Gesture Production Selectively Predicts Language Outcomes in Spanish-English Bilingual Children

Abstract

This longitudinal study (data collected from 2019 to 2023) examines the relation between Spanish-English bilingual Latino toddlers’ (n=46; F=22; M=24) early gesture production (M age=18.67 months; SD age=1.02) and later language skills (M age=36.87 months; SD age=0.81). Video recordings at child-age 18-months yielded counts of children’s speech and gesture production; the latter included gesture words (different meanings) and gesture sentences (gestures-plus-speech combinations). Multiple regression analyses revealed that gesture words and sentences at 18 months of age positively predicted word- and sentence-level skills at 36 months of age, respectively, but only in English. These relations held despite controlling for children’s speech production. These findings, that early gesture production selectively predicts language outcomes in bilingual children, suggest that gesture production may facilitate language-specific learning rather than reflecting a global communicative skill.

Even before they produce their first word, children use their hands to communicate (i.e., pointing to an object; Nicoladis, 2007). Nonverbal communicative behaviors (often with the hand or head) that convey meaning are referred to as gestures (Iverson & Goldin-Meadow, 2005; Rowe et al., 2022). With age, children’s gestures become integrated with speech. At the one-word stage, children can express a two-word idea by supplementing gesture with speech (e.g., pointing to an object, while saying “gimme”). Thus, children’s early gesture use can expand their communicative range (Iverson & Goldin-Meadow, 2005).

Moreover, developmental research reveals links between early gesture use and later language skills, at least for monolingual children (Colonnesi et al., 2010; Rowe et al., 2022). For example, the number of different word meanings that English speakers convey in gesture (i.e., gesture words) during toddlerhood predicts their English vocabulary skills in a positive direction right before preschool (e.g., 42-months of age; Rowe & Goldin-Meadow, 2009). Yet, different types of gestures differentially predict language outcomes. In English monolingual children, early gesture words do not predict their later sentence complexity. Sentence complexity is instead tied to the ways in which gestures are coordinated with speech. In other words, there is a one-to-one correspondence in terms of language domains, where gesture words (but not gesture sentences) predict later words, whereas gesture sentences (but not gesture words) predict later sentences. Of note, children can use gestures to complement the information provided in speech (say “cookie” and point to a cookie). Or children can convey more complex meanings in gesture, where information in gesture differs from, or supplements, the information in speech (say “eat” and point to a cookie) (Ozcaliskan & Goldin-Meadow, 2009). Past research shows that only the number of gestures that are supplemented with speech (i.e., gesture sentences) predicts sentence complexity (Rowe & Goldin-Meadow, 2009).

This finding of a one-to-one correspondence between language domains suggests that for English monolinguals, their use of gesture selectively predicts their later language skills, and thus that gesture does not reflect a global communicative skill (Goldin-Meadow, 2015). It has been argued that gesture words predict vocabulary—but not sentences—because the act of gesturing different word meanings plays an active role in children becoming good word learners (Rowe & Goldin-Meadow, 2009). By the same token, gesture sentences predict later sentences—but not vocabulary—because the act of expressing sentence-like meanings in gesture may play an active role in children becoming successful sentence learners.

Emerging bilingualism provides a unique window to further test the specificity of the relation between gesture and language development. Bilingual toddlers are in the process of acquiring two languages and typically at different rates (i.e., unbalanced bilingualism; Gámez et al. 2024; Hoff et al. 2012). If the same relations of gesture to language outcomes are evident in a bilingual’s two languages, despite their unbalanced skill levels, this would suggest a general language learning ability on the part of the child (i.e., global communicative skill). In contrast, if gesture provides children with opportunities to practice specific language features before they can produce them in speech, then it is possible that the relation of gesture to language outcomes would differ for unbalanced languages. For example, it is possible that within the same child, the relation of gesture to later language would only be evident in the weaker language (i.e., the one the child is actively working on) and not the stronger language (i.e., the one for which they receive more exposure/input). Or, while gesture words would predict later vocabulary in the weaker language, perhaps the relation of gesture sentences to later sentences would only be evident in their stronger language—because the weaker language is still only at the stage of vocabulary building. If gesture-to-language relations are evident in only one of the bilingual’s languages, and there is a one-to-one correspondence between language domains (e.g., gesture sentences, but not gesture words, predict later English sentences), this would add support to the claim that gestures predict language outcomes selectively. Thus, to probe the specificity of the gesture-to-language relation, we examine whether and how bilingual toddlers’ (18-month-olds’) use of gestures is associated with their later vocabulary and sentence-level skills, in both languages, that is, right before preschool (36 months of age).

Existing Literature on Bilingual Children’s Gesture Use

While a recent meta-analysis on pointing gestures and language development revealed a small, but significant overall effect size (i.e., r = .20; Kirk et al., 2022), the existing literature is limited in that it is primarily focused on monolingual populations. In fact, the authors of the meta-analysis did not identify any eligible research studies with sufficiently large samples of bilingual child to include in their analysis, signaling a lack of language diversity in the literature base.

The two existing studies that investigated the association between bilingual children’s gestures and language skills involved small sample sizes and revealed mixed results (Nicoladis, 1999; 2002), rendering their findings inconclusive. The motivation for these bilingual child gesture studies was not to explore the relation of early gesture use to later language skills, but their findings are nevertheless relevant here as they shed light on concurrent gesture and language associations. One longitudinal study (Nicoladis et al., 1999) included five French-English bilingual children, from 2 to 3.5 years old, living in one parent-one language households (i.e., the majority receiving French exposure from fathers and English exposure from mothers). To obtain gesture and language samples in French and English, children were video-recorded in separate play sessions with each parent; one session per parent/language. The video recordings were coded for the different types of gestures that children produced, including deictics (i.e., points) and iconics (i.e., representational gestures). Children’s mean length of utterance (MLU) served as the primary measure of language development—a production measure. The findings showed no significant relations between children’s MLU and deictics, at any age, and for either the English or French contexts. Of note, within both language contexts, children’s MLU was significantly correlated with their production of iconic gestures. The significant association between these measures concurrently makes sense given that MLU is considered a proxy for speech complexity and iconics are considered later emerging gestures, at an age when children’s speech is also more complex (Özçalişkan et al., 2014; Rowe & Goldin-Meadow, 2009).

Yet, a follow-up study (Nicoladis et al., 2002), involving eight 4-year-old French-English bilinguals, did not replicate the significant association between children’s iconics and MLU, nor was there a significant association between children’s gestures and their scores on standardized measures of receptive vocabulary (i.e., PPVT). What’s more, the non-significant relations between gesture and language outcomes (i.e., MLU, receptive vocabulary scores) held across both English and French contexts. The free-play sessions examined were designed to elicit only English or French language use by prompting native English or French speakers (i.e., caregivers or other adults) to interact in the desired language. These non-significant results for receptive vocabulary are especially noteworthy given that the above-mentioned gesture-language meta-analysis (Kirk et al., 2022) revealed a significant effect of language domain, such that studies including comprehension language measures typically revealed statistically significant results; production measures did not. These non-significant findings with receptive language measures scores across both languages, as well as the mixed findings for MLU (Nicoladis et al., 1999), thus warrant further investigation, especially longitudinally and in naturalistic settings where language context may alternate interchangeably.

