Event-related potential correlates of language change detection in bilingual toddlers

Highlights

* Bilingual toddlers detect a language change faster than their monolingual peers. * Unlike adult bilinguals, toddlers do not show signs of conscious stimulus revaluation mechanisms. * Language awareness and monitoring appears to develop after the age of three.

Abstract

Children raised in a bilingual environment are faced with the daunting task of learning to extract meaning from language input that can differ between caregivers but, depending on the social context, also within caregivers. Here, we investigated monolingual and bilingual toddlers’ brain responses to an unexpected language change. We presented 2–3 year old children with picture–word pairs and occasionally changed the language of the spoken word while recording event-related potentials (ERPs). In line with previous results obtained in adults, bilingual children differentiated between the languages of input faster than their monolingual peers, i.e., within 200 ms of spoken word onset, a time range previously associated with lexical access. However, while adult bilinguals displayed a late stimulus re-evaluation ERP response to a language change, no such modulation was found in bilingual toddlers. These results suggest that although bilingual individuals are sensitive to phonemic language cues already from an early age, language awareness and language monitoring mechanisms probably develop later in life.

1. Introduction

Learning to extract meaning from interconnected speech in infancy has been proposed to involve extracting statistical regularities in syllable or word boundaries (Mehler et al., 1988, Saffran et al., 1996). This learning process is for an important part guided by the prosody of the language, as observed in sleeping neonates (Shukla et al., 2011) but also in adults (Endress and Hauser, 2010), and is guided by the co-occurring visual information, for example, the face of the individual producing language (Teinonen et al., 2008).

One interesting question is whether learning to segment speech is affected when the input is rather variable as is the case when the child is brought up in a bilingual environment. Bilingual speech contains context-dependent differences in language usage (e.g., from different speakers and environment), but also frequent, unexpected language switches (Poulisse and Bongaerts, 1994, Myers-Scotton, 2005). Despite a particularly variable language input, bilingual children do not appear to differ from their monolingual peers regarding their total vocabulary size (Pearson et al., 1993, Paradis, 2010) and the age at which they reach critical milestones of language development (Pearson and Fernández, 1994, Petitto et al., 2001, Werker et al., 2009). Moreover, like monolingually raised infants, they display language experience dependent preference for each of their languages and discriminate them from non-native languages (Byers-Heinlein et al., 2010). Therefore, children in a bilingual environment may have different strategies in the way they process spoken language leading to efficient discrimination of the languages they are exposed to.

In the present study, we addressed this issue by testing whether toddlers raised bilingually perceive an unexpected language change differently than their monolingual peers. Indeed, 17-month-old bilingual infants are more flexible in accepting variation in pronunciation of familiar words than their monolingual peers (Mattock et al., 2010), and early exposure to a second language maintains or activates sensitivity to non-native phonetic contrasts (Conboy and Kuhl, 2011, Kuhl et al., 2003). Furthermore, being raised bilingually may affect other domains of (language) processing since 12-month-old bilingually raised infants are able to learn two new grammatical structures at once while their monolingual peers are not (Kovacs and Mehler, 2009).

Language switches may be particularly informative about bilingual processing because language characteristics must be noticed quickly to ensure comprehension in the wider context. Since language switches occur frequently and unexpectedly in the speech output of bilinguals (Poulisse and Bongaerts, 1994, Myers-Scotton, 2005) and given that phonotactic rules and word boundaries may differ between languages, a listener must be able to quickly adapt to a change in language. Consistent with this view, we recently showed using ERPs that bilingual adults detect a change in language faster than their monolingual peers (Kuipers and Thierry, 2010). ERPs are average electrical brain potentials recorded from the surface of the scalp that provide insight into human brain function with high temporal resolution. In our previous study involving Welsh–English bilingual adults, we observed a significant increase in ERP amplitudes approximately 200 ms after stimulus onset (P2 amplitude increase) when an unexpected language change had occurred, e.g., when Welsh words were occasionally heard within a stream of English words (Kuipers and Thierry, 2010). By comparison, monolingual participants distinguished between English and Welsh words with a stronger negativity from 350 to 450 ms after word onset, which is the time window traditionally associated with semantic integration (N400; Federmeier and Kutas, 2002). In addition, bilingual participants also displayed a P600 modulation, which is a positive peak arising 600 ms after stimulus onset indexing stimulus re-evaluation and memory updating (Osterhout and Holcomb, 1992). This suggests that only bilinguals re-evaluate verbal input when they encounter a language change.

