Text

theses

Previous research suggests people are remarkably good at processing sentences that contain novel words. For example, although we do not know what cratomize means, we know that the sentence The man cratomized the boy is grammatical (unlike The man cratomize the boy), and we can answer questions about it (e.g., who did the man cratomize?). This dissertation investigates the role of syntactic and morphological information in human sentence processing using relative clause sentences containing nonsense words, such as The actor who cratomized the critic impressed the director and The actor who the cratomer humiliated impressed the director. Experiments 1 and 2 reveal that human sentence processing is an automatic reflex which is unaffected by task requirements or presence of novel words. Subsequent experiments further examine processing of such sentences. Experiment 3 reveals that the impact of syntactic context is so great that, for certain syntactic positions, processing novel words bears no additional cost. Experiments 4-6 investigate how syntactic and morphological information interact. These experiments reveal that syntactic information plays a dominant role with morphology playing a very minor role, with incongruence between syntactic and morphological information always being resolved in favor of syntax. In addition to these behavioral studies, we propose two extensions of existing computational models of sentence processing that enable the models to process sentences with novel words. Our evaluations suggest that the integration of sentence processing models with models of word recognition is a promising future avenue of research. Furthermore, our analyses of English corpora reveal that derivational and inflectional suffixes tend to be infrequently and unreliably used in English, which may (partially) explain why morphological information plays such a minor role in English sentence processing. In the last section of the dissertation, we conduct cross-linguistic analyses that reveal an inverse relation between morphological and syntactic information. Specifically, languages with freer word-order constraints tend to be morphologically richer than languages with strict syntactic constraints, such as English. This hints at a possibility that morphology might play a greater role during sentence processing in languages that have richer morphology than English.

ETD_5737

Ph.D.

Includes bibliographical references

by Gaurav Kharkwal

doi:10.7282/T3NG4P4D

ETD doctoral

The author owns the copyright to this work.