Structural Expectations in Chinese Relative Clause Comprehension

Relative clauses (RC) are among the most well-studied constructions in the field of psycholinguistics. A wide variety of work explores a robust processing asymmetry such that subject relatives (SRs) are easier to process than object relatives (ORs). For example, English shows a subject advantage, as demonstrated by a number of studies involving different measures including: self-paced reading (King & Just, 1991), eye-tracking (Traxler, Morris & Seely, 2002), ERP (King & Kutas, 1995), fMRI (Just, Carpenter, Keller, Eddy & Thulborn, 1996); and PET (Stromswold, Caplan, Alpert & Rauch, 1996). A robust finding in the literature suggests that subject preference seems to be a universal processing phenomenon in RCs. Some pieces of evidence come from Dutch (Frazier, 1987), French (Frauenfelder, Segui & Mehler, 1980), German (Schriefers, Friederici & Kühn, 1995), Japanese (Miyamoto & Nakamura, 2003) and Korean (Kwon, Polinsky & Kluender, 2006). In order to account for the universal processing pattern of RCs, several theories are proposed, such as: WORD ORDER (Bever, 1970; MacDonald & Christiansen, 2002), the ACCESSIBILITY HIERARCHY (Keenan & Comrie, 1977), EXPERIENCE/FREQUENCY-BASED ACCOUNTS (Mitchell, Cuetos, Corley & Brysbaert, 1995; Hale, 2001), STRUCTURE-BASED APPROACHES (O’Grady, 1997; Hawkins, 2004) and WORKING MEMORY (Gibson, 2000; Lewis & Vasishth, 2005). Chinese RCs are valuable in testing those theories and have attracted much attention due to the controversial processing asymmetry in the literature. Inspired by Hale (2001), this work applies an information-theoretic linking theory, surprisal, as a complexity metric to derive the pattern of incremental processing difficulty in Chinese RCs. Under the principle that readers apply their knowledge of grammar in perceiving syntactic structures (Chomsky, 1965), this model investigates the role that frequencydrivened structural expectations play in parsing, especially when temporary ambiguities occur. It makes predictions which remain consistent with a universal SR preference in sentence comprehension.