Edited by: Manuel Perea, University of Valencia, Spain
Reviewed by: Jianfeng Yang, Shaanxi Normal University, China; Hsu-Wen Huang, City University of Hong Kong, Hong Kong
*Correspondence: Xingshan Li
This article was submitted to Language Sciences, a section of the journal Frontiers in Psychology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
In the current study, we used eye tracking to investigate whether senses of polysemous words and meanings of homonymous words are represented and processed similarly or differently in Chinese reading. Readers read sentences containing target words which was either homonymous words or polysemous words. The contexts of text preceding the target words were manipulated to bias the participants toward reading the ambiguous words according to their dominant, subordinate, or neutral meanings. Similarly, disambiguating regions following the target words were also manipulated to favor either the dominant or subordinate meanings of ambiguous words. The results showed that there were similar eye movement patterns when Chinese participants read sentences containing homonymous and polysemous words. The study also found that participants took longer to read the target word and the disambiguating text following it when the prior context and disambiguating regions favored divergent meanings rather than the same meaning. These results suggested that homonymy and polysemy are represented similarly in the mental lexicon when a particular meaning (sense) is fully specified by disambiguating information. Furthermore, multiple meanings (senses) are represented as separate entries in the mental lexicon.
Lexical ambiguity is one of several types of ambiguities and is widespread in human languages. There are two types of lexical ambiguities: homonymy and polysemy (Lyons,
Two different but closely related research questions regarding lexical ambiguity are how they are represented and how they are processed. Words are thought to be represented as lexical entries in the mental lexicon and each lexical entry contains the lexical information concerning the word, such as semantic, syntax, phonology, orthography, and so on (Levelt,
Previous studies conducted on alphabetical languages have shown that multiple meanings of a homonymous word are represented as separate lexical entries in the mental lexicon. Some researchers found that words with multiple meanings (homonymy) facilitated word processing in isolated word recognition studies. When using a lexical decision task, some studies found that response times (RTs) for homonymous words were shorter than for unambiguous control words (Rubenstein et al.,
In addition to evidence from isolated word recognition studies, eye-tracking studies have also provided evidence supporting separate representation for homonymy. In such studies, homonymous words were usually preceded by either a neutral or a dominant (subordinate) biasing sentence context. Readers spent longer fixation times on a balanced homonymy than an unbalanced one in a neutral context (Onifer and Swinney,
Many recent studies have investigated whether polysemous words are represented and processed differently compared with homonymous words. Some studies showed that they are represented and processed differently. Klepousniotou (
The differences between homonymy and polysemy were also observed in sentence reading tasks. Frazier and Rayner (
In contrast, others studies have showed that, like homonymy, polysemy is represented as separate lexical items. Recently, Foraker and Murphy (
Clearly, the issue of whether polysemous words are represented and processed similarly or differently compared with homonymous words is far from clear. It is important to note that contradictory findings about the representation and processing of homonymy and polysemy have come from different experimental tasks, and different tasks may investigate different processing stages of homonymy and polysemy. For example, a lexical decision task and a reading task may reflect different aspects of semantic processing of a word (e.g., there is also a semantic integration involved in a reading task). According to the assumption of constraint-based model (Spivey-Knowlton and Sedivy,
It is worth noting that most studies on the representation and processing of polysemous words have investigated alphabetical languages and less is known about how these two types of words are represented and processed in Chinese. Like English, Chinese has a large number of ambiguous words. For example, the word 火星 (
Most studies conducted on the Chinese language have focused on homonymous words and have showed that homonymous words are represented separately in the mental lexicon as in alphabetical languages (Ren et al.,
It is well known that alphabetical and non-alphabetical language systems have important cross-language differences. For example, a Chinese character contains more semantic information than an English letter (Hoosain,
The major goal of the current study was to explore whether homonymous words and polysemous words were represented and processed similarly or differently in Chinese reading. Addressing this question is important because it can improve our understanding of homonymy and polysemy in general, and also reveal how they are represented in the mental lexicon. In particular, we were interested in examining whether the differences observed between two lexical ambiguities in prior studies was due to representational differences. If the two kinds of words have similar mental representation, the differences observed in previous studies cannot be accounted for by mental representation differences. Investigating both homonymy and polysemy in the same study could reveal some more general rules governing word representation and processing.