One large-scale study of emerging gestures in infants of either Mexican or Dominican heritage, living in the U.S., shed further light on the unique contributions of gesture to bilingual language outcomes longitudinally (Tamis-LeMonda et al., 2012). It is worth noting that the overwhelming majority of participating families in that study spoke Spanish as their primary language and did not report consistent bilingual language exposure at home. However, the toddler’s 24-month receptive vocabulary scores included both the English and Spanish words they understood, reflecting a bilingual measure. The findings showed that the infants’ gestures at 14 months of age were not significantly associated with their receptive vocabularies at 24 months of age. In fact, Kirk and colleagues’ (2022) meta-analysis revealed less robust relations between gesture and later language outcomes for younger children (i.e., younger than 18 months of age), in comparison to older children (Kirk et al., 2022). An open question is thus whether a long-term gesture-to-language relation would be seen for bilingual toddlers (i.e., past infancy), with comprehension measures, and in their two languages.

Oral Language Skills in two Languages and Domains (Word- and Sentence-Level)

The upsurge of research on bilingualism in the past two to three decades (Kroll & Mendoza, 2022) has led to a now sizeable literature describing bilingual children’s oral language development, emphasizing mostly their vocabulary skills, and to a lesser extent, their sentence-level skills. That literature measures children’s oral language skills using standardized assessments, often in only one language. Much of the existing research describes bilingual children’s single-language skill performance as lower than monolingual children’s performance (e.g., Bialystok et al., 2010; Pearson et al., 1993). For example, when vocabulary knowledge is measured in only one language, bilingual children’s single language performance can be 1 to 2 standard deviations below monolingual norms (Hammer et al., 2014). There are mixed findings for the lesser-studied domain of syntax, with some findings showing that bilingual children score below monolingual norms, while others showing performance on par with monolinguals (Korade et al., 2022).

A reason that bilingual children may score lower than monolinguals on single-language assessments is that bilinguals rarely demonstrate balanced development in their two languages (Nicoladis et al., 1999). In other words, bilinguals tend to be more dominant in one of their languages, which for young children tends to be the language to which they have more exposure (Bohman et al., 2010; Grosjean, 2010; Treffers-Daller, 2016). In fact, a consistent finding in the bilingual literature is that children’s bilingual language outcomes are influenced by their amount of exposure to each language (Gámez et al. 2023; De Houwer 2011; Hoff et al. 2012). These findings are in line with research that shows that language input (e.g., number of words) is positively associated with later language outcomes for monolingual children (Huttenlocher, 1991; Rowe, 2012). Thus, in the current study on how gestures contribute to language outcomes in bilingual children, we consider children’s varying levels of exposure to both of their languages and in turn, unbalanced language skills.

The present study

The goals of this longitudinal study are two-fold. We aim to 1) describe gesture use, in terms of word meanings (gesture words) and sentence-like ideas (gesture plus speech combinations), for 18-month-old Spanish-English bilingual children with unbalanced skills in their two languages, and 2) examine the relation of their early gesture use to word- and sentence-level outcomes at 36 months of age. Specifically, we assessed children’s English and Spanish language outcomes to probe the specificity of gesture-to-language associations in bilingual children. Given that the existing literature shows that gestures predict language outcomes selectively in English monolinguals (Rowe & Goldin-Meadow, 2009), we argue that gesture provides children with opportunities to practice specific language features before they can produce them in speech. Thus, we hypothesized that the relation of gesture to language outcomes would differ for unbalanced languages, which we test with confirmatory hypothesis testing. However, if the same relations of gesture to language outcomes are evident in a bilingual’s two languages, despite their unbalanced skill levels, this would suggest a general language learning ability on the part of the child (i.e., global communicative skill).

The only other study that relies on longitudinal data to address these questions about early gesture-to-later language relations included the age range of 18 to 42 months, the latter which represents an age right before preschool. The ages included in this study are within this age range. The slightly different age at the upper end of the age range is justified by the need to control for bilingual children’s exposure levels. That is, because preschool entry can alter bilingual children’s exposure to their two languages, 36-months of age represents a time when exposure comes primarily from caregivers. Also, at child-age 36 months, we relied on standardized measures of children’s receptive language, both at the word and sentence-levels. The decision to use comprehension measures, in lieu of production measures, was guided by previous literature (outlined above) that suggests more robust gesture-to-language associations for comprehension than production measures (Kirk et al., 2022). In addition, the use of comprehension measures allows us to make comparisons between studies (e.g., Rowe & Goldin-Meadow, 2009).

At 18 months, we collected home-based video-recordings of toddlers’ gesture use, at the word level (i.e., in terms of their different word meanings) and sentence level (i.e., gesture plus speech combinations). To examine the word-level gesture-to-language associations for each language (i.e., English or Spanish), and because gestures can occur without speech, it was important to determine the language context of each word meaning associated with a child gesture. As noted, prior studies relying on gesture observations involved families that followed the one parent-one language exposure approach (Nicoladis et al., 1999). In those studies, to isolate language context, children were observed with the parent who exposed them to a particular language (i.e., English, or French). Other bilingual gesture studies explicitly restrict caregivers from speaking one of their languages to artificially create a single language context (Limia et al., 2019). The current study involved bilingual children who were exposed to both English and Spanish by both of their caregivers (though as discussed later, received more Spanish than English exposure), rendering past protocols inadequate for this type of diverse language environment. Thus, for our naturalistic design to be in line with previous bilingual gesture studies (e.g., Nicoladis et al., 1999), the language associated with children’s gesture meanings was based on the language that surrounded the gesture (e.g., the language used by the caregiver).

To examine the sentence-level gesture-to-language associations, we focused on children’s gestures that were accompanied by their speech, either in English or Spanish. Guided by previous studies that show a relation between gesture-speech combinations and later sentence-level outcomes (Rowe & Goldin-Meadow, 2009), we were particularly interested in gesture speech combinations where the child’s gesture supplemented the meaning conveyed by the speech, indicating a complex idea. To anticipate, in our analyses, the different word types (in English and Spanish) or mean length of utterance (MLU) in children’s speech (in English and Spanish) were included as control variables to account for differences in children’s spoken language skills. The following research questions guided this study:

1. How do Spanish-English bilinguals, with varying levels of Spanish and English exposure, use gestures at 18 months of age?

2. What is the relation of Spanish-English bilingual children’s early gesture use (at 18 months of age) to their later word- and sentence-level skills (at 36 months of age) within-languages (in English and Spanish)?

Method

Participants

Forty-six bilingual children (F = 22; M = 24) and their primary caregivers (F = 45; M = 1), from the midwestern United States, participated in this study. At the beginning of the study, children were approximately 18 months of age (M_age = 18.67 months; SD_age = 1.02) and 36 months of age (M_age = 36.87 months; SD_age = 0.81) at the end of this study. According to caregiver report, children were exposed to both Spanish and English by 18 months of age. Children were exposed to Spanish from birth and to English either from birth (n = 39) or by the first year (n = 3); four families did not give a date. In addition, the majority of children did not attend daycare or preschool (85% at 18 months; 70% at 36 months); 2 caregivers did not respond to this question at 36 months of age. The children who were in daycare at 18 months of age were exposed to either only Spanish (n = 4), or to both English and Spanish (n = 3) by the daycare provider. Of the children who were in daycare or preschool at the 36-month visit, the majority (n = 8) were exposed to both English and Spanish by the daycare or preschool provider, followed by only Spanish (n = 3), and only English (n = 2). All participating children were identified by their primary caregiver as Latino (Latino only n = 39; Latino & Caucasian/White n = 4; Latino & Black/African American n = 2, and Latino/Filipino/Black n = 1).

