Chapter 5 General Discussion
5.3 Dynamic process of sentence processing
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
c.
The movie that pleased the director received a prized the film festival.
d.
The movie that the director watched received a prized the film festival.
The thematic role assignment with animacy in the ORC with an animate head (b) and the SRC with an inanimate head (c) is against the bias for contrastive animacy role assignment. Therefore, these two conditions are more difficult than the other two.
However, since ORCs may be reduced in difficulty when the argument-verb relation is an irreversible one, the ORCs with an inanimate head (b) are further eased in difficulty. In sum, both biases took place in Traxler et al.’s (2002) experiment and the pure effect of animacy could not be established in their experiment.
In the present study, two experiments were designed to investigate these two biases separately. The bias of thematic role assignment that has effects when the argument-verb relation is irreversible can reduce the difficulty of syntactic complexity of both SRCs and ORCs, and the bias of thematic role assignment that exists when the arguments contrast in animacy strongly modulates syntactic ambiguities.
5.3 Dynamic process of sentence processing
The findings of the present experiment reveal the dynamic process of reading
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Chinese RCs as a sentence unfolds. In particular, readers face different sources of difficulty at different positions in a sentence. The sources of difficulty come from both syntactic cues and semantic cues, which lead or mislead readers to assign thematic roles. Comprehension difficulty is closely related to thematic role assignment.
The findings in Experiment 1 suggest that Chinese readers have to deal with syntactic ambiguities of SRCs and ORCs at different positions. This can be attributed to the unique syntactic properties of Chinese, which are very different from other languages in the world. Chinese is an isolating language and the sole syntactic mechanism depended upon is word order. Chinese has a canonical word order SVO, but several alternatives of word order may be motivated by pragmatics. Therefore, it is possible that semantic cues are also relied upon to assist comprehension. Moreover, Chinese is basically a VO language. Typologically, a VO language has a tendency to contain head-initial structures. However, relative clauses in Chinese are head-final structures. This combination of a VO order with a head-final RC is claimed to be unique typologically and is the result of geographical influence through language contact (Dryer, 1992, 2003). Thus, Chinese differs from other head-final languages, such as Japanese and Korean in terms of the combination and case marking. Readers of Japanese and Korean distinguish SRCs from ORCs according to case marking. In
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
the previous literature, they were simply classified as head-final languages like Chinese, but the distinctions between Chinese and the other head-final languages were ignored.
The unique syntactic properties of Chinese provide a chance to dissociate the memory-based accounts and the syntactic-processing-based accounts. Previous studies on English relative clause processing claimed that there is a steady preference towards SRCs. All the sentence processing models or accounts predict SRC preference for English and many other head-initial languages. However, these models or accounts make opposing predictions for Chinese. Moreover, the incremental process of sentence comprehension should be carefully considered because readers read word by word following a reading direction. Therefore, these predictions should specify the positions where the predicted effects occur.
The memory-based accounts, including the linear distance account (Hsiao &
Gibson, 2003) and the structural distance account (Lin, 2006), should be supported by evidence at the position where the filler-gap dependency can be fulfilled. Both of the accounts predict SRC preference at the RC region for English. Nevertheless, they make different predictions for Chinese: The linear distance account predicts ORC preference at the head noun while the structural distance account predicts SRC
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
preference at the head noun.
The syntactic-processing-based accounts include canonical word order hypothesis (Bever, 1970) and perspective-shifting theory (MacWhinney, 1977).
Though they are also related to working memory resources, they make predictions based on syntactic constraints. For English, canonical word order hypothesis predicts SRC preference at the RC region and perspective-shifting theory predicts SRC preference at the main verb. As for Chinese, the former predicts ORC preference at the pre-relativizer region; the latter predicts SRC preference at the main verb. These theories make different predictions for Chinese at different positions.
In the previous experiments on Chinese RC processing, different results derived opposing conclusions. While Hsiao & Gibson (2003) claimed that Chinese readers have ORC preference due to shorter linear distance, Lin (2006) argued that SRC preference is universal owing to shorter structural distance from the head to the subject gap. The distance factors should take effects at the head noun position because the position is where readers can establish the filler-gap dependency. However, the effects of these two studies actually occur at different positions. ORC preference was observed at the pre-relativizer position (Hsiao & Gibson, 2003) whereas SRC preference was observed at the relativizer and the head noun (Lin, 2006). Both studies
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
did not analyze the main verb region.
