Effect of speaker certainty on novel word learning in monolingual and bilingual children

Abstract

In the present study, we examined the effect of speaker certainty on word-learning performance in English-speaking monolingual (M_Age = 6.92) and Spanish-English bilingual (M_Age = 7.32) children. No group differences were observed when children learned novel words from a certain speaker. However, bilingual children were more willing to learn novel words from an uncertain speaker than their monolingual peers. These findings indicate that language experience influences how children weigh cues to speaker credibility during learning and suggest that children with more diverse linguistic backgrounds (i.e., bilinguals) are less prone to prioritizing information based on speaker certainty.

Many speaker-pertinent characteristics can cue credibility, including how knowledgeable, accurate, reliable, and certain a speaker is perceived to be. In general, children are more likely to learn new words from speakers portrayed as knowledgeable, accurate, reliable, and certain than from speakers portrayed as ignorant, inaccurate, unreliable, and uncertain (Bergstra et al., 2013; Birch, Vauthier & Bloom, 2008; Corriveau & Harris, 2009b; Koenig & Harris, 2005; Sabbagh & Baldwin, 2001). Children’s sensitivity to speaker characteristics develops early and matures relatively quickly during the preschool years. For instance, both younger and older preschoolers prefer to learn from a familiar informant or an informant most similar to them versus an unfamiliar informant (Corriveau & Harris, 2009a; Kinzler, Corriveau & Harris, 2011). Children are also aware of speaker credibility and use that information to guide their learning (Birch, Akmal & Frampton, 2010; Koenig & Echols, 2003; Koenig & Harris, 2005; Sabbagh & Baldwin, 2001). For example, Sabbagh and Baldwin (2001) showed that children were more successful when learning novel words from a speaker who verbally expressed certainty compared to a speaker who expressed uncertainty.

Not only are children able to identify a certain versus an uncertain speaker, children are also able to keep track of that information and use that information in later learning situations. For example, upon identification of a certain vs. an uncertain speaker, preschool children show preference for the certain speaker one week after certainty information about the speaker was provided (Birch, Vauthier & Bloom, 2008; Corriveau & Harris, 2009b). Birch, Vauthier and Bloom (2008) tested 3 and 4-year-old children on their ability to keep track of accuracy patterns of speakers when learning new words. Results revealed that children learned new words and new object functions from speakers who demonstrated certainty in the past but not from speakers who previously demonstrated uncertainty. Moreover, children can adjust their learning based on new information about the certainty or accuracy of their informant (Jaswal & Neely, 2006; Pasquini, Corriveau, Koenig & Harris, 2007; Scofield & Behrend, 2008). For instance, Jaswal and Neely (2006) found that when provided with a choice to learn new words from a child versus an adult, children consistently chose to learn new words from the adult. However, when provided with a choice to learn from a child with a history of being accurate versus an adult with a history of being inaccurate, children consistently chose to learn from the child. Additionally, children can extend their trust of an accurate speaker to other individuals associated with that speaker (Barth, Bhandari, Garcia, MacDonald & Chase, 2014).

Thus, there is ample evidence indicating that children are sensitive to speakers’ characteristics, and prioritize information provided by speakers they perceive as credible sources of knowledge over speakers they perceive as non-credible sources of knowledge. Prior research that has attempted to examine possible individual differences in children’s sensitivity to speaker cues focused on age effects, finding that older children are more sensitive to such cues than younger children. (Corriveau & Harris, 2009; Corriveau, Meints & Harris, 2009; Ganea, Koenig & Millett, 2011; Koenig & Harris, 2005). However, very little prior research exists regarding the possible variability in children’s sensitivity to speaker cues that may be due to the environments that the children occupy. One exception is a recent study that has examined the impact of Socio-Economic Status (SES) on children’s ability to learn novel words from speakers using complex versus simple syntax (Corriveau, Kurkul & Arunachalam, 2016). The findings were that children from low SES preferred to learn from an informant who used simple syntax, while their peers from higher SES preferred to learn from an informant who used more complex syntax. This study suggests that different environments may lead children to develop distinct cognitive strategies, leading children from different backgrounds to attend to speaker cues in distinct ways. Here, we ask a related question: Is it possible that children who grow up in bilingual vs. monolingual environments also differ in how they perceive speaker cues?

All prior work that has examined children’s sensitivity to speaker-related cues has focused exclusively on children from monolingual backgrounds. Monolingual children often have only limited experience with speakers who manifest variable levels of certainty that are not necessarily reflective of their credibility. It is possible that it is this experience that leads monolingual children to prioritize information from certain speakers. In contrast, bilingual children may have experience with speakers whose speech is characterized by verbal and non-verbal markers of uncertainty (such as higher use of pauses and slower speech rate, Krahmer & Swerts, 2005), and who are nevertheless reliable sources of information. This is because bilingual children are likely to be exposed to non-native speakers of their language(s) (Fernald, 2006; Place & Hoff, 2011), whose speech is often characterized by higher use of pauses and slower speech rate than native-speaker speech (Krahmer & Swerts, 2005). The hypothesis tested in the current study is that the experience with speakers who express uncertainty and who are nonetheless credible may lead bilingual children to weigh certainty cues differently than monolingual children.

