Chapter 4 Experiment Two: animacy configuration in RC processing
4.4.3 Bias of role assignment with contrastive animacy
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
re-fixate more frequently on the inanimate head nouns and spend longer times reading them during the first-pass reading. At the main verb, the analyses on regressions-out rate (ROR) suggested that readers have problem making the inanimate head be the subject of the main verb.
Therefore, at the regions before the head noun, the cause of the animacy effect is the animacy of the first noun (N1). The inanimate entities are conceptually less accessible so readers require more processing efforts to retrieve from their memory (c.f., Bock & Warren, 1985; Branigan, et al., 2008). At the head noun, readers encounter the same kind of difficulty from the inanimate nouns. At the main verb, the effects of AI difficulty may come from another problem: the inanimate noun in the AIs is assigned an agent role by the main verb. The bias of the role assignment with specific animacy leads to the animacy effects.
4.4.3 Bias of role assignment with contrastive animacy
Now, let us consider the interactions at different position in the sentence along with the main effects. At the pre-relativizer region, the interactions occurred in the analyses of GPT at W2, of TVT at W1 and of RRR at W1. The simple main effect tests of these interactions indicated that the AI-ORCs (d) were far easier than the other
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
three conditions because readers comprehend the animate subject plus the verb without difficulty. At this position, readers are aware of the syntactic ambiguity of SRCs but not yet aware of that of ORCs. Therefore, SRCs (a & c) are expected to be more difficult. Readers also confront difficulties reading the first two words of the RCs with an inanimate argument (a & b). Moreover, the inanimate agent in (b) is against the bias of thematic role assignment with an inanimate noun. The ORCs with an animate subject, however, is not at all difficult for readers since readers temporarily interpret it as a main clause with an animate agent. Thus, the interactions can be accounted for by the combined effects of these structural and semantic causes.
In general, condition (d), the ORCs with an animate agent is the easiest for readers at the pre-relativizer.
At the relativizer, there is a marginal main effect of animacy in the by-participant analysis of GPT. This effect may be the lag effect of processing the inanimate argument in the pre-relativizer region. An interaction was found in the by-participant analysis of GPT. This interaction indicated that the ORCs with an inanimate subject (b) was the most difficult, whereas the other three conditions did not differ from one another. This result cannot be predicted by the combined effects. Rather, it is important that the animacy of an argument fits with the thematic role it is assigned by
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
the verb. Thus, the bias of thematic role assignment is especially essential at this region because readers attempt to integrate the previous words at this region.
According to this bias, agents tend to be animate, and patients tend to be inanimate.
However, animate patients are also easily accessible since the inanimate agent is still absent at the relativizer. Therefore, among the four conditions, the inanimate agent in condition (b) is expected to cause greater interference than the other conditions, in which the same verb integrates with an inanimate patient (a), an animate patient (c), and an animate agent (d). Hence, the hypothesis at the relativizer is revised accordingly.
At the head noun, there are lag effects of SRC difficulty and also main effects of animacy showing greater processing efforts for inanimate head nouns. No interaction was found at this region, but the interaction at the main verb on FFD displayed a possible lag effect that may result from the thematic role’s animacy with contrastive animacy. The interaction suggested that the RCs with an inanimate agent and an animate patient (b & c) are the most difficult. The other two conditions (a & d) are easier because they comprise an animate agent and an inanimate patient. The former two are against the bias of thematic role assignment with contrastive animacy, as the latter two accord with the bias. Note that the combined effects at this region are not
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
supported by any evidence, so the effects are eliminated in the revision part.
At the main verb, only a main effect of animacy was found rather than the interaction on FFD. This main effect suggested the disadvantages of inanimate head noun as the agent of the main verb. The main effect of ORC difficulty was not observed in this experiment. However, it does not imply that readers do not have to solve the syntactic ambiguities of ORCs in this experiment. The impact of contrastive animacy on the syntactic processing is so great that subject-object asymmetry may be reduced. Thus, a greater number of interactions were observed in Experiment 2 than Experiment 1. If we look into the interactions found in this experiment and subject them to post hoc tests, different subject-object asymmetries still occurred at different positions as in Experiment 2.
