Object Label and Category Knowledge Among Toddlers at Risk for Autism Spectrum Disorder: An Application of the Visual Array Task

Abstract

Individuals diagnosed with autism spectrum disorder (ASD) demonstrate atypical development of receptive language and object category knowledge. Yet, little is known about the emerging relation between these two competencies in this population. The present study utilized a gaze-based paradigm, the visual array task (VAT), to examine the relation between object label and object category knowledge in a sample of toddlers at heightened genetic risk for developing ASD. Eighty-eight toddlers with at least one typically developing older sibling (low-risk; LR) or one older sibling diagnosed with ASD (high-risk; HR) completed the VAT at 17 (LR n=20; HR n=27) and/or 25 months of age (LR n=42; HR n=22). Results indicated that the VAT was both a sensitive measure of receptive vocabulary as well as capable of reflecting gains in category knowledge for toddlers at genetic risk of developing ASD. Notably, an early emerging difference in the relation between target label knowledge and category knowledge for the groups was observed at 17 months of age but dissipated by 25 months of age. This suggests that while the link between receptive vocabulary and category knowledge may develop earlier in LR groups, HR groups may potentially catch up by the second year of life. Therefore, it is likely meaningful to consider differences in category knowledge when conceptualizing the receptive language deficits associated with HR populations. During language learning, typically developing children are sensitive to the common features of category members and use this information to generalize known object labels to newly encountered exemplars. The inability to identify similarities between category members and/or utilize this information when learning new object referents at 17 months of age may be a potential mechanism underlying the delays observed in HR populations.

1. Introduction

In addition to core deficits in social communication and social interaction, individuals with autism spectrum disorder (ASD; American Psychiatric Association, 2013) often demonstrate atypical profiles of language (Boucher, 2012; Tager-Flusberg, Paul, & Lord, 2005) and category knowledge development (Gastgeb & Strauss, 2012; Mercado, Chow, Church, & Lopata, 2020). For typically developing children, a sensitivity to category information is central to the ability to generalize known object labels to newly encountered exemplars (Ferguson & Waxman, 2017). Thus, an inability to identify similarities between category members and/or utilize this information when learning new object referents may be a potential mechanism underlying differences in lexical acquisition (Arunachalam & Luyster, 2016) and, in particular, contribute to the reduction in vocabulary size observed for individuals with ASD compared to their typically developing peers (Kwok, Brown, Smyth, & Cardy, 2015; Rescorla & Safyer, 2021). Yet relatively little is known about the emergent relation between receptive vocabulary and category understanding in this population. Therefore, the present study explored the feasibility of examining both constructs using a single behavioral assessment, the visual array task (VAT; Hauschild, Pomales-Ramos, & Strauss, 2021), in a sample of toddlers at heightened familial risk for ASD.

Although delays in spoken language development are no longer included in the diagnostic criteria for ASD (DSM-5; American Psychiatric Association, 2013), aberrant language profiles among individuals with ASD remain prevalent (Tager-Flusberg, 2016). In fact, impairments and/or delays in spoken language are among the earliest, albeit not uniquely predictive, developmental signs of ASD (R. J. Luyster, Seery, Talbott, & Tager-Flusberg, 2011). Delays in language production (expressive language), such as onset of first spoken word, are particularly salient for parents of children later diagnosed with ASD and tend to be the first developmental concern reported to healthcare professionals (Coonrod & Stone, 2004). However, earlier emerging abnormalities in language comprehension (receptive language), such as response to spoken language, are also strong indicators of ASD in young children (Lord, 1995). Indeed, studies of infants who are at a heightened familial risk for developing ASD because they have an older sibling already diagnosed with ASD (Ozonoff et al., 2011; Sandin et al., 2014) have reported delays in receptive language that are present as early as 12 to 18 months of age (Landa & Garrett-Mayer, 2006; Mitchell et al., 2006; Zwaigenbaum et al., 2005). Furthermore, while language comprehension is advanced relative to production in typical populations (Fenson et al., 1994), the reverse relation, a relative expressive competency advantage, has often been identified for children with ASD (Charman et al., 2003; Luyster, Kadlec, Carter, & Tager-Flusberg, 2008; Luyster, Lopez, & Lord, 2007; Weismer, Lord, & Esler, 2010; but also see (Kwok et al., 2015).

One potential mechanism underlying these differences in lexical acquisition may be the influence of broader deficits in categorization characteristic of ASD (Mercado et al., 2020). For typically developing children, the ability to generalize known object category labels to newly encountered exemplars, known as extension (a marker of lexical comprehension and knowledge), is central to receptive vocabulary growth and reliant on a sensitivity to relevant commonalities between category members. Evidence of this noun-category linkage in lexical development emerges as early as nine months of age (Balaban & Waxman, 1997), with subsequent findings reporting on the ability of infants to generalize object labels based on perceptual, functional, and conceptual properties (for review, see Ferguson & Waxman, 2017). Indeed, this relation is bidirectional. While category labels may aid infants in learning or extending object names to specific exemplars, labeling objects can also promote attention to the common properties shared between category members. For example, naming basic level objects, such as a hat or a glove, with the superordinate category label “clothing” can lead infants to notice features shared between those objects like material and the ability to be worn, based on their function, and not merely their shape or appearance.

However, difficulties in forming central or prototypical representations of categorical boundaries commonly associated with ASD (Gastgeb, Dundas, Minshew, & Strauss, 2012; Gastgeb, Rump, Best, Minshew, & Strauss, 2009; Gastgeb, Strauss, & Minshew, 2006; Gastgeb, Wilkinson, Minshew, & Strauss, 2011; Gastgeb & Strauss, 2012; Klinger & Dawson, 2001) may limit the extension abilities of children with or at risk for ASD. In particular, a lack of an observed shape bias, the preference for mapping new word labels onto the shape of referents, has frequently been associated with reductions in generalizability and often implicated in reduced vocabulary size for children with ASD (Hartley & Allen, 2014; Potrzeba, Fein, & Naigles, 2015; Tek, Jaffery, Fein, & Naigles, 2008; Tovar, Rodríguez-Granados, & Arias-Trejo, 2020). Thus, an inability to abstract and/or utilize similarities between category members when learning new object referents may contribute to observed differences in lexical acquisition (Arunachalam & Luyster, 2016) as well as overall vocabulary size for children with ASD compared to their typically developing peers (Kwok et al., 2015; Rescorla & Safyer, 2021).