Most primary caregivers identified as Latino (n = 43; Latino only n = 42; Latino & Caucasian/White n = 1); the remaining identified as Caucasian/White (n = 3). The majority of Latino caregivers self-identified as second generation in the U.S. (58%; n = 25); the remaining identified as first generation in the U.S. (42%; n = 18). Most second-generation caregivers (84%; n = 21) identified with Mexican culture (Mexican American n = 14; Mexican n = 4; U.S. Mexican Guatemalan n = 2; U.S. Mexican Argentinian = 1); the remaining (16%; n = 4) identified with Guatemalan American (n = 2), Ecuadorian (n = 1), and Cuban Peruvian (n = 1) cultures. Similarly, most first-generation caregivers (89%; n = 16) identified with Mexican culture (Mexican n = 13; Mexican American n = 3); the remaining (11%; n = 2) identified with Colombian (n = 1) or Guatemalan (n =1) cultures. The 3 primary caregivers who identified as White/Caucasian identified with American (n = 2) or Greek Irish cultures (n = 1) and as 2^nd generation (n = 1) or above (4^th n = 1; 5^th n = 1) in the U.S. For these 3 families, the child’s secondary caregiver (who did not participate in the present study) identified as second-generation Latino (n = 3) and with Mexican (n = 2) or Ecuadorian (n = 1) cultures. Further, all families lived in predominately Latino neighborhoods (Latino M = 45%; SD = 20%; White M = 35%; SD = 19%; Black M = 15%; SD = 11%; Other M = 5%; SD = 11%) and annual household income varied from less-than-$15,000 to over-$100,000 (“less-than-$15,000 n = 3; “$15,000 to $49,999” n = 12; “$50,000 to $99,999” n = 14; “100,000-or-more” n = 16; one participant did not respond). Note that this was the final sample after 7 children were excluded from a sample of 53 either because their families dropped out of the study by this study’s last time point (i.e., attrition; n = 5), did not consent to video recording (n = 1), or the child was not speaking by age 2 (n = 1).

Materials

Children’s Reported Language Exposure.

A background and language questionnaire asked about children’s language exposure (Duursma et al., 2007). Specifically, one question asked the primary caregiver to indicate the languages used with the child at home. The responses to this question were set to the following 5-point scale: 5 = “Only Spanish,” 4 = “Mainly Spanish,” 3 = “English and Spanish Equally,” 2 = “Mainly English,” and 1 = “Only English.” Responses were recoded to represent a percentage of Spanish exposure (100%, 75%, 50%, 25%, and 0% Spanish, respectively) and percentage of English exposure (0%, 25%, 50%, 75%, and 100% English, respectively), separately.

Receptive Word-Level Skills.

The Peabody Picture Vocabulary Test, Fourth Edition (PPVT-IV; Dunn & Dunn, 2007) and the Test de Vocabulario en Imagenes Peabody (TVIP; Dunn et al., 1986) assessed children’s English and Spanish receptive vocabularies, respectively. These are standardized, norm-referenced assessments in which children point to a picture that corresponds to a named word, which increase in difficulty. The assessments yield raw scores that are converted to standard scores (M = 100; SD = 15). The publisher’s internal consistency reliability using the split-half method is .95 in English and .93 in Spanish.

Receptive Sentence-Level Skills.

The Sentence Structure subtest of the Clinical Evaluation of Language Fundamentals Preschool-2, English and Spanish versions (CELF; Wiig et al., 2004), assessed children’s comprehension of grammatically correct sentences. Children point to a picture that corresponds to a sentence that was read aloud to them. The sentences increase in length and complexity. Internal consistency reliability coefficient based on the split-half method is .79 for English and .76 for Spanish.

Video-recordings.

Caregiver-child interactions were video-recorded using handheld camcorders (Sony HDRCX405 HD).

Procedures

This study is part of a larger longitudinal study of bilingual language development in ~50 Spanish- and English-speaking Latino children for which data was collected between 2019 and 2023 (Gámez et al. 2023, 2024). For the current study, which involves the time points that correspond to child-ages 18 and 36 months, researchers interviewed primary caregivers, using demographic and language questionnaires. In addition, at the 18-month time point, researchers conducted a video-recorded observation in the home for ~75-minutes. Researchers were instructed to follow the child and keep both children and caregivers in the camera frame. Caregivers were instructed to go about their day as they typically would, engaging in a variety of activities, including toy play and meal-time, etc., mostly indoors, but also outdoors. At the 36-month time point, researchers administered standardized language assessments of children’s receptive vocabulary (always first) and sentence structure (always second) skills to children in English and Spanish, each language about a week apart. To administer the assessments, researchers used the language of the assessment; the family’s preferred language was used in all other interactions.

Of note, part of this study was conducted during the COVID-19 pandemic. Thus, to minimize exposure risk, the 18-month video recordings for 7 families took place only outdoors (backyard or nearby green space) and lasted between 30- to 45-minutes. Given this, when a video recording was not 75 minutes long, we prorated the language and gesture data based on the number of observation minutes missing (Rowe, 2012). Also, at the 36-month time point, all interviews were conducted by phone. Standardized language assessments were administered either virtually (during stay at-home orders; n = 4), outdoors (n = 13), indoors (when stay-at-home orders were lifted; n = 15), or both indoors and outdoors (n = 9). Five caregivers did not agree to one-on-one standardized testing (PPVT; CELF) for their child at the 36-month time point, yet all other data were collected, including home language use data as well as other child-level language assessments (which are not reported here as those analyses are beyond the scope of the present study). As is recommended (Jeličić et al., 2009), missing PPVT and CELF data for these 5 children were handled using multiple imputation (10 iterations; IBM Corp., 2010). To be clear, data were not imputed for the five children who dropped out of the study as we did not have any collected data for them at 36 months (see participant section). Thus, analyses were conducted only for the 46 children that participated in the current study. This study was approved by the Institutional Review Board at Loyola University Chicago (Project# 2467).

Spoken Language Measures.

Video-recordings were transcribed using modified CHAT conventions (Codes for Human Analysis of Transcription) of the Child Language Data Exchange System (CHILDES; MacWhinney, 2018). The transcription process involved identifying speakers (e.g., “*CHI:”), and breaking speech into utterances, which are units of speech bounded by breaks in the flow of speech, for example breaths, long pauses, or speaker switches. To distinguish between English and Spanish in the transcripts, transcribers indicated the base language of each file (e.g., Spanish) and marked the opposite language with a language switch code (e.g., [- English]). Transcribers were instructed to not correct any speech and to mark unintelligible words and utterances as such (i.e., unintelligible = “xx”). They were also instructed to represent children’s speech-like vocalizations phonetically and to indicate babbling as such (e.g., babbling = “ba@b”). When children did not use standard pronunciation of words (e.g., “ma” instead of “más” [“more”], transcribers annotated the transcript with the word’s standard spelling (e.g., “[: más]”). Repetitions were also marked in the transcript (e.g., “no [/] no [/] no”). Fully fluent Spanish-English bilingual undergraduate students underwent transcription training that consisted of working one-on-one with a reliable transcriber. The training culminated with a transcription reliability test that consisted of independently transcribing 5 files. Reliability was calculated as the mean number of lines that matched (in terms of symbols and words) over the total number of lines (note that unintelligible sounds and variations in phonetic spellings labeled as babbling were excluded from counts). Transcribers were deemed reliable when they achieved an average reliability score of 88% after transcription training. A second reliable transcriber reviewed all files for transcription accuracy.