While there is little prior work directly examining how bilingual vs. monolingual children respond to speaker cues during word learning, a few studies have tested how bilingual and monolingual children perceive speakers’ speech characteristics. For instance, Souza, Byers-Heinlein and Poulin-Dubois (2013) assessed monolingual and bilingual children’s preferences for native-accent versus foreign-accent speakers. Children were asked to choose the speaker that they would most likely want to be friends with, and both monolingual and bilingual children were found to prefer the native-accented speaker. However, while social preferences based on speaker identity may be similar in children from different language backgrounds, the degree to which speaker characteristics may influence learning outcomes may in fact fluctuate with children’s experience. This hypothesis finds indirect support in studies documenting differences in the mechanisms and the particular strategies that underlie word learning by monolingual vs. bilingual children (e.g., Byers-Heinlein & Werker, 2009; Davidson et al., 1997; Davidson & Tell, 2005; Houston-Price et al., 2010; Healey & Skarabela, 2008; Jaswal & Hansen, 2006).

Studies that have explored how monolingual vs. bilingual children approach word learning suggest that bilingual children make use of somewhat different strategies when learning new words than monolingual children (Brojde, Ahmed & Colunga, 2012; Davidson, Jergovic, Imami & Theodos, 1997; Davidson & Tell, 2005; Diesendruck, 2005; Fennell, Byers-Heinlein & Werker, 2007; Byers-Heinlein & Werker, 2009; Rosenblum & Pinker, 1983). For instance, bilingual children appear to adhere less to rules of mutual exclusivity – an assumption that each object corresponds only to a single label (Markman & Wachtel, 1988) – than monolingual children, and are more willing to accept more than one name for a single object (Davidson, Jergovic, Imami & Theodos, 1997; Davidson & Tell, 2005; Merriman & Kutlesic, 1993). Such findings have been hypothesized to be rooted in differences between bilingual and monolingual linguistic environments, and bilingual children’s earlier need to accept more than one name for a single object (e.g., Davidson & Tell, 2005).

Furthermore, there is a body of research that indicates differences between how bilingual and monolingual children interpret social and referential cues during communicative exchanges. In a series of studies, Yow and Markman (2011a; 2011b) demonstrated that bilingual children were more sensitive than monolingual children when interpreting paralinguistic (tone of voice) and non-verbal (gaze direction) cues to infer the speaker’s communicative intent. Similarly, Fan, Liberman, Keysar and Kinzler (2015) found that children exposed to multiple languages outperformed monolingual children on a social communication task requiring perspective taking to interpret the speaker’s intended meaning. More specific to word learning, Yow and Markman (2015) found that bilingual children outperformed monolingual children on a novel object identification task that required them to understand speakers’ referential intent, and Brojde et al. (2012) found that bilingual children appeared to attend to pragmatic cues, eye gaze in particular, more so than their monolingual peers when learning novel words. These findings have been posited to be rooted in bilingual children’s heightened need to attend to their communication partners because in the bilingual environment, the communication partners may speak different languages.

Current Study

The present study was designed to examine speaker certainty effects in word learning by monolingual English-speaking and Spanish-English bilingual children. The logic of positing that bilingual children may respond differently to speaker certainty cues than monolingual children was that bilingual children’s linguistic environment (vs. monolingual children’s linguistic environment) is characterized not only by the presence of two languages, but also by an increased likelihood that the input in the two languages will be delivered by non-native speakers. Such a hypothesis is supported by broad population statistics which indicate that more than half (53%) of Hispanic children in the U.S. have at least one parent who does not speak English very well (Murphey, Guzman & Torres, 2014). These children are therefore very likely to hear non-native English input on a regular basis, although the proportion of Hispanic children who encounter non-native speakers is likely to be even higher, with the exposure to non-native speakers in the family, the community, and the schools. Non-native speakers may be less confident in their language use, utilizing more qualifying terms, pauses, and slower speech rate – all indexes of a less certain speaker (Krahmer & Swerts, 2005). We hypothesized that because of bilingual children’s likely experience with less confident speakers, bilingual children may be more willing to accept speaker uncertainty during word learning than monolingual children.

To test this hypothesis, we recruited monolingual English-speaking and Spanish-English bilingual children between 5 and 9 years of age. We designed a word-learning task, where children were taught novel words by a certain and an uncertain speaker, and then had to retrieve the novel words in a confrontation-testing format. Therefore, unlike prior studies that have focused on speaker certainty cues, we tested children’s actual learning of the novel words, rather than their speaker preference. This made the task considerably more challenging than the tasks implemented in prior research on speaker certainty, thus requiring that we test children who were older than the children targeted by prior studies (e.g., Corriveau & Harris, 2009; Corriveau, Meints & Harris, 2008; Ganea, Koenig & Millett, 2011; Koenig & Harris, 2005). The wide age range of participants in our study was an advantage as it enabled us to consider possible age effects in our data. Given the results of previous work, we expected younger children to be more impacted by speakers’ certainty cues; relatedly, we expected the effects of bilingualism (should these be observed) to be particularly strong in the younger age group.