Given that both receptive vocabulary and categorization are early emerging processes, it is critical to study their emergent relation within the first two years of development. However, language comprehension and category knowledge can be particularly difficult constructs to measure in infants and young children. Language assessments based on parent report rely heavily on parent ability to infer language and vocabulary knowledge through observations of their children. There may be large differences in how accurately parents are able to deduce and report this information leading to potentially high levels of variability in the resultant data (Law & Roy, 2008; Tomasello & Mervis, 1994). Not only do parents differ in their abilities to infer language comprehension from their infant or toddler’s behavior, but also likely differ in their criteria of what it means for their child to recognize versus understand the referential meaning of a word. This may be particularly true for parents reporting on the younger siblings of children diagnosed with ASD as atypical language profiles exhibited by older children may skew parent perception of language norms.

In an attempt to reduce such bias, researchers have often relied on experimenter-administered assessments for measuring language (such as the Peabody Picture Vocabulary Test; Dunn, 2019) and category knowledge (such as the sequential touching procedure; Mandler, Fivush, & Reznick, 1987; Starkey, 1981). However, these assessments place high task demands on children to engage in pre-arranged tasks such as following a set of verbal instructions (e.g., “put your finger on the dog”) or manually manipulating a series of small objects. Additionally, these assessments are often dependent on successful interaction with an unfamiliar test administrator as well as the use of fine motor behaviors. These added requirements for successful assessment completion may become particularly challenging for individuals at risk for language delay, communication deficits, and motor deficits such as young children with or at risk for ASD (Iverson et al., 2019; Kasari, Brady, Lord, & Tager-Flusberg, 2013; Patten, Labban, Casenhiser, & Cotton, 2017; Skwerer, Jordan, Brukilacchio, & Tager-Flusberg, 2016).

Attention-based measures eliminate concerns surrounding reporter bias and significantly reduce task demands placed on participants. By presenting participants with two or more images and a single verbal label, experimenters can use changes in gaze-based behaviors (such as preferential looking to the labeled “target” image) to infer lexical knowledge without the need for direct engagement with the experimenter or gestural response. Previous work has also demonstrated the ability of attention-based measures of language comprehension to identify differences in receptive vocabulary between children and adolescents with ASD and typically developing controls (Bavin et al., 2014; Brady et al., 2014; Chita-Tegmark, Arunachalam, Nelson, & Tager-Flusberg, 2015; Skwerer et al., 2016; Venker, Eernisse, Saffran, & Weismer, 2013). Yet, these measures are not without their own limitations. Traditional iterations of these assessments, such as the Intermodal Preferential Looking Paradigm (IPLP; Gogate, Bolzani, & Betancourt, 2006; Golinkoff, Hirsh-Pasek, Cauley, & Gordon, 1987; Golinkoff, Ma, Song, & Hirsh-Pasek, 2013) and the looking-while-listening methodology (LWL; Fernald, Zangl, Portillo, & Marchman, 2008; Fernald, Pinto, Swingley, Weinberg, & McRoberts, 1998) primarily rely on a two-item, forced-choice paradigm. That is, they only require participants to discriminate and preferentially fixate a verbally labeled target item from one other “distractor” item. This reliance on a two-item stimulus pair limits experimenter interpretation of looking behavior as there is opportunity for task performance to be impacted by the degree of similarity between the target and alternate distractor item as well as by the potential use of elimination based problem-solving strategies (see Hauschild et al., 2021 for review).

A recently developed attention-based task, the visual array task (VAT; Hauschild et al., 2021), expands the size of the visual array used in traditional attention-based measures of receptive vocabulary from two to eight items. This eliminates concerns associated with forced-choice paradigms and provides researchers an opportunity to measure receptive vocabulary with more visually complex stimuli, that better approximate naturalistic word learning and labeling contexts. The expansion of the stimulus array to eight items also allows experimenters to more finely manipulate the relations between the single target item and seven distractor items and closely approximates stimulus sets used in sequential touching paradigms, a common method used to measure category knowledge development in infants and young children (Mandler et al., 1987; Ricciuti, 1965; Starkey, 1981). During these assessments children are presented with a set of small objects belonging to two or more distinct categories (e.g., four animals and four food items) and given several minutes to free play. Awareness of category relations are inferred when children spontaneously touch multiple object members of the same category in sequence. The eight-item array of the VAT allows for the creation of an analogous measure of sequential looking behavior. In an initial study evaluating this task, experimenters were able to infer lexical knowledge from measures of visual attention distribution (preferential fixation) and category knowledge from measures of systematic scanning (sequential looking behaviors akin to sequential touching) in a sample of 17- and 25-month-old toddlers. Findings indicated that the tested iteration of the VAT was a sensitive measure of both language comprehension and category knowledge for typically developing toddlers (reflected in the successful extension of target labels to other objects that share the same function, thus implicating an understanding of the meaning of the target word rather than simply word object-category matching; Hauschild et al., 2021).

To date, the VAT has only been used with a typically developing sample. However, given that it does not require participants to provide a verbal or gestural response, the VAT may be a particularly effective method of measuring language comprehension and category understanding for infants and toddlers with or at risk for ASD. Therefore, the present study aimed to examine the relation between object label and object category knowledge using the same iteration of the VAT tested by Hauschild and colleagues (2021) in a sample of 17- and 25-month-old toddlers at heightened familial risk for developing ASD (the younger siblings of children with ASD; high-risk). To better understand the relation between early word comprehension and category knowledge for the high-risk toddlers, we addressed three primary questions.

