The processing of homographs in Mandarin Chinese. A cross-modal lexical priming experiment

Table of Contents

1. Introduction

2. Models of lexical ambiguity resolution
2.1 The exhaustive access model
2.2 The interactive access model

3. Issues in lexical ambiguity resolution
3.1 Position of visual target in cross-modal priming
3.2 Types of lexical ambiguity
3.3 The relatedness of multiple meanings

4. The processing of Chinese homographs: an experiment
4.1 Method
4.2 Results
4.3 Discussion

5. Conclusion

Bibliography

1. Introduction

The phenomenon of lexical ambiguity is one of several types of ambiguities and is pervasive in almost all natural languages. When a listener hears an ambiguous word with multiple meanings in a sentence, do they activate all meanings associated with that ambiguous word automatically regardless of the preceding context, or do they make use of the preceding context to eliminate irrelevant meanings? This is one of the fundamental empirical questions in the study of language processing, i.e. lexical ambiguity resolution (Gorfein 1989).

Two main types of models have been suggested by the research literature for this issue. The first one is the exhaustive access model (Onifer and Swinney 1981; Swinny 1979), which proposes that all meanings of an ambiguous word will be accessed automatically following the occurrence of the word and that the context can only help to select the appropriate meaning at a later stage. The second contrasting model is the interactive access model, which assumes that the context can facilitate the contextually appropriate meaning of an ambiguous word at the initial stage of lexical access if the prior context provides a strong bias towards the appropriate meaning (Simpson 1981; Simpson and Krueger 1991). As the literature shows, what undoubted is that the context plays an important role in the selection of a contextually appropriate meaning of the ambiguous words, the main debate lies in how early the context effects take place in the lexical processing.¹

So far, these two models have been mostly examined in Indo-European languages, particularly in English. There has been just a scarcity of this line of research in Chinese², one of the most important East Asian languages, which differs significantly from Indo-European languages in terms of its unique psycholinguistic properties in phonological, lexical and syntactic structures (Li et al. 2006). On one hand, Chinese has a massive number of homophones on a lexical-morphemic level, which are a significant aspect of Chinese cultural customs. On the other hand, unlike languages that use alphabetic scripts (e.g. English), in which the orthography represents sounds through which a word's meaning is accessed, Chinese is a logographically-scripted language and the logographic orthography of a Chinese character represents its meaning primarily and the sound secondarily (Taylor and Taylor 2014), i.e. the logographic orthography of a Chinese character is more directly associated with its meaning. These properties can thus pose a good challenge to verify and complement the examination of these models in lexical processing from a different perspective.

For this reason, this paper will further investigate the processing of lexical ambiguity in Chinese. However, it will extend from the time course of context effects to also look at the semantic relatedness in ambiguous meanings, to explore whether it is an additional factor in resolving lexical ambiguity. Therefore, this paper will first present the models of lexical ambiguity resolution. Next, it will review three important factors that have been postulated to influence lexical ambiguity resolution other than context. In the meanwhile, previous findings regarding these factors will be discussed. In addition, this paper will report a cross-modal lexical priming experiment which focuses on Mandarin homographs of varying degrees of semantic relatedness. In this experiment, the time course of context effects and the effect of semantic relatedness on lexical ambiguity resolution will be examined.

2. Models of lexical ambiguity resolution

As mentioned in the introduction, there were two different models for lexical ambiguity resolution: the exhaustive access model and the interactive access model.

2.1 The exhaustive access model

According to the exhaustive or modular access model, language processing is made up of a set of independent processing modules that are functionally autonomous. In a modular system, sentential context at the discourse level of processing (a higher-level subsystem), cannot influence the access of a word's meaning at the lexical level (a lower-level subsystem); as a result, the flow is bottom-up (Ahrens 2001). Therefore, the exhaustive access model proposes that lexical processing is a modular and bottom- up process where all meanings are accessed first and the context can only assist with the selection of the most appropriate meaning at the post-access stage (Onifer and Swinny 1981). Some studies have shown that all meanings of an ambiguous word are momentarily available immediately following an ambiguity, even in the disambiguating contexts, and the appropriate one is selected for elaboration later on (Ahrens 1998, 2001; Onifer and Swinney 1981; Swinny 1979).

2.2 The interactive access model

The alternative view is the interactive or selective access model. According to this model, language processing system is composed of processing modules that are not functionally autonomous. Instead, it can interact and share information across levels. In other words, higher-level (sentential context) can give feedback to guide lower-level (lexical access) processes and lower the amount of “noise” in the processing system (McClelland 1987). Hence, the interactive access model argues that lexical processing is an interaction process restricted to the appropriate meaning of an ambiguous word in context. When the context of the preceding sentence provides a sufficiently strong bias towards a certain meaning, only that contextually supported meaning becomes activated (Simpson 1981). In other words, if the context is biased towards the primary meaning of an ambiguity, only the primary meaning is accessed. However, if the context is biased towards the secondary meaning of an ambiguity, only the secondary meaning is accessed. Some research offered evidence supporting this interactive access model in which context directs the early lexical selection of the appropriate meaning throughout the whole lexical processing (Simpson 1981; Li and Yip 1996, 1998; Shu et al. 2000; Wu und Shu 2002; Zhou et al. 2003).

3. Issues in lexical ambiguity resolution

Researchers investigating lexical ambiguity resolution have in the past found that multiple information, other than context, such as the position of visual target in cross- modal priming paradigm, different types of lexical ambiguity and the relatedness of alternative meanings might all influence the processing of lexical ambiguity.

