Human brain potentials to violations in morphologically complex Italian words

Neuroscience Letters 241 (1998) 83–86

Human brain potentials to violations in morphologically complex Italian words Matthias Gross a, Tessa Say b, Michael Kleingers a, Harald Clahsen b, Thomas F. Mu¨nte a , c ,* a

Department of Neurology, Medizinische Hochschule, Hannover, Germany Department of Language and Linguistics, University of Essex, Colchester, UK c Department of Cognitive Science, University of California San Diego, La Jolla, CA 92093-0515, USA b

Received 20 October 1997; received in revised form 12 December 1997; accepted 12 December 1997

Abstract Event-related brain potentials were recorded while 12 Italian-speaking subjects read correctly and incorrectly inflected verb forms. Participle forms of three types of verbs were investigated: 1st conjugation verbs (parlato ‘spoken’), 3rd conjugation verbs (dormito ‘slept’), and irregular 2nd conjugation verbs (preso ‘taken’). We compared correct and incorrect participle forms; the latter had stem formation errors and/or incorrect participle endings. Event-related potentials (ERP) showed different responses to incorrect regular and incorrect irregular participle forms: incorrect irregulars (*prendato instead of preso) elicited a widespread negativity, whereas incorrect regulars (*parlito, *dormato) produced no effect. This difference replicates previous results on German inflection and supports the linguistic distinction between lexically-based and rule-based inflection.  1998 Elsevier Science Ireland Ltd.

Keywords: Event-related potentials; Linguistic morphology; Italian; N400

A prevailing view in linguistics has been that two different kinds of mechanisms are involved in forming morphologically complex words: (1) lexical listing of irregular word forms such as drink, drank, drunk and (2) regular (affixation) rules. The latter concatenate an affix with a stem, for example an ending such as -ed with the stem walk to form walk + ed, and capture the regular truly productive aspects of word formation. The former do not involve any kind of affixation rules. Psycholinguistic experiments have produced substantial evidence for this distinction (review in [2]) and have led to a dual-mechanism model [11,17] which postulates separate mechanisms for directly accessing irregular forms stored in memory and for morphological structure building of inflected words from stems and affixes. Alternatively, single-mechanism computer simulations of inflectional morphology and other linguistic systems have been developed which operate without any directly implemented combinatorial principles * Corresponding author. Fax: +1 619 5341128; e-mail: [email protected]

[6,19]. Some researchers have taken these simulations to argue that the knowledge of language should better be represented in terms of connectionist networks with associatively linked units rather than in terms of grammars [5]. Against this background, the brain structures involved in morphological processing have recently been investigated by using modern neuro-imaging techniques [9,20]. The dualmechanism model predicts that regular affixation-based inflection should be processed by other brain structures than those dealing with irregularly inflected words which are stored in memory. The studies available are compatible with this view. Different areas of the brain have been shown to be active for regular and irregular past tense inflection in English [9], and in aphasia the two types of inflectional processes can be selectively impaired [12]. However, even though the results show some overlap they do not yet provide a coherent picture of brain activations in processing inflected words. Moreover since the techniques used (functional magnetic resonance imaging, and positron emission tomography) possess a high spatial but low temporal resolution, they do not directly tap into the rapid processes which

0304-3940/98/$19.00  1998 Elsevier Science Ireland Ltd. All rights reserved PII S0304- 3940(97) 00971- 3