First, do high-risk toddlers differ from their low-risk peers (the younger siblings of typically developing children; low-risk) in their ability to accurately identify the target item; thus, reflecting group differences in object label knowledge? Given the well-characterized early deficits in receptive language in infants and toddlers with ASD (Garrido, Watson, Carballo, Garcia-Retamero, & Crais, 2017), we predicted that toddlers in the high-risk group would demonstrate lower levels of target item identification than their low-risk peers. Although findings from the only previous work using an attention-based measure of receptive vocabulary did not identify differences between familial risk groups at 18 or 24 months of age (Chita-Tegmark et al., 2015), we believe that the added complexity of the eight item array task will induce risk group differences not observed by Chita-Tegmark and colleagues (2015) at these relative ages. It was also predicted that while the performance of the high-risk group may reflect gains in object label knowledge between 17 and 25 months of age, this trajectory of developmental improvement would not be as significant as that observed in the low-risk group.

Second, we addressed the question does the inclusion of a superordinate category contrast in the array differentially impact the attention distribution of high-risk and low-risk toddlers, reflecting group differences in object category knowledge? Hauschild, Pomales-Ramos, and Strauss (2021), reported an impact of category knowledge in measures of preferential item fixation and systematic scanning at both 17 and 25 months of age as well as relative growth in category knowledge between 17 and 25 months of age for low-risk toddlers. However, no prior work has examined measures of sequential looking in high-risk toddlers. Previous findings from tasks using measures of sequential touching in young children diagnosed with ASD (ages 3–6 years) indicated intact sequential touching behaviors reflective of basic and superordinate object category knowledge (Vitrano, 2015) that did not differ significantly from typically developing peers (Ungerer & Sigman, 1987). Yet, given that this evidence reflects sequential touching behaviors (not sequential looking behaviors) as well as the competencies of children with ASD older than the present sample, analyses related to risk group differences in category knowledge remained exploratory in nature; thus, no specific predictions were made about group differences in category knowledge.

Third, are there observable differences in the relation between object label knowledge and category knowledge for at-risk toddlers and their low-risk peers? Hauschild, Pomales-Ramos and Strauss (2021) reported correlation between measures of object label knowledge and category knowledge at 25 but not 17 months of age, indicating the emergence of a relation between receptive vocabulary and category understanding across the ages tested in this sample for low-risk toddlers. Given the reported difficulties in extension associated with ASD (Arunachalam & Luyster, 2016) and previous work indicating a weakened or overall lack of relation between receptive vocabulary and category knowledge for children diagnosed with ASD (Ungerer & Sigman, 1987; Vitrano, 2015), we predicted that a correlation between measures of object label knowledge and category knowledge will not be observed for the high-risk toddlers at either 17 or 25 months of age.

2. Method

2.1. Participants

Participants were recruited by the Autism Center of Excellence (ACE) at the University of Pittsburgh and drawn from a larger study conducted by the Center for Infant and Toddler Development (ITDC). The sample consists of infant siblings of children with ASD (high-risk infants; HR) and infant siblings of typically developing children (low-risk siblings; LR). HR participants had at least one older sibling with an ASD diagnosis confirmed by the Autism Diagnostic Observation Schedule-Generic (ADOS-WPS; Lord et al., 2001) and the Autism Diagnostic Interview-Revised (ADI-R; Lord, Rutter, & Couteur, 1994). LR participants had at least one typically developing older sibling as well as no first or second-degree relatives with ASD. Additional exclusion criteria for all participants were a birth weight less than 2500 grams, problems with pregnancy, labor or delivery, traumatic brain injury, prenatal illicit drug or alcohol use, and/or birth defects. Informed consent was obtained from the guardians of all study participants prior to participation and all study procedures were approved by the Institutional Review Board of the University of Pittsburgh and conform to Common Rule standards. Data from a substantial number of participants in the LR group have been previously reported in a prior publication (Hauschild, Pomales-Ramos, & Strauss, 2021). Data from participants in the HR group have not been previously reported.

The study sample included 88 toddlers that completed the VAT at 17 (LR n = 20; HR n = 27) and/or 25 months of age (LR n = 42; HR n = 22), resulting in the collection of data from a total of 111 unique task administrations. A subset of 23 toddlers (LR n = 14, HR n = 9), referred to here as the longitudinal cohort, completed the VAT at both 17 and 25 months of age (see Table 1).

Table 1.

Demographic information for the full sample and the longitudinal cohort

	Full Sample				LC
	17 Months		25 Months
Low-Risk	n	(M, SD) or (%)	n	(M, SD) or (%)	n	(M, SD) or (%)
Age 17 months (M, SD)	20	16.97, 1.90	-	-	14	17.04, 1.29
Age 25 months (M, SD)	-	-	42	24.75, 0.71	14	24.79, 0.60
Sex Assigned at Birth
Male (%)	13	(65%)	25	(60%)	10	(71%)
Female (%)	7	(35%)	17	(40%)	4	(29%)
Racial or Ethnic Minority (%)	1	(5%)	4	(10%)	1	(7%)
Maternal Education
High School (%)	0	(0%)	0	(0%)	0	(0%)
Some College or College Degree (%)	6	(43%)	12	(46%)	4	(36%)
Graduate of Professional School (%)	8	(57%)	14	(54%)	7	(64%)
Paternal Education
High School (%)	0	(0%)	0	(0%)	0	(0%)
Some College or College Degree (%)	7	(39%)	24	(60%)	6	(43%)
Graduate of Professional School (%)	11	(61%)	16	(40%)	8	(57%)
High-Risk	n	(M, SD) or (%)	n	(M, SD) or (%)	n	(M, SD) or (%)
Age 17 months (M, SD)	27	16.73, 0.77	-	-	9	17.09, 1.10
Age 25 months (M, SD)	-	-	22	25.46, 1.21	9	25.82, 1.27
Sex Assigned at Birth
Male (%)	17	(63%)	13	(59%)	7	(78%)
Female (%)	10	(37%)	9	(41%)	2	(22%)
Racial or Ethnic Minority (%)	7	(26%)	3	(14%)	1	(11%)
Maternal Education
High School (%)	3	(20%)	1	(9%)	1	(25%)
Some College or College Degree (%)	6	(40%)	6	(55%)	1	(25%)
Graduate of Professional School (%)	6	(40%)	4	(36%)	2	(50%)
Paternal Education
High School (%)	2	(8%)	0	(0%)	0	(0%)
Some College or College Degree (%)	13	(52%)	12	(60%)	5	(56%)
Graduate of Professional School (%)	10	(40%)	8	(40%)	4	(44%)

