combine with inflectional suffixes from unrelated languages, and in (5–7) code-switched inflectional suffixes follow unrelated roots: 1. English/Hebrew: (lixtov ‘write’)
You want me LIXTOVing all this?
2. English/Tagalog:
She’s in the back WALISing (walis ‘sweep’)
3. English/Urdu: Omar is PHEKing dirt on me! (phek ‘throw’) 4. Armenian/English:
BOOK-er-u metSen paner gese megmeg BOOK-pl-gen in-abl-def thing hab-say-3s one-one ‘She sometimes says things from her books.’
5. English/Swedish: Zoe play-TE with us (-te ⫽ Swedish past tense)
70
RACHELLE WAKSLER
6. Cebuano/English:
ga saka-ING siya sa hagdanan do go up-ING he det stairs ‘He is climbing the stairs.’
7. Indonesian/English: cowok yang bawa ferrari itu bisa dapet cewek-S cakep guy who drive ferrari det can get girl-S gorgeous ‘The guy who drives the ferrari can get gorgeous girls.’ Derivational suffixes involved in code-switching are exemplified in (8– 10): 8. English/Chinese What he did is totally un-DING-able (ding⫽accept, bear) 9. Indonesian/English bisa SORRY-in gua gak can sorry-causative me S-particle ‘Can you forgive me?’ 10. Korean/English koko nun jome boon we gi ga Shi-jang-TIC ha jan a koko is a little atmosphere is peasant-market-TIC S-particles Koko (a Korean cafe in Berkeley) is a little like a peasant-market. Full-listing theories cannot account for the data above, since the codeswitched forms are not lexical items in either language. These data also argue against giving inflectional affixes separate listings, but positing all derivational affixes within full-form listings in the mental lexicon. B. Prefixes, Circumfixes, and Reduplication Cross-linguistic evidence of prefixes and circumfixes that must be listed separately in the mental lexicon is provided below. Included in this section are examples of inflectional prefixes, which have not yet been addressed in the literature. (In English and many European languages, all prefixes are derivational, which might lead to the erroneous generalization that all prefixes are derivational in all languages.) Examples (11–15) provide a sampling of prefixes and circumfixes in codeswitching: 11. Derivational Prefix English/Russian: She is NE-reliable. (Russian ne- ⫽ negative prefix) 12. Derivational Prefix ⫹ Root, with Inflectional Circumfix German/Hungarian: Ich habe Tama´s EL-ge-GANCSOL-t. ‘I tripped Thomas’ (German: Ich habe Tamas ein Bein gestellt) (Hungarian: Elgancsoltam Tamast)
CROSS-LINGUISTIC EVIDENCE
71
el-gancsol ⫽ Hungarian stem (with derivational prefix) ‘trip someone’ ge ⫹ t ⫽ German past tense circumfix 13. Derivational Prefix ⫹ Circumfix-Inflected Stem German/Hungarian: Ich habe die Pillen schon BE-genommen ‘I already took the pills’ (German: ein-ge-nomm-en; Hungarian be-vettern ‘taken’) 14. Inflectional Prefix Tagalog/English: i-WALK mo namam si Zus present-WALK you please det Zeus (dog’s name) ‘Please walk Zeus.’ 15. Inflectional Prefix ⫹ Reduplication and Inflected Stem Tagalog/English: Saan si Jason? Nag-SWI-SWIMMING siya. where det Jason present-redup-SWIMMING he ‘Where is Jason? He’s swimming.’ Examples (11) and (14) show code-switching involving derivational and inflectional prefixes, respectively, demonstrating that full-listing of all prefixed forms is untenable. In (12), the Hungarian prefix and root are separated by the German circumfix, with alternations of Hungarian and German morphemes in one word, again precluding full-form listing. Examples like (13), in which the inflected German stem ‘genommen’ takes a Hungarian prefix, show that the Hungarian prefix and its root need separate listings in the mental lexicon (since the prefix was used without the root). In (15), the Tagalog inflectional prefix nag- accompanies a reduplicative prefix. Reduplication is a morphological process unseen in psycholinguistic studies to date. A reduplicative morpheme does not have a single underlying phonological form, but differs phonologically for each verb, as it is formed by reduplicating phonological information of the verb it attaches to. Thus the Tagalog form for ‘swimming’ is languy ‘swim,’ plus the inflectional prefix nag-, plus reduplication of the first part of the verb: nag-la-languy. In the code-switched example, nag- is followed by the reduplicated first syllable of the English verbal form SWIMMING. As the reduplication in (15) affects an English root, the reduplicated full-word form could not have been listed. Furthermore, (15) is evidence that a prefix without phonological realization must be included in the mental lexicon (further examples are shown below). The fact that the inflected English form appears suggests that SWIMMING is listed as a full-form in the mental lexicon. All other examples of morphological code-switching take what looks to be a single lexical item, be it a word or an affix, from the code-switched language. Also, the speaker would have put double tense/aspect marking on this word, as tense/aspect was al-
72
RACHELLE WAKSLER
