Explainability of neural networks for child language: Agent-First strategy in comprehension of Korean active transitive construction

This study investigates how neural networks address the properties of children's linguistic knowledge, with a focus on the Agent-First strategy in comprehension of an active transitive construction in Korean. We develop various neural-network models and measure their classification performance on the test stimuli used in a behavioural experiment involving scrambling and omission of sentential components at varying degrees. Results show that, despite some compatibility of these models’ performance with the children's response patterns, their performance does not fully approximate the children's utilisation of this strategy, demonstrating by-model and by-condition asymmetries. This study's findings suggest that neural networks can utilise information about formal co-occurrences to access the intended message to a certain degree, but the outcome of this process may be substantially different from how a child (as a developing processor) engages in comprehension. This implies some limits of neural networks on revealing the developmental trajectories of child language.

Research Highlights

• This study investigates how neural networks address properties of child language.
• We focus on the Agent-First strategy in comprehension of Korean active transitive.
• Results show by-model/condition asymmetries against children's response patterns.
• This implies some limits of neural networks on revealing properties of child language.

     

Prosodic modulations in child-directed language and their impact on word learning

Child-directed language can support language learning, but how? We addressed two questions: (1) how caregivers prosodically modulated their speech as a function of word familiarity (known or unknown to the child) and accessibility of referent (visually present or absent from the immediate environment); (2) whether such modulations affect children's unknown word learning and vocabulary development. We used data from 38 English-speaking caregivers (from the ECOLANG corpus) talking about toys (both known and unknown to their children aged 3–4 years) both when the toys are present and when absent. We analyzed prosodic dimensions (i.e., speaking rate, pitch and intensity) of caregivers’ productions of 6529 toy labels. We found that unknown labels were spoken with significantly slower speaking rate, wider pitch and intensity range than known labels, especially in the first mentions, suggesting that caregivers adjust their prosody based on children's lexical knowledge. Moreover, caregivers used slower speaking rate and larger intensity range to mark the first mentions of toys that were physically absent. After the first mentions, they talked about the referents louder with higher mean pitch when toys were present than when toys were absent. Crucially, caregivers’ mean pitch of unknown words and the degree of mean pitch modulation for unknown words relative to known words (pitch ratio) predicted children's immediate word learning and vocabulary size 1 year later. In conclusion, caregivers modify their prosody when the learning situation is more demanding for children, and these helpful modulations assist children in word learning.

Research Highlights

• In naturalistic interactions, caregivers use slower speaking rate, wider pitch and intensity range when introducing new labels to 3–4-year-old children, especially in first mentions.
• Compared to when toys are present, caregivers speak more slowly with larger intensity range to mark the first mentions of toys that are physically absent.
• Mean pitch to mark word familiarity predicts children's immediate word learning and future vocabulary size.

     

Does input influence uptake? Links between maternal talk,processing speed and vocabulary size in Spanish‐learning children

It is well established that variation in caregivers’ speech is associated with language outcomes, yet little is known about the learning principles that mediate these effects. This longitudinal study (n = 27) explores whether Spanish‐learning children's early experiences with language predict efficiency in real‐time comprehension and vocabulary learning. Measures of mothers’ speech at 18 months were examined in relation to children's speech processing efficiency and reported vocabulary at 18 and 24 months. Children of mothers who provided more input at 18 months knew more words and were faster in word recognition at 24 months. Moreover, multiple regression analyses indicated that the influences of caregiver speech on speed of word recognition and vocabulary were largely overlapping. This study provides the first evidence that input shapes children's lexical processing efficiency and that vocabulary growth and increasing facility in spoken word comprehension work together to support the uptake of the information that rich input affords the young language learner.     