Notes. LC = Longitudinal Cohort; % based on total number of respondents per item.

2.2. The Visual Array Task

2.2.1. Stimuli

Visual stimuli replicated those of Hauschild et al. (2021) and were eight item arrays depicting prototypical color illustrations of objects whose names were commonly known to 17- and 25-month-old toddlers (as determined by normative scores of the MacArthur-Bates Communicative Development Inventories [MB-CDI; Fenson, 2007] and the Mullen Scales of Early Learning [MSEL; Mullen, 1995]; Figure 1). Individual items belonged to one of six superordinate categories, also familiar to children of the sample’s age range: animals, clothing, food, furniture, vehicles, and utensils (Mandler, Bauer, & McDonough, 1991; Ross, 1980; Table 2). Each stimulus array included an equal number of items from two distinct superordinate categories. For example, four objects belonging to the superordinate category of animals and four objects belonging to the superordinate category of food. Item locations in each stimulus array were randomized using a 5 (width) × 4 (height) grid system, and each item was projected at a visual angle of 7.7° (width) × 7.7° (height).

Figure 1.

Example visual stimulus arrays depicting contrasts of items belonging to the A. animal and food as well as the B. clothing and utensil superordinate categories.

Table 2.

Items belonging to each of the six superordinate categories included in the present iteration of the visual array task.

	Superordinate Category
	Animals	Clothing	Food	Furniture	Utensils
Object Category Members	Cat	Hat	Bread	Door	Plate
	Dog	Pants	Banana	Table	Cup
	Horse	Shoe	Cookie	Bed	Spoon
	Bird	Shirt	Apple	Chair	Bottle

Each visual stimulus was paired with an audio recording of a natural female voice that instructed participants to look at a target item: “(Target label, e.g., “bird”). Look at the (target label). Where is the (target label)? (Target label). See the (target label)?”. The phrasing and tone of the auditory prompt remained constant throughout the testing session, with only the target label changing trial by trial. For example, during a trial in which the target item was a bird, participants heard “Bird. Look at the bird. Where is the bird? Bird. See the bird?”. Onset of the auditory stimulus was time-locked to the visual stimulus presentation, and the phrasing of the auditory prompt lasted the full duration of the trial presentation (10 seconds).

2.2.2. Apparatus

A standalone Tobii X120 eye tracker positioned in front of the participant, approximately 81 cm from the screen, was used to record eye movement data. Stimuli were projected onto the screen using Tobii Studio software (Version 2.0.6) while eye movements were recorded at a sampling rate of 60 Hz per second, accuracy of 0.5 degrees of visual angle, spatial resolution of 0.2 degrees, and drift of 0.3 degrees. A Tobii fixation filter (Olsson, 2007) provided by Tobii Studio software was used to convert raw eye movement data into fixations.

2.2.3. Procedure

The testing session took place in a quiet, dark room designed to mimic that of a small movie theater and minimize distractions. Stimuli were rear projected onto a large projection screen (69 cm × 91 cm) that was located approximately 152 cm from the participant. Toddlers were seated independently with their caregivers next to, but slightly behind their line of sight. Caregivers were asked not to point to, verbally label, or narrate the experience of viewing items on the screen. However, they were encouraged to respond to their toddler’s bids for attention and comfort the toddlers if needed.

Following a brief acclimation period during which toddlers were shown a popular cartoon, a standard calibration process was completed. During this process small moving targets (paired with toddler friendly music) prompted toddlers to shift their gaze to five predetermined points on the screen. Once the toddlers fixated one of the five points, an experimenter manually advanced the target to the next location. Calibration was considered successful once the Tobii eye tracker detected the right and left eye of the participant at all five target locations. This calibration process was repeated until successful calibration was obtained for each participant.

Throughout the testing session, toddlers viewed 12 unique stimulus presentation trials (visual array paired with an auditory prompt of the target item label), each approximately 10 seconds in length. To maintain the participant’s attention as well as reorient gaze to a central location on the screen, a short cartoon was played in between each trial. Trial advancement was experimenter controlled, ensuring that the task did not proceed to the next trial until the toddler was calm and attending to the screen. The presentation of all possible target objects (24) and six superordinate category contrast combinations (out of a possible 15) was counterbalanced across participant testing sessions according to the procedures reported by Hauschild et al. (2021).

2.4. Data Reduction

All computed variables reflect the total summation of participant looking data captured throughout the entirety of each 10 second trial. Trials in which toddlers failed to fixate at least one of the eight objects were eliminated from analyses (38/1,332; approximately 3%). A two-way ANOVA conducted for the full sample showed that the number of trials eliminated did not vary by age (F(1, 107) = 1.80, p = .18) or risk (F(1, 107) = .89, p = .37) group, nor was there a significant interaction (F(1, 107) = 1.44, p = .23).