To investigate whether language (change) detection and re-evaluation mechanisms are already in place at an early age, we tested toddlers using the same stimulus materials and a similar procedure as that used in adults (Kuipers and Thierry, 2010). In each trial, a picture was presented on a screen shortly followed by a spoken word that either matched the picture or was unrelated. The picture–word pairs were presented within an oddball paradigm with English as the frequent language (standard, 75%) and Welsh as the infrequent language (deviant, 25%).

Previous work on bilingual language processing in children has shown that, depending on the extent of exposure to each of their languages, bilingually raised toddlers display different brain signals for the two languages (Conboy and Mills, 2006). Furthermore, from 5 year old to early adulthood, bilinguals seem to show an increase in their ability to resolve conflict between languages (Kohnert et al., 1999, Kohnert and Bates, 2002). In the current study, we went further by testing a control group with the same materials, using an unexpected language switch, and using words presented in the context of pictures.

Given the well-established phonological discrimination skills of toddlers (Kuhl et al., 1992) and the observation of a more positive ERP response to unknown than known words in 13 to 20 month old monolingual infants (Mills et al., 1993, Mills et al., 1997), we expected an ERP modulation by language change around 200 ms after stimulus onset. However, since we previously observed that bilingual, but not monolingual adults showed a P2 modulation by language, we expected an earlier or stronger ERP modulation in the bilingual toddlers. Any group differences in the latency of discrimination between languages would be informative about the groups’ sensitivity to a change in language.

Although an oddball paradigm as we used here usually results in a mismatch negativity (MMN; Naatanen et al., 2007) reflecting detection of sound contrasts, here we did not expect to observe an MMN. The type of stimuli we used here (spoken words) had a far greater variability than the short sound stimuli usually used in MMN studies (but see Thierry et al., 2003, Vihman et al., 2007) and the number of trials available was far less than usually required in an MMN paradigm. In addition, we did not observe an MMN in our previous study (Kuipers and Thierry, 2010).

Regarding the ability of toddlers to re-evaluate the consequences of a language change, we expected a P600-like modulation as found in adults, because 2- and 3-year-old toddlers have been shown to display adult-like P600 amplitude modulations upon hearing a syntactic violation (Oberecker et al., 2005, Silva-Pereyra et al., 2007). The amplitude of the P600 in the two groups would be informative about the level of conscious awareness regarding the occurrence of a language change.

The manipulation of the semantic match between pictures and words was expected to provide insights into possible differences between the two groups in terms of semantic integration mechanisms. However, there were not enough artefact free epochs in the match and mismatch conditions considered separately to allow a comparison of this factor between groups. This was also the case in the previously reported adult study (Kuipers and Thierry, 2010), and similarly, we only report the language change results.

To summarise, the timing of an ERP modulation by language would be informative on the sensitivity to a change in language in each group, and the presence of a P600-type modulation would indicate the level of conscious awareness of the language change.