3.1 Position of visual target in cross-modal priming

Cross-modal priming (Swinny 1979) is a well-established and the most frequently used paradigm for the study of lexical ambiguity. It is this method by which the prior results were obtained. In this paradigm, participants are asked to listen to a sentence containing an ambiguous word (as a prime). At the instant the ambiguous word was heard by the participants, they should then respond to a visual target by making a lexical decision or by naming it. The visual targets are a word that is semantically related to the primary meaning or to the secondary meaning of the prime in the experimental condition and a control word that is unrelated to the meaning of the prime in the control condition. Thus, priming effects, measured by the differences in reaction time to the target in these two conditions, help researchers determine whether a particular meaning of the ambiguous words is activated at a certain time point of lexical access. If the time taken to recognize a visual target that is semantically related to the prime word is shorter than the time taken to recognize a target word that is not semantically related, this meaning of the prime word is considered activated (Onifer and Swinney 1981; Swinney 1979).

So far, relevant studies have produced different and often conflicting results by using this experimental method. If we take the experimental methodologies used to obtain the results into consideration, we can find that the position of the visual target in relation to the auditory perception of the prime word is of great importance for semantic activation (Ahrens 2001). Some studies placed the target word at the acoustic onset of the ambiguous word (Ahrens 1998, 2001; Wu und Shu 2002), some between the onset and the offset (e.g. Li et al. 2001; Li and Yip 1996, 1998; Shu et al. 2000; Wu and Shu 2002; Zhou et al 2003) and others at or after the offset position (Li et al. 2001; Li and Yip 1996, 1998; Wu and Shu 2002, Zhou et al. 2003).

Li and Yip (1996), for instance, found that both meanings of the ambiguity can be accessed when the visual target was presented at the onset of the prime words, which is an evidence for exhaustive access model, but that only the contextually appropriate meaning was accessed when the visual target was presented at or after the offset of the ambiguity, which gives support to interactive access model. Therefore, it is uncontroversial that the processor already sorts out the contextually appropriate meaning at the offset of the prime. However, inconsistent findings have been found regarding the contextual influence during the early stage of lexical access, i.e. at the onset or before the offset of the prime.

3.2 Types of lexical ambiguity

Lexical ambiguity can be divided into three types in all languages (Zhang et al. 2006). These are homophonic homographs (often called homographs, e.g. bow in English;

M kuang4jiang4 ‘shell frame' or ‘composition frame' in Chinese), heterophonic homographs (e.g. record in English; dao3che1 ‘interchange' or dao4che1 ‘back a car' in Chinese), and homophonic heterographs (often called homophones, e.g. raise- raze in English, ‘to be infected with inflammation' and ‘to speak' respectively in Chinese). In Chinese, homophones (both monosyllabic and disyllabic ones) are very frequent, but heterophonic homographs are not (ibid.) In terms of the types of lexical ambiguity investigated in Chinese literatures, homophones and homographs have respectively been used as the experimental stimuli.

Focusing on all the published findings of Chinese homophones in this line of research, all of them support the interactive access model, which argues for the immediate or early effects of context on meaning access (Li et al. 2001; Li and Yip, 1996; 1998; Shu et al. 2000; Zhou et al. 2003). For example, in their cross-modal priming studies, Li et al. (2002) investigated the processing of monosyllabic homophones in Cantonese (Experiment 1) and disyllabic homophones in Mandarin (Experiment 2). Their results show that when listeners hear a multiple-meaning Chinese homophone, they will automatically and rapidly make use of prior context to select the contextually appropriate meaning for the ongoing language processing. Convergent results are also found by another two recent eye-tracking studies on this research topic (Yip and Zhai 2018a, 2018b). Together with these relevant studies in Chinese homophone processing, it seems that Chinese homophones, which share the same phonology but differ in orthography, do not sufficiently lead to co-activation of meanings.

Among the studies that use homographs as stimuli, we can find mixed results contrasting the convergent finding on the processing of Chinese homophones. For example, Ahrens (1998, 2001) have tested on disyllabic nouns and verbs respectively to find out the processing of Chinese homographs in his cross-modal priming studies. The results indicated that both meanings of homographs were still accessed at the onset of the ambiguity when the sentential context is biased towards the second meaning, that is to say, meanings of Chinese homographs, which share both logographic and phonological representations, can be co-activated unconcerned of the prior context. This evidence presented supports the exhaustive access model.

However, Wu and Shu (2002), stood out as the only study that observed early contextual influence on semantic activation of Chinese homographs. Using the cross- modal priming paradigm, they examined the meaning activation of words like ft hua1 (‘spend' or ‘flower') in biased sentence contexts. They found that both primary and secondary meanings were activated more strongly in appropriate contexts than in inappropriate ones, demonstrating that context influences meaning access for homographs at a very early stage. But one of important property should be noted here: the homographs in their study were associated with distinct categories (e.g. the homograph ft hua1 can be an action verb ‘spend' or a noun ‘flower'). It is very likely that before this homograph appears in a sentence, the context already suggests the parse of an action verb or a noun in this sentence. In other words, if two meanings belong to different syntactic categories, one might think that the absence of multiple activation of alternative meanings could due to the inappropriate syntactic information (Zhang et al. 2006). Hence, the syntactic categories associated with an ambiguous word may be relevant to the diverse findings.

[...]

¹ Note that these two models disagree mainly on when the prior context plays a role, i.e. prelexical or postlexical. If one takes into account the variable of dominance or relative frequency of alternative meanings in the processing of lexical ambiguities, there are also the integration model (Rayner and Frazier 1989) and the reordered access model (Duffy et al. 1988). This paper, however, will not consider these effects in lexical ambiguity processing due to limited space.

² In this paper, “Chinese” is used as a generic term to refer to all dialects of Chinese, including Cantonese and Mandarin. In the discussion of specific experiments, however, Cantonese or Mandarin Chinese will be noted respectively.