Early, typical, and late talkers: an exploratory study on predictors of language development in the first two years of life

1.1 Late and early talkers

Late talkers are children who show a vocabulary delay in the absence of any sensory, cognitive, or neurological deficit (Buschmann, Jooss, Rupp, Dockter, Blaschtikowitz, Heggen & Pietz, 2008; Rescorla & Lee, 2000; Silva, 1980; Whitehurst & Fischel, 1994). For the aim of the present study, we adopted the most common criteria found in the literature and focused on 24 months children who have barely reached a 50-word vocabulary and/or use no word combinations (Fenson, Dale, Reznick, Thal, Bates, Hartung & Reilly, 1994; Dollaghan, 2013; Rescorla, 2002). It is well known that late talkers are not a homogenous group in terms of their developmental outcomes: some catch up, as captured by the “late bloomers” metaphor and some will experience persistent learning difficulties (Chilosi et al., 2019). Furthermore, language delay has a multi-dimensional impact on learning (see, for example, Paul et al., 1997; Rescorla, 2002, 2005, 2009; Rescorla & Dale 2013) and socioemotional development (see, for example, Paul, 1991; Carson, Klee, Carson & Hime, 1997, 1998; Irwin, Carter & Briggs-Gowan, 2002). Hence, the early investigation of factors associated with a language delay is crucial for the early intervention so that education pro-grams can be implemented to maximize positive child developmental outcomes across a variety of domains (see, for example, Buschmann et al., 2015; Roberts & Kaiser, 2015; St Clair et al., 2011). Although there is a significant interstudy variability in the analysis of language development in late talkers (Fisher, 2017), research findings have been able to identify a series of specific communicative and linguistic characteristics of late talkers. More specifically, different authors have shown that late talkers have a narrow phonological inventory of both vowels and consonants, mostly characterized by simple syllabic structures, unlike typical talkers of the same age (Paul & Jennings, 1992; Rescorla & Ratner, 1996; Mirak & Rescorla, 1998; Carson et al., 2003). Interestingly, differences between late and typical talkers are not strictly limited to linguistic variables. In fact, empirical studies have suggested that some communicative behaviors of late talkers, such as the different degree in gestural production, represented potential early predictors of delay in subsequent language development (Ellis & Thal, 2008; Thal et al., 2013). Studies on gestural communication in children with linguistic delay highlighted significant differences at 20 months between late and typical talkers, resulting in a disadvantage of late talkers in the production of gestures and the combination of gesture and gaze at mother’s face (Fasolo & D’Odorico, 2002). In the same perspective, we aim to explore the communicative behaviors of late and typical talkers, but investigating instead, the predictors of language development from the first months of life. Moreover, we do not limit our examination to typical vs. late talkers, as traditionally done in the extant literature. Indeed, with the aim of better identifying what language development predicts more broadly, we involve into our analysis early talkers too. In terms of productive vocabulary, early talkers are children who scored around the 90th percentile based on parent report of toddler’s expressive language (Fenson et al., 1994). Research has shown that early spoken language development can play a substantial role in oral and written language performance (Lyytinen, Poikkeus, Laakso, Eklund & Lyytinen, 2001; Lyytinen, Ahonen, Eklund, Guttorm, Kulju, Laakso & Viholainen, 2004a, b; Scarborough and Dobrich, 1990). Indeed, there are several studies on the early talkers regarding their learning outcomes during school-age years (see, for example, Crain-Thoreson & Dale 1992; Lyytinen et al., 2001, 2004a, b; Scarborough and Dobrich, 1990). However, to the best of our knowledge, no study has investigated the communicative behaviors of early talkers at the initial stage of language development.

