Lexical Development in Young Children With Autism Spectrum Disorder (ASD): How ASD May Affect Intake From the Input

Abstract

Purpose

Most children with autism spectrum disorder (ASD) have below-age lexical knowledge and lexical representation. Our goal is to examine ways in which difficulties with social communication and language processing that are often associated with ASD may constrain these children's abilities to learn new words and to explore whether minimizing the social communication and processing demands of the learning situation can lead to successful learning.

Method

In this narrative review of recent work on lexical development in ASD, we describe key findings on children's acquisition of nouns, pronouns, and verbs and outline our research program currently in progress aimed at further elucidating these issues.

Conclusion

Our review of studies that examine lexical development in children with ASD suggests that innovative intervention approaches that take into account both the social communication and processing demands of the learning situation may be particularly beneficial.

Presentation Video

https://doi.org/10.23641/asha.7324013

This research forum contains papers from the 2017 Research Symposium at the ASHA Convention held in Los Angeles, California.

Although not a defining feature of autism spectrum disorder (ASD), most children with ASD have below-age skills in structural language, or core language abilities including grammatical and lexical knowledge (e.g., Charman, Drew, Baird, & Baird, 2003; Ellis Weismer, Lord, & Esler, 2010; Luyster, Kadlec, Carter, & Tager-Flusberg, 2008). The presence of these structural language difficulties presents an important puzzle. Although they were originally thought to be core components of ASD (e.g., Kanner, 1943), research gradually suggested that, in fact, they might be better characterized as a consequence of primary difficulties with social communication, processing abilities, or interpretation of sensory information (e.g., Baron-Cohen, Leslie, & Frith, 1985; Groen, Zwiers, van der Gaag, & Buitelaar, 2008; Norbury, 2005; Pellicano & Burr, 2012; Tager-Flusberg, 1996). Thus, observed language difficulties may not be breakdowns in the linguistic system but rather a consequence of the fact that social communication and other abilities are typically heavily implicated in early language learning. This theoretical advance radically shifted the way that language is conceptualized in ASD, such that language impairment is no longer included in the Diagnostic and Statistical Manual of Mental Disorders criteria (American Psychiatric Association, 2013).

Primary difficulties with social communication are often accompanied by impairments in related systems—including memory, attention, and nonverbal cognition—which may affect language development by altering how children attend to and process the linguistic and extralinguistic input in their environment (see, e.g., Eigsti, de Marchena, Schuh, & Kelley, 2011, for a review). Critically, these social and cognitive deficits are not merely at play for any specific learning opportunity, but over the course of development, they are shaping learning in complex ways from the early months and years of life. That is, we must think of particular language difficulties identified in the preschool- or school-age years as stemming not only from social communication or other deficits observable at the time but also from early, strong, and cascading impacts of these deficits on prelinguistic and linguistic foundational skills.

Although there are many ways in which early difficulty with social communication and other abilities might have cascading impacts on language development, in the current review, we focus on the consequences of the fact that ASD affects children's abilities to learn from the linguistic input to which they are exposed. The linguistic input children receive from their caregivers is of critical importance to early language development. The quantity and quality of linguistic input predict language outcomes for both typically developing children (e.g., Hart & Risley, 1995; Hoff, 2003;Huttenlocher, Haight, Bryk, Seltzer, & Lyons, 1991; Rowe, 2012) and children with ASD (e.g., Bang & Nadig, 2015; Flippin & Watson, 2015; Haebig, McDuffie, & Ellis Weismer, 2013). However, for both populations, this input is not simply taken up veridically by the developing language system. Instead, their intake of the input and thus the linguistic representation that they form are constrained by their own linguistic and cognitive abilities. We have suggested that, to the extent that these abilities in children with ASD differ from those in typically developing children, their intake will differ as well and that focusing specifically on understanding children's intake will allow us to make strides in developing effective interventions (Arunachalam & Luyster, 2016).

In our research program, we have begun to explore these issues, with the goal of characterizing what kinds of input can lead to optimal intake for preschoolers with ASD. In what follows, we review work from our labs that sets a foundation for this program of research. We also describe ongoing expansions and elaborations in order to outline what we think are promising research directions, with preliminary data where available. These efforts are identified as work in progress so that readers view them as promising avenues rather than complete findings. ¹ (Of course, even with completed and published work, it is important to engage in replication, particularly given the heterogeneity of the ASD population.) Critically, in order to go beyond descriptive characterizations to studying mechanisms underlying language learning, it is important to use a range of behavioral tasks, which we have adapted largely from the literature on typical development.

We focus specifically on the acquisition of lexical meaning and on the role of two abilities that are critical in typically developing children and that are known to be challenging for children with ASD: social communication skills and language processing. Social communication is a hallmark deficit in ASD (American Psychiatric Association, 2013) and includes abilities related to the use of language in social interaction. Although language processing, or the cognitive processing required to understand and produce language in real time, is not typically viewed as a characteristic deficit of ASD, many studies have noted that individuals with ASD have poorer language processing skills than their age- and/or language-matched peers (e.g., Bavin et al., 2014; Chita-Tegmark, Arunachalam, Nelson, & Tager-Flusberg, 2015; L. M. Henderson, Clarke, & Snowling, 2011; Kamio, Robins, Kelley, Swainson, & Fein, 2007), which manifest as slower and/or less accurate language comprehension. Poor language processing in ASD is probably not a core deficit but likely stems from the cascading effects we have been discussing of early deficits in other abilities including information processing and/or nonverbal cognition.