At the N3 region, interactions were found on ROR and GPT. However, a mere factor of the thematic role’s assignment in the main clause cannot account for the findings. Rather, these findings suggested an overall pattern: IA-SRCs (a) are the easiest of all, while the other three conditions are as difficult. Since N3 is the last region of the whole sentence with an RC embedded in it, readers have to “wrap-up”
the sentence with an aim to integrate the RC and the main clause. Therefore, a combined effect can better explain the results. The effects combine three causes: the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
thematic role’s animacy in the RC with contrastive animacy, perspective shifting, and the thematic role’s animacy in the main clause. Whether the animacy of the arguments fits with the thematic role it is assigned is essential in syntactic integration of the RC and the main clause. It is also essential to integrate the RC and the main clause. The main verb plays an important role in the integration of the two clauses, and perspective shifting may lead to syntactic ambiguities of ORCs at this region.
Therefore, it is reasonable to combine the three causes as a predictor of the overall pattern. The overall pattern was also found on RRR at the pre-relativizer and on TVT at the relativizer. All the findings showing the overall patterns are headed with a star in Table 38.
4.4.4 Interim summary
In summary, readers meet syntactic ambiguities of SRCs in the pre-relativizer region as in Experiment 1, but ORC difficulty is not directed supported by main effects in this experiment. The syntactic ambiguities of ORCs are greatly modulated by the semantic cues of contrastive animacy. The overall pattern involves the effect of perspective shifting, which caused the syntactic ambiguities of ORCs at the main verb.
As for the effect of animacy, inanimate entities consumes more processing than
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
animate entities because inanimate entities are conceptually less accessible than animate ones. The findings supports the previous studies that it takes more efforts to retrieve inanimate entities from memory (c.f., Bock & Warren, 1985; Branigan, et al., 2008). Moreover, thematic role assignment highly correlates with animacy. Animate agents are more accessible than inanimate agents. Inanimate patients are more accessible than animate patients. Animate nouns tend to be agents, while inanimate nouns tend to be patients. This bias is intuitive and iconic, especially when the agent and the patient contrast in animacy. However, we should note that when only one argument was read by readers at the pre-relativizer region, an animate patient is not too bad, either. Readers may suppose both arguments are animate. Besides, most reversible verbs frequently take two animate arguments. In this experiment, the interactions suggested the effect of thematic role’s animacy in the RC or the combined effects of structural cues and semantic cues. As shown in Table 38, readers are cued syntactically and semantically during sentence processing. Some cues, whether syntactic or semantic, render temporal ambiguities for readers. The temporal ambiguities change as the sentence unfolds. Readers read incrementally and thus have to solve the ambiguities at different positions.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Chapter 5 General Discussion
5.1 Interplay of syntactic and semantic cues
This thesis investigated how semantic cues interact with syntactic cues in Chinese relative clause processing. In Experiment 1, the verb in the RC was manipulated to generate two types of semantic plausibility: reversible versus irreversible argument-verb relations. The arguments involved were animate, human common nouns. The arguments in the reversible relation were both plausible agents of the verb, but the ones in the irreversible relation were made distinct by allowing only one of them to be the plausible agent of the verb.
Readers have greater difficulty processing relative clauses with the reversible argument-verb relation than with the irreversible relation after they encounter the head noun. The irreversible relation provides the advantage to assign thematic roles easily through the bias that one argument should be the agent and the other the patient. In contrast, the arguments in the reversible relation compete with each other to be assigned as the agent. The competing arguments render confusion so readers need more efforts to solve the ambiguities of thematic role assignment and retain the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
correct interpretation. The semantic plausibility in the reversible relation does not help readers to assign thematic roles. Readers can only depend on syntactic cues of word order. When readers have to integrate the main clause at the main verb, requirement of perspective shifting for ORCs further confuses readers and consumes their working memory. Readers have to retain correct thematic roles for the relative clause and, meanwhile, assign correct thematic roles for the head noun in the main clause.