2.4.1. Areas of Interest (AOIs)

AOIs delineating a 190 × 190 square pixel region surrounding the eight items in each of the stimulus arrays were created using Tobii Studio software. Fixations (stabilized gaze on a single location) to these AOIs were then classified into three types: target fixations, target-category fixations, and other-category fixations. Target item fixations were defined as fixations within the AOI surrounding the item that was verbally labeled (1 per trial; e.g., “bird”). Target-category fixations were defined as fixations to any of the items belonging to the superordinate category of the target item (3 per trial; e.g., dog, horse, cat). Other-category fixations were defined as fixations to any of the items not belonging to the superordinate category of the target item (4 per trial; e.g., bread, apple, cookie, banana).

2.4.2. Visual Attention Distribution

To quantify attention distribution among items in each array, proportional fixation durations were computed. These proportions reflected the total amount of time toddlers spent looking within each AOI divided by the total duration of time spend looking to any of the eight AOIs, per stimulus trial. These base proportions were then used to compute the following three variables:

1. Proportion Target Fixation: Proportion of time spent fixating the target item averaged across the 12 trials.

2. Proportion Target-Category Fixation: Proportion of time spent fixating target-category items summed and averaged across the 12 trials.

3. Proportion Other-Category Fixation: Proportion of time spent fixating other-category items summed and averaged across the 12 trials.

2.4.3. Systematic Scanning

To measure the impact of category understanding on scanning behaviors, trained coders manually scored the number and sequence of visits (i.e., each time a participant’s gaze entered and exited an AOI) between the eight objects using a video playback of the eye tracking session. These visit sequences were then used to calculate two broad measures of sequential scanning behaviors: number of runs and run length. A run was operationalized as two or more sequential visits to AOIs belonging to the same superordinate category.

1. Number of Runs (NRuns):

   1. NRuns Target-Category: Number of runs made between items belonging to the superordinate category of the target item, summed across all 12 trials.
   2. NRuns Other-Category: Number of runs made between objects belonging to the superordinate category unrelated to the target object, summed across all 12 trials.
   3. NRuns Ratio: NRuns Target-Category divided by NRuns Other-Category. A value greater than 1 indicates a preference for making more frequent runs between items belonging to the target category.
2. Mean Run Length (MRL):

   1. MRL Target-Category: The length of all runs within the target category divided by the total number of runs, summed across all 12 trials
   2. MRL Other-Category: The length of all runs within the other-category divided by the total number of runs, summed across all 12 trials
   3. MRL Ratio: MRL Target-Category divided by MRL Other-Category. A value greater than 1 indicates a preference for making longer sequential runs between items belonging to the superordinate category of the target.

3.0. Results

Primary study analyses included data from participants in the longitudinal cohort in order to maximize power for detecting risk group differences at either 17 or 25 months of age. Means and standard deviations for the primary variables of interest are summarized in Table 3.

Table 3.

Eye tracking-derived variables for the full sample and longitudinal cohort.

	Full Sample						Longitudinal Cohort
	Low-Risk			High-Risk			Low-Risk			High-Risk
	N	M	SD	N	M	SD	N	M	SD	N	M	SD
17 months of age
Proportion of Fixation
Duration
Target	20	0.19	0.06	27	0.16	0.06	14	0.18	0.05	9	0.16	0.02
Target-Category	20	0.38	0.08	27	0.39	0.09	14	0.36	0.06	9	0.35	0.10
Other-Category	20	0.43	0.08	27	0.46	0.09	14	0.46	0.07	9	0.50	0.10
Number of Runs
Target-Category	20	14.50	3.85	27	14.47	6.47	14	14.36	4.07	9	12.56	5.00
Other-Category	20	12.00	4.10	26	13.46	5.67	14	12.86	4.20	9	12.22	6.50
Mean Run Length
Target-Category	20	3.42	0.68	27	3.39	0.82	13	3.58	0.75	9	3.32	0.91
Other-Category	20	3.44	0.81	26	3.26	0.77	13	3.70	0.87	9	3.18	0.90
NRuns Ratio	20	1.34	0.54	26	1.21	0.55	13	1.17	0.45	9	1.26	0.65
MRL Ratio	20	1.03	0.26	26	1.08	0.26	13	1.01	0.29	9	1.05	0.12
25 months of age
Proportion of Fixation
Duration
Target	42	0.28	0.09	22	0.26	0.09	14	0.27	0.08	9	0.26	0.10
Target-Category	42	0.36	0.05	22	0.35	0.06	14	0.37	0.06	9	0.35	0.06
Other-Category	42	0.36	0.07	22	0.40	0.09	14	0.36	0.08	9	0.40	0.06
Number of Runs
Target-Category	42	19.52	5.99	22	17.55	4.67	14	20.79	6.66	9	15.56	5.22
Other-Category	42	13.45	3.96	22	13.18	3.67	14	14.36	4.68	9	10.89	3.02
Mean Run Length
Target-Category	42	3.60	0.70	22	3.36	0.63	14	3.36	0.48	9	3.29	0.41
Other-Category	42	2.81	0.45	22	2.93	0.55	14	2.85	0.35	9	3.09	0.77
NRuns Ratio	42	1.48	0.43	22	1.39	0.39	14	1.48	0.41	9	1.46	0.42
MRL Ratio	42	1.30	0.28	22	1.18	0.28	14	1.19	0.21	9	1.11	0.27

Note: NRuns Ratio = number of runs ratio; MRL ratio = mean run length ratio

Demographic variables, including chronological age, sex assigned at birth, and whether participant families identified their child as a member of a racial or ethnic minority were compared between risk groups at 17 and 25 months of age using either chi-square tests or independent samples t-tests as appropriate. At 17 months of age, the risk groups were found not to differ on chronological age (t(45) = .87, p= .39), sex assigned at birth (χ²(1) = .02, p = .87), or their identification as a member of a racial or ethnic minority (χ²(1) = 3.78, p = .052). At 25 months of age, risk groups were found not to differ on sex assigned at birth (χ²(1) = .001, p =.97) or their identification as a member of a racial or ethnic minority (χ²(1) = .218, p = .64); however, a small but statistically significant difference was identified in chronological age (t(162) = −3.09, p = .003) indicating that toddlers in the HR group were on average approximately 0.71 months older than toddlers in the LR group.