Can Toddlers Learn Vocabulary from Television? An Experimental Approach

This study was inspired by the rise in television targeting toddlers and preverbal infants (e.g., Teletubbies, Baby Mozart). Overall, we investigated if very young children who are in the early stages of language acquisition can learn vocabulary quickly (fast map) from television programs. Using a fast mapping paradigm, this study examined a group (n = 48) of toddlers (15–24 months) and their ability to learn novel words. Utilizing a repeated measures design, we compared children's ability to learn various novel words in 5 different conditions. These included the presentation and identification of a novel word by an adult speaker via live presentation when the toddler was attending (i.e., joint reference), an adult via live presentation when the toddler was not attending, an adult speaker on television, and an edited clip from a children's television program (Teletubbies). Overall, the toddlers were most successful in learning novel words in the joint reference condition. They were significantly less successful in the children's program condition. Furthermore, there was a significant interaction between age and condition on children's performance. Both younger (15–21 months) and older (22–24 months) participants identified the target objects when they were taught the novel word by an adult speaker; however, it appeared that children under the age of 22 months did not identify the target item when they were taught the novel word via the television program.     

Spanish-English bilingual toddlers’ vocabulary skills: The role of caregiver language input and warmth

There is a well-documented link between bilingual language development and the relative amounts of exposure to each language. Less is known about the role of quality indicators of caregiver-child interactions in bilingual homes, including caregiver input diversity, warmth and sensitivity. This longitudinal study examines the relation between caregiver input (lexical diversity, amount), warmth and sensitivity and bilingual toddlers’ subsequent vocabulary outcomes. We video-recorded caregiver-child interactions in Spanish-English Latino homes when toddlers (n = 47) were 18 months of age (M = 18.32 months; SD = 1.02 months). At the 24-month follow-up, we measured children's vocabulary as total vocabulary (English, Spanish combined) as well as within language (Spanish, English). Results revealed that Spanish lexical diversity exposure at 18 months from caregivers was positively associated with children's Spanish and total vocabulary scores at 24 months, while English lexical diversity was positively associated with children's English scores; lexical diversity and amount were highly correlated. Additionally, caregivers’ warmth was positively associated with children's Spanish, English and total vocabulary scores. Together, these factors accounted for substantial variance (30–40%) in vocabulary outcomes. Notably, caregiver input accounted for more variance in single language outcomes than did caregiver warmth, whereas caregiver warmth uniquely accounted for more variance in total vocabulary scores. Our findings extend prior research findings by suggesting that children's dual language development may depend on their exposure to a diverse set of words, not only amount of language exposure, as well as warm interactions with caregivers. A video abstract of this article can be viewed at https://youtu.be/q1V_7fz5wog

Highlights

• Video-recorded observations of caregiver-child interactions revealed warmth and high sensitivity from Latino caregivers.
• Linguistically-detailed analyses of caregiver input revealed wide variation in the diversity of Spanish and English directed at 18-month-old bilingual toddlers.
• Bilingual toddlers’ vocabulary (single language, total) was positively associated with caregivers’ diverse input and warmth, thus extending prior findings on bilinguals’ amount of language exposure.
• Findings suggest that caregivers’ lexical diversity explains more variance in bilingual toddlers’ single language outcomes, whereas warmth explains more variance in total vocabulary scores.

     

Spanish-speaking caregivers’ use of referential labels with toddlers is a better predictor of later vocabulary than their use of referential gestures

Variation in how frequently caregivers engage with their children is associated with variation in children's later language outcomes. One explanation for this link is that caregivers use both verbal behaviors, such as labels, and non-verbal behaviors, such as gestures, to help children establish reference to objects or events in the world. However, few studies have directly explored whether language outcomes are more strongly associated with referential behaviors that are expressed verbally, such as labels, or non-verbally, such as gestures, or whether both are equally predictive. Here, we observed caregivers from 42 Spanish-speaking families in the US engage with their 18-month-old children during 5-min lab-based, play sessions. Children's language processing speed and vocabulary size were assessed when children were 25 months. Bayesian model comparisons assessed the extent to which the frequencies of caregivers’ referential labels, referential gestures, or labels and gestures together, were more strongly associated with children's language outcomes than a model with caregiver total words, or overall talkativeness. The best-fitting models showed that children who heard more referential labels at 18 months were faster in language processing and had larger vocabularies at 25 months. Models including gestures, or labels and gestures together, showed weaker fits to the data. Caregivers’ total words predicted children's language processing speed, but predicted vocabulary size less well. These results suggest that the frequency with which caregivers of 18-month-old children use referential labels, more so than referential gestures, is a critical feature of caregiver verbal engagement that contributes to language processing development and vocabulary growth.

