It is now well-established that lexical processing is influenced by semantic richness or the amount of semantic information a word possesses. A word’s semantic richness may be measured in multiple ways, including imageability, number of features, number of senses, number of associates (NoA), semantic neighborhood density (or number of semantic neighbors), and body–object interaction. A growing body of research has indicated that semantic richness facilitates lexical processing: semantically rich words are recognized more quickly and accurately than words that are more semantically impoverished (e.g., Buchanan et al., 2001; Duñabeitia et al., 2008; Pexman et al., 2008; Yap et al., 2011). These effects are observed across multiple dimensions and tasks (Yap et al., 2011, 2012), indicating that semantic richness is a robust predictor of word processing.

Several studies have also shown processing differences between semantically rich and impoverished words at the neurophysiological level (Kounios et al., 2009; Muller et al., 2010; Amsel, 2011; Laszlo and Federmeier, 2012; Rabovsky et al., 2012; Amoruso et al., 2013; Amsel and Cree, 2013). These studies used event-related potentials (ERPs), a technique with temporal resolution on the order of milliseconds, which is ideal for studying online language processing. These studies have focused primarily on the N400, a negative-going ERP component peaking at approximately 400 ms post-stimulus onset that is consistently modulated by semantic processing (Kutas and Federmeier, 2011).

Kounios et al. (2009) and Amsel and Cree (2013) found larger N400 amplitudes for words with a low number of features than for words with a high number of features, suggesting greater ease of processing of words with a high number of features. However, other studies investigating number of features have found the opposite effect (Amsel, 2011; Rabovsky et al., 2012). ERP studies that investigated the variable NoA have found larger N400 amplitudes for words with high than low NoA (Muller et al., 2010; Laszlo and Federmeier, 2012). The discrepancy in findings may be due to a number of factors including type of language task (lexical decision vs. semantic categorization), type of analyses (general linear model statistics vs. mixed-effects models), and the semantic richness variable under study (e.g., the effects of number of features may differ from the effects of NoA because they measure different aspects of the semantic information associated with a word).

To our knowledge, only one ERP study has investigated number of senses (Taler et al., 2013). Consistent with previous behavioral research, findings revealed faster reaction times for words with high than low number of senses during a lexical decision task in monolingual young adults. In addition, words with many senses exhibited smaller N400 amplitudes relative to words with few senses.

The majority of research into the effects of semantic richness on lexical processing has focused either on monolingual speakers or has not specified the language background of study participants. A large body of research, however, indicates differences in language function between monolinguals and bilinguals. Monolinguals consistently outperform bilinguals in language tasks (for a review, see Bialystok, 2009). Semantic organization has also been shown to differ between monolinguals and bilinguals. Antón-Méndez and Gollan (2010) found differences in associative effects between bilinguals and monolinguals: in a word association task, bilinguals’ responses were less typical than those of monolinguals (for example, in response to the target “bride,” bilinguals produced the item “dress” while monolinguals were more likely to produce “groom”). Bilinguals’ performance was similar to that of monolinguals when the associate was high frequency but not when the associate was low frequency. That is, bilinguals were more affected by frequency information than monolinguals. Similarly, Johns et al. (2016) demonstrate that semantic diversity information—that is, the number of contexts in which a word occurs—exerts a stronger effect on lexical decision performance in bilinguals than monolinguals.

In the current study, we aimed to extend previous research on semantic richness effects to a bilingual population, an area that remains unexplored in the literature. We used behavioral and ERP measures to examine sense relatedness effects in highly proficient bilingual speakers of English and French, and compare effects to those we previously reported for English-speaking monolinguals (Taler et al., 2013). Bilinguals completed testing in both English and French, allowing us to compare semantic richness effects in each language, as well as to compare bilinguals’ performance in first and second languages (L1 and L2). Because bilinguals have reduced exposure to each language, it is likely that the semantic knowledge possessed by bilingual speakers is more impoverished in each of their languages relative to monolingual speakers, as has been shown in children (Cremer and Schoonen, 2013). We thus hypothesized that semantic richness effects would be weaker in bilingual than monolingual speakers. We also predicted stronger effects of semantic richness in bilinguals’ L1 than in their L2.

Results were analyzed as follows: first, we ran a series of analyses comparing monolinguals to bilinguals in the English experiment. We then compared monolinguals to bilinguals in their first language (L1). Bilinguals’ performance in the English and French experiments was then compared. A final series of analyses was conducted comparing bilinguals in their L1 and L2. All analyses were conducted using SPSS Version 22. Results of each analysis are provided below.

