Cross-situational and ostensive word learning in children with and without autism spectrum disorder

Abstract

Numerous experimental studies have shown that infants and children can discover word meanings by using co-occurrences between labels and objects across individually ambiguous contexts—a phenomenon known as cross-situational learning. Like typically developing children, high-functioning school aged children with autism spectrum disorder (ASD) are capable of cross-situational learning. However, it is not yet clear whether cross-situational learning is similarly available to children with ASD who are younger and show a broader range of language and cognitive abilities. Using eye-tracking methodology, the current study provided the first evidence that preschool and early school-aged children with ASD can rely on cross-situational statistics to learn new words. In fact, children with ASD learned as well as typically developing children with similar vocabulary knowledge. In both groups, the children with the highest cross-situational learning accuracy were those who showed the best familiar word processing skills. Surprisingly, children in both groups learned words equally well in the cross-situational task and an ostensive word-learning task, which presented only a single label-object pairing at a time. In combination, these results point to similarities in the word learning abilities available to typically developing children and children with ASD.

Introduction

Word learning is a crucial part of language development, but identifying the meanings of new words is not always easy. This is true even in the case of object nouns—arguably one of the most straightforward cases of mapping word (label) to meaning (object). Determining correct label-object mappings can be difficult because there are often many things in the environment that a novel word could describe (Quine, 1960). Although children sometimes have access to adult cues that explicitly identify which object is being labeled (e.g., pointing, gaze), they also encounter many instances in which explicit cues are not available. In the absence of such cues, how might children determine the meanings of object nouns in a busy world filled with many things to see and hear? One possibility is that children discover word meanings by using co-occurrences between labels and objects across individually ambiguous contexts—a phenomenon known as cross-situational learning (Smith and Yu, 2008, Suanda et al., 2014, Yu and Smith, 2007). Even when the links between objects and labels are not immediately clear, these links may become clear as the same object and label co-occur over time. For example, if the words ‘duck’ and ‘ball’ are first presented in the absence of explicit cues that identify their referents, the learner may not yet know which label describes which object. However, it is possible to determine the correct label-object associations by attending to co-occurrences over time (e.g., that the round object is consistently visible when the word ‘ball’ is produced).

Experimental studies have shown that adults, young children, and even infants are capable of using this type of cross-situational information to learn the meanings of new words (Smith and Yu, 2008, Suanda et al., 2014, Vouloumanos and Werker, 2009, Yu and Smith, 2007). In a landmark study by Smith and Yu (2008), typically developing 12- and 14-month-old infants were exposed to a series of individually ambiguous trials, each of which presented two novel labels (e.g., bosa, gasser) and two unfamiliar objects. Within a given trial, no information was available to indicate which label-object associations were correct. The only way for infants to discover correct associations was by picking up on co-occurrence statistics—namely, which object was consistently visible when a given label was presented. Each label-object pairing occurred a total of 10 times during teaching. After less than 4 min, infants’ eye movements provided evidence that they had learned the new words on the basis of cross-situational statistics alone.

Over the past decade, cross-situational learning has continued to receive a great deal of empirical attention in typically developing infants, children, and adults, primarily in terms of the factors that affect it and the learning mechanisms that underlie it (Medina et al., 2011, Suanda and Namy, 2012, Trueswell et al., 2013, Vlach and Johnson, 2013, Yu and Smith, 2012). One current topic of considerable debate pertains to precisely how cross-situational learning is accomplished. Do learners gradually accumulate statistical associations between labels and objects (Smith and Yu, 2008, Yu and Smith, 2007, Zettersten et al., 2018), or do they propose and subsequently verify (or abandon) a single word-referent pairing (Smith et al., 2009, Trueswell et al., 2013, Woodard et al., 2016)? Or, does their learning falls somewhere in between associative learning and hypothesis testing, depending on contextual factors and individual abilities (MacDonald et al., 2017, Yurovsky and Frank, 2015)? Although this is a critically important issue, particularly in atypical development, the current study was not designed to differentiate among these possibilities and thus we do not discuss them extensively here. However, we return to the issue of gradual associative learning versus hypothesis testing in the Discussion, presenting potential interpretations of the current findings and avenues for future research.

Despite our growing understanding of cross-situational learning in typical development, we know very little about cross-situational learning in children with autism spectrum disorder (ASD)—a neurodevelopmental disorder characterized by deficits in social communication, repetitive behaviors and restricted interests (American Psychiatric Association, 2013), and wide variability in social, language, and cognitive abilities. Because of their unique behavioral profile, children with ASD present an opportunity to better understand the skills that support cross-situational learning, as well as the role that cross-situational learning plays in language development more generally. Investigating cross-situational learning in children with ASD will also shed light on the role of auditory-visual integration in statistical word learning. Numerous studies have identified differences in the way in which individuals with ASD integrate auditory and visual information (Iarocci & McDonald, 2006; but see Grossman, Schneps, & Tager-Flusberg, 2009). For example, visual information may have less of an influence on heard speech individuals with ASD than in individuals without ASD (Irwin, Tornatore, & Brancazio, & Whalen, 2012). In addition, individuals with ASD show “reduced multisensory speech perception for matched audiovisual stimuli” (Woynaroski et al., 2013, p. 2900), as well as a decreased preference for auditory-visual synchrony (Bebko et al., 2006, Grossman et al., 2015). Although it is not typically described in this way, associating auditory labels with visual objects across contexts can be considered a form of auditory-visual integration. Thus, cross-situational learning might be particularly difficult for children with ASD compared to TD children, particularly for children with ASD and co-occurring language delays.