The bilingual children in our study were all simultaneous bilingual children who acquired English and Spanish prior to 3 years of age. Nevertheless, there was significant variability in bilingual-experience characteristics within our bilingual sample, enabling us to examine how different bilingual-experience variables may impact children’s word-learning from a certain vs. an uncertain speaker. Of particular interest were the effects of bilingual children’s language proficiency, language exposure, and primary caregivers’ English language proficiency. Because we hypothesized that bilingual children’s language experience would impact their word-learning performance, we expected that children with more non-native exposure would be more willing to learn from an uncertain speaker.

Method

Participants

Thirty English monolingual (M_Age = 6.92, SD = 1.25, Range: 5.00–9.92; 14 males) and 51 Spanish-English bilingual children (M_Age = 7.32, SD = 1.36; Range: 5.08–9.92; 28 males) participated. All children were recruited from the Madison Metropolitan school district in Madison, WI. The bilingual group was oversampled due to inherent variability in bilingual language experiences and the difficulty associated with identifying a priori a homogenous group of bilingual children with fairly balanced skills across their two languages. Parents provided information regarding children’s developmental and educational history and completed the Language Experience and Proficiency Questionnaire (LEAP-Q; Marian, Blumenfeld & Kaushanskaya, 2007) about their own language background. All children were typically-developing and passed a bilateral hearing screening. For all children, socioeconomic status (SES) was indexed by the primary caregivers’ total number of years of education (M_Monolinguals = 17.65, SD = 2.80; M_Bilinguals = 13.53, SD = 3.53). Monolingual children had parents with significantly higher level of education compared to the bilingual children (see Table 1).

Table 1.

Participant Characteristics

	Monolinguals	Bilinguals	t-test
n	30	51
Age (years)	6.92 (1.25)	7.32 (1.37)	t (79) = −1.30
Parent Years of Education	17.65 (2.80)	13.53 (3.53)	t (79) = 5.46*
Non-verbal IQ	112.93 (18.32)	103.98 (13.51)	t (79) = 2.52
English Receptive Lang.	113.87 (1.83)	98.80 (14.13)	t (78) = 5.02*
English Expressive Lang.	115.13 (10.23)	87.98 (15.84)	t (79) = 8.40*
English Core Lang.	115.10 (10.63)	89.06 (15.63)	t (79) = 8.08*
Spanish Receptive Lang.	--	98.77 (13.07)	--
Spanish Expressive Lang.	--	89.43 (15.39)	--
Spanish Core Lang.	--	91.76 (14.69)	--

^*To account for multiple comparisons, p-value was divided by number of comparisons, yielding a criterion for significance of p < 0.01

The monolingual children were born in the U.S. and were exposed only to English in the home since birth. The majority (70%) of the monolingual children were Caucasian, 6% were African American, and 23% were multiracial. The monolingual children were not exposed to any other languages besides English. The bilingual children learned Spanish from birth (M_Months = 2.28, SD = 7.78) and learned English at an average age of 15 months (M_Months = 15.67, SD = 20.95). The majority of the bilingual children (47) were born in the U.S., one child was born in Spain, and the birthplace of three participants was undisclosed. Bilingual children were exposed to English 40.22% (SD = 29.17) of the time and to Spanish 59.77% (SD = 29.20) of the time, in a typical week, based on parent report. The majority (78%) of the bilingual children were Hispanic, 8% were Caucasian, 2% were African American, and 12% were multiracial.

Materials

Standardized measures.

The Visual Matrices subtest of the Kaufman Brief Intelligence Test (KBIT-2, Kaufman & Kaufman, 2004) indexed children’s nonverbal intelligence. The Clinical Evaluation of Language Fundamentals-Fourth Edition (CELF-4, Semel, Wiig & Secord, 2003) indexed children’s English language skills. The Clinical Evaluation of Language Fundamentals-Fourth Edition, Spanish (CELF-4 Spanish, Wiig, Semel & Secord, 2006) indexed bilingual children’s Spanish language skills. These measures revealed significant differences in English language abilities between bilingual and monolingual children (see Table 1). In addition, the bilingual children were balanced in their language knowledge as demonstrated by their scores on the English and Spanish standardized language assessment measures (CELF-4 English M_{Core-Language} = 89.45, SD = 15.69; CELF-4 Spanish M_{Core-Language} = 91.76, SD = 14.69; t (49) = −0.78, p = .44). Table 1 includes demographic and standardized assessment data for the two groups.

Word-learning task.

Fourteen novel words, following English phonological rules were selected from Gupta et al.’s (2004) database, and paired with pictures of highly familiar concrete nouns selected from the International Picture naming Database (Székely et al., 2004). The word-picture pairs were split into two lists of seven and across the two lists, the novel words were matched on length, stress patterns, neighborhood density, and phonotactic probability. See Table 2 for a full list of the novel words. Children learned one list of words in the certain condition, and one list in the uncertain condition. The auditory stimuli were recorded by two different female native speakers of English from the Midwest. All the stimuli were recorded in a soundproof booth at a 20 kHz sampling rate and were normalized to 70dB amplitude using Praat (Boersma & Paul, 2001).