1.2 The present study

To sum up, there has been an increase in studies interested in identifying communicative and linguistic factors related to the variability of the child’s expressive language. However, most of these studies involved children older than 10 months of age. Based on the literature we have identified two categories of predictors that can be fruitfully investigated starting from the first months of life and can be considered key competences underlying the subsequent productive vocabulary size (McCathren, Yoder & Warren, 1999; Vihman, Macken, Miller, Simmons & Miller, 1985; Brooks & Meltzoff, 2005, 2008; Carpenter, Nagell, Tomasello, Butterworth & Moore, 1998; Bates, Thal & Marchman, 1991; Iverson & Thelen, 1999; Lyytinen, Eklund & Lyytinen, 2005; Thal, Marchman & Tomblin, 2013; Goldin-Meadow and Alibali, 2013). With direct observational measures we explored the number of vocal behaviors, gaze at mother’s face and gestures produced by the child in interaction with the caregiver. Moreover, we evaluated actions, gestures and the receptive vocabulary size using the “Primo Vocabolario del Bambino (PVB)” questionnaire (Caselli, Bello, Rinaldi, Stefanini, & Pasqualetti, 2015) that is the Italian version of the MacArthur-Bates CDI (Fenson, Marchman, Thal, Dale, Reznick, Bates, 2007) filled out by parents. We have defined these predictors as interactive behaviors, as all these variables emerged in the social scene between child and adult (Tomasello, 1995). Besides, we considered the role of gender to control for its potential effect, while also exploring its interplay and contribution with the set of interactive behaviors considered. Given the relatively small sample size and the explorative nature of the current study, we adopted a descriptive approach. We aim to investigate, between the 3rd and the 18th month of age, which interactive behaviors best predicted the probability to be at 24 months old a typical, late, or an early talker, and at which time point this may occur. If early intervention is to be provided in the critical period for language development, it is necessary to investigate and identify early predictors in younger children as early as possible. Hence the goal of the present study is to add information on an evolutionary period sparsely considered in the extant literature, and that is the study of developmental trajectories of early and late talkers starting from early months.

2.1 Participants

This study included 30 monolingual Italian mother–child dyads. The dyads were selected from a larger and longitudinal study on parenting and infant development in Italy at University of Chieti before the lockdown due to Covid-19. The vocabularies of 10 late talkers (LT), 10 typical talkers (TT) and 10 early talkers (ET) were selected out of a pool of 80 MB-CDI forms (Caselli, Bello, Rinaldi, Stefanini, & Pasqualetti, 2015) completed by parents for children (42 males and 38 females) aged 24 months. Specifically:

Mother’s mean age was 34.20 (SD = 5.35). Concerning socioeconomic status, 80 % of them had a medium economic status, and 20% had a low economic income (LT: 70% medium income; ET: 80% medium income; TT: 90% medium income). In relation to educational level, 66.6% of mothers had a high school degree or less, and 33.3% a bachelor’s or master’s degree (LT: bachelor’s degree 40%, high school degree 50%, secondary school degree 10%; ET: bachelor’s degree 30%, high school degree 60%, secondary school degree 10%; TT: bachelor’s degree 30%, high school degree 70%).

2.2 Procedure

Participants were recruited at an Italian urban hospital within 2 days from the child’s birth, and mothers were asked to join a research project on infant development. The mothers who agreed to be part of the research study were then contacted 1 to 2 months later and invited to come to the laboratory of Developmental Psychology. Only mothers gave us their willingness to participate to the experiment in lab. A written informed consent (approved by the ethical committee) for participation in the study has been obtained by the mothers. Before testing, each mother–child dyads participated in a familiarization phase in a room next to the proper observational room, equipped with age-specific toys. The mother-child dyads were observed during a video-recorded semi-structured play session conducted in the laboratory when the child was 3, 6, 9 and 12 months old. At 3 and 6 months, the mother and the infant were observed during a video-recorded interaction session. The mother sat in a straight-backed chair without arms and was asked to “play and talk to your baby as you would at home”. The child was in front of the mother sitting on a bouncer at 3 months and on a highchair at 6. The sessions were videotaped. The camera was controlled by an experimenter present in the room and placed behind the mother. It framed both the infant’s face and mother’s face, thanks to a mirror positioned behind the infant’s station. The total session lasted approximately 4 minutes. At 9 and 12 months, mother and child were filmed while playing with a standard set of toys. The age-appropriate toy set represented objects with which all children were familiar in their everyday routines and were provided with the specific aim of stimulating communicative and interactive exchanges. At 9 months, at the beginning of the interaction, we provided cubes to build a tower, while in the middle of the interaction we provide a book with illustrations of animals. At 12 months, we provided an illustrated booklet with figures of animals and a set of toy food. The sessions were videotaped. The camera was controlled by an experimenter present in the room with a camera placed in front of the mother and the child. The interaction lasted approximately 6 minutes.