In the current study, we embedded unbalanced ambiguous words (homonymy and polysemy) in sentences and manipulated different parts of the sentences to provide different contexts for ambiguous words. The first part of the sentence was manipulated so that the target word was given either no clear biasing information (i.e., the neutral context condition) or bias toward either the dominant meaning (i.e., the dominant context condition) or the subordinate meaning (i.e., the subordinate context condition) of the target word. The critical target words were followed by a disambiguating region favoring either the dominant meaning (i.e., the dominant disambiguating condition) or the subordinate meaning (i.e., the subordinate disambiguating condition). A consistent condition was created when the prior context and disambiguating regions favored the same meaning of an ambiguous word, and an inconsistent condition was created when the disambiguating region supported a different interpretation than the prior context.
If polysemy is represented only as a core sense, and a specific sense is created online according to the underspecified core sense account, then no sense competition (i.e., longer reading times in the disambiguating region) should be observed in the inconsistent condition. On the other hand, if meanings (senses) showed a competition effect in the inconsistent condition, this would be strong evidence supporting a similar representation for both homonymy and polysemy (i.e., separate lexical representation). In addition, the contrasting ambiguity effects and processing differences observed in previous studies could not be explained on the basis of the representational difference between homonymy and polysemy.
Thirty participants (20 females and 10 males) were recruited from universities in Beijing near the Institute of Psychology, Chinese Academy of Sciences. Each was paid 45 Yuan (~9 U.S. dollars) to participate in this experiment. All were native Chinese speakers and had normal or corrected-to-normal vision. The participants' ages were from 18 to 28 years (
Two hundred potentially ambiguous words with two meanings were first selected from
Each of the ambiguous words was embedded in a sentence to assess the processing of ambiguous words in sentence reading. All experimental sentences were composed of two parts: a prior context and a disambiguating region. The prior context was manipulated to bias the target word toward either the dominant meaning (dominant context condition), subordinate meaning (subordinate context condition), or no biased information (neutral context condition). The disambiguating region followed the prior context and was consistent with either the dominant or subordinate meaning of the ambiguous word. The design was a 2 (word type: homonymy and polysemy) × 3 (context: dominant, subordinate, and neutral) × 2 (disambiguation: dominant and subordinate) within-participant design. This design created 3 consistent (when both prior context and disambiguating region were biased toward the same meaning or when the disambiguating region was biased toward the dominant meaning in the neutral context) and 3 inconsistent conditions (when prior context and disambiguating region were biased toward two divergent meanings, or when the disambiguating region was biased toward the subordinate meaning in the neutral context). Sample sentence materials are shown in Table
homonymy | Dominant | Dominant | 当宇航员小王看到火星出现的那一瞬间,不由得对宇宙的壮美惊叹不已。 |
(When the astronaut saw [the] |
|||
Dominant | Subordinate | 当宇航员小王看到火星出现的那一瞬间,不由得对将临的危险心生恐惧。 | |
(When the astronaut saw [the] |
|||
Subordinate | Dominant | 当消防员小王看到火星出现的那一瞬间,不由得对宇宙的壮美惊叹不已。 | |
(When the fireman saw [the] |
|||
Subordinate | Subordinate | 当消防员小王看到火星出现的那一瞬间,不由得对将临的危险心生恐惧。 | |
(When the fireman saw [the] |
|||
Neutral | Dominant | 当富商李成楠看到火星出现的那一瞬间,不由得对宇宙的壮美惊叹不已。 | |
(When the businessman Li Chengnan saw [the] |
|||
Neutral | Subordinate | 当富商李成楠看到火星出现的那一瞬间,不由得对将临的危险心生恐惧。 | |
(When the businessman Li Chengnan saw [the] |
|||
polysemy | Dominant | Dominant | 那位小说家说这个框架已经非常完美了,总体逻辑结构非常严谨清晰。 |
(The novelist said the |
|||
Dominant | Subordinate | 那位小说家说这个框架已经非常完美了,主体工程应该很快就能完工。 | |
(The novelist said the |
|||
Subordinate | Dominant | 那位工程师说这个框架已经非常完美了,总体逻辑结构非常严谨清晰。 | |
(The engineer said the |
|||
Subordinate | Subordinate | 那位工程师说这个框架已经非常完美了,主体工程应该很快就能完工。 | |
(The engineer said the |
|||
Neutral | Dominant | 那个郭月阳说这个框架已经非常完美了,总体逻辑结构非常严谨清晰。 | |
(Guo Yueyang said the |
|||
Neutral | Subordinate | 那个郭月阳说这个框架已经非常完美了,主体工程应该很快就能完工。 | |
(Guo Yueyang said the |
Prior context biased readers toward a dominant, subordinate meaning, or was neutral. Names of professions were employed in prior context to increase the activation level of intended meanings in different context conditions. For instance, 宇航员 (
Eighteen participants were recruited to rate the biasing of the disambiguating region, and they were given a target word plus the disambiguation region following it. They were instructed to guess the meaning of the target word based on the following disambiguating region by making a 2-alternative forced choice. Results showed that the mean percentage of choosing an intended biasing meaning of all sentences was 93%, indicating that disambiguating regions were biasing enough to activate the intended meaning.