To our knowledge, only one published study has investigated cross-situational learning in children with ASD. McGregor, Rost, Arenas, Farris-Trimble, and Stiles (2013) found that high-functioning 11-year-olds with ASD used cross-situational statistics to learn new words, thereby providing the first evidence that cross-situational learning mechanisms are available to at least some children with ASD. Furthermore, the children who were better cross-situational learners were those who had acquired stronger vocabulary skills outside the lab setting, leading McGregor and colleagues to hypothesize that cross-situational learning is more difficult for children with ASD who have weaker language abilities overall.

The findings of the study by McGregor et al. (2013) advance our understanding of cross-situational learning in several ways. First, the fact that children with ASD learned words based on cross-situational statistics alone demonstrates that cross-situational learning can operate successfully in the absence of strong social communication skills. Second, the correlation with vocabulary skills (but not age) suggests that cross-situational learning may play an important role in vocabulary development for children with ASD. However, the study by McGregor and colleagues included only older children with age-appropriate language and cognitive skills, leaving open the question of whether cross-situational learning is similarly available to children with ASD who are younger and show a broader range of language and cognitive abilities. In addition, it is not yet clear whether the link between cross-situational learning and language is specific, or whether it can be better explained by overall cognitive skills.

One critical issue that remains unexplored across both typical development and ASD is how cross-situational learning compares to word learning in ostensive (unambiguous) contexts that present only one word and one object at a time. From an information processing perspective, it is logical that tracking and integrating multiple co-occurrences across ambiguous contexts may be more difficult than associating a single word and with a single object. However, we are not aware of any published studies that have directly compared cross-situational and ostensive word learning in the same group of children. This is an important issue to investigate because it has implications for understanding how easily children can access particular learning mechanisms, what additional cognitive demands are placed on children in certain contexts, and what word-learning strategies children are most likely to rely on in everyday situations.

Research on language development in children with ASD has historically focused on characterizing these children’s existing language skills—namely, describing the delays they experience in certain domains and at certain points in development. The current study is part of a new line of work that emphasizes not what language skills children with ASD have acquired, but how they acquire these skills (Arunachalam and Luyster, 2016, Mayo and Eigsti, 2012, McGregor et al., 2013, Naigles et al., 2011, Venker et al., 2016). In the current study, we focus not only on the words that children know, but also on the word-learning abilities these children can access when exposed to words they have never heard before. Adopting this type of learning-based approach is beneficial because it allows us to answer new questions about language development in children with ASD, including: what learning mechanisms are intact, and where are the points of breakdown? In addition, focusing on a population with considerable variation in language and cognitive skills will shed light on the mechanisms that support word learning more generally, thereby complementing investigations of TD children.

The current study investigated cross-situational word learning in children with ASD (4–7 years old) and TD children (2–7 years old), matched on vocabulary knowledge. Children took part in a cross-situational eye-tracking task modeled after the task developed by Smith and Yu (2008). Each teaching trial presented two novel labels and two objects, with no information available within a single trial regarding which label described which object. Children also completed an ostensive word-learning task that presented label-object links explicitly (i.e., one label and one object at a time) and therefore did not require children to track multiple co-occurrences to determine word meaning.

We had three goals. The first was to determine whether young children with ASD were capable of cross-situational learning, and whether their learning differed from TD children matched on vocabulary knowledge. Given the general difficulties children with ASD have learning words (Charman et al., 2003, Luyster et al., 2007, McDuffie et al., 2012) and integrating auditory and visual information (e.g., Foss-Feig et al., 2010, Iarocci and McDonald, 2006), as well as the fact that our sample included children with language delays, we hypothesized that children with ASD would learn words in the cross-situational task but that they would show deficits relative to TD children. Our second goal was to determine whether individual differences in cross-situational learning were related to children’s language skills in either group. Given previous evidence of a relationship between cross-situational learning and language (McGregor et al., 2013, Smith and Yu, 2013), we hypothesized that children with stronger language skills would be better cross-situational learners—regardless of diagnostic group. Such a finding would point to similarities in the skills that may support cross-situational learning across different populations. Our third goal was to determine whether cross-situational word learning was more difficult than ostensive word learning for TD children or children with ASD. We hypothesized that children in both groups would learn words more easily in the ostensive word-learning task than in the cross-situational task, since the ostensive task presented only a single label and object at a time (whereas the cross-situational task presented two labels and objects in each trial).