Research Highlights

• We examined the frequency of referential communicative behaviors, via labels and/or gestures, produced by caregivers during a 5-min play interaction with their 18-month-old children.
• We assessed predictive relations between labels, gestures, their combination, as well as total words spoken, and children's processing speed and vocabulary growth at 25 months.
• Bayesian model comparisons showed that caregivers’ referential labels at 18 months best predicted both 25-month vocabulary measures, although total words also predicted later processing speed.
• Frequent use of referential labels by caregivers, more so than referential gestures, is a critical feature of communicative behavior that supports children's later vocabulary learning.

     

Knowing who knows: Metacognitive and causal learning abilities guide infants’ selective social learning

Given the widespread interest in the development of children's selective social learning, there is mounting evidence suggesting that infants prefer to learn from competent informants (Poulin‐Dubois & Brosseau‐Liard, Current Directions in Psychological Science, 2016, 25). However, little research has been dedicated to understanding how this selectivity develops. The present study investigated whether causal learning and precursor metacognitive abilities govern discriminant learning in a classic word‐learning paradigm. Infants were exposed to a speaker who accurately (reliable condition) or inaccurately (unreliable condition) labeled familiar objects and were subsequently tested on their ability to learn a novel word from the informant. The predictive power of causal learning skills and precursor metacognition (as measured through decision confidence) on infants' word learning was examined across both reliable and unreliable conditions. Results suggest that infants are more inclined to accept an unreliable speaker's testimony on a word learning task when they also lack confidence in their own knowledge on a task measuring their metacognitive ability. Additionally, when uncertain, infants draw on causal learning abilities to better learn the association between a label and a novel toy. This study is the first to shed light on the role of causal learning and precursor metacognitive judgments in infants' abilities to engage in selective trust.     

Children learn words easier when they are interested in the category to which the word belongs

The overall pattern of vocabulary development is relatively similar across children learning different languages. However, there are considerable differences in the words known to individual children. Historically, this variability has been explained in terms of differences in the input. Here, we examine the alternate possibility that children's individual interest in specific natural categories shapes the words they are likely to learn – a child who is more interested in animals will learn a new animal name easier relative to a new vehicle name. Two‐year‐old German‐learning children (N = 39) were exposed to four novel word–object associations for objects from four different categories. Prior to the word learning task, we measured their interest in the categories that the objects belonged to. Our measure was pupillary change following exposure to familiar objects from these four categories, with increased pupillary change interpreted as increased interest in that category. Children showed more robust learning of word–object associations from categories they were more interested in relative to categories they were less interested in. We further found that interest in the novel objects themselves influenced learning, with distinct influences of both category interest and object interest on learning. These results suggest that children's interest in different natural categories shapes their word learning. This provides evidence for the strikingly intuitive possibility that a child who is more interested in animals will learn novel animal names easier than a child who is more interested in vehicles.     

Adjacent and Non-Adjacent Word Contexts Both Predict Age of Acquisition of English Words: A Distributional Corpus Analysis of Child-Directed Speech

Children show a remarkable degree of consistency in learning some words earlier than others. What patterns of word usage predict variations among words in age of acquisition? We use distributional analysis of a naturalistic corpus of child-directed speech to create quantitative features representing natural variability in word contexts. We evaluate two sets of features: One set is generated from the distribution of words into frames defined by the two adjacent words. These features primarily encode syntactic aspects of word usage. The other set is generated from non-adjacent co-occurrences between words. These features encode complementary thematic aspects of word usage. Regression models using these distributional features to predict age of acquisition of 656 early-acquired English words indicate that both types of features improve predictions over simpler models based on frequency and appearance in salient or simple utterance contexts. Syntactic features were stronger predictors of children's production than comprehension, whereas thematic features were stronger predictors of comprehension. Overall, earlier acquisition was predicted by features representing frames that select for nouns and verbs, and by thematic content related to food and face-to-face play topics; later acquisition was predicted by features representing frames that select for pronouns and question words, and by content related to narratives and object play.     