Modulation of plausibility on syntactic processing is manifested in the re-analysis process. This implies that the irreversible relation helps readers retain the correct thematic role assignments so readers do not have to re-read the critical head noun so frequently even though perspective shifting does result in ambiguities. To be specific, readers have obvious difficulty in retaining the correct thematic roles for the reversible relation and in shifting perspectives for ORCs when they first reach the main verb. Therefore, at the main verb, both the semantic ambiguity of thematic role assignment and the syntactic ambiguity of perspective shifting make readers garden-path. The most ambiguous part is the thematic role assignment of the head noun because it is critical in solving both ambiguities. The ORCs with the irreversible relation do not suffer from the semantic ambiguity. Instead, the bias of thematic role assignment helps readers retain the correct interpretation in the relative clause easily.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Readers have more cognitive resources to deal with the syntactic ambiguities so the recovery from syntactic ambiguities speeds up, which is consistent with the previous English study (Traxler et al., 2002).
The findings of Experiment 1 concerning the subject-object asymmetry showed that readers apparently detect the embedding structure of Chinese SRCs before meeting the relativizer due to non-canonical word order. The disambiguating position of SRCs hence occurs earlier than that of ORCs. Readers have obvious processing difficulty in solving the syntactic ambiguities of ORCs only when they reach the main verb. The cause of such a comprehension difficulty apparently comes from perspective shifting of the head noun in the ORCs. As soon as readers reach the main verb, both syntactic cues and semantic plausibility initiate their influence on sentence comprehension. The interplay between syntactic and semantic factors is especially obvious when readers attempt to integrate the sentence after they meet difficulties.
In Experiment 2, robust interplay between syntactic cues and semantic cues were observed. The verb in the RC was controlled to be identical across conditions, but the arguments of the verb were manipulated instead to be one animate and one inanimate.
The contrast in animacy keeps the arguments markedly distinct. As a result, the effects of syntactic cues were not as apparent as in Experiment 1.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
At the pre-relativizer, only one argument is met, so the contrast between the two arguments in animacy has not yet occurred. Readers are aware of the subject gap due to the non-canonical word order and have to deal with the syntactic ambiguities of SRCs in this region. The disambiguating positions of SRCs still occur at the pre-relativizer as shown by the regression-out rate (ROR). However, the main effect of SRC difficulty also occurs at the head noun in the measures of first-pass time (FPT) and go-past time (GPT). The results still support the canonical word order hypothesis (Bever, 1970). The word order of ORCs is not misplaced though a relativizer is inserted. The word order of SRCs is transposed as V-O-S and readers have to re-order so as to assign thematic roles. It is very likely that the reversible relation with contrastive animacy across all conditions make the accomplishment of filler-gap dependency difficult. Therefore, readers spend more FPTs and GPTs to assign the thematic role of the head noun. Before the relativizer is met, both the syntactic ambiguities of SRCs and the semantic information collaborate to lead readers. The word order cues and the animacy cues influence readers’ thematic role assignment of the first noun before the readers meet the relativizer. Correct interpretations should be based on the word order information. Meanwhile, the correlation of thematic role assignment and animacy provide a bias that reinforces the correct interpretation or
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
misleading information. The correlation of thematic role assignment and animacy is manifested at the relativizer.
The syntactic ambiguities of ORCs are manifested at the main verb in Experiment 1. In this experiment, main effects showing ORC difficulty were not found. Contrastive animacy configuration seems to have greater impact on readers’
comprehension than syntactic ambiguities at the main verb, where the two clauses should integrate. At the main verb, though ORCs trigger higher regressions after the first-pass reading than SRCs, the difference did not reach significance. Rather, the animacy of the agent in the main clause is more essential for readers. Inanimate agents elicit more regressions than animate agents. Therefore, the correlation between thematic role assignment and animacy matters most.