Another 15 participants were presented with prior context until the target word, then asked to write down a word that fit into the prior context. If the target word was given, then the predictability was 1; otherwise, the predictability was 0. Results showed that only one item was predicted by 80% of the participants and almost 90% of the target words were completely non-predictable. The mean predictability of target words was 0.02.
To ensure that the whole sentence could be understood by readers, we recruited another 18 participants to rate the readability of the sentences. They were asked to read the sentences and to respond “yes” if they could understand it and “no” if they could not. The results showed that 96% of all sentences could be understood by readers.
Participants' eye movements were recorded using an eye-tracking system. A chinrest was used to minimize head movement during the experiment. Participants read sentences (which were presented horizontally from left to right on a single line) on a 21-inch CRT monitor (resolution: 1024 × 768 pixels; refresh rate: 150 Hz) connected to a Dell PC. They were seated 58 centimeters away from the computer; at this distance, one character subtended a visual angle of ~0.6°. Viewing was binocular, but only the right eye was monitored.
Before the experiment, participants were given a brief introduction to the eye tracker as well as instructions for the experiment. The eye tracker was calibrated and validated at the beginning of the experiment. Participants were required to look at a white dot randomly presented in a horizontal middle line of the screen. Calibration was conducted as necessary during the experiment. The calibration error was smaller than 0.5° of the visual angle. Each participant read 144 sentences (72 containing homonymous words and 72 containing polysemous words). In addition, 144 filler sentences with the same sentence structure were included randomly among the experimental sentences to prevent participants from becoming aware of lexical ambiguity. All sentences were presented randomly. In order to familiarize participants with the experimental procedure, 10 practice sentences were presented before the test session. Participants were instructed to read these sentences silently for comprehension, and comprehension questions were asked after about 30% of the sentences were read. All sentences were displayed on one line on the computer screen. Participants were instructed to fixate on a white dot presented in the middle of the screen for drift check at the beginning of each trial. Next, a white square (1° × 1°) was presented at the position of the first character of the sentence before the sentence was shown. Participants pressed a button on the bottom box to start the next trial. The whole experiment lasted about 90 min with a few short breaks.