Table 2.

Word-Learning Stimuli

Session 1			Session 2
Version	Novel Words	Familiar Object	Version	Novel Word	Familiar Object
A Certain	Betis	Shovel	A Uncertain	Bomoge	Helmet
	Gabek	Bed		Bonede	Church
	Konete	Hat		Dosene	Shoe
	Konove	Table		Gateke	Truck
	Kudile	Balloon		Kafane	Butterfly
	Patol	House		Tebon	Cat
	Tinuf	Couch		Timok	Witch
B Certain	Bomoge	House	B Uncertain	Betis	Witch
	Bonede	Table		Gabek	Cat
	Dosene	Balloon		Konete	Truck
	Gateke	Hat		Konove	Church
	Kafane	Couch		Kudile	Shoe
	Tebon	Bed		Patol	Helmet
	Timok	Shovel		Tinuf	Butterfly
C Certain	Betis	Shovel	C Uncertain	Bomoge	Helmet
	Gabek	Bed		Bonede	Church
	Konete	Hat		Dosene	Shoe
	Konove	Table		Gateke	Truck
	Kudile	Balloon		Kafane	Butterfly
	Patol	House		Tebon	Cat
	Tinuf	Couch		Timok	Witch
E Uncertain	Bomoge	Helmet	E Certain	Betis	Shovel
	Bonede	Church		Gabek	Bed
	Dosene	Shoe		Konete	Hat
	Gateke	Truck		Konove	Table
	Kafane	Butterfly		Kudile	Balloon
	Tebon	Cat		Patol	House
	Timok	Witch		Tinuf	Couch
F Uncertain	Betis	Witch	F Certain	Bomoge	House
	Gabek	Cat		Bonede	Table
	Konete	Truck		Dosene	Balloon
	Konove	Church		Gateke	Hat
	Kudile	Shoe		Kafane	Couch
	Patol	Helmet		Tebon	Bed
	Tinuf	Butterfly		Timok	Shovel
G Uncertain	Bomoge	Helmet	G Certain	Betis	Shovel
	Bonede	Church		Gabek	Bed
	Dosene	Shoe		Konete	Hat
	Gateke	Truck		Konove	Table
	Kafane	Butterfly		Kudile	Balloon
	Tebon	Cat		Patol	House
	Timok	Witch		Tinuf	Couch

Note. Versions D and H were excluded due to a programming error. The novel words and picture pairs within each version were presented in random order.

Children were told that they would learn words from a new language. In the certain condition, the female speaker introduced herself by saying the following, “Hi, my name is Sarah. Today, I am going to teach you some new words. These words are from a language I learned as a baby. I know this language very well. Please listen carefully and try to remember these new words.” She taught the new words in a confident voice, with no pauses or disfluencies. In the uncertain condition, a different female speaker introduced herself by saying the following, “Hi, my name is Jenny. Today, I am going to teach you some new words. These words are from a language I started learning two weeks ago. I do not yet know this language very well. Please listen carefully and try to remember these new words.” She hesitated as she taught the words, and her speech included fillers (e.g., umm, ah) and pauses. All cues to speaker certainty were auditory in nature, and the children never saw the faces of the speakers.

The pairings of novel words and pictures were counterbalanced across lists. Furthermore, speaker identity was counterbalanced across children, such that for half of the children, speaker 1 was certain, and for half of the children, speaker 2 was certain. The two word lists were taught during two different sessions, with order of condition counterbalanced across participants. The task included an exposure phase and a testing phase, both presented in English. In the exposure phase, children learned to associate the novel words with their paired objects. A novel word was presented twice, and the picture stayed on the computer screen for 6 seconds. In the testing phase, children were required to recognize correct word-object parings. Each novel word was paired once with a correct picture, and once with an incorrect picture, for a total of 14 trials; trial order was randomized for each child. The testing phase included only the pictures and the words that the children were previously exposed to in the exposure phase. The child pressed one button if the pairing was correct, and a different button if the pairing was incorrect. The testing phase was presented immediately following the learning phase.

Coding and Analyses.

Given the large number of children who performed below chance in the certain condition (the condition expected to be easier for the children), the data were analyzed in two steps: First, analyses were conducted for all children and then analyses were conducted only for children who performed above chance in the certain-speaker condition (classified as leaners). To identify learners, accuracy data in the certain-speaker condition were coded in terms of sensitivity scores (A’), as outlined by Stanislaw and Todorov (1999). A hit rate and a false alarm rate were calculated for each condition. A hit was defined as a child indicating the novel word matched the picture when this was indeed the case. A false alarm was defined as a child indicating the novel word was the name of the picture they saw when in fact the picture and the word mismatched. Sensitivity scores range between 0 and 1, where 1 indicates perfect sensitivity. A sensitivity score of .5 and below indicates performance at chance. Children were designated as learners if their sensitivity scores in the certain condition exceeded an A’ score of .50 and non-learners if their sensitivity scores in the certain condition were an A’ of .50 and below. This demarcation resulted in 14 learners and 16 non-learners in the monolingual group and 25 learners and 26 non-learners in the bilingual group (see Table 3).

