Chapter 2 Literature Review
2.4 Processing of Chinese ambiguous words
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
meaning is a verb, which needs more time to reach a high activation level, resulting in a somewhat delayed meaning competition. In contrast, when an NV homograph is embedded in a subordinate-biased context, the context-unintended meaning is a noun, reaching a high activation level fast, leading to an immediate meaning competition.
As a result, the SBE for VN homographs may be less obvious than that for NV homographs in a subordinate-biased context.
2.4 Processing of Chinese ambiguous words
With different types of ambiguous words and experimental paradigms, studies on Chinese lexical ambiguity resolution have also obtained discrepant results. Using lexical-semantic ambiguous words (i.e., NN/VV), some researchers demonstrated the selective access of multiple meanings of Chinese ambiguous words (任桂琴, 韩玉昌,
& 于泽, 2008; 周治金 & 陈永明, 2006; 陳怡蓉, 2008; 韩玉昌, 于泽, & 李立洁,
2009), while others found the multiple access in Chinese lexical ambiguity resolution (任桂琴 et al., 2008; 周治金 & 陈永明, 2006; 陳怡蓉, 2008; 楊芝瑜, 2010; 盧 怡璇, 2012; 韩玉昌 et al., 2009). Similar to English findings, meaning dominance
and contextual bias jointly influence Chinese lexical ambiguity resolution (任桂琴 et al., 2008; 楊芝瑜, 2010; 盧怡璇, 2012; 韩玉昌 et al., 2009). In a neutral context,
multiple meanings of Chinese balanced homographs were activated approximately at
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
the same time and competed with each other, while processing difficulty occurred in the disambiguating region following biased homographs when the subordinate meaning was context-intended, indicating that the dominant meaning was selected initially (任桂琴 et al., 2008; 韩玉昌 et al., 2009). In a biasing context, the SBE
was also observed for Chinese biased homographs. Using a normal reading task and a visual world paradigm, 盧怡璇 (2012) conducted two eye-tracking experiments to
investigate how Chinese biased homographs are resolved in a subordinate-biased context. In her Experiment 1, three types of target words were manipulated:
low-frequency homographs (A), low-frequency unambiguous words (LF), and high-frequency homographs (HF). The results showed the significant SBE (A > LF) in gaze durations and second-pass times on the target as well as in go-past time and rereading probability in the post-target region. In her Experiment 2, the time course of Chinese lexical ambiguity resolution was examined. The results showed that both the dominant and the subordinate meaning were activated prior to the acoustic onset of post-target. Taken together, the two experiments demonstrated that the dominant meaning was still activated even in a subordinate-biased context, supporting the reordered access model. Finally, the resolution of Chinese ambiguous verbs was also delayed. 韩 玉 昌 et al. (2009) observed a delayed SBE for Chinese biased
homographic verbs in a subordinate-biased context. The SBE was delayed until the
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
second-pass reading on the disambiguating information following the target.
Research on the semantic resolution of Chinese SCA words has also obtained inconsistent findings and supported different accounts. As in English, a delayed resolution of Chinese SCA words was also observed. Using a moving window paradigm, 武 宁 宁 and 舒 华 (2003b) found longer reading times in the
disambiguating region that followed the SCA words compared to controls, no matter when the dominant or the subordinate meaning was context-intended. Nevertheless, some studies supported the constraint-satisfaction models. Using a normal reading task, 武宁宁 and 舒华 (2003b) observed longer reading times and more regressions
from the disambiguating region only when the subordinate meaning was context-intended, indicating an immediate resolution of Chinese SCA words. Another interesting research did not find the results completely consistent with either the delay model or the constraint-satisfaction models. 張亞旭, 劉友誼, 舒華, and 孫茂松
(2003) suggested that the strength of probabilistic constraints can determine how syntactic category ambiguities are resolved. When the probabilistic constraints are weak, the syntactic parsing principles (i.e., to construct the simplest structure) can mediate the resolution of syntactic category ambiguity. In their study using a moving window paradigm, two types of Chinese SCA words (i.e., VN: 交代; NV: 包裝) were preceded by a neutral context and followed by a verb-biased context (e.g., 最後
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
村長終於給了馮莉一個交代事件發生經過的機會). One of the important results
showed that NV caused larger processing difficulty than VN in the disambiguating regions, indicating that the probabilistic constraints (e.g., frequency of meaning) can modulate the resolution of syntactic category ambiguity. In addition, when the probabilistic constraints are weak3, the processing difficulty was still observed in the disambiguating regions, demonstrating an effect of syntactic parsing principles.