What Is the Buzz About Iconicity? How Iconicity in Caregiver Speech Supports Children's Word Learning

One cue that may facilitate children's word learning is iconicity, or the correspondence between a word's form and meaning. Some have even proposed that iconicity in the early lexicon may serve to help children learn how to learn words, supporting the acquisition of even noniconic, or arbitrary, word–referent associations. However, this proposal remains untested. Here, we investigate the iconicity of caregivers’ speech to young children during a naturalistic free-play session with novel stimuli and ask whether the iconicity of caregivers’ speech facilitates children's learning of the noniconic novel names of those stimuli. Thirty-four 1.5-2-year-olds (19 girls; half monolingual English learners and half bilingual English-Spanish learners) participated in a naturalistic free-play task with their caregivers followed by a test of word-referent retention. We found that caregivers’ use of iconicity, particularly in utterances in which they named the novel stimuli, was associated with the likelihood that children learned that novel name. This result held even when controlling for other factors associated with word learning, such as the concreteness and frequency of words in caregiver speech. Together, the results demonstrate that iconicity not only can serve to help children identify the referent of novel words (as in previous research) but can also support their ability to retain even noniconic word-referent mappings.     

Probing Lexical Ambiguity: Word Vectors Encode Number and Relatedness of Senses

Lexical ambiguity—the phenomenon of a single word having multiple, distinguishable senses—is pervasive in language. Both the degree of ambiguity of a word (roughly, its number of senses) and the relatedness of those senses have been found to have widespread effects on language acquisition and processing. Recently, distributional approaches to semantics, in which a word's meaning is determined by its contexts, have led to successful research quantifying the degree of ambiguity, but these measures have not distinguished between the ambiguity of words with multiple related senses versus multiple unrelated meanings. In this work, we present the first assessment of whether distributional meaning representations can capture the ambiguity structure of a word, including both the number and relatedness of senses. On a very large sample of English words, we find that some, but not all, distributional semantic representations that we test exhibit detectable differences between sets of monosemes (unambiguous words; N = 964), polysemes (with multiple related senses; N = 4,096), and homonyms (with multiple unrelated senses; N = 355). Our findings begin to answer open questions from earlier work regarding whether distributional semantic representations of words, which successfully capture various semantic relationships, also reflect fine-grained aspects of meaning structure that influence human behavior. Our findings emphasize the importance of measuring whether proposed lexical representations capture such distinctions: In addition to standard benchmarks that test the similarity structure of distributional semantic models, we need to also consider whether they have cognitively plausible ambiguity structure.     

Pragmatics aid referent disambiguation and word learning in young children and adults

How do children succeed in learning a word? Research has shown robustly that, in ambiguous labeling situations, young children assume novel labels to refer to unfamiliar rather than familiar objects. However, ongoing debates center on the underlying mechanism: Is this behavior based on lexical constraints, guided by pragmatic reasoning, or simply driven by children's attraction to novelty? Additionally, recent research has questioned whether children's disambiguation leads to long-term learning or rather indicates an attentional shift in the moment of the conversation. Thus, we conducted a pre-registered online study with 2- and 3-year-olds and adults. Participants were presented with unknown objects as potential referents for a novel word. Across conditions, we manipulated whether the only difference between both objects was their relative novelty to the participant or whether, in addition, participants were provided with pragmatic information that indicated which object the speaker referred to. We tested participants’ immediate referent selection and their retention after 5 min. Results revealed that when given common ground information both age groups inferred the correct referent with high success and enhanced behavioral certainty. Without this information, object novelty alone did not guide their selection. After 5 min, adults remembered their previous selections above chance in both conditions, while children only showed reliable learning in the pragmatic condition. The pattern of results indicates how pragmatics may aid referent disambiguation and learning in both adults and young children. From early ontogeny on, children's social-cognitive understanding may guide their communicative interactions and support their language acquisition.

Research Highlights

• We tested how 2-3-year-olds and adults resolve referential ambiguity without any lexical cues.
• In the pragmatic context both age groups disambiguated novel word-object-mappings, while object novelty alone did not guide their referent selection.
• In the pragmatic context, children also showed increased certainty in disambiguation and retained new word-object-mappings over time.
• These findings contribute to the ongoing debate on whether children learn words on the basis of domain-specific constraints, lower-level associative mechanisms, or pragmatic inferences.