At 15-, 18- and 24-months, the parents filled the questionnaire MB-CDI. The “Word and Gesture – WG” version was administered at 15 months, while the “Words and Sentences – WS” version at 18 and 24 months.

2.3 Measures

2.3.1 Interactive behaviors in the first year of life

Two trained transcribers, working independently one from the other and using CHAT (CHILDES system) format (MacWhinney, 2000), transcribed all the interactive behaviors at 3, 6, 9 and 12 months directly from the mother – child recorded interaction. The second transcriber was not informed of the subject’s group assignment until the transcription was completed. The interactive behaviors were coded with a multi-dimensional schema:

Vocal productions. Each production was counted as a single unit of analysis when separated by at least one second. Crying vocalizations and sounds of discomfort were not considered. The frequency of the child’s vocal productions per minute was calculated. Child’s vocal productions were coded as follows: vocalization (vowel sound typically produced open-mouthed), grunt (consonant sound produced by closed or semi closed mouth), onomatopoeia (sounds that reproduce the sound of animals, or the noise produced by vehicles or actions), babbling (syllabic sound consisting of a consonant and a vowel), proto words (sounds similar to words that take on specific meaning in certain contexts).

Gaze at mother’s face. The unit of analysis corresponds to a child’s gaze at mother’s face. Every time that the child looked at the mother’s face, the gaze was coded. Gaze at the mother’s body, hair, hands, and gaze to environment was not considered. The frequency of child’s gaze at mother’s face per minute was calculated.

Gestural productions. The unit of analysis corresponded to a child gestural production. Every gesture accompanied by other behaviors that signal the communicative intent (i.e., direct gaze to the interlocutor) was coded. The frequency of child gestural productions per minute was calculated. Gestural productions included deictic gestures: pointing (extending the index finger in the direction of an object, an event, or a person), giving (put an object towards the hand of the interlocutor), showing (put the object along the line of sight of the interlocutor). Moreover, we included representative gestures: iconic gestures (gestures which refer to objects, persons, or events, reproducing physical or functional characteristics), conventional (gestures with a culturally defined meaning and form e.g., waving bye, asking for silence pressing the index finger against the lips).

2.3.2 Reliability

Transcription reliability was calculated on 25% of the total videos. Point-by-point interactive behaviors comparisons were made, and the intercoder reliability ranged from 0,80 to 0,95. To maintain data consistency, only the material produced by the first transcriber was used for the final analysis. Inter-rater reliability for each predictor was computed by having a second rater that independently scored the subject performance on each measure. The inter-rater reliability was coded on 10% of the sample; the inter-coder reliability ranged from 0.83 to 0.94. Disagreements were solved with discussion to identify a unique score.

2.3.3 Interactive behaviors in the second year of life

The interactive behaviors in the second year of life at 15-, 18- and 24-months were assessed using the scores assigned by parents at the CDI. Specifically, we considered:

CDI Actions and Gestures. We considered the number of actions and gestures produced by the child reported by the mother at 15 months. This category included communicative gestures, games and routines, actions with objects, symbolic game (i.e., pretending to be mom or dad, imitating the actions of the adult, and playing pretend with objects). Data was missing for a male in the LT group for CDI actions at 15 months.