ready represented with Tagalog prefixes. Thus a fully-parsed model is not favored by the code-switching data. C. Infixes Infixation, the morphological process in which an affix is inserted inside a root, has not yet been seen in the psycholinguistic literature (because languages that use this morphology have not yet been examined for lexical representation). In (16–18), examples from Tagalog show infixes that must be separately listed in the mental lexicon. 16. B-in-LANKET-an mo ba yung bata? BLANKET (tense infix -in-, suffix -an) you Q-marker det child ‘Did you put the blanket on the child?’ 17. B-in-OTTLE niya yung manga BOTTLE (tense infix -in-) she det mango ‘She bottled (pickled) the mangoes.’’ 18. C-um-O-COLOR siya COLOR (Reduplicated first syllable CO, plus aspect infix -um-,) she ‘She is coloring.’ In (16) and (17), the tense infix is inserted after the first consonant of the code-switched roots. In (18), the first syllable of the root COLOR is reduplicated, and then the infix is added inside the reduplication. Since the roots are English, (which has neither reduplication nor infixation), the infixed forms cannot have full-form listings. D. Nonconcatenative Morphology Some languages include discontinuous morphemes which are interleaved inside roots. In Hebrew, many verbal inflectional and derivational morphemes are vowels which are inserted between the consonants of the verb roots. In (19–20), nonconcatenative Hebrew tense/aspect morphemes are interleaved with English roots: 19. Hebrew/English: Aiza raiah tov, tasmili. ‘What a good smell, smell [it].’ root sml from English ‘smell’ with Hebrew future, 2f nonconcatenative morphology 20. Hebrew/English: dupaz ‘it was deposited’ root dpz from English ‘deposit’ with Hebrew -u-a- noncatenative passive morphology Examples like (19 and 20) cannot be handled by a full-listing theory, since vowels from Hebrew morphemes are combined with consonants from En-
CROSS-LINGUISTIC EVIDENCE
73
glish morphemes. Moreover, access of just the root consonants from the English verb suggests that McCarthy’s (1981) model of separate morpheme tiers may represent a parameter in the mental lexicon selectively activated by those languages using nonconcatenative morphology. II. NOVEL FORMS AND ERRORS
In this section, English novel forms and morphological errors are used to investigate categories of morphemes requiring parsed vs full-word listings. The novel examples below typify the data observed: 21. I wonder if I don’t feel that spark because I’m just not spark able right now. 22. Your work was wonderful and I definitely want to use you again for a new project. Not now, but future ly. 23. I didn’t have a
pre -idea about it.
24. I’ll wait till the phone is
un busy . . .
These examples require separately listed affixes, as the outputs are novel forms. Particularly interesting in these data is that no examples using Level I affixes (e.g., -ity, -ory, -al) were found in the corpus of ⬎90 novel items. All suffixes appearing in the novel data were Level 2 suffixes in English, suggesting that a correspondence between lexical levels and separately-listed morphemes is one that bears investigation. The errors in (25 and 26) exemplify the incorrect affix data: 25. I want to motiv 26. He used to be real
ize (motivate) my dog to work harder. un sensitive . . . uh . . . insensitive.
Such errors support a model with (at least some) separately listed roots and affixes. However, the multiply-affixed errors in (27–30) suggest that affixed stems were incorrectly selected from the mental lexicon and then affixes were added to derive the appropriate grammatical categories: 27. to my horrification (horror) 28. I am a good observationalist (observer). 29. The proprietariness (propriety) of this method . . . 30. The play was one of great whimsicality (whimsy). In (27), for example, if the already affixed ‘horrify’ were selected from the mental lexicon instead of the root ‘horror’, an affix to make the form a noun would be needed in the context. Otherwise, the root would have already fit (unaffixed) into the context. Such data suggest that some full-listing forms occur in the mental lexicon, (though they do not preclude an account in
74
RACHELLE WAKSLER
which the first affix was mistakenly applied, and then the second affix was syntactically required). III. CONCLUSION
The cross-linguistic data presented here support a dual-listing model of lexical representation, in which some whole words are listed in the mental lexicon, and some individual morphemes are listed (e.g., Pinker, 1991; Frauenfelder & Schreuder, 1992). Also, the data can be used to posit three morphological parameters in the mental lexicon: 1. Automatically/selectively listed suffixes Some languages may have automatic listing of separate morphemes throughout the language, while others list only some affixes separately. The novel data suggest that level ordering (or some of the properties contributing to level ordering) may be useful in predicting which morphemes receive separate listing. 2. Position with respect to root PRE/POST/IN/CIRCUM The expanded set of affixes provided here supports a four-way categorization for affix position. 3. ⫹/⫺ morphological tiers This parameter predicts that languages differ in the possibility of needing separate morphological tiers for roots and affixes in the mental lexicon. REFERENCES Butterworth, B. 1989. Lexical access in speech production. In W. Marslen-Wilson (Ed.), Lexical representation and process. Cambridge, MA: MIT Press. Dell, G., & O’Seaghdha, P. 1992. Stages of lexical access in language production. Cognition, 42, 287–314. Frauenfelder, U., & Schreuder, R. 1992. Constraining psycholinguistic models of morphological processing and representation: The role of productivity. In G. Booij and J. van Marle (Eds.), Yearbook of Morphology. Dordrecht: Kluwer. McCarthy, J. 1981. A prosodic theory of nonconcatenative morphology. Linguistic Inquiry, 12, 373–418. Pinker, S. 1991. Rules of language. Science, 253, 530–535. Taft, M., & Forster, K. 1975. Lexical storage and retrieval of prefixed words. Journal of Verbal Learning and Verbal Behavior, 14, 638–647. Taft, M. 1994. Interactive-activation as a framework for understanding morphological processing. In D. Sandra and M. Taft (Eds.), Morphological structure, lexical representation and lexical access. Hillsdale, NJ: Erlbaum.