Generally, their findings supported the constraint-satisfaction models. However, it is unclear whether the syntactic parsing principles or the probabilistic constraints guided the readers to select the noun meaning for NV. Thus, to conclude the probabilistic constraints determine how syntactic category ambiguity is resolved, this study warrants a type of strongly-biased VN (i.e., the verb meaning is far more frequent than the noun meaning). If processing difficulty for strongly-biased VN can be found instantly in the target region, it means both the syntactic parsing principles and the probabilistic constraints are considered simultaneously, which can strongly buttress the constraint-satisfaction models.
To sum up, the resolutions of Chinese lexical-semantic and syntactic category ambiguity are still unclear. Although some researchers have investigated how meaning dominance and contextual bias interact during Chinese lexical ambiguity
3 According to the authors, the mean proportion of noun to verb meaning of the VN was 0.9; thus, the VN used in this experiment were near balanced.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
resolution, the findings diverge and bolster different accounts. Moreover, there is a lack of thorough examination on how different combinations between meaning dominance and syntactic category are resolved during sentence comprehension.
Therefore, the present study attempts to clarify the influence of syntactic category and semantic constraints on the resolution of Chinese lexical-semantic and syntactic category ambiguities. Experiment 1 aims to examine whether the syntactic category constraints can determine the semantic resolution of Chinese SCA words and to investigate whether syntactic category of alternative meanings of Chinese homographs would influence the SBE during lexical ambiguity resolution. Four types of disyllabic homographs (NN, VV, VN and NV) were embedded into sentences involving syntactic category and semantic biases toward the subordinate meaning.
Experiment 2 aims to directly examine whether the semantic resolution of Chinese SCA words can be determined by the syntactic category constraints alone. The four types of disyllabic homographs were embedded into semantically neutral sentences containing only syntactic category information consistent with the subordinate meaning. Frequency-matched unambiguous words were also used as control words, which can fit into the same sentence frames constructed for their corresponding homographs.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
42
Experiment One:
Syntactic category and semantic constraint effects on lexical ambiguity resolution
The aims of Experiment 1 were (1) to examine whether the syntactic category constraints can determine the semantic resolution of Chinese SCA words and (2) to investigate whether syntactic category of alternative meanings of Chinese homographs can influence the SBE during lexical ambiguity resolution. The purpose was to assess the availability of syntax-first account and constraint-satisfaction account on explaining the semantic resolution of syntactic category ambiguity. In Experiment 1, low-frequency biased homographs were embedded in sentences in which the sentential context was syntactically and semantically biased toward the subordinate meaning. Frequency-matched unambiguous words were used as control words. Previous studies demonstrated that gaze duration (GD), second-pass time, and spillover were important eye-movement measures for the subordinate bias effect (SBE) (Folk & Morris, 2003; 盧怡璇, 2012). According to syntax-first models, the SBE (A
> UA) was expected for NN and VV homographs because the syntactic category constraint cannot distinguish between the alternative meanings and both meanings were activated and competed. The SBE may last until the post-target region. In
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
addition, the inherent properties of verb may cause delayed resolution of VV homographs as compared to NN homographs. In contrast, for VN and NV homographs, only the syntactic category appropriate meaning was activated because the syntactic processor was able to select the meaning. Thus, no SBE was predicted in the target and post-target regions. However, according to constraint-satisfaction models, the SBE was predicted for four types of ambiguous words because multiple meanings were activated. The SBE may last until the post-target region because the subordinate-biased context conflicted with the activated dominant meaning. In addition, the SBE would be immediately modulated by syntactic category of alternative meanings. The SBE for VN homographs might be less obvious than that for NV homographs, since the dominant-verb meaning is activated more slowly than the dominant-noun meaning.
3.1 Method
3.1.1 Participants
Forty undergraduate and graduate students (9 males and 31 females) aged between 19-26 years old (mean age = 21) were paid to participate in Experiment 1.
All participants were native speakers of Mandarin Chinese and had normal or corrected-to-normal vision. None had participated in any previous norming study.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
3.1.2 Materials and design
Eighty biased homographs, including 4 types (NN, VV, VN, and NV), and 80 corresponding unambiguous words (NNN, VVV, NVN, and VNV) were used as target words in Experiment 1. Word-form frequency, word stroke, and frequency of the first character (C1F) obtained from Academia Sinica Balanced Corpus of Modern Chinese (Academia Sinica Balanced Corpus, 2004) were matched for homographs and unambiguous words (see Table 3). The two-way analysis of variance (ANOVA) with word type and ambiguity as independent variables was conducted for the word properties. For word-form frequency, there was a significant difference among word types, F(3, 152) = 4.72, p < .01. Post hoc comparisons using the Bonferroni test revealed that the frequency of VN was significantly higher than that of NN (p < .01) and that of VV (p < .05). There was no significant difference between homographs and unambiguous words, and the interaction was non-significant (Fs < 1, ps > .3). For word stroke and C1F, the main effects and the interaction were non-significant (Fs < 3, ps > .07).