We begin by exploring how reducing the social communication demands of a learning situation may affect intake of the input with respect to children's acquisition of nouns, pronouns, and verbs. We address this broad swath of the lexicon because much research on word learning focuses on nouns labeling objects, and we consider it critical to think about how different kinds of words present different kinds of learning challenges. Then, because of the role of language processing in verb learning specifically, we focus on how language processing ability may modulate intake for verb learning.

Social Communication

There is evidence that minimizing social pressure or social demands supports performance in ASD for skills such as learning to imitate sounds or interpret a social situation (DeThorne, Johnson, Walder, & Mahurin-Smith, 2009; Pepperberg & Sherman, 2000, 2002; Pierce, Glad, & Schreibman, 1997). These observations spark an interesting hypothesis for word learning: Perhaps, limiting the social demands of a vocabulary learning situation might enhance performance for children with ASD. Of course, children must learn to use vocabulary in social situations, and minimizing social demands cannot be a cure-all. Still, it could be that by presenting new vocabulary in situations in which children need not participate in a social interaction, we can help them form foundational lexical representations that they can later build upon.

On the other hand, a second possibility is that removing the social context shuts down learning altogether. After all, language is a tool for social communication, and if we limit the social aspects of language teaching, we might also be removing the relevance and motivation of the language information. We might predict, then, that children would fail to learn new words unless there is some social communicative relevance (e.g., Fennell & Waxman, 2010; Ferguson & Waxman, 2016, 2017; A. M. Henderson, Sabbagh, & Woodward, 2013). Children with ASD show reduced attention to and preference for speech (e.g., Kuhl, Coffey‐Corina, Padden, & Dawson, 2005). It may be that they will be more likely to attend to speech when communicative relevance is clear.

In the following sections, we ask whether, when we reduce the social demands of the learning situation, children can nevertheless begin to acquire word meanings and how performance compares to more standard word-learning contexts. We discuss some recent and ongoing work on acquisition of nouns, pronouns, and verbs in typically developing children and children with ASD.

Nouns

Early word-learning studies on language learning in ASD focused on a difference between ASD and typical development; put broadly, whereas typically developing children seemed to rely heavily on social cues such as a speaker's direction of gaze, children with ASD seemed not to (e.g., Baron-Cohen, Baldwin, & Crowson, 1997). In Baron-Cohen et al.'s landmark study, children with ASD assigned a novel label to a toy that was their own focus of attention rather than the toy in the focus of attention of the person uttering the novel word. However, later research substantially qualified the conclusion that children with ASD cannot use social cues for word learning, finding that, in many circumstances, they can learn a name for a novel object when it is presented ostensively as part of a social interaction (e.g., Akechi et al., 2011; Luyster & Lord, 2009; McDuffie, Yoder, & Stone, 2006; Parish-Morris, Hennon, Hirsh-Pasek, Golinkoff, & Tager-Flusberg, 2007; Venker, Kover, & Ellis Weismer, 2016). For example, Luyster and Lord (2009) presented children with ASD with a mean age of 2.5 years (much younger than the 7- to 12-year-olds studied by Baron-Cohen et al.) with two novel objects. The experimenter presented each novel object in turn, addressing the child directly while moving the object in the child's field of vision. For one of the objects, the experimenter provided a novel name (e.g., “That's a toma”), and for the other, she simply commented on it (e.g., “Wow, look at that!”). At the test, the child was asked to select the toma from among the two objects and a third novel distractor; overall, they succeeded, with 14 of the 21 children choosing the labeled object. They also succeeded when they were holding their own distracting toy while the experimenter labeled the target object; like typically developing children and unlike children in Baron-Cohen et al.'s study, 19 of 21 children with ASD were able to match the novel label to the object that was the experimenter's focus of attention rather than the toy they were holding. Interestingly, in Parish-Morris et al.'s (2007) study of 5-year-olds with ASD involving a similar task, children were able to map novel words to objects that they preferred or found interesting but not to objects they did not prefer. These findings suggest that the difficulties children with ASD have with social communication skills act as a constraint or filter shaping their intake from the input; they are not necessarily an impermeable barrier to intake.

In typical development, children can often learn new word meanings even in the absence of social cues such as eye gaze. As early as 18 months, laboratory studies have shown that typically developing toddlers can learn a new word simply by attending to a conversation between two adults (e.g., Akhtar, 2005; Akhtar, Jipson, & Callanan, 2001; Floor & Akhtar, 2006; Gampe, Liebal, & Tomasello, 2012; Shneidman, Buresh, Shimpi, Knight-Schwarz, & Woodward, 2009). Interestingly, there is some evidence that this ability does not play a substantial role in typically developing children's real-world language acquisition; the amount of overheard speech children hear does not predict their vocabulary size, although the amount of child-directed speech they hear does (Shneidman & Goldin-Meadow, 2012; Weisleder & Fernald, 2013). However, this in-principle ability is still important to study in children with ASD because one-on-one interactions may not be as effective for children with ASD as typically developing children because of their difficulties with social communication, and thus, there may still be room for overhearing interactions to boost vocabulary acquisition. Further, play-based intervention situations for children with ASD often permit a relatively high degree of control, similarly to experimental lab situations, in factors such as the number and types of exposures a clinician provides of a new word. Therefore, even if the real world does not provide supportive exposure in overhearing situations, an intervention situation might be able to.

To see whether children with ASD can learn words by overhearing others' conversations, in a recent study, we tested children with ASD in a novel noun-learning task (Ghofrani, Luyster, & Arunachalam, 2017; Luyster & Arunachalam, 2018). We adapted Akhtar et al.'s (2001) paradigm, used with typically developing 2-year-olds, in which children are exposed to a novel noun either in a direct interaction with an experimenter or as overhearers of a conversation between the experimenter and another adult. We chose this paradigm because typically developing children succeeded in both conditions, indicating that the task is well suited to ask whether children with ASD differ from typically developing children.