The evidence of the present eye movement experiments supports the view that processing RCs involves a dynamic process of sentence comprehension rather than a simple preference of RC type. Readers have to deal with the non-canonical word order at the pre-relativizer for SRCs and then to cope with the perspective-shifting at the main verb for ORCs. The findings are primarily consistent with both previous studies. The SRC difficulty found at the head noun in Experiment 2 can be argued to support the linear distance account (Hsiao & Gibson, 2003). Readers seem to hold the gap information longer for SRCs than for ORCs and suffer from more intervening discourse referents for the filler-gap dependency. However, the linear distance account cannot directly predict the disambiguating positions of Chinese RCs and it is not supported by the findings in Experiment 1. In Experiment 1, no main effect of RC type occurs at the head noun, which is the critical region for the prediction of the linear distance account. Canonical word order hypothesis can better explain both experiments since SRCs involve a non-canonical word order with transposed S and O.
Though we did not find ORC difficulty at the relativizer and the head noun as Lin (2006) did, the relativizer and the head noun regions were more often re-visited for ORCs than for SRCs as shown by re-reading rate.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
In fact, all cognitive processing is constrained by limited working memory resources. It is possible that linear distance and structural distance both impose constraints to a certain extent. When they make consistent predictions to a language like English, a preference for SRCs or ORCs can be consistently found. Yet, when they produce conflicting predictions to a language like Chinese, such a preference may not be strong or contradictory findings may even occur at different positions.
Then, the syntactic-processing-based accounts can better explain the findings than the memory-based accounts. The unique syntactic properties of Chinese allow us to dissociate the working memory constraints and syntactic processing constraints through positions. The present findings suggested that syntactic processing constraints are more essential than working memory constraints. They are also more directly connected to thematic role assignment.
The semantic cues, as well as the structural cues, contribute to such a dynamic processing with an incremental nature of sentence processing. Comprehension difficulty does not come from the information newly encountered, but from the accumulation of the new information and its previous material. As we predicted, the effect of plausibility occurred after the head noun when the head noun forms a complete relation with the verb and the other argument in the RC. The reversible
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
argument-relation consumes more processing efforts than the irreversible relation. To our surprise, one reverse effect was found at the pre-relativizer region on first-pass time, suggesting that irreversible relation is more difficult than the reversible when only one argument is encountered. The reason may be that the irreversible relation carries more specific meanings so draws more attention in the first-pass time. This unexpected finding adds vividness to the dynamic process.
In Experiment 2, the manipulation of animacy also led to a dynamic process, just as we hypothesized. At the pre-relativizer, AI-ORCs were found to be the easiest due to its salient pattern of an animate subject preceding a verb. The other three conditions suffer from the non-canonical word order, the animacy of the argument encountered, and the unfitting animacy to the role (an inanimate agent plus a verb). Later at the relativizer, the fact that IA-ORCs bear an inanimate agent with a verb aggravated the processing effort and makes them the most difficult conditions. At the head noun, the AI-ORCs again become the easiest because the inanimate head quickly fits into the object gap and serves as an inanimate patient. The distinctness of an animate agent and an inanimate patient and the saliency of such a combination possibly speed up the thematic role assignment. At the main verb and the N3 region, the overall pattern suggested that the IA-ORCs and AI-SRCs suffer greatly from their violation of the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
thematic role bias: Animate nouns are salient agents and also eligible patients, but inanimate nouns are only salient patients, but not salient agents. AI-ORCs are numerically easier than the two and their difficulty mainly comes from making the inanimate head noun as the agent of the main verb and the perspective-shifting of the head. The IA-SRCs were the easiest and they are significantly easier than the former two conditions. When reading them, readers do not have to shift perspectives and the thematic roles fit with the salient animacy. In sum, Experiment 2 once again unveiled a dynamic process of reading Chinese RCs and the sources of difficulty can be both structural cues and semantic cues. Contrastive animacy keeps the arguments distinct with a degree greater than the verb manipulation in Experiment 1. Thus, the modulation of animacy on syntactic processing is greater, as compared with the semantic plausibility.