The freq and MLU commands of the Computerized Language Analysis (CLAN; MacWhinney, 2018) software of CHILDES (MacWhinney, 2018) were used to derive children’s word types and mean length of utterance in words (MLUw), respectively. Both commands exclude unintelligibles, and for MLUw specifically, repetitions and instances of singing (e.g., “clean up [/] clean up”; “baby shark”) were excluded. Instances of babbling were not included in word counts, nor were sounds like “um,” and “ah,” but onomatopoeic sounds were counted as words (Pan et al., 2005; e.g., “guau guau” for “dog” in Spanish; woof-woof in English). Each child received two sets of word-level (word types) and sentence-level (MLUw) scores, one for each language: (1) number of different words in Spanish (Spanish Word Types), (2) number of different words in English (English Word Types), (3) mean length of utterance in Spanish (Spanish MLUw), (4) mean length of utterance in English (English MLUw). CLAN was also used to derive caregivers’ (1) total number of words in Spanish (Spanish Tokens) and (2) total number of words in English (English Tokens). These measures represent children’s observed amount of Spanish and English input, respectively.

Gesture Type Coding.

Gestures were defined as nonverbal communicative (often hand and head) movements that are symbolic in nature (Iverson & Goldin-Meadow, 2005; Rowe et al., 2022). Gestures were identified in the video recordings, and each was categorized according to the following gesture types: deictic, hold-ups, reach-fors, iconics, and conventional. Deictic gestures consisted of pointing at a person, item, or location in the immediate environment. Index-point and palm-points were both coded as pointing gestures. The only gestures to manipulate objects were hold-up gestures; these consisted of holding an item up to show or bring attention to it, thus were primarily communicative. Hold-up gestures are considered a potential precursor to pointing gestures, but they are typically grouped together with deictics for simplicity purposes (Cameron-Faulkner et al. 2005; Özcaliskan & Goldin-Meadow, 2009). Reach-for gestures, sometimes referred to as reaches or ritualized reaches (Cameron-Faulkner et al., 2005; Iverson & Goldin-Meadow, 2005), signify wanting something and consisted of an extended arm, without holding the item. Iconic gestures represented an idea, action, or the shape of an object, for example, flapping arms to indicate bird. Conventional gestures have culturally or universally pre-established meanings, for example, waving to communicate a greeting. Each child received 1) a score representing the number of times they used a gesture type and 2) a score representing the number of total gestures (all types combined). The proportion of each gesture type was derived using the total number of gestures from each category over the total number of gestures produced. Coders passed a reliability test, scoring better than 87% agreement, after their gesture coding training. Reliability was calculated as the number of correctly identified gesture types over the total number of gestures.

Gesture Words Coding.

Next, coders identified each gesture’s meaning, that is, what the child is trying to communicate with the gesture (e.g., point to cup means “cup”). We focused on identifying the number of different gesture word meanings for each language (i.e., English types and Spanish types). To be more in line with previous studies (Nicoladis et al., 1999), we coded for gesture meanings as occurring within an English, Spanish, or code-mixed context. To do so, coders identified the language that immediately followed the gesture, typically provided by the primary caregiver, but also the other caregiver. An utterance (or word) in English (or Spanish) that followed a gesture was coded as an English (or Spanish) context. Utterances that included both English and Spanish were coded as a code-mixed context. A relatively small number (0.04) of gestures occurred in a code-mixed context, and because the present study is not intended on conducting bilingual analyses, we excluded all code-mixed meanings. Thus, only responses to gestures in an exclusively English context or Spanish context were included here. Each unique meaning was counted. Each child received two sets of gesture word scores: (1) number of different gesture word meanings in Spanish (Spanish Gesture Words), (2) number of different gesture word meanings in English (English Gesture Words).

Gesture plus Speech Sentences Coding.

Separately, transcripts of children’s speech, together with the video recordings, were coded for whether speech was accompanied by a gesture. If speech and gesture occurred together, these instances were coded for in terms of whether the words and gestures conveyed different meanings. When the gesture’s meaning was different from the word’s meaning, this was considered a supplementary gesture. For example, if a child said the word “mom” while pointing to a cup, this gesture would be coded for as a supplementary (i.e., different meaning) gesture as its meaning “cup” would be different form the meaning “mom” expressed in speech. Thus, the use of a supplementary gesture with a spoken word conveys sentence-like ideas; the gesture conveys one idea and speech another (Rowe & Goldin-Meadow, 2009). The number of gesture-plus-speech combinations conveying sentence-like ideas represented the measure of gesture sentences. Each child received two gesture sentence scores: (1) number of gesture plus (different) speech combinations in Spanish (Spanish Gesture Sentences), (2) number of gesture plus (different) speech combinations in English (English Gesture Sentences).

Data Analysis Plan

Descriptive statistics were used to address the research question regarding bilingual children’s use of gestures (RQ #1). To address the research question regarding the relations of gesture to language outcomes (RQ #2), we relied on a series of multiple regressions with gesture words and gesture sentences as the main predictor variables for word-level and sentence-level skills, respectively. As noted, we added the control variable Word Types to word-level models and MLUw to sentence-level models to account for differences in spoken language as well as pinpoint the specific role of gesture (Rowe & Goldin-Meadow, 2009). The control variable Age was added to account for differences in the timing of when the language assessments were administered (due to the COVID-19 pandemic). Preceding the inferential analyses, descriptive statistics are reported of children’s raw scores for spoken language and standardized assessments.

Results

Children’s Unbalanced Language Exposure and Speech at 18 months

Table 1 shows that, as a group, children were exposed to “mostly Spanish” by their primary caregivers (closer to ~75% Spanish; ~4 on the language exposure scale). This was the case at both 18 and 36 months of age. In fact, a paired-samples t-test showed no significant difference in primary caregiver’s reported exposure across time, Spanish exposure t(45) = −1.78, SE = .02, CI = [−.07, .02], p > 0.05. Indeed, on average, caregivers used mostly Spanish during the 75-minute observation, t(45) = 5.52, SE = 235.57, CI = [825.12, 1774.04], p < 0.05. Specifically, the majority of caregivers (n = 36) used more Spanish (M = 2193.36; SD = 960.93) than English Tokens (M = 237.66; SD = 309.54); 10 caregivers used more English (M = 1683.29; SD = 929.65) than Spanish Tokens (M = 620.85; SD = 442.37).

Table 1.