Mean accuracy of the comprehension questions was 93%, indicating that participants could understand these sentences well. Trials with blinks in the target-word region, pre-target-word, or post-target-word regions and more than three blinks were discarded from the analysis, resulting in 12% of the trials being excluded. In addition, fixations shorter than 80 ms or longer than 1000 ms were also excluded from our analyses. Prior to statistical analysis, trials with a reading time beyond three standard deviations for each participant and each condition were excluded from the analysis. Mean fixation durations in milliseconds and standard errors for each condition are reported in Table
First pass time | Dominant | Dominant | 286 (8) | 282 (8) | 1198 (30) | 1236 (35) |
Dominant | Subordinate | 280 (9) | 289 (13) | 1103 (39) | 1180 (42) | |
Subordinate | Dominant | 297 (8) | 285 (8) | 1116 (43) | 1117 (30) | |
Subordinate | Subordinate | 290 (8) | 287 (8) | 1133 (34) | 1141 (40) | |
Neutral | Dominant | 287 (8) | 286 (8) | 1134 (37) | 1176 (32) | |
Neutral | Subordinate | 284 (8) | 294 (11) | 1070 (36) | 1064 (37) | |
Total time | Dominant | Dominant | 362 (16) | 359 (23) | 1782 (50) | 1867 (70) |
Dominant | Subordinate | 439 (24) | 384 (24) | 2041 (68) | 2035 (54) | |
Subordinate | Dominant | 404 (14) | 395 (18) | 1891 (54) | 1945 (56) | |
Subordinate | Subordinate | 415 (20) | 386 (17) | 1867 (52) | 1921 (49) | |
Neutral | Dominant | 404 (20) | 399 (19) | 1815 (46) | 1798 (55) | |
Neutral | Subordinate | 427 (19) | 410 (20) | 1947 (50) | 1905 (56) | |
Regression-out probability | Dominant | Dominant | – | – | 0.62 (0.05) | 0.59 (0.05) |
Dominant | Subordinate | – | – | 0.72 (0.05) | 0.64 (0.04) | |
Subordinate | Dominant | – | – | 0.69 (0.05) | 0.69 (0.05) | |
Subordinate | Subordinate | – | – | 0.69 (0.05) | 0.64 (0.04) | |
Neutral | Dominant | – | – | 0.67 (0.05) | 0.59 (0.05) | |
Neutral | Subordinate | – | – | 0.72 (0.05) | 0.67 (0.04) |
We mainly analyzed eye-movement measures in the target-word region (the two-character target word) and the disambiguating region (the disambiguating region). See Figure
We reported the following eye-movement measures: (1) first-pass time
A series of 2 (word type: homonymy, polysemy) × 3 (prior context: dominant, subordinate, and neutral) × 2 (disambiguation: dominant disambiguating and subordinate disambiguating) ANOVA were carried out with word type, context condition, and disambiguating condition as within-participant factors, and with participants (
For first-pass time, no main effects (
Total times on homonymous words (
Importantly, there was significant interaction between prior context and disambiguation [
For first-pass time, main effects of prior context [
In addition, we found that in some conditions (e.g., dominant-subordinate condition) first-pass times in the disambiguating region were numerically longer for polysemous words than for homonymous words. However, this difference did not statistically significant [
For total time, there was a main effect of prior context [
Although no three-factor interaction was found in the disambiguating region (
Moreover, prior context also affected total times on the subordinate disambiguating region [
In addition, neither the interaction between word type and prior context [
Regression-out probability was higher in the homonymy condition (
Moreover, the interaction between prior context and disambiguation was also significant [
The current study investigated whether the processing and representation of homonymous words and polysemous words are similar or different in Chinese reading. We manipulated two different parts in a single sentence: (i) the prior context preceding target words, and (ii) the disambiguating region following the target words. We compared these two lexical ambiguity words (i.e., homonymy and polysemy) within one experiment.
We found similar data patterns for homonymy and polysemy. When the prior context and disambiguating region created bias toward the same meaning or sense (the consistent condition), readers spent less total time on the target word as well as on the disambiguating region and made fewer regressions out from the disambiguating region. However, when the prior context and the disambiguating region created bias toward divergent meanings (the inconsistent condition), longer total times on the target word and disambiguating regions, and more regression-outs were observed in the disambiguating region, as indicated by significant interaction between context and disambiguation. The observed significant interaction between prior context and disambiguation showed that different meanings (senses) tend to compete with each and the processing was hindered in the inconsistent condition.