Table 3.

Demographic and Standardized Assessment Information for Learners vs. Non-Learners

Monolinguals				Bilinguals
	Learners	Non-Learners	t-test	Learners	Non-Learners	t-test
n	14	16		25	26
Age (years)	7.26 (1.02)	6.64 (1.38)	t (28) = 1.38	7.70 (1.34)	6.96 (1.31)	t (49) = 1.99†
Parent Years of Education	17.50 (2.68)	17.78 (2.97)	t (28) = −0.27	14.00 (3.43)	13.08 (3.64)	t (49) = 0.93
Non-verbal IQ	118.00 (20.38)	108.50 (15.62)	t (28) = 1.44	107.52 (14.17)	100.58 (12.15)	t (49) = 1.88†
English Receptive Lang.	115.86 (10.36)	112.13 (11.26)	t (28) = 0.94	102.00 (11.19)	95.85 (16.05)	t (49) = 1.56
English Expressive Lang.	116.86 (11.68)	113.63 (8.88)	t (28) = 0.86	91.00 (12.94)	85.08 (17.98)	t (49) = 1.35
English Core Language	116.57 (10.78)	113.81 (10.66)	t (28) = 0.69	92.24 (11.90)	86.00 (18.25)	t (49) = 1.44
Spanish Receptive Lang.	--	--	--	99.96 (12.25)	97.62 (13.96)	t (49) = 0.40
Spanish Expressive Lang.	--	--	--	89.60 (16.61)	89.25 (14.37)	t (47) = 0.20
Spanish Core Language	--	--	--	92.88 (14.09)	90.58 (15.51)	t (47) = 0.90

^†Marginally significant at .051= p < 0.06

Logistic mixed effect models were constructed in R, version 3.3.1 (R Core Team, 2015) using the lme4 packages (Bates, Maechler, Bolker & Walker, 2015) to analyze the accuracy data, first for all the children, and then for children identified as learners. Contrast coding was used for the dichotomous predictor variables for ease of interpretation. Trials in the certain condition were coded as 0.5 and trials in the uncertain condition were coded as −0.5. Contrast coding was also used for group, with monolinguals coded as −0.5 and bilinguals coded as 0.5. The effect of group and certainty on accuracy were considered first. Age, SES, and English language skills were entered into the model as covariates because all three variables have been reported to impact word learning (Bion, Borovsky & Fernald, 2013; Lee & Burkam, 2002; Nash & Donaldson, 2005; Nelson, Welsh, Trup & Greenberg, 2011). All three variables were centered for ease of interpretation. The model included random certainty-by-participant intercept and random slope. Raw parameter estimates from the logistic mixed effect regression models and the odds ratio to quantify effect size of significant effects are reported.

Results

For the participant sample as a whole, the analyses revealed only a significant effect of group (b = 0.42, SE = 0.17, χ²(1) = 6.04, p = .01), such that bilingual children outperformed monolingual children on the word-learning task, across both conditions. The effect of certainty was not significant (b = −0.001, SE = 0.10, χ²(1) = 0.000, p = .99), and neither was the interaction between group and certainty (b = −0.21, SE = 0.21, χ²(1) = 1.01, p = .32). See Table 4 for the full model.

Table 4.

Logistic Mixed-Effects Model for the Whole Group

	Estimate	SE	z
Intercept	0.02	0.06	0.39
Group	0.42	0.17	2.46*
Certainty	−0.00	0.10	−0.01
Age	0.17	0.04	3.79**
SES	0.02	0.02	1.08
English language skills	0.01	0.00	2.62**
Group × Certainty	−0.21	0.21	−1.00

Note. Monolinguals were coded as the reference group.

^*p < 0.05;

^**p < 0.01

When only learners were considered, the analyses revealed a significant effect of group (b = 0.54, SE = 0.20, χ²(1) = 6.52, p = .01), and a significant effect of certainty (b = 0.41, SE = 0.14, χ²(1) = 6.50, p = .01). The interaction between group and certainty was not statistically significant (b = −0.42, SE = 0.28, χ²(1) = 2.09, p = .15). See Table 6 for the full model. However, the interaction term was strong enough to merit a follow-up analysis examining certainty effects within each group. For monolinguals, results revealed that the effect of certainty was significant (b = 0.63, SE = 0.23, χ²(1) = 7.67, p = .005). Conversely, for bilinguals, results revealed that the effect of certainty was not significant (b = 0.20, SE = 0.17, χ²(1) = 1.26, p = .26).¹

Table 6.

