Semantic Representations in Bilingual Speakers


Previous studies have shown that concreteness, valence, and arousal are factors that can influence the semantic representations in monolingual speakers, but the factors were confounded and it's still unclear if the factors also apply to bilingual speakers. The Modified Hierarchical Model by Pavlenko (2009) suggests that L2 proficiency might also play a role, but L2 proficiency were not well measured in previous studies. The studies on Chinese speakers are also limited.

Research Questions / Hypotheses

1, Question: whether or not there is a difference in the semantic representations used by bilingual speakers depending on whether they are doing the task in Chinese (L1) or English (L2).    Hypothesis: There will be differences in semantic representations between speakers who perform the task in different languages.    2, Question: whether or not these differences in semantic representations can be predicted by the lexico-semantic factors (concreteness, valence, and arousal).    Hypothesis: Lower concreteness, higher arousal, and higher absolute values of mean-centered valence (i.e., either very negative or positive, more emotional) will be associated with more distinct semantic representations between languages.    3, Question: whether or not differences in semantic representations can be predicted by L2 proficiency.    Hypothesis: Bilingual speakers with lower L2 proficiency will have more distinct semantic representations from native English speakers than bilinguals with higher L2 proficiency.


65 native English speakers and 52 Chinese-English bilinguals completed the study.    17 participants were excluded because their correlation was less than 1.5SD of the mean correlation. 9 participants were removed because their first language was not Mandarin Chinese or English. 2 bilinguals were removed because they used translation software    during the relatedness judgment task.


The first task was a relatedness judgment task. Participants saw a cue word and were asked to select three strongest related words out of 35 alternatives. Chinese-English bilinguals did the task in either Chinese or English and native English speakers did the task in English.    After that, Participants did a Lextale task to measure their L2 proficiency and completed the LEAP-Q.


The rated similarity between words within each language was analyzed using multidimensional scaling. We found that there would be differences in semantic representations between speakers who perform the task in different languages.    A regression model found that concreteness was a stable predictor of the semantic representations in the predicted direction. Valence was also a stable predictor but in the reverse direction. Arousal was not a significant predictor.    A multilevel regression model found no effect of L2 proficiency, but a quantile split found a small effect of L2 proficiency in the predicted direction from the extremes of our dataset.


The results of MDS are consistent with previous studies.    The effect of concreteness is consistent and stable, whereas the effect of valence contradicts previous studies, which suggests the role of valence might be more complex than what researchers thought before. We found no predictive effect of arousal, suggesting that arousal is not an influential factor in semantic representations.    Bilinguals with higher L2 proficiency tend to construct more similar semantic representations to native English speakers, but the effect might be small.    The current study introduces a large list of stimulus words and two distinct languages, Chinese and English, which make our findings more generalizable. The results also provide evidence for the asymmetry in the MHM model.