At the object of the main verb, readers have to integrate the two clauses in order to comprehend the whole sentence ultimately. In Experiment 1, no significant effect was found at the same location. In Experiment 2, readers seem to have less difficulty wrapping up the SRCs with an animate head. The results can still be explained by the bias of thematic role assignment with animacy. Both the roles’ animacy in the relative clause and in the main clause conforms to the bias for the SRCs with an animate head.
Besides, readers do not encounter perspective shifting for SRCs.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Similar to Experiment 1, thematic role assignment is the key to comprehension in Experiment 2. Readers judge the thematic roles of the arguments based on both syntactic and semantic cues.
5.2 Bias of thematic role assignment
Our hypotheses for both experiments are based on some biases of interpreting semantic cues, which are based on real world and probabilistic information. In Experiment 1, there is a bias of thematic role assignment that assigns a certain kind of person to be the agent and another type of person to be the patient of an irreversible verb. This bias has its root in real life. Although both arguments of interest are animate, human common nouns, only one combination of the arguments and the verb looks plausible to readers. Therefore, while the arguments in the reversible relation suffer from the similarity-based interference (Gordon, et al., 2006), the ones in the irreversible relation suffer less from it. The bias motivates a rather precise thematic role assignment in the irreversible relation so the “similarity” of their lexical properties is decreased. So is the confusability of the two arguments. Though the arguments here are not from different noun categories as similarity-based interference was proposed, the bias of thematic role assignment in irreversible relation leads to a
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
similar effect.
However, the similarity-based interference cannot explain the results in Experiment 2. Since the design of contrastive animacy makes the arguments dissimilar, the confusability should decrease for each condition regardless of the linear sequence of the animate / inanimate nouns. The results showed that there is a dynamic up-and-down alternation of the comprehension difficulty across conditions and that IA-ORCs and AI-SRCs are generally the most difficult while IA-SRCs the easiest. If the similarity-based interference works, all of the conditions in Experiment 2 should have reduced in difficulty or there is a steady modulation force of the contrastive animacy on RC difficulty as Experiment 1 showed. The fact is the modulation occurred out of the expectation of similarity-based interference. Animacy seems to have a rule of its own.
There is an animacy bias of thematic role assignment when contrastive animacy takes place. Our hypotheses for Experiment 2 were based on the bias: Animate nouns prefer to be agents, while inanimate nouns prefer to be patients; animate patients are salient enough (though not as salient as the previous two), but inanimate agents are not salient. Note that this bias works best when the verb can take both animate and inanimate nouns as both agents and patients. In Experiment 2, this kind of verb was
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
chosen and controlled across conditions. As a result, this hypothesis supports the findings observed at different positions. At the pre-relativizer, only one of the argument occurred with a verb. The sequence of the argument (N1) and the verb (V1) determines which thematic role the argument should be: If the argument follows the verb, it is in the object position and surfaces as a patient in the active voice sentence;
otherwise, it is in the subject position and surfaces as an agent. At this moment, the second argument is missing, so the animacy is not contrastive yet. Readers encounter difficulty when the inanimate noun in IA-ORCs serves as an agent but there is no such difficulty when the animate noun in AI-SRCs serves as a patient. Later when the head noun, the second argument, occurs, the animate head in IA-ORCs finds it against the bias to be a patient when an inanimate agent is in place. Similarly, the inanimate head of the AI-SRCs finds it difficult to fit into the agent role. Therefore, IA-ORCs and AI-SRCs are the most difficult among the four conditions for Chinese readers.
The AI-ORCs and IA-SRCs both have salient fit of animacy and thematic roles, so they are easier than the former two. IA-SRCs are the easiest since they do not suffer from perspective-shifting as AI-ORCs do.
Traxler et al. (2002) proposed that readers prefer to treat sentential subjects as the subject of relative clauses in the initial interpretation. The initial syntactic analysis
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
is correct for SRCs so processing proceeds smoothly. When readers read ORCs, the
is correct for SRCs so processing proceeds smoothly. When readers read ORCs, the