84

M. Gross et al. / Neuroscience Letters 241 (1998) 83–86

are involved in inflecting words. Electrophysiological methods, specifically event-related brain potentials (ERPs), appear to be better suited for the temporal analysis of different kinds of inflectional phenomena and were therefore used in the present study. ERPs have been shown to be sensitive to different aspects of language processing, e.g. semantic [10], syntactic [7,8,14] and morphological levels [16,21]. In a recent series of experiments we investigated the effects of morphological violations in the German participle [16] and noun plural systems [21]. In these experiments regular and irregular words were presented with either the appropriate suffix or the ending of the other morphological class (e.g. regular participles: correct, durchgetanz-t; incorrect, *durchgetanz-en; irregular participles: correct, *eingelad-en; incorrect, eingelad-et), thus creating morphological violations. Only violations of the irregular verbs and nouns (i.e. overregularizations) gave rise to a rather widespread negativity with a left anterior maximum. These data were taken to indicate differences in the neuronal processing of regular and irregular words and hence were interpreted as supporting the dual mechanism model of morphology [2,11,17]. In particular, the similarity of the negativity for irregular incorrect stimuli to ERP effects obtained for syntactic violations (e.g. number mismatches between verbs and nouns, see [7,13,14]) suggested that it might indicate morphological structure building. This paper deals with brain potential data from past participles in Italian to determine whether the effects found for German inflection generalize to typologically different languages. In Italian, regular participles (such as parlato ‘spoken’, dormito ‘slept’, etc.) are formed by an affixation rule which concatenates the ending -t- with a regular verb stem. A gender/number marker is added to all participles, regular and irregular, -o being the default masculine singular. Verb stems consist of two segments, a root (parl-, dorm-) plus a so-called theme vowel (-a-, -i-, -u-). Irregular participles (such as prendere/preso, ‘taken’; rompere/rotto, ‘broken’) involve phonologically modified stems and cannot be decomposed into root, theme vowel and participial ending. Thus regular participles appear to involve two different processes: stem formation (concatenation of root and theme vowel) plus affixation of -t-. Irregular participles involve neither process. The dual-mechanism model leads us to expect processing distinctions between Italian participles formed by -t- affixation and irregular participles formed by stem modification. Orsolini and Marslen-Wilson [15], however, claim that this is not the case as they did not find significant differences between regular and irregular inflected verb forms in cross-modal priming and production tasks. However, the design of these experiments does not clearly differentiate between the formation of stems and rule-based affixation. Here we provide brain potential evidence for a regular/irregular distinction in Italian participle formation that parallels previous findings in German [16]. Specifically, we will show that only incorrect irregular par-

ticiples such as *prendato produce an ERP-effect (i.e. a widespread negativity) when compared to the correct form preso, whereas incorrect regulars are not associated with such an effect. Twelve healthy, native speakers of Italian (one lefthanded, eight women, aged 22–35 years) gave informed consent to participate. Stimuli were selected using two Italian frequency dictionaries [1,4] and comprised 110 1st conjugation verbs (lemma/participle frequency per million words 35.47/8.64), 110 3rd conjugation verbs (frequencies 36.87/8.85), 110 irregular 2nd conjugation verbs (frequencies 34.31/8.85) as critical stimuli, an additional 330 verbs as filler stimuli and 130 proper nouns as targets. The three classes of critical words, matched closely for word length, appeared either in their correct participle form or an incorrect participle form. Two types of incorrect participles were formed: (1) stem formation errors, by using the theme vowel -a- of the 1st conjugation for 3rd conjugation stimuli or by using the theme vowel -i- of the 3rd conjugation for 1st conjugation stimuli; (2) overgeneralization errors, by incorrectly concatenating the 1st conjugation theme vowel -aand the regular participial affix -t- to the stem of an irregular 2nd conjugation verb. This error was chosen since it is the one made most often by Italian children [18]. Two stimulus lists were compiled containing the stimuli in random order (see Table 1). The lists were constructed such that each critical item was presented in its correct form in one list and in its incorrect form in the other. Each list was shown to half of the subjects. The stimuli were presented one word at a time (duration 300 ms, random onset asynchrony 1400–2200 ms, height 0.8°, width 1.5–4.3° visual angle) in yellow letters against a blue background. The subjects’ task was to read the words silently and to press a button held in the dominant hand whenever a noun was presented. The EEG was recorded (bandpass 0.01–100 Hz, AD resolution 4 ms) with tin-electrodes mounted in an Electro-Cap from all standard sites of the 10/20 system plus 10 additional intermediate sites with reference electrodes placed on both mastoid processes (algebraic mean). The electrooculogram was recorded for artifact correction purposes [3]. Averages Table 1

Class

Number

Example

1st: correct 1st: incorrect 3rd: correct 3rd: incorrect Irregular/correct Irregular/incorrect Filler infinitive/correct Filler infinitive/incorrect Filler past tense Filler present tense Noun (target)

55 55 55 55 55 55 55 55 110 110 130

Amato Rubito Pulito Ferato Risposto Prendato Lambire Riconoscire Succedemmo Insistiamo Paese