     

Parent coaching at 6 and 10 months improves language outcomes at 14 months: A randomized controlled trial

Previous studies reveal an association between particular features of parental language input and advances in children's language learning. However, it is not known whether parent coaching aimed to enhance specific input components would (a) successfully increase these components in parents' language input and (b) result in concurrent increases in children's language development. The present randomized controlled trial assigned families of typically developing 6‐month‐old infants to Intervention (parent coaching) and Control (no coaching) groups. Families were equivalent on socioeconomic status, infants' gender, and infants' age. Parent coaching took place when infants were 6 and 10 months of age, and included quantitative and qualitative linguistic feedback on the amount of child‐directed speech, back‐and‐forth interactions, and parentese speech style. These variables were derived from each family's first‐person LENA recordings at home. Input variables and infant language were measured at 6, 10, and 14 months. Parent coaching significantly enhanced language input as measured by two social interaction variables: percentage of speech directed to the child and percentage of parentese speech. These two variables were correlated, and were both related to growth in infant babbling between 6 and 14 months. Intervention infants showed greater growth in babbling than Control infants. Furthermore, at 14 months, Intervention infants produced significantly more words than Control infants, as indicated by LENA recordings and parent report via the MacArthur‐Bates Communicative Developmental Inventory. Together, these results indicate that parent coaching can enrich specific aspects of parental language input, and can immediately and positively impact child language outcomes. A video abstract of this article can be viewed at: https://youtu.be/7wqR28gPiwo     

Foreign accents reduce false recognition rates in the DRM paradigm

ABSTRACT

More cognitive resources are required to comprehend foreign-accented than native speech. Focusing these cognitive resources on resolving the acoustic mismatch between the foreign-accented input and listeners’ stored representations of spoken words can affect other cognitive processes. Across two studies, we explored whether processing foreign-accented speech reduces the activation of semantic information. This was achieved using the DRM paradigm, in which participants study word lists and typically falsely remember non-studied words (i.e. critical lures) semantically associated with the studied words. In two experiments, participants were presented with word lists spoken both by a native and a foreign-accented speaker. In both experiments we observed lower false recognition rates for the critical lures associated with word lists presented in a foreign accent, compared to native speech. In addition, participants freely recalled more studied words when they had been presented in a native, compared to a foreign, accent, although this difference only emerged in Experiment 2, where the foreign speaker had a very strong accent. These observations suggest that processing foreign-accented speech modulates the activation of semantic information.

Highlights

• The DRM paradigm was used to explore whether semantic activation is reduced when processing foreign-accented speech.

• Across two experiments, false recognition of non-studied semantic associates was lower when word lists were presented in a foreign accent, compared to native speech.

• The above results suggest semantic activation may be reduced when processing foreign-accented speech.

• Additionally, it was found that when the foreign speaker had a mild accent, correct recall of studied words was uninfluenced. If the foreign speaker had a strong accent, however, correct recall of studied words was reduced.

     

Representing word meaning and order information in a composite holographic lexicon

The authors present a computational model that builds a holographic lexicon representing both word meaning and word order from unsupervised experience with natural language. The model uses simple convolution and superposition mechanisms (cf. B. B. Murdock, 1982) to learn distributed holographic representations for words. The structure of the resulting lexicon can account for empirical data from classic experiments studying semantic typicality, categorization, priming, and semantic constraint in sentence completions. Furthermore, order information can be retrieved from the holographic representations, allowing the model to account for limited word transitions without the need for built-in transition rules. The model demonstrates that a broad range of psychological data can be accounted for directly from the structure of lexical representations learned in this way, without the need for complexity to be built into either the processing mechanisms or the representations. The holographic representations are an appropriate knowledge representation to be used by higher order models of language comprehension, relieving the complexity required at the higher level.     