Logistic Mixed-Effects Model for the Learners Only

	Estimate	SE	z
Intercept	0.37	0.08	4.78
Group	0.54	0.20	2.63**
Certainty	0.41	0.14	2.85**
Age	0.06	0.06	1.04
SES	0.02	0.02	0.91
English language skills	0.01	0.01	2.37*
Group × Certainty	−0.42	0.29	−1.45

^*p < .05

^**p < .01

Post-hoc analyses

Age Effects. Children between the ages of 5:0 to 7;11 were selected to represent a younger age group. This resulted in 23 monolinguals, of whom 9 were learners, and 35 bilinguals, of whom 14 were learners. See Table 7 for the younger children’s demographic information. See Table 8 for the full statistical model. For the younger sample overall, analyses revealed a significant effect of group (b = 0.54, SE = 0.21, χ²(1) = 6.60, p = .01); however, the effect of certainty was not significant (b = 0.002, SE = 0.11, χ²(1) = 0.0003, p = .99), and the interaction between group and certainty was not significant (b = −0.19, SE = 0.21, χ²(1) = 0.80, p = .37). For learners only, analyses revealed significant effect of group (b = 0.57, SE = 0.28, χ²(1) = 3.98, p = .04), and a significant interaction between group and certainty (b = −0.76, SE = 0.37, χ²(1) = 4.05, p = .04). The effect of certainty was not statistically significant (b = 0.32, SE = 0.19, χ²(1) = 2.87, p = .09). The significant group by certainty interaction led us to assess the effect of certainty separately in monolinguals and bilinguals. For monolingual children, the effect of certainty was significant (b = 0.70, SE = 0.26, χ²(1) = 7.15, p = .007). However, for bilingual children, the effect of certainty was not significant (b = −0.08, SE = 0.27, χ²(1) = 0.09, p = .76).

Table 7.

Demographic and Standardized Assessment Information for Younger Monolingual and Bilingual Learners

	Monolinguals	Bilinguals	t-test
n	9	14
Age (years)	6.66 (0.70)	6.70 (0.80)	t (21) = −0.14
Parent Years of Education	17.67 (3.32)	14.79 (4.06)	t (21) = 1.78
Non-verbal IQ	111.89 (23.23)	104.71 (16.60)	t (21) = 0.87
English Receptive Lang.	114.67 (10.65)	102.86 (9.32)	t (21) = 2.81*
English Expressive Lang.	118.22 (14.17)	90.64 (10.18)	t (21) = 5.44*
English Core Language	118.56 (13.05)	91.21 (9.74)	t (21) = 5.75*

^*To account for multiple comparisons, p-value was divided by number of comparisons, yielding a criterion for significance of p < 0.01.

Table 8.

Logistic Mixed-Effects Model for the Younger Children

	Estimate	SE	z
Intercept	0.36	0.10	3.45
Group	0.57	0.28	1.99*
Certainty	0.32	0.19	1.69
Age	0.05	0.12	0.41
SES	0.02	0.02	0.95
English language skills	0.01	0.00	1.18
Group × Certainty	−0.75	0.37	−2.01*

^*p < .05

Bilingual Variables.

Five different variables were created to assess different aspects of the bilingual environment and their impact on word learning from a certain vs. a non-certain speaker. The five variables were: bilingual proficiency ratio, length of bilingualism, bilingual exposure ratio, primary caregiver English proficiency, and primary caregiver English accent rating. Refer to Table 9 for a full description of the bilingual variables. Five different models were constructed, one for each bilingual variable. The models were ran on the whole bilingual sample. Each model included word learning accuracy as the outcome variable and certainty by the bilingual variable interaction. Each model also always included certainty by participant random intercept and random slope as predictor variables. See Table 10 for the results of each model. For four of the five bilingual variables (bilingual proficiency ratio, length of bilingualism, bilingual exposure ration, and primary caregiver English accent rating), analyses revealed non-significant interactions between bilingual experience and certainty. However, for parent English proficiency, results revealed a significant interaction between certainty and parent English proficiency rating (b = - 0.11, SE = 0.05, χ²(1) = 5.20 p = .02). Follow-up analyses revealed a significant effect of certainty (b = 1.01, SE = 0.47, χ²(1) = 4.64, p = .03) for children whose parents rated their English speaking abilities at 5 (adequate) or below. However, the effect of certainty was not significant (b = 0.32, SE = 1.03, χ²(1) = 0.09, p = .75) for children whose parents rated their own English speaking abilities above 5.

Table 9.

Definitions of Bilingual Variables

Variable Name	Definition	Mean (SD)	Range
Bilingual Proficiency Ratio	English language score ÷ Spanish Language score	1.0 (0.26)	0.47 – 1.65
Length of Bilingualism	(Chronological age - age of second language acquisition) ÷ chronological age	0.81 (0.22)	0.28 – 1.0
Bilingual Exposure Ratio	Weekly % English exposure ÷ weekly % Spanish exposure	1.09 (0.74)	0.15 – 4.0
Primary Caregiver English Proficiency	Self-rating of English speaking ability on a scale from 0 to 10	5.55 (2.85)	1.0 – 10.0
Primary Caregiver English Accent Rating	Self-rating of English accent on a scale from 0 to 10	4.84 (2.86)	0.0 – 9.0

Table 10.