Further, and critically for our research question, the condition in which the child is directly addressed by the experimenter strongly resembles a word-learning task in which children with ASD, too, can succeed (Luyster & Lord, 2009). Specifically, the first of Luyster and Lord's conditions described above, in which children were not given their own distracting toy to hold but were directly interacted with by the experimenter, strongly resembles the direct interaction (“addressed”) condition of Akhtar et al. (2001).

Thus, using a variant of Akhtar et al.'s (2001) paradigm permits us to ask, given that we expect children with ASD who have similar characteristics to those tested in Luyster and Lord (2009) to succeed when the context is a social interaction, can they also succeed in the less socially demanding condition? In Luyster and Arunachalam (2018), we taught 4- and 5-year-old children with ASD (N = 13) two novel nouns, each in either an addressed condition or an overheard condition. In the addressed condition, the child is directly spoken to by an experimenter, who introduces three novel objects. She pulls each object out of an opaque bucket, one by one. About two of the objects, she says, “Look at this one. I like this one.” About the third, she says, “Look at this toma. I like this toma,” labeling the object with a novel noun. (The identity of the object and the order in which it is introduced are counterbalanced across children.) She gives the child the three objects to play with briefly and then replaces them in the buckets and repeats the training.

In the overheard condition, the experimenter introduces three new novel objects and a new novel word and recites the same script. However, this time, the experimenter speaks to another adult. She first says to the child, “I'm going go over here to play with my friend, and then I'll be right back,” before moving away to sit across from and interact with another adult. The child is not given any specific instructions to watch the interaction and is not prevented from getting up to find a toy or interact with the parent. This is not an ostensive word-learning situation for the child, who is not directly involved in the interaction, but otherwise, the same teaching elements are present if the child chooses to attend to the adults' interaction.

The testing phase is identical in both conditions. After each novel word is taught, the experimenter places the three novel objects on a tray and asks the child to, for instance, “get the toma” (a different novel label is used across conditions). The child is invited to place the object in a “penguin”—a waste bin with a spinning lid decorated to look like a penguin and designed to motivate children to select an object when asked (see Luyster & Lord, 2009).

Our results showed that children with ASD chose the target novel object at test significantly more often than would be expected by chance in both the addressed and overheard conditions. Thus, our primary goal of asking whether at least some children with ASD can learn from overheard speech demonstrates that, indeed, they can. Thus, with respect to our two hypotheses—that reducing social communication demands would enhance learning because it reduces social pressure or that it would prevent learning because it would remove the motivation to learn—our results rule out the second. Further, we found no difference in success rate between the two conditions; thus, there was no evidence that learning in one condition was easier than in the other (though given the high level of performance overall, there was insufficient variance to permit detection of such differences).

Our goal in this study was to explore whether learning from overhearing is a learning mechanism that is available to at least some children with ASD—that is, that it is not substantively impaired by core deficits of the ASD diagnosis. However, in future work, we plan to test a more heterogeneous sample of children and to take an individual differences approach to see if there are differences between conditions and specifically to test the prediction that, for some children, learning in an overhearing situation might be enhanced relative to learning when children are directly addressed. Understanding which children benefit from this strategy will help to inform interventions.

Another important question for future work will be whether children perform just as well in the overheard condition when they themselves have the opportunity to hold a different object from the ones being labeled. Baron-Cohen et al. (1997) found that children with ASD had difficulty drawing attention away from an object they were engaged with to attend to an object the experimenter was labeling, but Luyster and Lord (2009), working with younger children with less severe intellectual impairment, found much better success in a very similar task. Given that the children we studied here more closely resemble children in Luyster and Lord (2009) than those in Baron-Cohen et al. (1997), we suspect that the presence of a distracting toy in itself should not hinder the ability to attend to others' conversations to learn new word meanings. This remains to be tested, however.

Pronouns

Our experimental work suggests that at least some children with ASD do have a general ability to learn vocabulary in overhearing contexts. However, the strongest benefits of this ability may not be for acquiring names for objects. Overhearing is likely to be particularly beneficial for acquiring other kinds of words, most notably personal pronouns. The referent of these deictic words shifts depending on the context—you can refer to anyone who is being addressed by the speaker, and I, to anyone who is speaking. One-on-one interactions alone provide poor evidence about these deictic shifts; a child who only interacts with her father may think that you is another name for herself and I is another name for her father, particularly if she is preverbal herself and does not produce either. The opportunity to overhear others' interactions provides important evidence about how pronoun referents shift depending on the identity of the speaker and addressee. For typically developing children, Oshima-Takane and colleagues have found that second-born typically developing children are more skilled with personal pronouns than first-born typically developing children, presumably because they have more opportunities to overhear pronouns for which they themselves are not the addressee (Oshima-Takane, 1988; Oshima-Takane & Benaroya, 1989; Oshima-Takane, Goodz, & Derevensky, 1996; Oshima-Takane, Takane, & Shultz, 1999). A multiparty conversation, in particular, allows the child to recognize that the referents of these words shift dynamically as the conversation unfolds.