Descriptive Statistics

Child Variable	Mean	SD	Min	Max
Age	18.67	1.02	17.72	22.88
Time I (18 months)
Time II (36 months)	36.87	.81	35.51	39.42
Percent Language Exposure (Spanish Reported by Caregiver)
Time I (18 months)	69.32%	23.42%	25%	100%
Time II (36 months)	71.74%	28.26%	25%	100%
Time I (18-month) Data
Spanish
Word Types	25.03	23.72	2.00	132.00
Mean Length of Utterance (words)	1.07	0.08	1.00	1.36
Gesture Words	26.91	16.57	2.00	85.23
Gesture Sentences	8.21	8.63	0.00	37.00
Input (from Caregiver)	1851.51	1089.67	43.00	4560.00
English
Word Types	15.01	19.89	1.00	101.00
Mean Length of Utterance (words)	1.05	0.06	1.00	1.18
Gesture Words	9.13	14.86	0.00	66.00
Gesture Sentences	4.59	7.79	.00	35.00
Input (from Caregiver)	551.93	781.55	2.50	2992.00
Time II (36-month) Data
Spanish
TVIP Raw	11.38	7.33	1.00	30.00
TVIP Standard Score	94.54	11.65	77.00	124.00
CELF Raw	9.55	3.01	2.00	14.00
CELF Scaled Score	9.49	2.38	3.00	13.00
English
PPVT Raw	32.37	14.73	3.00	70.00
PPVT Standard Score	87.98	15.06	51.00	113.00
CELF Raw	7.23	2.78	3.00	13.00
CELF Scaled Score	6.66	2.40	2.00	12.00

Note. CELF = Clinical Evaluation of Language Fundamentals; PPVT = Peabody Picture Vocabulary Test; TVIP = Test de Vocabulario en Imagenes; Percent Language Exposure in English is the inverse of Percent Language Exposure in Spanish.

A paired-samples t-test also showed that children’s spoken language was unbalanced, such that they produced on average ~10 more Spanish Word Types than English Word Types, t(44) = 3.411, SE = .08, CI = .12-.46, p < 0.01. That is, during the 75-minute observation at 18 months, children used, on average, 25 different words in Spanish, and 15 in English. However, children’s MLUw in Spanish and English did not differ statistically, t(44) = 1.288, SE = .01, CI = [−.01, .05], p > 0.01. They used about one word per utterance in each of their languages; though the mean being slightly above one suggests that this group was past the one-word stage and had started combining more than one word to create utterances. Of note, the spoken language scores were transformed using the logarithmic function because they showed positive skew (i.e., Skewness > 1.5). In the case of English Word Types and MLUw, there remained one outlier (> 2 SD’s from the mean) due to one child not using any English in the video recording, and thus, the English scores from this one child were excluded from any analyses involving English. (Table 1)

Children’s Gesture Words and Sentences and Relations with Speech at 18 months

Table 2 shows the proportion of each gesture type produced. Approximately 40% of identified gestures were coded as deictics, with all children producing deictic gestures during the 75-minute observation. Conventional were the second most occurring type of gestures (25%), followed by reach-for gestures (18%) and hold-up gestures (17%). Iconics were the least occurring gesture type (1%), with 67% children producing no iconics during the 75-minute observation at 18 months. (Table 2)

Table 2.

Proportion of Types of Gesture Produced at Child-Age 18 months

	Mean	SD	Min	Max
Deictic	.38	.21	.05	.91
Conventional	.25	.16	.00	.63
Reach For	.18	.12	.00	.56
Hold Up	.17	.13	.00	.59
Iconic	.01	.02	.00	.06

In terms of gesture production by language context, Table 1 shows that children used about 27 different gesture words in a Spanish context, and about 9 in an English context. Also, they used about 8 gesture sentences (i.e., gesture-plus-speech combinations) in Spanish, and about 5 in English. Paired-samples t-tests showed that these differences were statistically significant (p’s < 0.001). This pattern of more gesture production in Spanish than English is consistent with this sample’s primarily Spanish input. In fact, paired-samples t-tests showed that children who received a greater amount of caregiver input in Spanish than English input, as measured from the video recordings, used significantly more Spanish gesture sentences (M = 8.86; SD = 8.20) than English gesture sentences (M = 2.84; SD = 5.64) (p < 0.05). Inversely, the children who received more caregiver English than Spanish input, as measured from the video recordings, used significantly more English gesture sentences (M = 10.72; SD = 11.10) than Spanish gesture sentences (M = 3.5; SD = 5.21) (p < 0.05). Moreover, children who received more caregiver input in Spanish than in English used significantly more Spanish gesture words (M = 29.50; SD = 17.34) than English gesture words (M = 4.81; SD = 8.80) (p < 0.05). Also, while children who received more caregiver input in English than in Spanish used more English gesture words (M = 24.24; SD = 21.46) than Spanish gesture words (M = 18.06; SD = 11.10), this different was not statistically significant (p > 0.05). Together, these results suggest that gesture use differs as a function of children’s language input, that is, children who are exposed to more Spanish than English, tend to gesture more in Spanish, than English, and vice versa. (Table 3)

Table 3.

Correlations Between Vocabulary, Sentence Structure Scores and Percent Spanish Language Exposure at Child-Age 36 months

	English PPVT (Vocabulary)	English CELF (Sentence Structure)	Spanish TVIP (Vocabulary)	Spanish CELF Sentence Structure
English PPVT (Vocabulary)	--
English CELF (Sentence Structure)	.522***	--
Spanish TVIP (Vocabulary)	−.026	.101	--
Spanish CELF Sentence Structure	−.015	−.160	.460***	--
Percent Spanish Exposure	−.515***	−.263*	.244*	.240*

^*p < .05

^**p < .01

^***p < 0.001.

Note. CELF = Clinical Evaluation of Language Fundamentals; PPVT = Peabody Picture Vocabulary Test; TVIP = Test de Vocabulario en Imagenes: English n = 45; Spanish n = 46.

Further, the positive and significant Pearson correlations in Table 4 indicate that children’s spoken language and gesture measures were correlated within language (p’s < 0.05). However, separate Pearson correlations revealed no statistically significant relations between gesture and speech measures across languages (p’s > 0.05). Of note, given that gesture scores were right skewed (i.e., Skewness > 1.5), these data were transformed using a log transformation. To do so, a constant was added to correct for the zeros that resulted from gesture coding by context (i.e., all children gestured). Thus, log transformed scores were used for all analyses.

Table 4.

Correlations Between Gesture and Language Measures at Child-Age 18 months

Spanish	Gesture Words	Gesture-plus-Speech	Word Types	MLUw
Gesture Words	--
Gesture-plus-Speech	.536***	--
Word Types	.485**	.708***	--
MLUw	.168	.298*	.388**	--
English	Gesture Words	Gesture-plus-Speech	Word Types	MLUw
Gesture Words	--
Gesture-plus-Speech	.638***	--
Word Types	.570***	.688***	--
MLUw	.322*	.469**	.715***	--

^*p < .05

^**p < .01

^***p < 0.001.

Note. MLUw = Mean Length of Utterance in Words; English n = 45; Spanish n = 46.