Bilingual Variables: Logistic Mixed-Effects Model for the Bilingual Sample

	Estimate	SE	z
Bilingual Proficiency
Intercept	−0.13	0.36	−0.37
Certainty	0.00	0.51	0.00
Bilingual Proficiency	0.24	0.35	0.70
Bilingual Proficiency × Certainty	−0.11	0.49	−0.22
Length of Bilingualism
Intercept	0.09	0.33	0.26
Certainty	−0.80	0.46	−1.73
Length of Bilingualism	0.03	0.40	0.07
Length of Bilingualism × Certainty	0.86	0.55	1.55
Bilingual Exposure Ratio
Intercept	0.32	0.15	2.05*
Certainty	−0.08	0.22	−0.37
Bilingual Exposure Ratio	−0.19	0.11	−1.63
Bilingual Exp. Ratio × Certainty	−0.02	0.17	−0.13
Primary Caregiver English Proficiency
Intercept	0.03	0.23	0.14
Certainty	0.54	0.30	1.76
PC English Proficiency	0.01	0.04	0.34
PC Eng. Proficiency × Certainty	−0.11	0.05	−2.28*
Primary Caregiver English Accent Rating
Intercept	−0.12	0.20	−0.63
Certainty	−0.39	0.29	−1.35
PC English Accent	0.05	0.04	1.32
PC English Accent × Certainty	0.06	0.05	1.24

^*p < 0.05

Discussion

Consistent with prior research (Bergstra et al., 2013; Sabbagh & Baldwin, 2001), monolingual children in the present study were more likely to learn from a certain speaker than from an uncertain speaker, although the results were true only for the monolingual children who performed above chance in the certain-speaker condition. The new finding was that bilingual children were more willing to accept uncertainty when learning novel words than monolingual children. However, once again, this was only true for the bilingual children who performed above chance in the certain-speaker condition. It is clear then that our findings are only relevant to the children who were sensitive to the cues we implemented to index speaker certainty.

In studies indicating bilingual advantages for processing social-pragmatic cues, differences between bilingual and monolingual linguistic environments have been implicated. Similarly, differences in bilingual vs. monolingual linguistic environments are likely at the root of the differences in word learning in the present study. Bilingual children are likely to experience more diversity in their linguistic input than monolingual children. They are exposed not only to multiple languages, but also to speakers who may be characterized by a wider range of language abilities including non-native speakers (Fernald, 2006; Place & Hoff, 2011; Unsworth, 2016). Non-native speakers may demonstrate verbal and non-verbal uncertainty (e.g., non-native speakers tend to be less fluent than native speakers; e.g., Krahmer & Swerts, 2005) that does not reliably cue credibility. Consequently, bilingual children may be more likely than monolingual children to experience situations where uncertain speakers are nonetheless reliable sources of new knowledge. Although we do not have data regarding children’s exposure to non-native speakers, we do have evidence that bilingual children in the present study experienced more exposure to non-native input than monolingual children. Parents of the bilingual children classified as learners rated their non-native English accent on average as 5.27 (SD = 2.83) on the LEAP-Q indicating moderate accent. In contrast, parents of the monolingual children classified as learners rated their non-native English accent on average as 0.00 (SD = 0.00) indicating they were native English speakers. Parents with limited English ability may use Spanish only when communicating with their children; however, the vast majority of the bilingual children in the present study (84%) were exposed to English in the home. Thus, we argue that the bilingual children in the present study were more accepting of the uncertain speaker and were more willing to treat novel words produced by the uncertain speaker as reliable information because of their experience with uncertain non-native speakers.

Our findings also revealed that the bilingual children who had parents with poorer English speaking abilities were impacted by certainty cues, while the effect of certainty was not significant for bilingual children whose parents had adequate English speaking abilities. These results likely stemmed from the fact that parents with poorer English speaking skills did not use English with their children, and communicated exclusively or mostly in Spanish in the home. The majority of parents who rated their own English skills as “poor” indicated that their child was mainly exposed to Spanish in the home (22 out of 30 parents). In contrast, of the parents who rated their language skills as adequate and above, about half reported that their child was mainly exposed to English and the other half indicated that their child was exposed to both languages at home. These data provide strong, albeit indirect, evidence that the children whose parents had better English speaking skills were more likely to be exposed to English spoken by non-native speakers in the home. It is this experience that likely led them to treat certainty cues differently. Notably, other bilingual variables such as the child’s own language proficiency, language exposure, and length of bilingualism did not influence word-learning performance. Therefore, it appears that the bilingual child’s environment, and in particular their interaction with non-native speakers of English, shapes their responses to speaker certainty cues. However, in future studies, it will be important to capture children’s experience with non-native input in a more explicit manner.