In ASD, personal pronouns are notoriously difficult. The most common features listed in descriptions of atypical pronoun use in ASD are pronoun avoidance (e.g., using one's own name instead of a pronoun) and pronoun reversal (e.g., using you when referring to oneself; e.g., Jordan, 1989; Kanner, 1946), although recent evidence from a small group of 15 children with ASD indicates that pronoun reversal, at least, is relatively infrequent (Naigles et al., 2016). Because pronouns shift reference depending on the identity of the speaker and addressee, mastery of pronouns may call upon nonlinguistic abilities including perspective-taking, theory of mind, and understanding of discourse roles—abilities that are often compromised in children with ASD (e.g., Boucher, 2003; Loveland, 1984; Tager-Flusberg, 1994; Tager-Flusberg & Sullivan, 1994).

Difficulty with pronouns appears to be related both to the language delay associated with ASD (Naigles et al., 2016), as well as social–cognitive factors such as perspective-taking. Given the potential role that overhearing others' conversations plays in acquiring pronouns, we are exploring the nature of the input to which children with ASD are exposed when it comes to pronouns. In corpus work, we find that infants at higher risk for ASD (due to having an older sibling with ASD diagnosis) hear proportionally fewer pronouns than infants at lower risk (with an older sibling who does not have an ASD diagnosis; He, Luyster, Hong, & Arunachalam, 2018). Our results indicate that this is because caregivers of high-risk infants are spending proportionally more of their utterances calling the infant's name to get his or her attention—perhaps, this could be in response to a lack of responsiveness on the part of the infant, as high-risk infants have been shown to be less socially responsive than low-risk infants (e.g., Ozonoff et al., 2010; Zwaigenbaum et al., 2005).

In work in progress, we are asking if there are similar effects of having opportunities to observe third-party interaction as have been found for typically developing children by Oshima-Takane and colleagues (Oshima-Takane, 1988; Oshima-Takane & Benaroya, 1989; Oshima-Takane, Goodz, & Derevensky, 1996; Oshima-Takane, Takane, & Shultz, 1999) using a survey asking parents about first- and second-person pronoun errors in their preschoolers with ASD and about the amount of time their child spends in one-on-one interactions as compared with interactions with two or more people. Preliminary results indicate that parent report of first- and second-pronoun errors in preschoolers with ASD is indeed correlated with the amount of time the child spends in one-on-one interactions as compared with interactions with two or more people (Georgeson, 2018; Fitch, Georgeson, Netburn, Arunachalam, & Luyster, 2018). One possible interpretation of this finding is that children with more severe ASD symptoms spend more time in intensive applied behavior analysis (ABA) therapy, which is typically one-on-one, and that therefore one-on-one interaction is correlated with more severe ASD. However, in a follow-up analysis, we found that children in ABA therapy do not make more pronoun errors than children not in ABA therapy. These preliminary results are suggestive that a research program examining in detail how and with whom children spend their time can shed light on how access to different kinds of interactions might affect pronoun development.

However, one interesting open question pertains to children's abilities in comprehension. Few studies have examined pronoun comprehension in children with ASD, and these only involved the experimenter and the child (Jordan, 1989; Lee, Hobson, & Chiat, 1994), meaning that the design was not set up to test whether children understood how pronouns shift reference in real time depending on the identity of the speaker and addressee. Therefore, these studies do not reveal whether children with ASD can correctly comprehend pronouns when they are not themselves the referent of the pronoun.

This is a critical gap to fill. Understanding comprehension will be critical to understanding the cause of pronoun difficulty in production. However, it is also critical for our current topic of how social communication deficits might impact word learning. If learning from overhearing is a strength, we might expect pronoun comprehension to be relatively strong, as children can learn about this by witnessing third-party interactions. This extends our hypothesis to see if prior real-world multiparty experience affects pronoun knowledge, whether or not they make errors in production.

To begin to address this important gap, we have ongoing work in progress probing children's comprehension of first- and second-person pronouns. We describe the procedure in some detail here because we think it is a particularly promising one for understanding pronoun comprehension and adaptable to other populations and other pronoun types than the ones we are studying so far.

We adapted a clever paradigm that was recently developed to study pronoun knowledge in typically developing 2-year-olds (Moyer, Harrigan, Hacquard, & Lidz, 2015). In our version, children with ASD aged 4 to 6 years (N = 14), with a mean language equivalency score of 3;6 on the Expressive Language Scale of the Mullen Scales of Early Learning: AGS Edition (Mullen, 1995) and typically developing controls, roughly matched on expressive language (N = 12; data collection in progress) and played a sticker-finding game with two experimenters. At the beginning of the game, the child and each of the experimenters were assigned a box of a different color. (Each also wore a colored nametag to help children remember which box belonged to which person.) On each trial of the game, the experimenters raised a board between themselves and the child and hid a sticker in one of the three boxes. They then removed the board, and one experimenter provided a verbal clue as to the location of the sticker. The verbal clue included either a first-person pronoun (“It's in my box”) or second-person pronoun (“It's in your box”). For the latter, the experimenter either directed her gaze to the child while providing the clue or to the other experimenter (who had “forgotten” the location of the sticker). The child was then invited to look for the sticker, and the box he or she chose served as the dependent variable. The design of the game meant that the child heard the second-person pronoun in reference to three individuals (the child and each experimenter) and the first-person pronoun produced by, and in reference to, two individuals (each of the experimenters). We chose this task because of its minimal task demands—the child is never asked to speak—and highly motivating sticker reward structure.