CDI Productive and Receptive Vocabulary Size. We considered the receptive vocabulary size of the child (the number of words understood by the child) as reported by the mother at 15 and 18 months. At 24 months, we considered the productive vocabulary size (the number of words produced by the child) which was used as a criterion for inclusion in one of the 3 groups (late, typical, and early talkers). The CDI Receptive vocabulary measure at 18 months was missing for two children in the LTs group (1 male, 1 female) and one female in the ETs group.

2.4 Analytic plan

For the analysis of data, given the relatively small sample size, we avoided adopting a Null-Hypothesis Significance Testing Approach, rather we commented on the effect sizes describing patterns of findings (Wasserstein et al., 2019). After having explored mean and standard deviation values as well as bivariate correlations among study variables, we ran and compared a series of multinomial logistic regression models including group membership (early, typical, and late talkers) as the outcome variable, and as predictors vocal production, gaze at mother’s face, gesture production, CDI actions and gestures, and CDI receptive vocabulary at the different time points investigated. We included one predictor at time in the regression models, to explore the differential effect of each variable across the different time points considered. As vocal production, gaze at mother’s face and gesture production were collected across multiple time points, we used Akaike Weights to identify what time point was more relevant for each specific variable. Akaike Weights represent a measure of the strength of adaptation for each model and the probability, conditioned to the set of models considered, that the same model can predict new data (McElreath, 2016; Vandekerckhove Matzke & Wagenmakers, 2015; Wagenmakers & Farrell, 2004). Akaike weights range from 0 to 1 and the higher the value, the more the model comparatively describes the data accurately. In each multinomial regression model, as groups were unbalanced regarding gender, we included gender as a predictor to simultaneously control for its impact as well as exploring its contribution. As the sample was relatively homogenous in terms of socio-economic and educational background, we did not include any other control variable in the model. Linear predicting probabilities of belonging to a specific group as a function of the predictors considered at different time points were plotted and graphically explored (Figure 1-5). Finally, as a follow-up analysis, for the best model selected we provided the probability of belong to a specific group for low (< 25°), medium (< 25° < 75°) and high (> 75°) levels of the predictor. All analyses were run with the statistical software R version 4.2.3, package ‘nnet (Ripley, Venables & Ripley, 2016) for multinomial regression analysis and gglpot2 (Wickham, 2016) for the graphical representation of findings.

Figure 1. Predicted vocal production probabilities for males and females at 3, 6, 9, and 12 months, plotted against vocal frequency.

3.1 Preliminary analyses

Means, SDs, and correlation values among variables of interest are reported in Table 1 and Table 2.

3.2 Multinomial regression models

Akaike Weights derived from the comparison of multinomial regression models, as reported in Table 3, suggested that early vocal production at 3 months of age was more able to predict subsequent 24 months group membership than vocal production at 6, 9, and 12 months. In relation to gaze at mother, the model receiving most support was the one including gaze at 6 months, suggesting that gaze at the mother in this specific developmental period was more predictive of language development than not gaze at 3, 9, and 12 months of age. Regarding the gestural production, its assessment at 12 months was more predictive of group membership than gestural production at 9 months. Hence, vocal production at 3 months, gaze at mother at 6 months, gestural production at 12 months comparatively emerged as the most informative markers of later vocabulary productions, and hence group membership, at 2 years of age. A more in-depth analyses of findings for each predictor is reported in the paragraphs below.

3.2.1 The role of child vocal production

As evidenced above, vocal production at 3 months of age was more predictive of group membership at 24 months compared to vocal production at months 6, 9 and 12. This result suggests that vocal productions that occur early in life are potentially first key predictors of verbal productivity at 24 months. More specifically, the graphical exploration of findings (see Figure 1) showed that the higher the vocal production at 3 months of age, the lower was the probability of belong to the LTs group with a stronger effect for males (see Figure 1) – which were still predominant in the LTs group. Predictive probabilities of belonging to each group for low (below the 25th percentile), medium (between the 25th and 75th percentile) and high (above the 75th percentile) levels of vocal production at 3 months (Table 4). The pattern identified at 3 months of age was comparable to that at 12 months, though – coherently with the fact that this model was less supported by Akaike weights - the effect was less strong. At 9 months of age, vocal production was not predictive of group membership while at 6 months the higher the vocal production, the higher the probability for females to be in the ET group. No predictive effect for LT belonging was evident.