M. Gross et al. / Neuroscience Letters 241 (1998) 83–86

were obtained for 1024 ms epochs beginning 100 ms prior to presentation of the stimuli. Mean-amplitude measures were taken in the 250–500 and 500–750 ms time windows and subjected to repeated measures analyses of variance (ANOVA) for midline, parasagittal and temporal electrode rows separately. As an omnibus ANOVA with verb-type (1st vs. 3rd vs. irregular), correctness (correct vs. incorrect), hemisphere (left vs. right; parasagittal and temporal only) and site (3–5 levels anterior to posterior) revealed complex verb-type by correctness by topography interactions, these were followed up by separate ANOVAs for each verb-type. Only these latter analyses are reported here. The HuynhFeldt correction for non-sphericity was used and P-values shown are corrected. Performance in the noun detection task was fast (mean ± SD 616 ± 50 ms) and accurate with 96 ± 2.4%. The grand average ERPs for the three different verb classes are shown in Fig. 1. It appears that only the irregular verbs show a pronounced effect for the incorrect (regularized) participle forms. These are associated with a more negative waveform beginning 250 ms after stimulus onset. For the 1st and 3rd conjugation (regular) verbs, on the other hand, only small differences between correct and incorrect forms are observed. This is corroborated by Fig. 2 showing the mean amplitude in the 500–750 ms time-window. Again, only the irregular verbs show a marked difference between correct and incorrect participles. Statistically, main effects of correctness were obtained in both time-windows for the irregular verbs (d.f.(1,11); 250–500 ms: parasagittal, F = 5.33, P , 0.05; midline, F = 8.68, P , 0.015; 500–750 ms: midline, F = 9.59, P , 0.01). The lateralization of this effect towards the right in the temporal region gave rise to correctness by hemisphere interactions for these electrode rows (d.f.(1,11): 250–500 ms: F = 9.61, P , 0.01; 500–750 ms: F = 5.96, P , 0.04). By contrast, no significant effects

Fig. 1. Grand average (n = 12) for the three midline electrodes. Incorrect irregular participles are associated with a more pronounced negative going waveform starting at about 250 ms and extending for the rest of the recording epoch. By contrast, the regular (1st and 3rd conjugation) verbs only show small differences.

85

Fig. 2. Distribution of the incorrectness effects. Depicted are the mean amplitudes in the 400–700 ms time window for the parasagittal electrode rows. Regular (1st and 3rd conjugation) verbs show very similar amplitudes for correct and incorrect conditions while irregular verbs show a widespread negativity for the incorrect condition.

were obtained for the 3rd conjugation verbs, while for the 1st conjugation verbs there was only one significant interaction (temporal electrodes, 250–500 ms, correctness by site, F(2,22) = 3.68, P , 0.05) reflecting a slightly more negative waveform for the incorrect participles at the right anterior temporal site. These data are similar to our previous findings for morphologically complex words in German [16,21] in that only the morphologically incorrect irregular verbs gave rise to a significant and widespread negativity. While the orthographical dissimilarity between the correct and incorrect for the Italian irregulars (c.f. preso/*prendato) was greater than for the regulars (dormito/*dormato), this was not the case in the German materials used previously [16]. We can thus eliminate orthographic dissimilarity as a cause for the observed negativity. By contrast, incorrect regular participles such as *dormato or *parlito did not produce any major ERP differences compared to their corresponding correct forms again corroborating our previous findings [16]. The lack of an ERP effect for regulars (in spite of the subjects’ awareness of the errors) might partially be explained by the fact that they do not require a reanalysis assumed to underlie the positivity seen for syntactic errors [7,13,14]. A shallow processing of the items oriented towards the endings can be ruled out, since (a) the incorrect irregulars, while having a correct verb ending, did give rise to a sizeable ERP effect and (b) no ERP effects emerged for incorrect regulars in our parallel German experiment in spite of tasks that necessitated a deep semantic analysis [16]. In summary, the brain again appears to honor the regular/ irregular distinction with different brain potentials emerging for the two morphological clusters. Further experimentation presenting morphological violations of complex Italian words in different contexts (e.g. sentences and stories) is needed to address the topographical differences between the negativities to irregular verbs in Italian and German.