Data collection is still in progress, and these preliminary results should be read cautiously, but the pattern thus far is striking. On first-person trials, children with ASD performed well (86% correct), regardless of which experimenter produced the “my” statement. Thus, the children understood that first-person pronouns shift reference depending on who the speaker is and could use voice and/or gaze cues to identify which experimenter was speaking. On second-person pronoun trials, performance was poorer overall. On those trials on which the child was the intended referent of the pronoun, performance was again high (86%), but on trials on which one of the experimenters was the intended referent, they identified the correct box only 36% of the time. These errors reflected a large “self” bias; when children chose incorrectly, they usually selected their own box even though they were not the addressee of the clue. Typically developing controls, roughly similar in language level, showed a similar pattern of errors but had much higher success rates overall. We further coded these second-person pronoun trials for whether the child had directed his or her gaze toward the speaker during the clue, to see where the speaker was looking, because, in order to succeed on these trials, children had to notice that the speaker's attention was directed to the other experimenter and not to themselves. We found that children with ASD looked up at the speaker on 72% of these trials, while typically developing controls did so on 91% of trials.

Our data thus far, from a comprehension task rather than a production task, support prior evidence that personal pronouns, particularly second-person pronouns, are challenging for children with ASD. However, they may not differ categorically from typically developing children; it may rather be a matter of degree. It could also be that poor performance with second-person pronouns, resulting in a bias to interpret the referent of the pronoun as themselves, is the result of a complex interplay of factors that are at work throughout the course of development. For example, executive function may play a role, particularly in our task, in which children may have had to inhibit a bias to reach for their own box, which was likely the most salient box for them (see Ellis Weismer, Kaushanskaya, Larson, Mathée, & Bolt, 2018, for discussion of executive function in children with ASD).

Data from children with ASD who are acquiring American Sign Language, in which personal pronouns are indexical points—the signer points to himself or herself for first-person singular pronouns and directly to the addressee for second-person pronouns—add an interesting wrinkle. Although personal pronouns in American Sign Language are arguably more referentially transparent than in English, children with ASD also demonstrate pronoun avoidance (Shield, Meier, & Tager-Flusberg, 2015). This too supports the notion that the challenges of learning the pronoun system are not only related to social attention or echolalia; rather, the difficulty can manifest in both comprehension and production and in both transparent and nontransparent systems.

This provides further evidence that the challenges of pronoun acquisition are not solely driven by social attention or social communication but also by children's language skill (Naigles, Kelty, Jaffery, & Fein, 2011). However, critically, it is important to consider how children's behavior in the moment adds up to learning over time, as their linguistic representations are beginning to form. Failure to visually attend to a speaker (either one speaking directly to the child or to a third social partner) or even to inhibit a self-bias in the moment during the task may not alone explain children's performance. However, consider that the child likely shows these same difficulties repeatedly in daily life when they hear personal pronouns, potentially missing numerous opportunities throughout early childhood to gain evidence about what the pronouns mean and how they shift reference. Thus, social communication difficulties will have cascading effects on language.

Verbs

A second kind of word whose acquisition should be particularly helped by overhearing situations and/or minimal social demands is verbs. Overhearing contexts may be particularly beneficial for verb learning because verbs are unlikely to be learnable simply by observing the world when the word is uttered and, instead, are thought to be learned by attention to linguistic context (e.g., Gleitman, 1990), in some cases even without access to the verb's visual referent (e.g., Yuan & Fisher, 2009). Thus, for verb learning, it may be that overhearing others' conversations may increase opportunities to acquire verbs.

Linguistic context is critical for verb acquisition because, in many situations in which an unfamiliar verb is uttered, children may witness multiple simultaneous events that are each lexicalized by different verbs—for example, chasing and fleeing nearly always co-occur, as do buying and selling, and giving and receiving (e.g., Fisher, Hall, Rakowitz, & Gleitman, 1994). Further, verbs are often uttered when their referent events are not currently ongoing (Gleitman & Gleitman, 1992; Tomasello & Kruger, 1992), exacerbating the problem of identifying their meaning.

A solution to this problem, at least for typically developing children, is a mechanism known as syntactic bootstrapping, in which children attend to the linguistic context in which the verb appears to glean aspects of its meaning (Gleitman, 1990). Linguistic context, over and above joint attention, can provide critically useful information. For example, in a visual scene in which a dog chases a cat, the sentence context “The dog is blicking the cat” allows us to disentangle the chasing and fleeing interpretations and assign the verb “blick” the meaning “chase,” whereas the sentence context “The cat is blicking the dog” allows the reverse. By two years of age, typically developing children can use the syntactic bootstrapping mechanism in novel verb-learning tasks. Naigles (1990) found that 2-year-olds who heard a novel verb in a transitive linguistic context (e.g., “The duck is gorping the bunny”) preferred to look to a causative scene (e.g., duck pushes bunny into a squat) as compared with those who had heard the verb in an intransitive context (e.g., “The duck and the bunny are gorping”).

Thus, hearing verbs in linguistic contexts is particularly important for their acquisition. However, if children can attend to and learn from overheard speech, the number of linguistic contexts in their everyday experience that can serve as language input vastly increases. Therefore, it is critical to examine how children might be able to engage in syntactic bootstrapping in overhearing contexts, over and above how they might be able to learn nouns in overhearing contexts. Yuan and Fisher (2009) presented linguistic contexts during a videotaped dialogue between two actors instead of as a voiceover while the referent events were ongoing and then tested typically developing children on whether they had extracted information about the novel verb from the dialogue and could use it to identify the appropriate referent event at test. Their paradigm was very similar to Naigles (1990) and other studies on syntactic bootstrapping, except that, instead of simultaneously seeing candidate visual events and hearing the novel verb, they heard the novel verb first, in conversation, and then saw the candidate visual events in the test phase. Here too, children succeed (see also Arunachalam & Waxman, 2010).

