The present research investigates the composition and representation of complement coercion in Mandarin Chinese. Complement coercion is the phenomenon, in which certain verbs (e.g., start), requiring an event-denoting complement, are combined with an entity-denoting complement (e.g., book), as in The author started a book. The semantic type mismatch between the verb and the complement is supposed to be repaired by a coercion operation, i.e., adding an event sense to the complement (e.g., writing the book) to satisfy the selectional restrictions of the verb. The phenomenon has been well-studied in European languages such as English and German, but is relatively under-research in languages outside the European language family, such as Mandarin Chinese. This research, using Chinese as a medium, examines how complement coercion expressions are mentally represented and processed online, via a self-paced reading experimental paradigm. One pilot and three main studies have been undertaken to answer three crucial but under-debate questions.

The Pilot Study (Chapter 3) explores whether Chinese sentences, composed of an entity complement following an event-selecting verb, exhibit different processing behaviors from those composed of an entity complement following a non-event-selecting verb. The event-selecting verbs were chosen from different semantic classes. The non-event-selecting verbs were those specifying either a preferred action, or a non-preferred but plausible action that readers infer when they read the expressions with event-selecting verbs. Another type of sentences were formed with a preferred verb inserted between the event-selecting verb and the entity complement. The results showed that the expressions with an event-selecting verb required longer reading times to process than the other three types of expressions. These results suggested that sentences involving complement coercion required more comprehension computations.

Study 1 (Chapter 4) sheds light on a typical event-selecting verb class, i.e., aspectual verbs (e.g., 开始 kāi-shǐ “start”), to test whether sentences with aspectual verbs induce reading time slowdowns as reported in the Pilot Study. This sentence type was compared with two other types including either a preferred or non-preferred verb. The reading time data showed that aspectual verb sentences were read slower than the other two types of sentences, which was mainly reflected at the two words immediately after the complement noun. The findings indicated that aspectual verbs semantically required an event-denoting complement, and triggered processing cost when they were combined with a complement referring to a physical individual. This study confirmed the argument selectional properties of the aspectual verbs, i.e., taking an event object.

Study 2 (Chapter 5) puts emphasis on another verb class, i.e., psych verbs (e.g., 享受), which was included previously into the set of event-selecting verbs, to examine whether expressions with this class of verbs engender a similar processing profile as the aspectual verb class. The analyses revealed that sentences with psych verbs and control verbs were read with shorter reading times than those with aspectual verbs at the two regions directly after the complement noun. Such findings manifested that psych verbs, differentiating from aspectual verbs, might not exclusively require an event as their internal argument. Thus, no additional effort was needed when readers processed compositions composed of a psych verb plus an entity complement. This study called into question most of the previous metalinguistic analyses and empirical observations that psych verbs and aspectual verbs shared a similar selectional properties, and induce processing cost when paired with an entity noun.

Study 3 (Chapter 6) seeks to determine whether the detected processing cost from sentences with aspectual verbs is caused by the coercion-driven semantic process, above and beyond the unpredictability-driven surprisal of the entity noun following an event-taking verb. This study contrasted an aspectual verb condition and a control condition, with another neutral condition, in which the predictability of the complement noun was lower than the control condition but, crucially, similar to that in the aspectual verb condition. The reading time results illustrated that aspectual verb sentences triggered much longer reading times than the other two conditions of sentences at the word after the complement noun. At the subsequent region, the apparent reading time difference was only observed between the aspectual verb and control verb sentences. The results suggested that aspectual verb sentences were taxing to comprehend than the neutral verb and control verb sentences. The findings supported the coercion-related compositional process undertaking on the semantic representation of the complement.

To summarize, the findings confirmed the processing difficulty triggered by complement coercion cross-linguistically. While the evidence of processing difficulty was found for the aspectual verb sentences, no such evidence was found for the psych verb sentences. This posed a question to the previously assumed argument that the two classes of verbs shared similar selectional properties, and exhibited parallel processing profiles when they were combined with an entity complement. The research further showed that the observed processing cost for aspectual verbs was largely due to the coercion-related semantic compositional process above and beyond the surprisal effect.

Taking on an empirical approach, the dissertation provided the much-needed empirical evidence to the theoretically puzzling issues toward the complement coercion phenomenon. It ultimately contributed to the understanding of the representation and processing of sentences with complement coercion.