3.1. Accuracy of target identification

3.1.1. Do HR and LR toddlers differ in their ability to accurately identify the target object?

In order to determine if there were any differences between the HR and LR toddlers with respect to the proportion of target fixation, one-way ANOVAs were conducted for the full sample that included risk (HR vs. LR) as a between factor for each age point. Results indicated that at 17 months of age the LR group (M = 0.19, SD = 0.06) spent a significantly longer proportion of their fixation duration looking to the target object than the HR group (M = 0.16, SD = 0.06; F(1, 45) = 4.50, p = 0.04). In contrast, at 25 months of age, there were no significant differences in the proportion of fixation duration to the target object made by the LR (M = 0.28, SD = 0.09) and HR toddlers (M = 0.26, SD = 0.09; F(1, 62) = 1.15, p = .29). This suggests that HR toddlers demonstrate a reduction in preferential looking to the target object compared to LR infants at 17 months of age but catch up by 25 months of age.

To determine whether participants fixated the target object greater than would be expected by chance, one-sample t-tests were conducted for both the 17- and 25-month groups that compared the proportion of fixation duration of the target object to the value of 1/8 or 0.125. This value of 1/8 assumes equal distribution of fixation duration between all eight items of the array. Since the one-way ANOVA determined that there was a significant group difference at 17 months of age, separate one-sample t-tests were conducted for the LR and HR groups. At 17 months of age both the LR (t(19) = 4.84, p < .001) and HR (t(26) = 2.79, p = .01) groups fixated the target object for a proportion of duration greater than chance. This finding indicates that although the LR group fixated the target object proportionally longer than the HR group at 17 months of age, both groups are able to identify and preferentially fixate the target object. Because the one-way ANOVA determined that there were no significant differences in the proportion of fixation duration to the target object between LR and HR toddlers at 25 months of age, the two risk groups were combined. Results revealed that at 25 months of age toddlers spent a significantly greater proportion of time looking to the target object than would be expected by chance (t(63) = 13.12, p <.001).

3.1.2. Do HR and LR toddlers demonstrate different developmental trajectories of target identification?

To test for group differences and developmental trajectories of target fixation, a two-way repeated measures ANOVA was conducted for the proportion of target fixation duration that included age (17-months vs. 25-months) as a within-subjects factor and risk (LR vs. HR) as a between-subjects factor. A significant main effect for age (F(1,21) = 31.77, p < .001), indicated that from 17 (M = 0.17, SD = 0.04) to 25 months of age (M = 0.27, SD = 0.09), regardless of risk, toddlers increased the proportion of time they spent fixating the target object. There was neither a significant main effect of risk (F(1,21) = 0.56, p < .45), nor a significant interaction between age and risk (F(1,21) = 0.18, p < .67).

3.1.3. Do target identification abilities correlate across age?

Results of Pearson’s correlations indicated that the LR group showed a strong correlation between the proportion of target looking at 17 months of age and the proportion of target looking at 25 months of age (r(12) = .61, p = .02). No such correlation was found for the HR group (r(7) = −.14, p = .72). This likely reflects a more consistent ability of toddlers in the LR group to accurately identify the target item over time (i.e. ranking of individual ability that remains consistent over time or consistent high ability across individuals and the two time points) and greater overall heterogeneity in the developmental trajectories of target identification among toddlers in the HR group.

3.2. Impact of category contrast on attention distribution

3.2.1. Do HR and LR toddlers differ in their proportion of other-category object fixation?

Similar to the analyses that were conducted to evaluate preferential target looking, separate one-way ANOVAs examining the proportion of other-category fixation duration were conducted for each age group that included risk (HR vs. LR) as a between-subjects factor for the full study sample. Results indicated that there were no differences by risk group at 17 months of age (F(1,46) = 1.54, p = .22). Regardless of risk status, toddlers spent approximately 45% (M = 0.45, SD = 0.09) of their time fixating the four objects that did not belong to the target’s superordinate category. A one-sample t-test comparing this mean to the chance value of 4/8 or 0.5 indicated that toddlers were looking to the other-category items significantly less than would be expected assuming equal fixation distribution between all eight objects (t(46) = −4.36, p < .001). This observed reduction in attention to other-category items may be a by-product of preferential looking toward the target-item or indicate that toddlers direct their attention away from items not belonging to the same superordinate category as the labeled item, reflecting an ability to rule out other-category objects as possible referents.

A marginal group difference was found for other-category object looking at 25 months of age (F(1, 62) = 3.69, p = .06). At 25 months of age, HR toddlers (M = 0.40, SD = 0.09) looked longer to the other-category objects than the LR toddlers (M = 0.36, SD = 0.07). HR toddlers spent 40% of their time looking to the other-category objects while LR toddlers looked to those objects for only 36% of their time. To determine if the participants were fixating the other-category items less than would be expected by chance (0.5) separate one-sample t-test were conducted for LR and HR toddlers at 25 months of age. Results indicated that both LR (t(41) = −12.53, p < .001) and HR toddlers (t(21) = −5.70, p < .001) looked to the other-category objects for a duration that was less than chance. This suggests that although there was a marginal difference between the LR and HR toddlers, both groups again reduced looking to the other-category objects in favor of increasing looking to the target object and/or within-category members at 25 months of age, thus demonstrating knowledge of superordinate category membership and organization.