There are several aspects of the findings that bear scrutiny when interpreting the results. First, over half of the children in each group did not perform above chance on the word learning task in the certain condition. We were unable to determine what distinguished the children who learned vs. children who did not learn, although we administered a wide array of language and cognitive measures to the children. The overall low rate of learning is likely due to the difficulty level of the task where children learned seven novel words within a short period of time while similar studies required children to learn only two to three novel words (Koenig & Harris, 2005; Sabbagh & Baldwin, 2001). The high rate of non-learners may also be due to the fact that the word learning task was not interactive. Future studies examining speaker certainty cues in the context of bilingualism would do well to emulate prior monolingual experiments in which children saw, or interacted with the speakers (e.g., Bergstra et al., 2013; Corriveau & Harris, 2009; Koenig & Harris, 2005). Second, the nature of our task, which required retention of novel words, dictated that we test older children than the children typically tested in studies examining the effects of speaker cues on word learning. Notably, while prior studies suggest that sensitivity to speaker cues increases with age (Corriveau & Harris, 2009; Corriveau, Meints & Harris, 2009; Ganea, Koenig & Millett, 2011; Koenig & Harris, 2005), our analyses revealed that the younger monolingual children showed a more robust pattern of speaker-certainty effects than the overall sample. Further, the interaction between speaker-certainty and group (monolingual vs. bilingual) was stronger when only the younger children were considered. Notably, even our younger children were older than the children typically tested in prior studies. Thus, future work should test younger children and consider the possibility that speaker cues become less relevant to a word learning task that requires retention, especially once the memory system becomes more functional.

Third, while young monolingual children have been shown to differentiate speakers based on the speaker cues we implemented here (e.g., Bergstra et al., 2013; Sabbagh & Baldwin, 2001), we did not obtain direct data regarding bilingual children’s ability to differentiate a certain from an uncertain speaker. Therefore, future work should explicitly assess whether bilingual children are sensitive to the pragmatic and acoustic cues to speaker certainty. Fourth, the cues used in the present study may have cued multiple aspects of speaker identity. That is, although previous studies have successfully used the cues presented in our study to cue certainty (e.g., Bergstra et al., 2013; Sabbagh & Baldwin, 2001), we cannot be assured that these cues were communicating certainty, competency, or both. Nevertheless, we can be assured that these cues communicated differences in speaker characteristics, and bilingualism appeared to influence how these speaker characteristics were perceived. Fifth, future studies should assess whether bilingual children attend more closely to the uncertain speaker than monolingual children, perhaps as the result of showing more empathy to the uncertain speaker. Such a possibility is in line with work indicating that bilingual children are particularly finely tuned to the perspective of others (Bassetti, 2007; Fan et al., 2015; Yow & Markman, 2015). Last, the bilingual children in the present study outperformed the monolingual children on the word-learning task, consistent with prior work suggesting bilingual word-learning advantages (Kaushanskaya, Gross & Buac, 2014; Kaushanskaya & Marian, 2009; Kalashnikova et al., 2014). However, our statistical models accounted for the effects of SES as well as language, thus compensating for the fact that the bilingual children were characterized by lower SES and lower English language skills than the monolingual children. In this, our sample of bilingual children is representative of the Spanish-English bilingual child population in the U.S. (Camarota, 2012; Hoff et al., 2012; Vagh, Pan & Mancilla-Martinez, 2009). Nevertheless, future work should attempt to separately assess the effect of SES and language ability, and compare monolingual and bilingual children from high SES and monolingual and bilingual children from lower SES on their sensitivity to speaker cues. Such a comparison would be particularly worthwhile, as SES appears to influence children’s ability to attend to speaker cues when learning novel words (Corriveau, Kurkul & Arunachalam, 2016).

In conclusion, although it is possible that differences other than the linguistic environment (e.g., cultural; social; family-dynamics-related; etc.) may contribute to the patterns of word-learning performance observed here, the findings of the current study indicate that bilingual and monolingual children weight speaker characteristics differently. Unlike monolingual children, bilingual children do not appear to prioritize information from certain speakers over information from uncertain speakers. It remains to be seen whether these differences between bilingual and monolingual children’s treatment of speaker uncertainty may have cascading, long-term repercussions for children’s linguistic and cognitive growth.

Figure 1.

Predicted proportion correct for monolingual and bilingual learners adjusting for English language skills, age and SES. Error bars represent 1 standard error.

Figure 2.

Predicted proportion correct for monolingual and bilingual younger learners adjusting for English language skills, age and SES. Error bars represent 1 standard error.

Table 5.

Demographic and Standardized Assessment Information for Monolingual and Bilingual Learners

	Monolinguals	Bilinguals	t-test
n	14	24
Age (years)	7.26 (1.02)	7.70 (1.34)	t (37) = −1.08
Parent Years of Education	17.50 (2.68)	14.00 (3.43)	t (37) = 3.29*
Non-verbal IQ	118.00 (20.38)	107.52 (14.17)	t (37) = 1.89
English Receptive Lang.	115.86 (10.36)	102.00 (11.19)	t (36) = 3.78*
English Expressive Lang.	116.86 (10.36)	102.00 (11.18)	t (37) = 6.19*
English Core Language	116.57 (10.78)	92.24 (11.90)	t (37) = 6.32*

^*To account for multiple comparisons, p-value was divided by number of comparisons, yielding a criterion for significance of p < 0.01