‧
Examples of word materials and means (and standard deviations) of word properties for each condition in Experiment 1
Word
type Ambiguity Word
Dominant / Subordinate meanings Word property Dominant Subordinate Frequency Stroke C1F
NN
Note. A = ambiguous words; UA = unambiguous words; C1F = frequency of the first character
Sentences were designed in such a way that both the preceding and the succeeding sentential context were semantically and syntactically biased toward the subordinate meaning of the homographs. For semantic bias, a word semantically related to the subordinate meaning of homographs and the control words was embedded in the preceding context. For syntactic category bias, we used a pre-target word to constrain the syntactic category of the targets to the syntactic category of
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
homographs’ subordinate meaning. Pre-target, target and post-target regions were all disyllabic Chinese words. A total of 80 sentence frames were constructed where the homograph and the corresponding unambiguous word can fit the context into the same position. The beginning of target words was located in the range of 14th to 18th character in the sentence containing 25 to 27 characters. Table 4 shows the examples of target words and experimental sentences. Two experimental lists of trial sequences were established in such a way that the homograph-embedded and the corresponding unambiguous word-embedded sentences were not in the same list. Each list consisted of 90 sentences, including 40 homograph-embedded sentences (10 for each type), 40 unambiguous word-embedded sentences (10 for each group) and 10 filler sentences.
The order of experimental sentences was randomized in each list, and each sentence was presented only once for each participant. The experiment consisted of five blocks of 18 trials, beginning with 2 filler trials. Around one third of trials were followed by a true-false comprehension question.
‧
Examples of targets and sentences for each condition in Experiment 1
Sentence
Type Conditions Preceding context Target Succeeding context
NN
A 由於強烈颱風肆虐，屋外的猛烈
Due to the terrible typhoon, outside the house there was strong
風聲
The manager of bank asks the accountant
算帳 to do accounts
之後要立即向他報告 帳目。
and then report to him.
When diagnosing, the doctor said the expired medicine had lost
效力
the efficacy 足 以 對 抗 病 菌 。 to fight the germs.
UA
功效 the efficacy
NV
A 這名受過專業訓練的員警已經成功
The trained police officer has successfully
Note. A = ambiguous words; UA = unambiguous words
Prior to eye-tracking experiment, six norming studies were conducted to ensure the appropriate manipulations of word semantics and sentential contexts. For word semantics, ambiguous words should be biased homograph, which has two distinct meanings with biased dominance (relative frequency of use). For sentential contexts, both syntactic and semantic context should be biased toward the subordinate meaning of homographs. Table 5 shows the summarized results of all norming studies in Experiment 1. A total of 160 Chinese sentences were chosen for Experiment 1,
‧
including 80 homograph-embedded (20 for each type) and 80 corresponding unambiguous word-embedded sentences (see Appendix A).
Table 5
Means (and standard deviations) of words and sentences in all norming studies for Experiment 1
UA Identical to the corresponding homographs 4. Semantic bias A 0.92 (0.07) 0.94 (0.06) 0.9 (0.12) 0.95 (0.07)
UA － － － －
5. Syntactic category of target word
A 0.99 (0.05) 1 (0.02) 0.93 (0.13) 0.87 (0.19) UA 0.99 (0.02) 0.99 (0.02) 0.99 (0.04) 0.92 (0.22) 6. Plausibility
(1 = not plausible; 7 = very plausible)
A 5.7 (0.72) 5.53 (0.68) 5.82 (0.46) 5.65 (0.67) UA 5.95 (0.54) 5.76 (0.54) 5.81 (0.49) 5.8 (0.47)
Note. A = ambiguous words; UA = unambiguous words
3.1.2.1 Norming study 1: Meaning dominance
This norming study aimed to pick appropriate Chinese biased ambiguous words of four types (NN, VV, VN and NV) and unambiguous control words for the present experiments. The dominant and the subordinate meaning of biased ambiguous words were determined by 40 participants’ responses. Due to insufficient qualified biased ambiguous words after the rating, another group of 40 participants were invited to rate
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
the meaning dominance for the additional ambiguous and unambiguous words. All participants’ native language was Mandarin Chinese.
