Input matters: Multi-accent language exposure affects word form recognition in infancy

To become mature communicators, infants must learn to recognize words across contexts. This can be particularly difficult when confronted with speakers of unfamiliar accents. Indeed, laboratory studies suggest that infants initially struggle in such situations.^1–3 What does that mean for children who are routinely exposed to multiple variants of their native language? Imagine, for instance, a child born to an American English-speaking mother and an Irish English-speaking father. While her mother may label the yellow vehicle that takes her big brother to school as a bus, her father's pronunciation will sound more like boss, leaving the child to deduce that mother's bus and father's boss refer to the same object even though her mother's pronunciation of bus and boss label two separate referents. How does language development in this child differ from that in a child growing up in a family where both parents speak in the same accent?

Unfortunately, we currently have very little data to assess this question. Most developmental word recognition studies to date have focused on linguistically homogeneous populations, and those studies testing more heterogeneous populations typically do not consider variation as a predictive factor. The few existing studies examining the impact of accent variation on early word recognition have tested children around their second birthday.^4,5 By this age, however, children's vocabularies are rapidly expanding and phonological constancy (presumably a prerequisite to coping with accents) already appears to be present.¹ Hence, the greatest impact of accent variability is likely to be observed earlier in infancy. It is thus crucial to evaluate younger infants' word form recognition abilities as a function of their exposure to different accents.

Although theories of early speech perception differ in their perspectives regarding the mechanics underlying word recognition, all predict that variability in the input should affect early word form recognition.⁶ That is, whether due to more fine-grained word representations^7,8 or to more sophisticated signal-to-word mapping skills,² exposure to greater variation in word forms should enhance children's ability to recognize novel word tokens.^9,10 At the same time, however, greater distances between word tokens might lead to greater difficulty extracting commonalities. As a result, infants may require exposure to more tokens to start treating them as the same word. By extrapolation, then, extensive exposure to variability early in life might have a dramatic effect on the developmental trajectory of children's word recognition capabilities.

This Headturn Preference study tests the impact of accent variation on 12.5-month-olds' word form recognition. Infants listen to known and nonsense words spoken in the local accent. Past work using this design has revealed preferences for known over nonsense words.^1,2,11,12 Here, we test monolingual children who either consistently receive Canadian English language input (low variability group) or a mixture of Canadian English and a second accent (high variability group). We predict that the amount of accent variability an infant experiences at home will impact the developmental trajectory of word form recognition in the laboratory.

Two groups of typically developing monolingual English-learning 12.5-month-olds were recruited from the Greater Toronto Area, all receiving at least 90% of their language input in English. Infants in the low variability group (N = 20; age range: 373–395 days; mean age: 383 days; 9 boys) were almost exclusively exposed to the dominant regional variant (on average ∼96% Canadian-accented English), whereas infants in the high variability group (N = 20; age range: 363–403 days; mean age: 383 days; 12 boys) were exposed to multiple variants of English (on average only ∼33% Canadian-accented English, ∼65% other variants of English). The additional variant, spoken by either their parent(s) or a caregiver with whom they spent at least 32 h a week, could either be a native (e.g., Irish) or a foreign (e.g., Polish) accent. Importantly, however, the total input of English did not differ between the two groups (99% vs 98%, respectively), as established by a detailed language questionnaire. Groups were further matched on maternal education level (as a proxy for socioeconomic status) as well as reported vocabulary size (see Table 1).¹³ An additional three infants were tested, but excluded from the analysis due to experimenter error (1) or fussiness (2). All infants received a small toy as a thank you gift.

Eight word lists (described elsewhere² in detail) were used, four containing words typically known by 12.5-month-olds (daddy, bottle, diaper, mommy, grandma, kitty, ball, dog, bath, kiss, cup, shoe), and four containing unknown words (koddy, dimma, dapper, mitty, guttle, shammy, bog, bap, deuce, kie, koth, brall). Words were recorded by a female Canadian English speaker who had exclusively lived in the Greater Toronto Area. Word types were closely matched for average word length (known words: 559 ms; nonsense words: 579 ms) and average pitch (known words: 358 Hz; nonsense words: 378 Hz). Within each list, all words were presented twice. Word order differed between lists. All lists lasted 34.5 s.