2. Materials and methods

2.1. Participants

Thirty-three monolingual English and 28 Welsh–English toddlers participated in the study that was approved by the Ethics Committee of Bangor University. Participants received a present and their caregiver received £15 towards travel expenses. Eighteen participants of each group had a sufficient number of artefact free trials (>20 per condition) to be included in the analyses. The parents filled in the British English version of the McArthur-Bates Communicative Development Inventory (CDI; Hamilton et al., 2000) within a week of the testing session. Three parents of the monolingual children and 4 parents of the bilingual children failed to return the completed CDI. The monolingual English children (9 females; mean age 31.3 ± 2.8 months) had an average score of 91% CDI score. The bilingual children (10 females; mean age 28.6 ± 2.8 months), scored 78% on the English CDI and 75% on a Welsh adaptation of the CDI. The bilingual children's CDI scores in each of the two languages was within 20–80% of their total CDI score (Pearson et al., 1993) and parents reported that their child had equal to near-equal exposure to Welsh and English. The total vocabulary score (Pearson et al., 1993) for the English children was 279 ± 15 and for the bilingual children (English and Welsh words combined – cognates counted once) was 354 ± 76. Neither the age difference between groups was not significant (p > .2), nor was the numerical difference in known words between the groups (p > .07). The English monolingual children had had no noteworthy exposure to Welsh at the time of testing.

2.2. Stimuli

Thirty-six pictures of highly familiar objects were selected from an in-house picture database. Basic-level English and Welsh picture names were recorded by a native Welsh–English bilingual woman without apparent accent in either of the two languages. In addition, the English picture names were also recorded by two native English women. The sound files were resampled at 44.1 kHz, 16 bit encoding, mono. Average duration was 658 ms ± 0.03 for English and 847 ms ± 0.03 for Welsh words and the word lists did not differ in mean amplitude (p > .06). Mean familiarity and concreteness ratings for English stimuli were: 577 ± 12.0 and 605 ± 5.8, respectively, on a scale from 100 to 700, the mean Kucera–Francis frequency (Kucera and Francis, 1967) was 99 ± 41.4, the mean age of acquisition was 1.95 years ± 0.1, and the mean number of phonemes was 3.1 ± 0.3 (Coltheart, 1981). The Welsh words were translation equivalents of the English words and, as such, they were assumed matched for familiarity, concreteness, frequency, and age of acquisition in the absence of a Welsh corpus (but see Pena, 2007).

2.3. Design and procedure

The children were seated on their caregiver's lap about 1.8 m from a screen on which a wall-mounted projector displayed the picture stimuli. Pictures spanned a maximum of 9° of visual angle. Each trial consisted of the presentation of a picture followed 500 ms later by a spoken word played via loudspeakers set in front of the participant at an intensity of 60–68 dB. The picture remained on the screen for 2 s, that is, more than the duration of the word in all cases followed by an 800 ms inter stimulus interval (blank screen; Fig. 1). In the match condition, pictures were paired with their English or Welsh name. In the unrelated condition, pictures and words were re-paired avoiding semantic, phonological and orthographic overlap between the picture name and the spoken word. Thus, each picture was presented 8 times paired with a spoken word in English or Welsh that either matched 50% or mismatched 50% the identity of the picture. The stimuli were presented in an oddball paradigm with 3, 4 or 5 standards (English words, 75%) before each deviant (Welsh word, 25%). The ongoing presentation of the stimuli was interrupted by the experimenter if the child looked away from the screen, in which case a television character and a sound were presented to orient the child's attention back to the screen.

Fig. 1.

Trial procedure of the adapted oddball paradigm.