Relations of bilingual children’s use of gesture words at 18 months on their Spanish and English receptive vocabularies at 36 months

Table 1 shows wide variability in children’s receptive vocabulary scores in English (PPVT) and Spanish (TVIP) at 36-months of age. Note that we did not make direct comparisons between language assessments as the raw scores are set to different scales and the standard scores were derived from the norms for monolingual, not bilingual children. To help explain this variance in vocabulary scores, we built a series of regression models, using each language as an outcome: English PPVT and Spanish TVIP. The word-level models included word-level predictor variables in the same language as the outcome (e.g., Spanish language predictors for Spanish outcomes; English language predictors for English outcomes). Part A of Table 5 shows that the main predictor variable English Gesture Words at 18 months was positively associated with English PPVT scores at 36 months, explaining 26% of the variance in scores. Part B of Table 5 shows that English Gesture Words remained a significant and positive factor in predicting English PPVT scores (p < 0.05), even when controlling for children’s English Word Types, which was not a significant factor (p > 0.05). This model explained 29% of the variance in scores. Part C shows a similar pattern of results when including the non-significant control factor, Age (p > 0.05).

Table 5.

Multiple Regression Models Predicting Receptive Vocabulary in English and Spanish at Child-age 36 months

	English Receptive Vocabulary (PPVT)	Spanish Receptive Vocabulary (TVIP)
Parameter Estimates	B (SE)	B (SE)
Part A
Intercept	22.55 (3.15)***	−0.68 (5.16)
Gesture Words	14.51 (3.71)***	8.82 (3.70)*
R2	.263	.115
Effect size f2	.357	.130
Part B
Intercept	18.46 (4.63)***	−3.41 (4.92)
Gesture Words	11.43 (4.49)*	3.61 (3.95)
Word Types	6.37 (5.30)	7.81 (2.87)**
R2	.287 **	.245
Effect size f2	.403	.325
Part C
Intercept	55.20 (92.00)	−6.11 (46.08)
Gesture Words	11.26 (4.56)*	3.89 (3.82)
Word Types	6.73 (5.43)	7.16 (2.78)*
Age (months)	−1.01 (2.51)	0.08 (1.23)
R2	.290	.246
Effect size f2	.408	.326

^*p < .05

^**p < 0.01

^***p < 0.001.

Note. Main predictor variables (Gesture Words, Word Types) were measured at child-age 18 months, whereas Age was measured at child-age 36 months; Spanish models n = 46; English models n = 45.

In contrast, while Spanish Gesture Words significantly and positively predicted Spanish TVIP scores in a model without any controls (Part A), it was no longer significant after controlling for Spanish Word Types, that is, spoken language (Part B). Instead, Spanish Word Types significantly predicted Spanish TVIP scores in a positive manner, even after controlling for Age (Part B), which was not a significant factor (p > 0.05). Spanish Word Types explained about 25% of the variance in scores.

Relations of bilingual children’s use of gesture sentences at 18 months on their Spanish and English sentence structure at 36 months

Table 1 also shows variability in children’s sentence structure (CELF) scores in English and Spanish at 36-months of age. To help explain this variance in scores, we built a series of regression models, using each language as an outcome: English CELF and Spanish CELF. The sentence-level models included sentence-level predictor variables in the same language as the outcome (e.g., Spanish language predictors for Spanish outcomes; English language predictors for English outcomes). Part A of Table 6 shows that the main predictor variable English Gesture Sentences at 18 months was positively associated with English CELF scores at 36 months, explaining 23% of the variance in scores. Part B of Table 6 shows that English Gesture Sentences remained a significant and positive factor in predicting English CELF scores (p < 0.05), even when controlling for children’s English MLUw, which was not a significant factor (p > 0.05). This model explained 24% of the variance in scores. Part C shows a similar pattern of results when including the non-significant control factor, Age (p > 0.05); this model explained 25% of the variance in scores.

Table 6.

Multiple Regression Models Predicting Sentence Structure in English and Spanish at Child-age 36 months

	English Sentence Structure (E-CELF)	Spanish Sentence Structure (S-CELF)
Parameter Estimates	B (SE)	B (SE)
Part A
Intercept	5.94 (0.52)***	8.45 (.87)***
Gesture Sentences	2.80 (.79)**	1.43 (.98)
R2	.227	.046
Effect size f2	.293	.048
Part B
Intercept	5.81 (.54)***	8.30 (.89)***
Gesture Sentences	2.47 (.90)**	1.14 (1.03)
MLUw	13.38 (16.60)**	13.40 (14.34)
R2	.239	.065
Effect size f2	.314	.070
Part C
Intercept	−7.20 (17.80)	−.17 (21.35)
Gesture Sentences	2.49 (.90)**	1.29 (1.11)
MLUw	13.77 (16.71)	13.79 (14.52)
Age	.35 (.48)	.23 (.57)
R2	.249	.068
Effect size f2	.331	.073
Part D
Intercept	5.11 (.67)***	6.00 (1.34)*
Gesture Sentences	1.93 (.93)*	.35 (1.06)
MLUw	15.63 (16.26)	10.51 (13.86)
Percent Language Exposure	2.95 (1.68)	4.308 (2.04)*
R2	.292	.155
Effect size f2	.413	.183

^*p < .05

^**p < 0.01

^***p < 0.001.

Note. MLUw = Mean Length of Utterance in Words; Main predictor variables (Gesture Sentences, MLUw) were measured at child-age 18 months, whereas Age and Percent Language Exposure were measured at child-age 36 months. Spanish models n = 46; English models n = 45.

In contrast, the main variable Spanish Gesture Sentences did not significantly predict Spanish CELF scores (p > 0.05; Part A). The control variables MLUw and Age were also not significantly associated with Spanish CELF scores (p’s > 0.05; Parts B and C). What’s more, these models explained less than 7% of the variance in scores.

Post-hoc Analyses

In sum, our main multiple regression analyses showed statistically significant early gesture-to-later language relations for English, but not Spanish. In particular, early gesture words in an English context predicted later English words, while early gesture sentences in English predicted later English sentences. These findings show that gesture production selectively predicts language outcomes in English. Moreover, these findings align with the hypothesis that gesture is playing an active role in helping children learn language, for example, by providing children with opportunities to practice specific language features before they can produce them in speech. Although not central to our aims, to further probe the specificity of these gesture-to-language relations within-languages, we conducted follow-up analyses that examined the gesture-to-language relations across languages.

Specifically, if the role of gesture is to provide children with opportunities to practice specific language features that they do not yet have in speech, then the relation of gesture to later language should only be evident within, and not across languages. In contrast, if early gesture use represents a global communicative skill, then the relation of gesture to later language should be evident not only within languages, but also cross-linguistically. In fact, bilinguals have been shown to leverage their lexical knowledge in their first language to learn new word labels in their second language, thus suggesting cross-linguistic relations (Cummins, 2000; Castilla et al., 2009; Genesee, 2019).

To examine across-language gesture relations, we built word-level models that included word-level predictor variables in the opposite language from the outcome variable. Specifically, the model with English PPVT as the outcome included the predictor variables Spanish Gesture Words, Spanish Word Types, and Age. The model with the Spanish TVIP as the outcome included the predictor variables English Gesture Words, English Word Types, and Age. None of the across-language variables were significant predictors of English PPVT scores (p’s > 0.05) or Spanish TVIP scores (p’s > 0.05).

To examine across-language gesture relations, we built sentence-level models that included sentence-level predictor variables in the opposite language from the outcome variable. Specifically, the model with English CELF as the outcome included the predictor variables Spanish Gesture Words, Spanish Word Types, and Age. The model with the Spanish CELF as the outcome included the predictor variables English Gesture Words, English Word Types, and Age. None of the across-language variables were significant predictors of English PPVT scores (p’s > 0.05) or Spanish TVIP scores (p’s > 0.05).

