University of Cambridge
Centre for Speech, Language and the Brain

Centre for Speech, Language and the Brain

Department of Psychology

Skip navigation

You are in:  Home » Publications » Published abstracts & posters

Publications

Published abstracts and posters

11th International Conference on Cognitive Neuroscience (ICON XI)
Palma de Mallorca, 2011
Proceedings of ICON XI, pp. 118

Parsing Sentences are Unlikely: Corpus-based Analyses of the Neural Processing of Verbs.

Devereux, B.J., Korhonen, A., and Tyler, L.K.

Verb subcategorization frame (SCF) preferences affect sentence processing by placing constraints on how potential arguments can be incorporated into the emerging syntactic representation (e.g. McDonald, 1994). Most studies have used norm-based measures of syntactic preference based on subjective responses, but these may not accurately reflect verbs' SCF usage frequencies. Here we take a different approach, asking whether verb SCF data derived from large cross-domain corpora determine patterns of neural activation as listeners process verbs in sentences. In an fMRI study, subjects listened to sentences containing locally syntactically ambiguous phrases (e.g. insulting neighbours). The verb participle functioned as either a gerund (insulting neighbours is not encouraged) or an adjective (insulting neighbours are not respected). From the VALEX lexicon (Korhonen et al., 2006), we obtained SCF frequency distributions for each verb. In a Principal Components Analysis (PCA) we extracted 6 components. Component 1 accounted for most variance (29%), and reflected the likelihood of the verb taking a direct object complement. We expected verbs with low values on Component 1 (e.g. yawn) to show a preference for the adjectival reading, since the following noun is unlikely to function as the verb's theme. Verbs with these properties produced greater activation in bilateral IFG, LpMTG and LIPL, consistent with this prediction. This indicates a role for SCF preference in sentences where the syntactic role of the verb is ambiguous. Since the VALEX measure reflects properties of verbs alone (independent of the other words in the sentence), the results demonstrate that access of verbs' lexico-syntactic representations specifically is an effective heuristic in ambiguity resolution.