Factors affecting relative clause processing in Mandarin

Current models of sentence processing make contrasting predictions regarding the processing of head-final relative clauses (RCs) in Chinese, but existing research has found mixed results. It is not yet clear (i) whether subject-extracted RCs are easier to process than object-extracted RCs; and (ii) whether classifiers in a dislocated position before the RC can facilitate processing. This dissertation investigates how extraction site, animacy configuration, and classifier positioning guide real-time parsing of Chinese RCs. Assuming a correlation between frequency of occurrence and ease of processing, I analyzed the frequency pattern of these factors in the Chinese Treebank 5.0 corpus, then formulated and experimentally tested hypotheses to account for the patterns observed.;Chapter 1 and 2 introduce the issues and review the relevant literature. Chapter 3 presents a corpus analysis, focusing on extraction type (subject-extracted vs. object-extracted) and animacy of the head and embedded nouns. The results show that subject-extracted RCs are more frequent than object-extracted RCs and suggest three Animacy Preference Constraints (APCs): (i) head nouns that are RC subjects tend to be animate; (ii) head nouns that are RC objects tend to be inanimate; (iii) in both cases, the animacy of the head and embedded nouns tend to contrast.;Chapter 4 presents three self-paced reading experiments to test whether the hypothesized APCs influence ease of RC parsing. The results show that animacy configurations modulate the processing load induced by Chinese RCs: subject-extracted RCs are easier to process than object-extracted RCs when the animacy configuration is inversely contrastive (i.e., RC subject = inanimate, RC object = animate).;Chapter 5 focuses on classifiers as possible cues for upcoming RCs. The corpus reveals an asymmetrical pattern of classifier distribution in subject-extracted and object-extracted RCs. I propose two processing principles related to anticipatory processing and lexical access. These principles are supported by one eye-tracking and two reading-time experiments. Results suggest that pre-RC classifiers help the human parsing system identify head-final RC structures efficiently.;I consider the findings in light of four major theoretical accounts, and suggest that the parsing of Chinese RCs best fits the probabilistic, expectation-based constraint-satisfaction model.