Exploring What Is Encoded in Distributional Word Vectors: A Neurobiologically Motivated Analysis

The pervasive use of distributional semantic models or word embeddings for both cognitive modeling and practical application is because of their remarkable ability to represent the meanings of words. However, relatively little effort has been made to explore what types of information are encoded in distributional word vectors. Knowing the internal knowledge embedded in word vectors is important for cognitive modeling using distributional semantic models. Therefore, in this paper, we attempt to identify the knowledge encoded in word vectors by conducting a computational experiment using Binder et al.'s (2016) featural conceptual representations based on neurobiologically motivated attributes. In an experiment, these conceptual vectors are predicted from text-based word vectors using a neural network and linear transformation, and prediction performance is compared among various types of information. The analysis demonstrates that abstract information is generally predicted more accurately by word vectors than perceptual and spatiotemporal information, and specifically, the prediction accuracy of cognitive and social information is higher. Emotional information is also found to be successfully predicted for abstract words. These results indicate that language can be a major source of knowledge about abstract attributes, and they support the recent view that emphasizes the importance of language for abstract concepts. Furthermore, we show that word vectors can capture some types of perceptual and spatiotemporal information about concrete concepts and some relevant word categories. This suggests that language statistics can encode more perceptual knowledge than often expected.     

Look who's talking: speech style and social context in language input to infants are linked to concurrent and future speech development

Language input is necessary for language learning, yet little is known about whether, in natural environments, the speech style and social context of language input to children impacts language development. In the present study we investigated the relationship between language input and language development, examining both the style of parental speech, comparing ‘parentese’ speech to standard speech, and the social context in which speech is directed to children, comparing one‐on‐one (1:1) to group social interactions. Importantly, the language input variables were assessed at home using digital first‐person perspective recordings of the infants' auditory environment as they went about their daily lives (N =26, 11‐ and 14‐months‐old). We measured language development using (a) concurrent speech utterances, and (b) word production at 24 months. Parentese speech in 1:1 contexts is positively correlated with both concurrent speech and later word production. Mediation analyses further show that the effect of parentese speech‐1:1 on infants' later language is mediated by concurrent speech. Our results suggest that both the social context and the style of speech in language addressed to children are strongly linked to a child's future language development.     

Learning Proper Names and Count Nouns: Evidence From 16- and 20-Month-Olds

In 4 experiments, we examined 16- and 20-month-old infants' understanding of proper names and count nouns. In each experiment, infants were taught a novel word modeled linguistically as either a proper name (e.g., "DAXY") or a count noun (e.g., "a DAXY") for a stuffed animal shown on a puppet stage. This animal was moved to a new location on the stage, and a second identical-looking animal was placed in the original animal's starting location. We then assessed infants' looking behavior in response to the word they had heard. At 20 months (Experiments 1 and 3), but not at 16 months (Experiments 2 and 4), infants were significantly more likely to look first at the original object in response to hearing the word in the proper name condition than in the count noun condition or in a baseline condition in which they heard no word. The results suggest that distinct and appropriate form–meaning links involving proper names and count nouns emerge in children's language between 16 and 20 months.     

Numerical morphology supports early number word learning: Evidence from a comparison of young Mandarin and English learners

Previous studies showed that children learning a language with an obligatory singular/plural distinction (Russian and English) learn the meaning of the number word for one earlier than children learning Japanese, a language without obligatory number morphology (Barner, Libenson, Cheung, & Takasaki, 2009; Sarnecka, Kamenskaya, Yamana, Ogura, & Yudovina, 2007). This can be explained by differences in number morphology, but it can also be explained by many other differences between the languages and the environments of the children who were compared. The present study tests the hypothesis that the morphological singular/plural distinction supports the early acquisition of the meaning of the number word for one by comparing young English learners to age and SES matched young Mandarin Chinese learners. Mandarin does not have obligatory number morphology but is more similar to English than Japanese in many crucial respects. Corpus analyses show that, compared to English learners, Mandarin learners hear number words more frequently, are more likely to hear number words followed by a noun, and are more likely to hear number words in contexts where they denote a cardinal value. Two tasks show that, despite these advantages, Mandarin learners learn the meaning of the number word for one three to six months later than do English learners. These results provide the strongest evidence to date that prior knowledge of the numerical meaning of the distinction between singular and plural supports the acquisition of the meaning of the number word for one.