These results showed that different meanings of a homonymous word are separately represented. In the inconsistent condition, readers had to retrieve another meaning when a contextually inappropriate meaning was previously accessed, indicating that different meanings were incompatible with each other. This finding was in line with results in studies of alphabetical languages, supporting a separate lexical account for homonymy (Rubenstein et al.,
In addition, we also found longer first-pass times in the disambiguating region in the consistent condition than that in the inconsistent condition for both homonymy and polysemy. The increased processing time observed under the consistent condition suggested a wrap-up process (Just and Carpenter,
It should be noted that some eye movement measures were different between homonymy and polysemy in some conditions. For example, for polysemous words, in the dominant-subordinate condition, first-pass times on target words were longer than that for homonymous words and a similar pattern was also observed in the disambiguating region. However, these differences did not reach statistical significance in both regions. In addition, we also found that total times on homonymous words were significantly longer than those on polysemous words in dominant-subordinate condition. These differences cannot be easily interpreted as that the processing of the processing of homomymous words was different from the processing of polysemous words. Especially given that the two types of ambiguous words were embedded into two different sentence frames in our current study, thus those differences may be caused by sentence frames. Again, we did not find a three-factor interaction involving word type, thus these “differences” cannot necessarily reflect the processing difference between homonymy and polysemy.
Although semantic relatedness among senses of polysemy is higher than relatedness among the senses of homonymy, these senses are fully specified when a sentence provides abundant context information. For instance, a Chinese polysemous word 创伤 can refer to psychological trauma as well as physical wound. These interpretations refer to two totally different ideas in spite of their semantic relatedness. Therefore, it was difficult for readers to integrate two different senses during sentence reading.
From this aspect, the separate representation account and the underspecified core sense representation account may not be completely incompatible. The critical difference lies in whether the disambiguating information is abundant enough to specify one particular meaning (sense) of ambiguous words. It is highly likely that semantic competition occurs when different meanings (senses) are fully specified and activated by different disambiguating regions in one sentence (as in the current study). However, homonymy and polysemy may be processed differently in a situation in which meanings (senses) are underspecified or only partially specified when the contextual information is not abundant or biased enough. This under-specification of meanings (senses) could explain why different data patterns for homonymy and polysemy were found in previous studies (Frazier and Rayner,
Frazier and Rayner's study found a garden-path effect on homonymy but not on polysemy when the disambiguating context followed target words (Frazier and Rayner,
Moreover, whether a meaning (sense) is fully specified can be modulated by the experimental tasks as well. Previous isolated word-recognition studies found different ambiguity effects for homonymy and polysemy and attributed this difference to inner representational difference between homonymy and polysemy. Difference in tasks employed in these studies might account at least partially for the differences in results. Our primary aim in this study was to explore whether the mental representations of homonymy and polysemy are similar or different when they are completely specified. The significant and stable interaction effect between context and disambiguation for total time in the disambiguation region in the current study provided firm evidence that the processing was facilitated in the consistent condition because readers could successfully integrate the initial selected meaning with the rest of the sentence, while processing was hindered in the inconsistent condition since different meanings were biased by prior context and disambiguating regions. These results provided some evidence that homonymy and polysemy had very similar lexical representations when they were fully specified by abundant context, with each meaning (sense) represented as a separate lexical entry in the mental lexicon. It should be noted that only biased polysemous words were included in our study, it still not clear whether biased and unbiased polysemous words have the same inner lexical representations in the mental lexicon (i.e., separate lexical representation). More studies are needed to further investigate this issue.
In the current study, we examined the representation of ambiguous Chinese words—specifically, those exhibiting homonymy and polysemy. We found a separate lexical representation for both homonymy and polysemy with their meanings (senses) were saliently represented separately in the mental lexicon, and this finding is in line with those of Klein and Murphy (
In summary, we did not find evidence supporting a different lexical representation account for Chinese homonymous and polysemous words. Instead, our data suggest that they may have separate representations in the mental lexicon when they are fully specified by disambiguating information during sentence reading.
Conceived and designed the experiments: WS, XL. Performed the experiments: WS. Analyzed the data: WS, XL. Wrote the paper: WS, XL.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This research was supported by a grant from the Natural Science Foundation of China (31571125), and was supported by Beijing Advanced Innovation Center for Imaging Technology (BAICIT-2016018).
1The term “meanings” here referred to multiple distinct and unrelated lexical entries of homonymy, while the term “senses” referred to multiple semantic related lexical entries of polysemy.
2Noted that this measure is generally termed as “gaze duration” when the region of interest is an individual word. For consistency, we used the term “first-pass time” here.
3This measure calculates regressions to the target-word region and other parts of the sentences.