In a recent follow-up with typically developing children, we saw these positive findings as an opportunity to investigate whether children can establish representations for new verbs in an even less social context (Arunachalam, 2013). Instead of dialogues between two actors, 2-year-old typically developing children heard the novel verbs in transitive or intransitive linguistic contexts simply as ambient noise, while they viewed an unrelated silent animation or played quietly with toys. The novel verbs occurred in a list of unconnected sentences read in an adult-directed speech register with list-reading prosody. This scenario was intended to minimize any cues that the auditory information was “for the child” or “for learning.” Then, children participated in an eye-tracking test to see if they had formed a representation for the novel verbs, using the same kinds of test events as in prior syntactic bootstrapping studies. See Figure 1. Indeed, those who had heard the verb in transitive linguistic contexts preferred to look to the causative scene as compared with those who had heard intransitive contexts.

Figure 1.

Schematic depiction of stimuli from one trial of Arunachalam (2013) with typically developing children and Horvath et al. (2018) with children with autism spectrum disorder. Reprinted from Cognition, 129, S. Arunachalam, “Two-Year-Olds Can Begin to Acquire Verb Meanings in Socially Impoverished Contexts,” pp. 569–573, Copyright © 2013, with permission from Elsevier.

Given the minimal social communication demands of attending to the teaching materials (and because we used eye tracking, the minimal task demands of demonstrating a response), we thought this would be an interesting paradigm to use with children with ASD. Prior work demonstrates that children with ASD can use syntactic bootstrapping in verb learning (Naigles et al., 2011; Shulman & Guberman, 2007). There are also positive findings of abilities to engage in statistical learning and other kinds of implicit learning from ambient linguistic streams (e.g., Eigsti & Mayo, 2011; Foti, De Crescenzo, Vivanti, Menghini, & Vicari, 2015; Mayo & Eigsti, 2012; Obeid, Brooks, Powers, Gillespie-Lynch, & Lum, 2016). Further, children with ASD are able to use linguistic information describing an object's change of location in the absence of visual access to that object to make inferences about its new location (Fitch, Valadez, Ganea, Carter, & Kaldy, 2018), suggesting that they can attend and learn from language presented in the absence of visual referents.

Therefore, we recently asked whether children with ASD could also learn verbs from linguistic contexts given this kind of nonsocial input (Horvath, McDermott, Reilly, & Arunachalam, 2018). Children with ASD aged 2;1 (years;months) to 4;5 (N = 34, mean age = 3;3), with a mean receptive language equivalency score of 2;4 and expressive language equivalency score of 2;2 (Mullen Scales of Early Learning: AGS Edition; Mullen, 1995), participated. They heard novel verbs in either transitive or intransitive linguistic contexts while viewing the same silent animations or playing quietly with toys as in Arunachalam (2013). They too, succeeded, with those in the transitive condition preferring the causative scene as compared with those in the intransitive condition. Thus, this study not only replicated previous findings that children with ASD can engage in syntactic bootstrapping (Naigles et al., 2011; Shulman & Guberman, 2007), but it also showed that they can do so in the absence of a social-communicative context highlighting the relevance of the situation for the child.

Children's abilities to use syntactic bootstrapping in verb learning are remarkable and particularly important given the role that verbs play in the lexicon. Verbs permit the transition from single words to longer utterances and the expression of more complex ideas; not surprisingly, verb knowledge predicts typically developing children's grammar outcomes better than noun knowledge (Hadley, Rispoli, & Hsu, 2016). However, the ability to use linguistic context to acquire word meanings also presents a puzzle, perhaps particularly for children with ASD. Syntactic bootstrapping requires children to parse the linguistic context efficiently. Events in the world are fleeting, and a child who is slow to process language and identify the event that is being described may miss the event altogether. Many children with ASD show slower language processing than typically developing peers (e.g., Bavin et al., 2014; L. M. Henderson et al., 2011; Kamio et al., 2007). In Horvath et al. (2018) described above, children were presented with a string of 27 sentences including the novel verb, but a naturalistic learning situation is unlikely to have this kind of high-density exposure. If a child only hears one or two repetitions of a novel verb in a single learning situation, is learning still possible, and are there particular kinds of linguistic context that are better than others? We consider this question to be an extremely productive one to explore because it could lead to clear guidelines for how best to introduce new vocabulary, not only for children with ASD but also for other children who have poor language processing skills, such as children with developmental language disorder (previously known as specific language impairment; see Bishop, Snowling, Thompson, Greenhalgh, & CATALISE consortium, 2016). Thus, in what follows, we consider the role of language processing in learning. We continue our focus on verbs because of the importance processing plays in syntactic bootstrapping, but processing abilities can also play a role in acquiring other kinds of vocabulary (e.g., Fernald, Marchman, & Hurtado, 2008).

Language Processing

Our research with typically developing children has clearly shown that indeed, certain contexts are better than others, and the processing burden associated with the context does affect learning. For example, with typically developing 2- and 3-year-olds, we have found that introducing a new verb with just pronouns, as in, “Look, he's pilking it,” is less supportive of learning than introducing the verb with content nouns, as in, “Look, the boy is pilking a balloon”—this is true even if the visual scene depicts only one possible referent for each pronoun (Arunachalam & Waxman, 2011, 2015). This suggests that young learners require a certain amount of information to decipher what an unfamiliar verb might mean and that content nouns provide such information. Of course, pronouns are not always harmful for learning; they can support children's identification of a novel word's phonological form and grammatical category (e.g., Mintz, 2003). However, in this particular learning scenario, children required more semantic information than was available in pronouns alone to discern the verb's meaning. Other informative words in the linguistic context can also help to provide this critical semantic information; for example, Syrett, Arunachalam, and Waxman (2014) found that the manner-of-motion adverb “slowly” can boost performance in a pronoun context (e.g., “Look, he's pilking it slowly”), whereas other adverbial expressions that do not encode manner of motion (“nicely,” “right now”) did not; presumably, the adverb focused children's attention on the motion in the visual scene, which in turn helped them map the novel verb to this motion itself.