Given the consistent non-significant findings for Spanish CELF scores, we also ran post-hoc analyses to further investigate the factors that could help explain the observed variance in CELF scores. Specifically, we built regression models that included the variable Language Exposure (Table 6: Part D), which replaced the Age variable (at 36 months) to ensure an acceptable ratio of predictors to observations (Snijders, 2003). Note that while our analyses did not reveal significant differences between reported language exposure at 18 and 36 months, there is literature to suggest that bilinguals’ language exposure patterns can change across time (Treffers-Daller, 2016), and thus, in these models, we relied on the Language Exposure variable from child-age 36 months. In a model, predicting Spanish CELF scores (Part D, Spanish), the percentage of Spanish exposure was significantly associated with outcomes in a positive direction (p < 0.05); this model explained 16% of the variance. Further, when the percentage of English exposure was added to a model predicting English CELF outcomes (Part D, English), it was not a significant predictor (p > 0.05), yet English Gesture Sentences remained statistically and positively significant (p < 0.05). Cross-language analyses revealed statistically insignificant omnibus regression models for Spanish CELF (p > 0.05) and English CELF outcomes (p > 0.05).

Notably though, Kendall’s Tau correlations showed that Language Exposure scores were significantly associated with each outcome (Table 3). Thus, for example, the negative correlations between Language Exposure (in Spanish), English PPVT, and English CELF scores indicate that more English exposure is tied to higher English outcomes. Note also that when we attempted to build multiple regression models with the Language Exposure score for the vocabulary outcomes, the VIF scores were all above 1.5, suggesting collinearity amongst variables. Because the Gesture Words scores were derived from coding for language context, this variable and Language Exposure may have been explaining some of the same variance in outcome scores.

Discussion

Past developmental research (Rowe & Goldin-Meadow, 2009) shows that monolingual English-speaking children’s early gestures, specifically those that correspond to distinct communicative meanings (i.e., gesture words), are positively associated with their later vocabulary outcomes. Moreover, monolingual children’s gesture-speech combinations (i.e., gesture sentences), which typically emerge later in development (Iverson & Goldin-Meadow, 2005), predict syntactic complexity. Bilingualism poses a unique opportunity to isolate linguistic trajectories associated with gestures at the word- and sentence-levels, especially in children acquiring each language at different rates (i.e., unbalanced bilinguals). If gesture-language outcomes are similar across a bilingual’s two unbalanced linguistic systems, this would suggest that gesture development may be a general communicative skill. However, if each language has distinct gesture-language trajectories, this would suggest that gesture does not merely reflect a global communicative skill. For example, gesture production may selectively predict language outcomes in one language, but not the other. Our study’s main aim was to examine whether and how bilingual children’s use of gestures at 18 months of age relates to their vocabulary and sentence-level skills within-languages, at 36 months of age.

Bilingual Children’s Gesture and Speech Production

To probe the specificity of the relation between early gesture and later language skills, we first examined Spanish-English bilingual children’s gesture production at 18 months of age, from video-recordings of naturally occurring language interactions in their homes. Consistent with the prior gesture literature (Kirk et al., 2022; Nicoladis et al., 2002), we found that the majority of children’s gestures were classified as deictics (40%), that is, pointing to convey different word meanings. Children used few iconics at 18 months of age (1%), which is overall consistent with the literature on monolingual children’s gestures, where significant increases in iconics are observed from 22 to 30 months (Özçalişkan et al., 2014; Özçalişkan & Goldin-Meadow, 2011).

Because the bilingual children in our sample were exposed to both English and Spanish, it was important to examine gesture use as a function of language context, that is, English or Spanish. In naturalistic settings, bilingual interlocutors can switch from one language to the other, depending on the speaker, activity, etc. (De Houwer, 2019). This dynamic nature of language use necessitated a fine-grained analysis of the language context that surrounded children’s gesture use. In doing so, we found that, on average, this sample of children--who were exposed more to Spanish than English--used about 27 different gesture meanings in a Spanish context, and about 9 in an English context. These figures are in line with a past study on 18-month-old monolingual children’s gesture production that similarly relied on video-recordings of caregiver-child interactions (Rowe & Goldin-Meadow, 2009). Differences in observation times between that study and ours restricts us from directly comparing children’s gesture production (our study = 75 minutes; 90 minutes in Rowe & Goldin-Meadow, 2009). However, a per minute ratio conversion of their reported means (33.6 English gesture words in 90 minutes, p. 183) and ours, specifically where we combined children’s English and Spanish gesture counts (i.e., English + Spanish = 36.04 in 75 minutes), indicates that in both studies, children used a gesture every two minutes (~.40 to .50 gestures per minute).

In addition, at 18 months of age, children combined gestures with different words to convey sentence-like ideas. It is worth noting that the group’s average MLUw indicates that children were mostly producing 1-word utterances (in the oral modality), and this was the case in both English and Spanish. What’s more, there was no significant difference in MLUw (spoken language) between languages. Yet, children did combine more gestures with words in Spanish than English. On average, during the 75-minute observation, children produced about 8 gesture-plus-Spanish-speech combinations and about 5 gesture-plus-English-speech combinations. Thus, as has been suggested before, gesture may serve to extend children’s communicative range, that is, allowing them to convey more complex ideas with gesture than they can with speech alone (Iverson & Goldin-Meadow, 2005). Our study further adds that children produce more gesture-plus-speech combinations in the language that they are exposed to most. That is, the majority of children in our study received more Spanish input during the 75-minute recording and produced more gesture-plus-Spanish combinations. Recall that the group of children who received more English than Spanish input, during the 75-minute recording, produced more gesture-plus-English speech combinations.

Gesture and Language Outcome Relations

A main contribution of the current longitudinal study is that we examined whether the variability in bilingual children’s language outcomes related to their early gesture use, thus extending prior bilingual gesture studies focused on investigating children’s concurrent gesture and language production (Nicoladis et al., 1999; 2002). Our findings replicate and extend past study findings with monolinguals (Rowe & Goldin-Meadow, 2009), showing that bilingual children’s gesture words at 18 months of age significantly predicted their word-level skills at 36 months of age. Moreover, bilingual children’s gesture sentences at 18 months of age significantly predicted their sentence-level skills at 36 months of age. These relations held even after controlling for children’s spoken language.

The novelty and uniqueness of our study findings with bilingual children is that gesture-to-later-language relations (at the word- and sentence-levels) were found only in English, not in both languages. Instead, children’s Spanish language outcomes at 36 months of age were related to spoken language measures (for word-level skills) or language exposure (for sentence-level skills). In this study, primary caregivers reported on their child’s language exposure at home, and more exposure to Spanish was related to higher performance on the standardized measure used to proxy sentence-level skills (sentence structure). While our model building prevented us from testing the relation between language input and later word-level skills in Spanish, our correlational analyses, indeed, showed that language input and word-level outcomes were positively associated in Spanish. In fact, a consistent finding in the bilingual literature is that bilingual children’s language outcomes are influenced by their amount of exposure to each language (Gámez et al. 2023; Hoff et al. 2012).