Infants were tested in the Headturn Preference Procedure. Children sat on their parents' lap in a three-sided area inside a dimly-lit double-walled IAC booth. Attached to the pegboard panels in front and to the side were three lights at eye level. Loudspeakers were positioned behind the side lights, out of view from the infants. A camera just above the center light allowed the experimenter to monitor the infant's behavior from outside the booth. The center light flashed at experiment onset. Once the infant oriented towards this light, the experimenter pressed a button ceasing the light, and one of the side lights was triggered. A head turn toward the flashing light initiated a word list, which played until the infant looked away for two seconds or until the maximum trial length of 34.5 s was reached. Infants listened to all eight lists, presented randomly with the restriction that lists of the same type could not occur more than twice in a row. Presentation side was randomized and not associated with list type. To limit parental influence, parents wore closed headphones playing masking music. The experiment lasted 2–3 min.

Mean orientation time to lists of known and nonsense words was computed (see Fig. 1). As accent type (native vs foreign) did not yield any main effects or interactions in Experiment 1 or 2, this factor was excluded from further analyses. A 2 × 2 analysis of variance (ANOVA) with word status (known vs nonsense words) as a within-participant factor and accent exposure group (low vs high variability) as a between-participant factor revealed a main effect of word status [F(1,38) = 4.971; p = 0.032; $ηp2$ = 0.116], which was modulated by an interaction between word status and exposure group [F(1,38) = 4.492; p = 0.041; $ηp2$ = 0.106]. Two-tailed dependent samples t tests were conducted to further examine this interaction. The effect of word status reached significance for the low-variability group [t(19) = 3.174, p = 0.005, Cohen's d = 0.710, mean difference = 4.38 s, 95% CI (1.49, 7.28)], but not for the high-variability group [t(19) = 0.76, p = 0.941, mean difference = 0.11 s, 95% CI (–2.97, 3.18)]. That is, only infants with exposure predominantly to Canadian English listened longer to known than to nonsense words.

Language input outside the lab thus clearly affects infants' recognition of familiar word forms in the laboratory. Given that multi-accent input is so common, claims that children recognize regionally accented words in this test paradigm by the end of their first year of life may only be true for a subset of language learners (i.e., those with relatively uniform input). Experiment 2 examines when multi-accent infants start to recognize word forms spoken by a Canadian-accented speaker.

An additional two groups of typically developing English-learning children were tested. All met the same language criteria as the high variability group in experiment 1. The younger group were 14.5-month-olds (N = 20; age range: 431–459 days; mean age: 447 days; 12 boys), and the older group were 18-month-olds (N = 20; age range: 530–566 days; mean age: 550 days; 11 boys). An additional eight infants were tested, but excluded from the analysis due to parental interference (1), experimenter error (1), or fussiness (6). All infants received a small toy as a thank you gift.

The materials, design, procedure, and questionnaires were identical to Experiment 1. To best assess when infants with mixed accent input learn to recognize word forms in the Canadian accent, orientation times to known and nonsense word lists from all age groups (including the 12.5-month-olds from Experiment 1) were analyzed. A repeated measures ANCOVA with word status (known vs nonsense word) as a within-participant factor and age (in days) as a covariate was conducted on these data. This revealed a marginally significant main effect of word status [F(1,58) = 3.281; p = 0.075; $ηp2$ = 0.054], as well as a main effect of age [F(1,58) = 7.671; p = 0.008; $ηp2$ = 0.117], but more importantly, an interaction between word status and age [F(1,58) = 5.159; p = 0.027; $ηp2$ = 0.082], suggesting that infants' looking preference changed over time (see Fig. 1). Two-tailed dependent samples t-tests analyses conducted for each of the older age groups revealed that only the older infants recognized the words [t(19) = 2.378; p = 0.028; Cohen's d = 0.532; mean difference = 4.80 s, 95% CI (0.57, 9.03) for the 18-month-olds and t(19) = 0.178; p = 0.861; mean difference = 0.25 s, 95% CI (−2.73, 3.23) for the 14.5-month-olds]. Thus, not until around 18 months of age do infants exposed to substantial accent variation in their language input begin recognizing the locally dominant versions of word forms in this test paradigm. This aligns with findings from laboratory studies showing that 18-month-old (but not younger) infants' word recognition benefits from experience with multiple accents in the laboratory prior to test.¹⁴ 