However, at the same time, we have also found that too much information can be harmful (Arunachalam, Leddon, Song, Lee, & Waxman, 2013; He, Kon, & Arunachalam, 2018; Kon, Göksun, Bagci, & Arunachalam, 2015). We suspect that because information increases processing load, children may not be able to benefit from information that uses up a significant amount of their processing capacity. Processing load refers to the resources (e.g., working memory) required to process a linguistic context. Many features of linguistic context affect processability, including frequency (e.g., Rayner & Duffy, 1986) and brevity (e.g., Hawkins, 2004), and for adults, pragmatic felicity (i.e., appropriateness to the situation; Wilson & Sperber, 1993). Children's lexical and phrasal processing abilities continue to develop through early childhood (e.g., Fernald, Pinto, Swingley, Weinberg, & McRoberts, 1998; Trueswell, Sekerina, Hill, & Logrip, 1999). It could be that positioning detailed semantic information before the novel verb uses up processing resources that the child needs in order to learn the novel verb downstream in the sentence. For example, adding modifiers to a content noun, as in, “The pretty round balloon is pilking,” can hinder learning (He, Kon, & Arunachalam, 2018; Kon et al., 2015).

Our synthesis of these and other experimental results is that linguistic context can facilitate acquisition of new verb meanings but only when the linguistic context is balanced between informativeness (i.e., providing sufficient useful information) and processability (i.e., imposing a sufficiently small processing burden; He & Arunachalam, 2017). Of course, multiple repetitions of the context may reduce processing load overall, but if none of the contexts is individually helpful for learning, simple repetition may not help. If we consider linguistic contexts as falling on a spectrum of informativeness and processability, we should be able to identify optimal windows encompassing contexts that support verb learning. See Figure 2. Because children with ASD often struggle with language processing, the optimal windows on this spectrum of linguistic contexts may be smaller than for typically developing children, encompassing only the ones easier to process of those contexts that fall within the typically developing child's window. Other child factors, including nonverbal cognitive abilities, will undoubtedly affect the size and positioning of the optimal window as well.

Figure 2.

Schematic depiction of a continuum of linguistic contexts, from those that are easy to process but convey little information to those that convey a lot of information but are more difficult to process. Where contexts fall on this continuum, as well as which parts of the continuum best support children's learning, will differ depending on the learning situation and the learner's abilities.

We believe that a crucial avenue for future research is therefore to understand what conditions best support acquisition of vocabulary—and particularly of verbs—so that these conditions can be optimized in treatment. Although our own work in this area has primarily focused on the informativeness and processability of linguistic contexts, many other factors such as timing, dose frequency, and intensity of exposure matter as well (see, e.g., Warren, Fey, & Yoder, 2007, for discussion). For example, although there is little research precisely manipulating these factors in children with ASD, it is likely that they need more intensive exposure than typically developing children for successful word learning (e.g., Luyster & Lord, 2009). Although the conditions that best support learning for children with developmental language disorder may carry over somewhat to ASD, it will also be important to investigate these factors for ASD specifically, as the social communication difficulties associated with the disorder will have specific impacts on the effectiveness of various strategies.

Another aspect of this larger research program is to examine the conditions that parents provide when interacting naturalistically with their children. If linguistic context supports learning, but only when the context achieves a balance between informativeness and processability, a natural question is whether parents produce new verbs in balanced contexts—or more broadly, whether parents are sensitive to the informational requirements and processing limitations their child brings to the task of word learning.

We have begun to study this issue using a new experimental paradigm in which parents produce unscripted descriptions of an object or event, and children's eye gaze is recorded as they seek to identify which object or event of the visual array is being labeled and described (Arunachalam, 2016). In this paradigm, parent and child sit together in front of an eye-tracking monitor displaying an array of several pictures or videos, and they play a “finding” game. (The parent wears laser glasses that block the wavelengths used by the eye-tracker, so that the parent can see the screen clearly, but his or her gaze is not tracked.) On each trial, one of the pictures or videos is the “target,” and the parent's job is to label or describe it such that their child correctly identifies it. In our work with typically developing children with this paradigm, we have found that parents' language is beautifully adapted to support comprehension: Caregivers overwhelmingly chose easy-to-process descriptions if the task was particularly difficult, whereas they were much less discriminating with easier tasks (Arunachalam, 2016).

This paradigm is especially suited to children with ASD because the dependent measure of children's performance is eye gaze, and therefore, they are not required to vocalize or point, although they are encouraged to do so to keep the gamelike quality at the forefront. Indeed, in our pilot work, we found that some children with ASD identified the target visually but provided no verbal or pointing feedback to their parent that they had understood, leaving the parent unsure of the child's level of understanding. Thus, this paradigm allows us to see what features of parent speech best support children's understanding, as well as how well parents seem to gauge their child's level and needs.

In work in progress using this paradigm, we are focusing on verb learning by showing visual arrays, including video clips of events that can be labeled with verbs that are known to parents but unlikely to be familiar to preschool-aged children (taken from vocabulary lists for ages 6 years and above, e.g., “zigzag”; He, Luyster, & Arunachalam, 2018). See Figure 3. Parents confirmed on a vocabulary checklist that their child had minimal or no familiarity with the verbs.

Figure 3.

Screenshot from one trial. The target event is the “zigzagging” event in the top left corner.