Stated differently, the specific associations observed for English, but not Spanish, provides further support to previous findings in the literature on the selective role of gesture. Why would we have seen a predictive role for gestures in English and not Spanish, which was instead, related to exposure and spoken language measures? Within the same learner, gesture might be most helpful for the language that is in a transitional stage (i.e., gesture might provide children with communicative resources before they can express the same ideas in speech; Goldin-Meadow, 2011), in this case English. As noted, as a group, the children in the current study had more exposure to Spanish and a greater number of Spanish word types, than English exposure and word types. Thus, for Spanish, where our sample had stronger language abilities, again within the learner, gesture might have ceased to play as strong of a supportive role. Indeed, the prior gesture literature suggests that gesture might play a greater role as a forerunner in initial stages of language learning compared to later stages, where children flesh out existing structures or skills in their repertoire (Ozcaliskan & Goldin-Meadow, 2009).

There is also developmental literature to suggest that the predictive value of gestures, versus speech, may depend on developmental timing (Choi, Wei, & Rowe, 2021). That is, there might be a window of development where children’s gesture production predicts their later language skills, for example, when children cannot express themselves well verbally and thus, depend on gestures to a greater extent to communicate. Then, when children can better express themselves verbally, speech might take over gestures as a stronger predictor of language skills. In other words, for the current sample, we may have missed the window of time when gestures were predictive of their Spanish skills because speech had already taken over as a stronger predictor, even by 18 months of age. The current study findings are informative and add to the literature in suggesting that the role of gesture might vary within a single learner. Specifically, we highlight that, the role of gesture, or the particular window when gesture may play the strongest role, depends on sample characteristics, including the skill level of the learner on a given task.

What might be the mechanisms that underlie these gesture-to-language relations? There might be multiple mechanisms underlying the specific associations. Gesture might simply index children’s communicative abilities, but moreover producing different types of gestures might play an active role in children’s language development (Goldin-Meadow, 2015). That is, gestures may give children opportunities to practice specific words and sentence constructions that then later appear in speech. Indeed, experimentally manipulating children’s gestures leads to later changes in their speech (LeBarton et al., 2015). In support of this hypothesis that gesture plays an active role in language learning, our follow-up analyses showed no statistically significant cross-linguistic gesture-to-language relations. That is, results showed no significant relation for word-level or sentence-level English language outcomes when gesture words and sentences were derived from a Spanish context, and vice versa. Relatedly, correlational analyses of the gesture and speech measures that were used in this study showed that gesture and speech were positively associated with each other only within language, not across languages (i.e., Spanish measures were not correlated with English measures). The finding that the relation of gesture to language is evident within-language, but not across languages, supports the idea that gesture provides children with opportunities to practice specific language features that they do not yet have in speech.

The lack of cross-language effects in this study may seem inconsistent with the long line of research on cross-linguistic influences in the domains of vocabulary learning (e.g., Castilla et al., 2009; Floccia et al., 2018) and syntactic development (Gámez and Vasilyeva 2020; Serratrice et al., 2022), but we believe that this finding aligns with the mechanistic explanation of cross-language effects (Cummins, 2000; Genesee, 2019). In particular, the existing cross-linguistic literature suggests that bilingual children can leverage their knowledge in one language to learn and use the other language (Genesee, 2019). It is important to note that the cross-linguistic influence literature primarily focuses on speech. For example, when a child is familiar with a concept and has a word label in one language, this aids in her/his understanding of the unknown word label in the other language, thus facilitating the mapping of a new word label to that concept (Cummins, 2000). It is thus possible that we did not find cross-linguistic influences because the children in this study had not yet learned the concepts that they were referring to with their gestures. Indeed, some scholars argue that children gesture when they are on the verge of learning a new concept (Goldin-Meadow, 2009). Perhaps cross-linguistic influences would have been evident if the children in this study had been at later stages of language learning when their gesture production was assessed, for example, beyond the 1-word stage, when they hold deeper conceptual and word knowledge.

Another mechanism that may underly these within-language gesture-to-language relations is that children’s gesture production elicits responses from their caregivers, and it is their responses that facilitate language learning. In fact, research shows that child gestures serve to elicit responses from caregivers, who supply the spoken words or constructions that the gestures are meant to convey (Goldin-Meadow et al., 2007). In doing so, caregivers provide models for conveying thoughts with words, which is in line with theoretical models of language development (Snow, 1994; Tomasello, 2000; Vygotsky, 1978). Caregivers’ input may thus be what facilitates language learning in young children. In fact, caregiver responses relate to children’s gesture use, which then, predicts their vocabulary in both monolingual (Özçalişkan & Goldin-Meadow, 2005) and bilingual samples (Limia et al., 2019).

Limitations and Future Directions

Examining caregivers’ responses in terms of whether they supplied a spoken word or sentence for children’s gestures was beyond the scope of the present study. That investigation would require a different design. For example, to see if the word or sentence that caregivers supplied at an earlier time point appeared in children’s speech at a later time-point, a different research design would be required where caregivers’ responses to children’s gestures are examined at one time point and then children’s speech are examined at a later time point.

However, we did pay attention to the language that interlocutors used when children gestured. This was done to determine the language context of each gesture (i.e., English, or Spanish), given that this study involved bilingual children exposed to both English and Spanish by the same caregiver and during a continuous 75-minute observation. As noted previously, isolating language context has been more straightforward in past gesture studies with bilinguals whose caregivers naturally used only used one language with the target child (e.g., one parent-one language method; Nicoladis et al., 1999). In this study, language context was determined by coding for the language used by caregivers’ when responding to children’s gestures. This approach was guided by past study findings that show that child gestures elicit high rates of maternal responses (van der Klis, Adriaans, & Kager, 2003). Nevertheless, future studies would do well to consider caregivers’ speech that appears before children’s gesture, especially if the gesture-to-later language relation is due to children’s practice of the modeled language. Further, to test the hypothesis that gesture plays an active role in language learning by providing children with language practice, future studies are needed that employ experimental designs, for example, that experimentally increase bilingual children’s gestures when learning new words in each of their languages (see LeBarton et al., 2015), or in the language they are exposed to the least.

Further, future research with bilingual samples exposed to both English and Spanish from a single caregiver might attempt to isolate language context by conducting more purposeful sampling of language throughout the day. This is because different activities may require the use of one language over the other. Moreover, while bilingual children know two languages, different interlocutors may prompt them to use each language to different degrees.

Also, this study relied on commonly used measures of receptive vocabulary (PPVT/TVIP) and sentence structure (CELF). Together with its Spanish counterpart, the use of the English receptive vocabulary measure allowed for comparisons to past study findings with the same measure (i.e., Rowe & Goldin-Meadow’s, 2009). Moreover, we calculated children’s standard and scaled scores as additional evidence for children’s unbalanced bilingual status. However, it is important to note that these measures were normed on monolingual children. Thus, caution should be used when making larger claims about children’s relative standing to their peer group. Given this, we used raw scores instead of standard or scaled scores for the analyses in this study.

In sum, the study findings represent an important step toward better understanding the specificity of the relation between gesture and bilingual language development. Our study findings suggest that gesture use does not reflect a global communicative skill. Instead, gesture use selectively predicts later language skills in bilinguals, specifically in one language and not the other. In the case of our unbalanced Spanish-dominant bilinguals, early gesture use predicted later language skills in English only. Thus, the role of gestures in language outcomes may vary and depend on the learners’ skill level in each language.