86

M. Gross et al. / Neuroscience Letters 241 (1998) 83–86

These differences notwithstanding, these results provide electrophysiological evidence for processing differences between regular and irregular words. As the ERP effects for these morphological classes differ in kind rather than in size (as would be predicted by connectionist accounts); they most likely correspond to the distinction between rulebased and lexically-based morphological processes made by the dual mechanism account. Affixation of -t in German and in Italian thus involves a rule by which an affix is combined with a stem. Verb stems, however, are lexically listed. Our studies, both on Italian and on German, show that only overapplication of inflectional rules yield ERP-effects. The technical assistance of Berdieke Wieringa and Jobst Kilian is gratefully acknowledged. This work was supported by grants from the Hermann and Lilly Schilling Foundation to T.F.M., and the German Science Foundation to H.C. [1] Bortolini, U., Tagliavini, C. and Zampolli, A., Lessico di Frequenza della Lingua Italiana Contemporanea, IBM Italia, Milano, 1971. [2] Clahsen, H., Eisenbeiss, S. and Sonnenstuhl, I., Morphological structure and the processing of inflected words, Theor. Linguist., 22 (1997) 121–156. [3] Dale, A.M., Source localization and spatial discriminant analysis of event-related potentials: linear approaches (Dissertation), University of California, San Diego, CA, 1994. [4] De Mauro, T., Mancini, F., Vedovelli, M. and Voghera, M., Lessico di Frequenza della Lingua Italiana Parlata, Etas Libri, Milano, 1993. [5] Elman, J., Bates, L., Johnson, M., Karmiloff-Smith, A., Parisi, D. and Plunkett, K., Rethinking Innateness. A Connectionist Perspective on Development, MIT Press, Cambridge, MA, 1996. [6] Elman, J., Learning and development in neural networks: the importance of starting small, Cognition, 48 (1993) 71–99. [7] Friederici, A.D., Hahne, A. and Mecklinger, A., The temporal structure of syntactic parsing: early and late effects elicited by syntactic anomalies, J. Exp. Psychol. Learn. Mem. Cogn., 22 (1996) 1219–1248. [8] Hagoort, P., Groothusen, J. and Brown, C., The syntactic positive shift (SPS) as an ERP measure of syntactic processing, Lang. Cogn. Proc., 9 (1993) 439–483. [9] Jaeger, J.J., Lockwood, A.H., Kemmerer, D.L., Van Valin, R.D., Murphy, B.W. and Khalak, H.G., A positron emission tomo-

[10]

[11]

[12] [13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

graphic study of regular and irregular verb morphology in English, Language, 72 (1996) 451–497. Kutas, M. and Hillyard, S.A., Reading senseless sentences: brain potentials reflect semantic incongruity, Science, 207 (1980) 203–205. Marcus, G.F., Brinkmann, U., Clahsen, H., Wiese, R. and Pinker, S., German inflection: the exception that proves the rule, Cogn. Psychol., 29 (1995) 189–256. Marslen-Wilson, W. and Tyler, L., Dissociating types of mental computation, Nature, 387 (1997) 592–594. Mu¨nte, T.F., Heinze, H.J., Matzke, M., Wieringa, B.M. and Johannes, S., Brain potentials and syntactic violations revisited: no evidence for specificity of the syntactic positive shift, Neuropsychologia, (1998) in press. Mu¨nte, T.F., Matzke, M. and Johannes, S., Brain activity associated with syntactic incongruencies in words and pseudowords, J. Cogn. Neurosci., 9 (1997) 300–311. Orsolini, M. and Marslen-Wilson, W., Universals in morphological representation: evidence from Italian, Lang. Cogn. Proc., 12 (1997) 1–47. Penke, M., Weyerts, H., Gross, M., Zander, E., Mu¨nte, T.F. and Clahsen, H., How the brain processes complex words: an ERP study of German verb inflection, Cogn. Brain Res., 6 (1997) 37–52. Pinker, S. and Prince, A., Regular and irregular morphology and the psychological status of rules of grammar. In L.A. Sutton, C. Johnson and R. Shields (Eds.), Proceedings of the 17th Annual Meeting of the Berkeley Linguistics Society, Berkeley, CA, 1991. Pizzuto, E. and Caselli, M.C., The acquisition of Italian morphology: implications for models of language development, J. Child Lang., 19 (1992) 491–557. Rumelhart, D.E. and McClelland, J.L., On learning the past tenses of English verbs. In J.L. McClelland, D.E. Rumelhart and the PDP Research Group (Eds.), Parallel Distributed Processing: Explorations in the Microstructure of Cognition, Vol. 2, Psychological and Biological Models, MIT Press, Cambridge, MA, 1986, pp. 216–271. Ullman, M., Corkin, S., Pinker, S., Coppola, M., Growdon, J. and Locascio, J., Which neural structures subserve language? Evidence from inflectional morphology, J. Cogn. Neurosci., 9 (1997) 289–299. Weyerts, H., Penke, M., Dohrn, U., Clahsen, H. and Mu¨nte, T.F., Brain potentials indicate differences between regular and irregular German noun plurals, NeuroReport, 8 (1997) 957–962.