Animacy In Sentence Processing Across Languages: An Information-Theoretic Prospective

ANIMACY IN SENTENCE PROCESSING ACROSS LANGUAGES: AN INFORMATION-THEORETIC PROSPECTIVE Zhong Chen, Ph.D. Cornell University August 2014 This dissertation is concerned with different sources of information that affect human sentence comprehension. It focuses on the way that syntactic rules interact with non-syntactic cues in real-time processing. It develops the idea first introduced in the Competition Model of MacWhinney in the late 1980s such that the weight of a linguistic cue varies among languages. The dissertation addresses this problem from an information-theoretic prospective. The proposed Entropy Reduction metric (Hale, 2003) combines corpus-retrieved attestation frequencies with linguistically-motivated grammars. It derives a processing asymmetry called the Subject Advantage that has been observed across languages (Keenan & Comrie, 1977). The modeling results are consistent with the intuitive structural expectation idea, namely that subject relative clauses, as a frequent structure, are easier to comprehend. However, the present research takes this proposal one step further by illustrating how the comprehension difficulty profile reflects uncertainty over different initial substrings. It highlights particular disambiguation decisions that contribute to processing difficulties found in object relative clauses in English, Italian and Chinese. More importantly, this dissertation examines the role that the universally applicable animacy cue plays in understanding relativized structures. Based on the frequency distribution of animacy in treebanks, this line of modeling work not only provides finer-grained explanations for the animacy effect at the head noun reported in previous experimental literature, but also contributes to integrate an important functional notion into the formal linguistic framework in a unique way. Incorporating functional features like animacy allows us to explore rich, cognitively-plausible grammars in human sentence comprehension