Processing and parameter setting in language acquisition: A computational approach

Children beginning to speak English omit subjects of sentences more often than they omit objects in mandatory contexts. The reason for this has been widely debated in the literature but has not been satisfactorily resolved. Proponents of a grammatical explanation (e.g., Hyams & Wexler 1993) have argued that no existing processing account has been able to explain the asymmetry between subject and object omission. However, most parameter-based accounts have difficulty in accounting for the different argument omission patterns in the speech of children learning pro-drop and topic-drop languages, or they present learnability problems. In Chapter Two, I review arguments on both sides and show why the evidence favors a processing account.^ In Chapter Three, I offer a model of sentence generation which accounts for early subject omissions as a processing effect. The model incorporates basic assumptions of Principles-and-Parameters theory. Phrasal units are assembled from the matrix verb, working from the head, to its complement, and finally to its specifier. The processing load is incremented whenever lexical insertion and attachment operations occur. When processing capacity is exceeded, elements currently assembled are output to the phonological processor. Because complements are attached before specifiers, subjects are more likely to be dropped than objects.^ In Chapter Four, I report the results of my analyses of the speech transcripts of three children, available on the CHILDES database (MacWhinney & Snow 1985), which confirm the model\u27s predictions. Findings include a steady increase in overt subject use and Mean Length of Utterance (MLU), and a correlation between the two; an increase in the relative proportion of overt pronominal subjects, but with productive use of different pronominals emerging at different times; early problems with person deixis; few wh-questions with overt wh-terms in early transcripts; some wh-questions with missing wh-terms; no clear examples of wh-questions with preposed wh-terms and missing subjects; no correlation between subject omission and the absence of overt tense- or agreement-marking.^ In Chapter Five, different crosslinguistic patterns of early argument omission are accounted for as the combined effect of limited processing capacity and early, accurate parameter setting.