One hundred and eight Chinese disyllabic ambiguous words and 140 Chinese disyllabic unambiguous words were collected from various resources, including 現代 漢語多義詞詞典 (袁暉, 2001), 中文多字多義詞自由聯想常模 (胡志偉, 陳貽 照, 張世華, & 宋永麒, 1996), 詞彙歧義解困的次要語義偏向效應再視：中文 多義詞的眼動研究證據 (盧怡璇, 2012), and Academia Sinica Balanced Corpus of
Modern Chinese (Academia Sinica Balanced Corpus, 2004). The unambiguous words
were collected which were suitable to be embedded in the same position as the corresponding ambiguous words in sentences. 88 ambiguous words and 140 unambiguous words were rated in this norming study.4 Two lists were constructed, and 20 participants were assigned to rate one of the lists. The ambiguous words and their corresponding unambiguous words were assigned to different lists. All words in each list were presented in a randomized order. Some examples of the questionnaire are provided in Appendix B.
Participants used the words in the list one at a time to generate a comprehensible sentence according to the first meaning that came to their mind. In addition, they were
instructed that position of the target words in generated sentences was not restricted
4 20 out of 108 collected ambiguous words that have been rated in the Interpretation Preference Task conducted by 盧怡璇 (2012) were not rated again in the present norming study.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
and that the meaning of target words in generated sentential context should be clear enough. Five practices were given to participants before the main task, and the rating could be completed around one hour.
To obtain participants’ first interpretation of each word, we examined the generated sentences to check the meaning of target words in each sentence based on the following online resources, Revised Chinese Dictionary and Chinese Wordnet (CWN) . Each word’s meaning dominance was calculated as the proportion of
participants’ first interpretation. If all participants recognized a word with the same interpretation, then the meaning dominance of the word would be 1. Ambiguous words were regarded as biased if the meaning dominance was above 0.7.5 Their dominant meaning was the interpretation with which at least 70% of the participants recognized the words, and subordinate meaning was the one with which less than 30%
of the participants recognized them. Unambiguous words were chosen only when the meaning dominance was 1.6 Overall, sixty-two biased ambiguous words and 114 unambiguous words passed the meaning dominance rating.
3.1.2.2 Norming study 2: Meaning relatedness
The aim of meaning relatedness rating was to ensure that the dominant and the
5 For VN ambiguous words, they were regarded as biased if the meaning dominance was above 0.6 because of insufficient materials. Only 2 words’ meaning dominance was 0.65, and 1 was 0.6.
6 1 unambiguous word was chosen whose meaning dominance was 0.75 because there was no other suitable candidate unambiguous word for the corresponding ambiguous word.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
subordinate meaning of ambiguous words were semantically unrelated. Twenty undergraduate and graduate students were paid to participate in the rating. All participants’ native language was Mandarin Chinese, none of whom had participated in the Meaning Dominance Rating.
Sixty biased ambiguous words (8 NN, 11 VV, 20 VN, and 20 NV) obtained from Norming study 1 were used to construct two understandable sentences for their dominant and subordinate meanings respectively. A questionnaire was constructed, in which the ambiguous words, meaning definitions, and sentences were presented (see Appendix C). Four lists were generated with different randomized word orders, and each was rated by 5 participants.
Participants read one ambiguous word at a time and then read two definitions of both the dominant and the subordinate meaning of the ambiguous word. They continued to read two example sentences. Subsequently, they were asked to rate the semantic relatedness between the two meanings on a 7-point scale (where 1 = not related; 7 = much related). Three practices were given to participants before the main
task, and the rating could be completed around half an hour.
The results showed that most ambiguous words were qualified homographs, that is, their meaning relatedness was below 3.5. Including 20 qualified biased homographs having been rated in Lu’s study, 80 homographs (20 for each type) were chosen for
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
the eye-tracking experiment. The one-way ANOVA on meaning relatedness showed no significant difference across four types of homographs, F(3,76) = .34, p = .80.
3.1.2.3 Norming study 3: Word predictability and syntactic category bias
A cloze task was used to ensure that both biased homographs and unambiguous words were unpredictable from the preceding context. In addition to word predictability, syntactic category bias rating was conducted to evaluate the syntactic category constraint from the preceding context. The sentences were rated by 40 participants. Due to insufficient qualified sentences after the rating, another group of 40 participants were invited to rate the additional sentences. Eighty undergraduate and graduate students were paid to participate in Norming study 3－6. All participants’
native language was Mandarin Chinese, and none of them had participated in Norming study 1 and 2.
There were 80 selected word pairs of homograph and unambiguous word as well
There were 80 selected word pairs of homograph and unambiguous word as well