2.4. Data acquisition

Event-related potentials were continuously sampled at 1 KHz and band-pass filtered between 0.1 and 200 Hz from 22 Ag/AgCl electrodes placed according to the 10–20 convention (Fp1, Fp2, F7, F3, Fz, F4, F8, T7, C3, Cz, C4, T8, P7, P3, Pz, P4, P8, O1, OZ, O2, right mastoid) referenced to the left mastoid. Impedances were kept below 10 kΩ. Off-line EEG recordings were band-pass filtered between 0.3 and 20 Hz using a zero phase shift filter and re-referenced to the average of left and right mastoids. Due to data loss at peripheral electrode sites (due to e.g., the child pulling on electrode leads and/or resting of the head against the caregiver), we selected 9 electrodes for the statistical analysis (F3, Fz, F4, C3, Cz, C4, P3, Pz, P4), which cover the same region as in the analysis of the previously tested adults (Kuipers and Thierry, 2010). Artefacts were removed using a ±30 μV artefact rejection procedure applied to the reference channel. The outcome of the artefact rejection procedure was visually inspected and remaining artefacts (which could include eye blinks) were manually removed. In the latter manual procedure, FP1 was used for the monitoring of eye blinks. For the analysis on the timing of language change detection, EEG waveforms were averaged from epochs of −100 to 700 ms relative to the onset of the spoken word and baseline corrected in reference to pre-stimulus activity. To investigate possible late ERP modulations (up to 1 s), we also analysed mean amplitude differences from 700 ms to 1 s after word onset. This long post-stimulus epoch was not used for the analysis of earlier (<600 ms post stimulus onset) mean amplitude modulations, because with increasing epoch length, the number of artefacts free trials reduces resulting in a loss of statistical power. We prevented a difference in the number of sweeps between the standard and deviant conditions by using only those standard trials (English) that immediately preceded deviant trials (Welsh). The percentage of words spoken by the monolingual women included in the analysis was 33.8% and 32.8% for the bilingual and monolingual groups respectively. The average number of sweeps in the Welsh–English bilingual group was 28 ± 4.9 for the standards, 29 ± 4.4 Welsh words and for the English monolingual group: 36 ± 6.1 English words and 36 ± 5.4 Welsh words.

2.5. Statistical analysis

To determine the time point at which the ERP response to English words started to differ from Welsh words we performed t-tests on mean amplitudes of 50 ms long bins at every millisecond for each group. Next, we performed correlation analyses in the two groups to investigate when, relative to the onset of the target stimulus, the two groups displayed similar topographies.

3. Results

The distribution of the difference between the languages was mostly left-to-right while differences from anterior to posterior were very small. Hence, we averaged over left, middle and right electrodes to achieve maximal power to detect reliable differences in the language effect (Fig. 2). The ms-by-ms t-tests on mean amplitude bins revealed that the difference between languages was indeed strongly lateralised to the left hemisphere, and that the bilinguals showed an earlier distinction between English and Welsh words. For the bilinguals, mean amplitude differences were significant in the left hemisphere only from 50 ms to 131 ms (31 consecutive significant t-tests), corresponding to the P1, and from 190 to 301 ms (61 consecutive significant t-tests) with a maximum effect between 227 and 277 ms, corresponding roughly to the time window of the P2. The monolinguals did not distinguish between languages until 293–419 ms (67 consecutive significant t-tests) after word onset in the left hemisphere only with a maximal effect at 317–367 ms. In the epoch extending from 700 ms to 1 s after stimulus onset (not displayed in Fig. 2) differences in mean amplitudes were not significant (a more negative waveform for Welsh words than English words displayed by the bilinguals only, from 1060 to 1300 ms; p > .06 at the electrode of maximum sensitivity, Pz).

Fig. 2.

ERP recordings for English and Welsh words. Average ERP amplitudes for left (F3, C3, P3), middle (Fz, Cz, Pz) and right (F4, C4, P4) electrodes in the monolingual English children (top) and the Welsh–English bilingual children (bottom). Shaded bars indicate significant modulations by language.

To test whether the distinction between languages in bilinguals may have been due to a similar cognitive process as the later distinction in monolinguals, we analysed the topography of the language effect (individual ERP waveforms for English subtracted from the Welsh) in both groups. We performed ms-by-ms correlation analyses on ERP amplitude of the language effect at each channel. We made 3 comparisons based on the observation that the maximal language effect in the monolinguals was about 100 ms later than in the bilinguals. We correlated the language effect of the bilinguals with the monolinguals in an early time-window (200–300 ms), a late time window (300–400 ms), and across these time windows (200–300 ms in the bilinguals with 300–400 ms in the monolinguals (Fig. 3). The correlation of the topography across time windows was in general higher than within the early and late time windows and significant (one-sided tests) from 231 to 352 ms. This means that the topography of the early language effect in bilinguals from 231 to 352 ms corresponded well to the language effect observed 100 ms later (331–352 ms) in monolinguals. The only time at which the topography of the language effect was similar across groups when using the same temporal reference was at the end of the late (300–400 ms) time window.