3.2.2. Do HR and LR toddlers demonstrate different developmental trajectories of other-category object fixation?

To test for group differences in developmental trajectories of other-category fixation a two-way repeated measures ANOVA was conducted for the proportion of other-category fixation duration that included age (17-months vs. 25-months) as a within-subjects factor and risk (LR vs. HR) as a between subjects factor. A significant main effect of age (F(1,21) = 16.89, p < .001) indicated that from 17 (M = 0.47, SD = 0.08) to 25 months of age (M = 0.38, SD = 0.08), regardless of risk, toddlers decreased the proportion of time they spent fixating the four other-category objects. There was neither a significant main effect of risk (F(1,21) = 2.70, p < .12), nor a significant interaction between age and risk (F(1,21) = 0.02, p < .90). It is possible that toddlers either become more efficient in sustaining attention to the target-item or at shifting attention away from other-category objects with age regardless of ASD risk, signifying gains in superordinate category understanding.

3.2.4. Do HR and LR toddlers differ in their sequential looking behaviors during the VAT?

One-way ANOVAs examining the number of runs (NRuns) target category, the NRuns other category, mean run length (MRL) target category and MRL other category including risk (HR and LR) as a between-subjects factor were conducted for each age to identify any differences in the visual scanning patterns between groups. There were no significant differences identified between the risk groups at either 17 or 25 months of age.

One-way ANOVAs including risk (HR and LR) as a between-subjects factor were conducted for each age to identify any group differences in the NRun ratio and MRL ratio variables. No differences by risk-group were found.

3.2.5. Do patterns of sequential looking change across development?

To test for group differences in the longitudinal cohort two-way repeated measures ANOVAs were conducted for NRuns target category, NRuns other-category, MRL target category and MRL other category that included age (17-months vs. 25-months) as a within subjects factor and risk (LR vs. HR) as a between subjects factor. A significant main effect of age (F(1,21) = 9.99, p < .01) as well as a marginal main effect of risk (F(1,21) = 4.07, p = .06) was observed for NRuns target category variable. Regardless of risk, between 17 (M = 13.65, SD = 4.44) and 25 months of age (M = 18.74, SD = 6.55), toddlers increased the number of runs that they made within the target category. Regardless of age, HR (M = 14.06, SD = 2.71) toddlers made marginally fewer runs within the target category than LR toddlers (M = 17.57, SD = 4.73). There was no interaction observed between risk and age for the NRuns target category variable. There were no significant main effects or interactions observed for the NRuns other-category or MRL target category variables. A marginal main effect of age was also observed for the MRL other category variable (F(1,21) = 3.39, p = .08). This suggests that, regardless of risk, between 17 (M = 3.44, SD = 0.90) and 25 (M = 2.94, SD = 0.55) months of age toddlers may reduce the length of runs between objects that belong to the superordinate category unrelated to the target object. A significant main effect of risk or a risk by age interaction was not observed for the MRL other-category variable.

3.3. Relation between receptive vocabulary and category knowledge

3.3.1. Do measures of object label knowledge correlate with measures of category knowledge for the HR and LR groups?

To identify possible relations between measures of object label and object category knowledge, correlation matrices were computed separately for HR and LR toddlers in the full sample at each age point. The eye tracking variables included in the matrix were: (1) proportion of target fixation duration, (2) proportion of other-category fixation duration, (3) NRuns ratio, and (4) MRL ratio (Table 4).

Table 4.

Correlation matrices reporting relations between variables indexing object label and category knowledge for the low-risk and high-risk samples. Pearson’s correlation coefficients for the 17-month-old groups are displayed on the left half of each chart and coefficients for the 25-month-old groups are displayed on the right.

	17 months of age				25 months of age
Low-Risk Sample	(1)	(2)	(3)	(4)	(1)	(2)	(3)	(4)
(1) Proportion of Target Fixation	-	−.46*	.11	.07	-	−.79**	.68**	.32*
(2) Proportion of Other-Category Fixation		-	−.69**	−.47*		-	−.69**	−.46**
(3) Number of Runs Ratio			-	−.03			-	.35
(4) Mean Run Length Ratio				-				-
High-Risk Sample	(1)	(2)	(3)	(4)	(1)	(2)	(3)	(4)
(1) Proportion of Target Fixation	-	−.37	.38	−.02	-	−.77**	.44*	.44*
(2) Proportion of Other-Category Fixation		-	−.45*	−.21		-	−.41	−.68**
(3) Number of Runs Ratio			-	−.38			-	.09
(4) Mean Run Length Ratio				-				-

Note

^*p<.05

^**p<.001.

At 17 months of age, a significant correlation between the proportion of target fixation duration and the proportion of other-category fixation duration was observed for the LR group. This indicates that as LR toddlers increased looking to the target object, they decreased the amount of time they spent fixating the four other-category objects. No significant correlations between the proportion of target fixation and the three measures of category knowledge were observed for the HR group.

At 25 months of age, there were significant correlations between proportion of target fixation and all three measures of category knowledge for both the HR and the LR toddlers. A strong negative correlation was observed between the proportion of target fixation duration and the proportion of other-category fixation duration for both the LR and HR toddlers. Again, this indicates a reciprocal relation between time spent looking to the target object and time spent looking to the four other-category objects. A moderate positive correlation was observed between the proportion of fixation duration to the target item and the NRuns ratio for the LR and HR toddlers. This finding indicates that as toddlers increase time spent looking to the target item they are also increasing the proportion of runs that they make within the target category. Finally, a moderate to low positive correlation between the proportion of target fixation duration and the MRL ratio was observed for the LR and HR toddlers. This indicates that as toddlers increased their proportion of time spent looking to the target object, they also increased the length of runs they made between objects belonging to the target category compared to items belonging to the other superordinate category.