3.2.2 The role of gaze at mother’s face

Concerning the role of gaze at mother Akaike weights (Table 3) suggested that a critical time point was 6 months, as depicted in Figure 2. Predicting probabilities of group membership for low, medium, and high values of gaze at mother’s face at 6 months are reported in the supplementary material. As it was for vocal reactivity at 3 months, the higher the number of gazes at mother at 6 months of age, the lower was the probability of belonging to the LTs group with the strongest effect for males. A comparable pattern emerged at 3 months, though with a less strong effect. At 9 and 12 months, gaze at mother’s face seems progressively less significant in predicting group membership, for both males and females. Hence, results suggest that gaze at mother at 6 months of age comparatively represents a better predictor (compared to gaze at mother at the other investigated time points) of language productivity at 24 months, and it might particularly help in identifying males that are more likely to become late talkers. The probability values of belonging to a group was then explored for the low, medium, and high values of the gaze at mother’s face at 6 months (Table 5).

Figure 2. Predicted gaze probabilities at mother's face for males and females at 3, 6, 9, and 12 months, against gaze frequency.

3.3.3 The role of child gestural production

For what pertains to gestural production, at 9 months few frequencies other than 0 emerged in the production of gestures, and the frequency per minute of gestures at 9 months seems to be relatively insignificant in predicting the group membership. A key time point for the prediction of group membership was age 12 months (see Table 3 and Figure 3). In males, as the frequency per minute of gestures produced increases, the probability of belonging to the group of late talkers decreases. For both males and females, the higher the gesture production at 12 months, the higher was the probability to be an early talker. As for the other variables, the probability values of belonging to a group was then explored for the low, medium, and high values of the gestural production at 12 months, (Table 6).

Figure 3. Predicted gestural production probabilities for males and females at 9 and 12 months, plotted against gestural frequency.

3.3.4 The role of CDI actions and gestures

In relation to actions and gestures, (Figure 4), results suggested that the higher the number of gestures and actions at 15 months, the higher the probability of being in the ETs, and the lower the probability of being in the LTs, with a comparable trend for males and females. The probability values of belonging to a group was explored for the medium and high values of the actions and gestures model at 15 months (Table 7).

Figure 4. Predicted CDI action and gesture probabilities at 15 months for males and females, plotted against count.

3.3.5 The role of CDI receptive vocabulary size

Finally, we investigated the role of receptive vocabulary at 15 and 18 months. As reported in the plot of predicted probabilities (Figure 5), as the receptive-vocabulary size increases, the probability of being ETs increases too, with a similar trend for both male and female, while the probability of belonging to TTs and LTs decreases. The probability analysis of receptive vocabulary size is in Table 8. The probabilities of belonging to the group for low, medium, and high values of the variables of interest were then explored, considering the mean of the receptive-vocabulary size at 15 and 18 months when possible, or only one time point if data were available for only one assessment phase (Figure 6).

Figure 5. Predicted CDI receptive-vocabulary probabilities at 15 and 18 months for males and females, against count.

Figure 6. Plot variable correlation.

Note: V is the abbreviation of Vocal at 3, 6, 9 and 12 months; G is the abbreviation of Gaze at 3, 6, 9 and 12 months, GE is the abbreviation of gestural production at 9 and 12 months, AG is the abbreviation of MB-CDI at 15 months, RV is the abbreviation of MB-CDI Receptive Vocabulary size at 15 and 18 months, and PV is the abbreviation of MB-CDI Productive Vocabulary size at 24 months.