Fig. 3.

Correlation coefficients of the topography of the language effect in both groups. (a) Plotted are the ms-by-ms Pearson correlation values of the topographies in an early time window (200–300 ms after word onset), a late time window (300–400 ms after word onset), and across time windows (200–300 ms in bilinguals vs. 300–400 ms in monolinguals). (b) Significance level of the correlations.

4. Discussion

We addressed the issue of whether the brain response to an unexpected change in language differs between bilingual and monolingual children. We have previously shown that adults bilinguals show faster language switch detection than their monolingual peers (Kuipers and Thierry, 2010). These adult bilinguals also showed ERP responses to Welsh words relating to monitoring and memory updating, which suggests that they were consciously aware of the language change.

Here, the children in both language groups tested displayed a larger positivity in their ERP response to Welsh words than to English words, which indicates that both groups perceived Welsh words differently to English words. However, the bilingual's ERP response distinguished between languages well within the first 200 ms after word onset, whereas the monolinguals’ response became significant only around 300 ms. Hence, the bilingual brain of children differentiates between the languages of input faster than the monolingual one and this resembles the pattern found in adults. In other words, individuals brought up bilingually have a fast language distinction capacity from an early age which persists into adulthood. The likely function of such a language change-detection mechanism probably relates to the safeguarding/optimisation of speech comprehension since different languages have different phonology-to-meaning mappings (with the exception of cognates1 ).

The similarity in topography of the maximum language effects in both groups suggests that language change detection may have originated from similar neural populations in the two groups, albeit 100 ms later in monolinguals than in bilinguals. Since the time-window around 200 ms after stimulus onset is associated with lexical access (van den Brink et al., 2001, Strijkers et al., 2009), the language effect in bilinguals suggests that activation of lexical representations in a lexicon not-in-use may underlie the P2-like effect observed in bilinguals. We also observed a significant correlation in topography between groups towards 400 ms which most likely relates to the onset of semantic integration (Federmeier and Kutas, 2002). Hence, both groups may rely on similar neural networks at the stage of semantic integration.

The language effect in the monolinguals on the other hand, seemed to resemble the word familiarity effect. Unfamiliar words have previously been shown to induce more positive, left lateralised ERP responses in monolingual children from approximately 200 to 500 ms after stimulus onset (Mills et al., 1997, Thierry et al., 2003, Vihman et al., 2007). Hence, we replicated the familiarity effect in both timing and lateralisation in the monolingual children. Note that in monolingual adults, the word familiarity effect is no longer observed and instead, pronounceable non-words elicited a N400 response (Chwilla et al., 1995), which is what we found in our study with adults (Kuipers and Thierry, 2010).

The increased P600 amplitude in response to unexpected Welsh words observed in bilingual adults was not found in either of the toddler groups. This suggests that the neither group consciously re-evaluated the auditory input after an unexpected language change. Hence, although bilingual toddlers perceive a language switch quickly after word onset, they seem less aware of the change in language than adult bilinguals. Instead, the bilingual toddlers may have a pragmatic approach to processing language, that is, they may devote relatively little attention to the language spoken at any one time provided they can access semantics.

5. Conclusion

In line with previous results in adults, bilingual toddlers display language change detection mechanisms absent in monolinguals and operating within the first 300 ms after a language change has occurred. Thus, it seems important for speech perception that the language spoken is quickly identified enabling the correct mapping from sound to meaning. Furthermore, late re-evaluation/working memory updating mechanisms previously observed in adult bilinguals and seemingly absent in bilingual toddlers suggest that processes of language monitoring and meta-linguistic awareness develop later in life. Therefore, children may be less aware than adults of the language spoken at any one time even though they quickly and automatically adapt to a change in language-specific phonological input.