On each trial during the task, parents were asked to encourage the child to identify the target scene. They were given a card with the target verb printed on it and told, “if possible, please use the word on the card” (wording adapted from Clark, 2010). Otherwise, they were not given instructions about what to say or how much to say. We assessed both the properties of the parents' speech and how quickly children looked to the target scene. Parents' utterances were evaluated along the same dimensions that have been important in novel verb-learning tasks with typically developing children—whether the event participants are labeled with content nouns or pronouns and whether any modifiers are present.

Our results thus far reveal a fascinating developmental trajectory in parent utterances. Parents of young typically developing children, at 2 years of age (N = 8), tended to produce the novel verbs almost exclusively with content nouns as subjects (e.g., “The doggy is zigzagging”); they rarely produced pronouns or modifiers. Parents of older typically developing children, ages 3 and 4 years (N = 6), showed more variety in their utterances, using contexts similar to the full spectrum of those depicted in Figure 2. They often used pronouns (e.g., “Someone is zigzagging”) and sometimes used modifiers as well (e.g., “The silly black and white doggy is zigzagging”). With respect to children with ASD, we tested children aged 3 and 4 years (N = 7) who had below-age language ability. The parents of these children behaved more like parents of typically developing 2-year-olds, almost exclusively producing content nouns.

Next, we examined children's latencies to identify the target. Because of the minimal variance in the parent speech in the ASD group, we collapsed across all children to examine whether there were differences in latency to find the target given different kinds of parent productions. Indeed, we found that contexts with content nouns—the context type favored by parents of younger typically developing children and children with ASD as compared with parents of older typically developing children—yielded significantly shorter latencies than contexts with pronouns. (There were insufficient contexts with modifiers to permit comparison.)

This pattern suggests that at least some parents of children with ASD are sensitive to their child's language level and produce language that is tuned to the child's language ability rather than chronological age. This has been demonstrated previously in prior work with respect to global characteristics of language such as the frequency of adult speech, lexical diversity, and mean length of utterance (e.g., Bang & Nadig, 2015; Warren et al., 2010), but here, our goal is to examine this with respect to the specific features of informativeness and processability when introducing unfamiliar verbs.

This paradigm, which allows us to collect data both about parent input and, simultaneously, children's processing of that input, holds strong promise for understanding language ability in children with ASD and, in particular, individual differences among this diverse group. Ultimately, we may also be able to use this paradigm to provide feedback directly to parents and to inform them about the kinds of speech that most effectively supports their child's processing and learning, at least within a specific and constrained learning goal (e.g., forming a lexical representation for a specific vocabulary word). On a methodological note, the use of eye tracking to monitor and evaluate children's language comprehension provides an excellent window into their abilities, potentially revealing greater knowledge than standard assessments (e.g., Tager-Flusberg & Kasari, 2013; Venker & Kover, 2015).

Conceptualizing children's language comprehension as related to both informativeness and processability allows us to predict, for any given situation, what kinds of linguistic contexts will best support learning. Contexts that achieve a balance between informativeness and processability—a balance that may manifest differently for different children, depending on factors such as their own processing ability—will be optimal for learning. Naturally, these two features are not the only relevant ones, and how they manifest will differ across properties of the situation (e.g., the complexity of the event, the number of candidate meanings the child is entertaining) and properties of the child (e.g., developmental level with respect to several abilities including language and nonverbal cognition). We are just at the beginning stages of understanding precisely how these abilities affect what constitutes optimal contexts for children's learning. Although corpus research (e.g., Bang & Nadig, 2015; Warren et al., 2010), supported by our own preliminary work described above, indicates that parents show some natural abilities to tailor the language input they provide to their child's language level, future research is warranted to examine how we can further refine these abilities and help parents produce optimally supportive speech.

General Discussion

Recent work within our research program reveals that young children with ASD can mine their environment for useful linguistic information, even learning from language that is not directly addressed to them. Importantly, however, not all input is equally supportive of learning. For children with ASD, to maximize learning, it is important to consider how social-communicative difficulties and slow language processing constrain learning. Children's intake is not simply a copy of the input; rather, the input is filtered through, or constrained by, these and other linguistic and nonlinguistic abilities.

Success in lab-based tasks such as the ones we have reviewed can demonstrate that children with ASD have access to learning mechanisms that can enable vocabulary building. Nevertheless, language abilities are below age level in most of these children. Why, if children with ASD are able to learn new words in these tasks, do they present with poor vocabularies? One possibility we have suggested is that our laboratory tasks provide input that is highly accessible to children with ASD, particularly in not requiring them to make heavy use of abilities they struggle with, such as social attention. Research that brings elements of these laboratory tasks into the clinic and the home will be critical in advancing our theoretical and practical understanding of this possibility. We hope that advances in research, and ultimately in evidence-based practice—both on the part of caregivers and professionals—can support learning in children with ASD by calibrating demands on social attention and language processing, ultimately increasing children's intake from the input and improving both short- and long-term outcomes.

Funding Statement

This article stems from the 2017 Research Symposium at ASHA Convention, which was also supported by the National Institute on Deafness and Other Communication Disorders of the National Institutes of Health under Award R13DC003383. Research described in this publication and/or preparation of this article was also supported by the National Institute on Deafness and Other Communication Disorders under Award Number K01DC013306 to Sudha Arunachalam, by Autism Speaks Grant 8160 to Sudha Arunachalam, by a Charles H. Hood Foundation Child Health Research Award to Sudha Arunachalam, by an Emerson College Faculty Advancement Fund Grant to Rhiannon Luyster, and by an American Speech-Language-Hearing Association Advancing Academic-Research Careers Award to Rhiannon Luyster.