To examine what drives this development, a standard multiple linear regression analysis was carried out to predict the difference in orientation time between known and nonsense words based on age (in days), log-transformed receptive vocabulary scores, the proportion of exposure to Canadian English, and maternal education level. Although age and log-transformed vocabulary size both correlated with this difference score [r(58) = 0.286; p = 0.027 and r(58) = 0.438; p < 0.001, respectively], only the log-transformed vocabulary scores added significantly to the model [F(4,57) = 4.335; p = 0.004; R²= 0.25]. Thus, vocabulary score (β = 0.374; p = 0.005) was a better predictor than age (β = 0.143; p = 0.282), proportion of exposure to Canadian English (β = 0.180; p = 0.145), and maternal education (β = 0.094; p = 0.437), suggesting that the ability to recognize words in the local accent may develop as a function of receptive vocabulary size.

Research on infant speech perception differentiates between children learning one versus multiple languages, but rarely considers how exposure to accents might influence language acquisition. Here, we examined how accent variability affects infants' language development by testing two groups of monolinguals: one with routine exposure to primarily Canadian English and one with exposure to Canadian English as well as at least one other variant of English. Our results demonstrate a dramatic influence of home accent exposure on infants' word form recognition in the laboratory. Infants with little accent variation in their home environment readily recognized familiar word forms in Canadian English by 12.5 months of age, but infants exposed to variable accents failed to show a reliable preference for words over non-words spoken in Canadian English until 18 months of age. This performance difference was observed despite our mono- and multi-accent groups being well-matched in vocabulary size and socioeconomic status.

How can we explain this finding? One possibility is that children receiving more accent variability in their home environment may simply be approaching our task differently than the children receiving less accent variability. Perhaps infants accustomed to hearing more variability are interested both in words they recognize and in words they do not, rendering preference studies inappropriate to study their word form recognition abilities. However, there are two reasons that call this explanation into question. First, these children do show a preference, just not until 18 months. And second, bilinguals (who might logically be expected to react similarly to our mixed-accent group) have been found to display looking preferences in studies using similar procedures as early as 11 months of age.¹⁵ 

Alternatively, only the amount of exposure to Canadian English might contribute to infants' recognition of Canadian-accented words. As infants from the high variability group receive less exposure to Canadian English, they might not have reached the required threshold enabling them to recognize words in the Canadian accent. This, too, seems unlikely. First, if this were the case, the proportion of Canadian accent exposure within this multi-accent group should have predicted infants' performance in this task (which it did not). Second, as 15-month-olds have been shown to accommodate unfamiliar accents after as little as two minutes of exposure,² it seems extremely improbable that monolingual 14.5-month-old infants who receive an average 38% of their language input in Canadian English have not yet passed that minimum.

A final explanation for our findings then, is that children with more exposure to accent variation in their daily environment create word representations^7,8 (or signal-to-word mapping strategies²) that are qualitatively different from those of infants who receive less variable input. Because of differences in the pronunciation of words across accents, the initial word representations of infants with multi-accent language exposure may, for instance, be less constrained than those of their single-accent peers. While this may not prevent infants from recognizing words under naturalistic listening conditions (e.g., understanding their Canadian-accented mother as well as their Irish-accented father, and potentially even a less familiar neighbor), coping with unfamiliar speakers in the absence of any context (as is characteristic of the paradigm employed here) might be too difficult.¹⁶ Note that if this were the case, then the fully developed word recognition skills of (older) infants hearing multiple accents will likely be more flexible than those of similarly aged infants receiving uniform accent input. As a result, children with exposure to accent variability might later excel in situations where they need to cope with unfamiliar accents.

Taken together, this study demonstrates a clear impact of daily accent variability on word form recognition in the regionally dominant accent for infants 12.5 to 14.5 months of age. This suggests that even very early on in life, monolingual infants do not form a homogeneous group. Discrepancies in experimental findings among labs in different locations could hence be due to the specific language environments experienced by children in their home settings. Thus, accent exposure may not only have profound implications for theories of early language processing, but with the infant speech perception community moving toward conducting large-scale cross-lab replications,¹⁷ it should also be included as a predictor of infants' early linguistic abilities.

This work was supported by grants from the Natural Sciences and Engineering Research Council of Canada and the Social Sciences and Humanities Research Council awarded to E.K.J. Part of this work was presented at the 2013 Society for Research in Child Development meeting, Seattle, WA and at the 2013 Workshop on Infant Language Development workshop, Donostia – San Sebastián, Spain. We thank Melissa Paquette-Smith for assistance with testing.