The present study examined the effect of one measure of semantic richness—the number of senses (NoS) possessed by a word—on behavioral and ERP responses in English-speaking monolinguals and highly proficient bilingual speakers of English and French. Participants completed a lexical decision task with high- and low-NoS items in English (all participants) and French (bilingual participants only). We compared monolingual performance to bilingual performance first in English, and then in the first language (English for 10 participants and French for eight participants). Bilingual performance was then compared in the two languages (English vs. French and L1 vs. L2).

Overall, shorter RTs and higher accuracy were observed for high- than low-NoS items in the English task across all participants. Similarly, a larger amplitude N400 was revealed for low- relative to high-NoS items in right lateral sites; this difference was larger in monolinguals than bilinguals. When bilinguals’ L1 performance was compared to monolingual performance, shorter RTs and higher accuracy were again observed for high- relative to low-NoS items, and low-NoS items elicited greater N400 amplitude compared to high-NoS items. RT effects were stronger in monolinguals than bilinguals even when bilingual performance in the L1 was used in the analyses.

These findings indicate that both monolinguals and bilinguals exhibit semantic richness effects in a lexical decision task, as predicted; these effects are observed in both the English and the French tasks. However, the high-NoS advantage was more robust in monolinguals than bilinguals, as predicted: in the English-only analysis, larger differences in the ERP response were observed between conditions in monolinguals relative to bilinguals, and in the L1-only analysis, RT effects of NoS were stronger in the monolingual group. While accuracy effects were stronger in the bilingual group, this finding is likely due to ceiling effects in the monolingual group (accuracy exceeded 97% in both conditions in this group).

The results thus confirm our hypothesis that semantic richness effects would be observed in both monolinguals and bilinguals, but that they would be stronger in the former than the latter group. These findings are consistent with previous research indicating differences in semantic function in monolinguals and bilinguals in children (Cremer and Schoonen, 2013). Specifically, we argue that semantic representations may be more impoverished in bilinguals relative to monolinguals, possibly due to less language experience in each of their languages. This account is consistent with the frequency lag hypothesis (Gollan et al., 2008), which holds that bilingual disadvantages in language processing result from lower experience with words in one language relative to monolinguals. Because any given word form thus has lower relative frequency for bilinguals than monolinguals, lexical retrieval is slower and less accurate. We propose that semantic representations are also less elaborated in bilinguals than monolinguals as a result of this reduced language experience. Specifically in the present case, bilinguals may have been exposed to fewer senses of a polysemous lexical item, leading to reduced effects of number of senses.

We also examined bilinguals’ performance in each of their languages. Behaviorally, bilinguals showed shorter RTs and higher accuracy in English than French. As predicted, high NoS words were recognized more quickly and accurately than low NoS words. N400 effects were also consistent: low NoS words elicited larger N400 amplitudes than high NoS words. Interestingly, this effect was stronger in French than English even though behavioral performance was superior in English. In order to tease apart the language effects, we then analyzed bilinguals’ performance in their first and second languages. Overall, bilinguals’ behavioral performance was stronger in the first than the second language, and the predicted NoS effects were observed overall in both behavioral and ERP measures, albeit slightly more strongly in the second language.

These results suggest a complex picture of language dominance in this highly proficient bilingual sample, with stronger behavioral performance in English in the context of stronger ERP effects in French. Assessing language dominance is complex because language proficiency is multifactorial (Sheng et al., 2014); the present findings suggest that neural and behavioral performance on a semantic task may reveal different patterns of dominance. We suggest that the N400 NoS effect, which was stronger in French than English, reflects superior semantic knowledge in French, while superior behavioral performance in English may reflect stronger lexical- or decisional-level performance in English. It is particularly interesting that NoS effects are stronger overall in the second language; this finding indicates that the native language is not necessarily the language in which semantic representations are most elaborated. It is of course possible that the differing performance in English and French may be due to differences in the experiments themselves. For example, the overall degree of semantic richness may differ between English and French. This confound is unavoidable, because stimulus lists must differ across the two languages. However, the fact that neural and behavioral responses were dissociated in the two languages suggests that the differing performance is not due purely to differences in test materials. Future research should examine these effects in other language pairs and in speakers of differing degrees of dominance in order to assess the generalizability of the current findings to other bilingual populations.

In sum, as predicted, we found effects of NoS across both participant groups, and these effects were stronger in monolinguals than in bilinguals. This finding suggests that semantic representations are more elaborated in monolinguals than bilinguals, consistent with previous research on frequency effects (Gollan et al., 2008). Moreover, we found dissociations across languages in bilinguals, with stronger behavioral NoS effects in English and stronger ERP NoS effects in French. This finding points to the dissociation in different processing stages in the two languages, and suggests that different aspects of linguistic performance may be stronger in each of a bilingual’s two languages.

VT: Intellectual conceptualization, analyses, and manuscript drafting. RL: Intellectual contribution, data collection, analyses, and manuscript drafting. SK: Intellectual contribution, analyses, and manuscript drafting.

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.