4.0. Discussion

4.1. Accuracy of target identification

As predicted, an early emerging difference in the primary measure of object label knowledge, proportion of target fixation, was observed between the risk groups at 17 months of age. That is, the HR toddlers were found to spend proportionately less time fixating the target object than the LR toddlers. Contrary to our prediction, a significant difference between groups was not observed at 25 months of age. This suggests that HR toddlers were less proficient than their LR peers in identifying the target object at 17 but not 25 months of age, which may be indicative of early deficits in lexical growth for HR toddlers that diminish over time. Given the number of observed differences in receptive vocabulary size between risk groups in the broader literature (Garrido, Petrova, Watson, Garcia-Retamero, & Carballo, 2017; Jones, Gliga, Bedford, Charman, & Johnson, 2014; Tager-Flusberg et al., 2005) and that deficits in receptive language associated with ASD persist throughout childhood and into adulthood (Boucher, 2012; Tager-Flusberg et al., 2005), it is more likely that the lack of an observed difference between the older groups reflects particular characteristics of the stimuli used in the current iteration of VAT. In order to optimize the likelihood of validating the paradigm in a sample of toddlers at familial risk for ASD, the objects included in the current iteration of the VAT were carefully chosen to be commonly understood by toddlers in the tested age ranges. Furthermore, the stimulus arrays and target object labels included in the task did not vary across testing sessions. As task difficulty did not increase between testing sessions, the present study may not have been able to capture a persistent difference that likely exists between the groups at 25 months of age.

No other differences pertaining to target fixation duration were observed between the groups. As predicted, both the HR and LR toddlers fixated the target item for a longer duration than would be expected by chance at 17 and 25 months of age. Furthermore, analyses of the longitudinal cohorts failed to identify differential trajectories of target fixation across age between the HR and LR groups.

This study is the first to test the VAT in a sample of toddlers at familial risk for ASD; however, previous work has examined the receptive language abilities of HR toddlers using a gaze-based task (Chita-Tegmark et al., 2015). Using the looking-while-listening procedure, Chita-Tegmark and colleagues (2015) compared the receptive vocabularies of toddlers at heightened familial risk for ASD to their low-risk peers at 18, 24 and 36 months of age. Unlike the present study, the findings of this work suggested a later emerging difference in target fixation between the risk groups at only 36 months of age. Discrepancy in the findings of these two studies may be due to two distinct underlying factors. First, the added complexity of the eight-item arrays used in the VAT may have induced risk group differences at 17 months of age not observed by Chita-Tegmark and colleagues (2015) at 18 months of age using a two-item paired comparison. Second, the stimulus sets employed by Chita-Tegmark and colleagues included both early-acquired target nouns (e.g., bottle and shoe), similar to those used in the present study, as well as late-acquired target nouns (e.g., weasel and kiwi). The inclusion of late-acquired nouns likely resulted in a comparative increase in task difficulty that provided Chita-Tegmark and colleagues (2015) with an overall greater variability in task performance that led to a more sensitive measure of receptive vocabulary difference for at risk toddlers in this later developmental age group. Future research examining a more exhaustive assortment of vocabulary words as well as a broader range of developmental time points needs to be conducted in order to better delineate a developmental trajectory of receptive language growth associated with familial risk for ASD.

4.2. Impact of category contrast on attention distribution

Differences between groups for the proportion of other-category fixation followed the opposite pattern. While no differences between risk groups were observed at 17 months of age, a marginally significant difference was observed at 25 months of age. This indicated that toddlers in the HR group fixated the four other-category items marginally longer than toddlers in the LR group. One interpretation of this finding is that toddlers in the HR group needed to look longer at the other-category objects to rule them out as possible referents than toddlers in the LR group. While it is unknown whether this decreased difficulty in rejecting other-category objects for the LR group is due to semantic, perceptual, functional or even conceptual differences between the target object and other-category objects, these are all features that contribute to category formation and organization. Therefore, a preference to fixate the target object as well as target-category objects over other-category objects is indicative of superordinate category knowledge and organization. However, due to the small but statistically significant difference in chronological age noted between risk groups at 25 months of age, this observed marginal difference must be interpreted with caution. Future work should aim to replicate this finding using a sample more closely matched on chronological age.

No differences were observed between the groups in the full sample with respect to our measures of sequential looking. A main effect was identified in the longitudinal cohort that indicated HR toddlers made fewer runs between target-category members regardless of age. However, they were just as likely as their LR peers to demonstrate a preference for making runs within the target category. This may indicate that the sequential looking behaviors of the HR group are impacted to a lesser degree by hearing the verbal target label. While the HR toddlers still make more runs within the target category than the other category, this preference does not appear to be as strong as that of toddlers in the LR group.

4.3. Relation between object label knowledge and category knowledge

Another early emerging difference between risk groups was observed between the relation of the study’s measure of object label knowledge (proportion of target fixation) and one of the study’s measures of object category knowledge (proportion of other-category fixation). At 17 months of age, there was a negative correlation observed between the proportion of target fixation and proportion of other-category fixation for the LR group suggesting a reciprocal relation between these variables. That is, as toddlers in the LR group increased their attention to the target object, they decreased their attention to other-category objects, demonstrating a recognition of the superordinate category relations tested. No correlation between these measures of object label knowledge and category knowledge was observed for the HR group at 17 months of age.

No differences between risk groups were observed at 25 months of age. Regardless of risk, significant correlations between the measure of object label knowledge and the three measures of object category knowledge (proportion of other-category fixation, number of runs ratio, and mean run length ratio) were found for both the LR and HR groups. This finding suggests that the development of the relation between object label knowledge and object category knowledge in HR infants may be delayed compared to their LR peers. However, they demonstrate an ability to catch up by 25 months of age, as measured by this iteration of the VAT.

4.4. Conclusions

Early delays in receptive language are consistently reported for children that later go on to receive an ASD diagnosis (Landa & Garrett-Mayer, 2006; Mitchell et al., 2006; Zwaigenbaum et al., 2005). In the present study, a difference in target identification between risk groups was observed at 17 but not 25 months of age. One interpretation of these findings is that toddlers at high-risk (HR) for ASD may demonstrate early delays in receptive vocabulary, reflected in smaller lexicons, but quickly catch up to their low-risk (LR) peers. However, the iteration of the VAT used in the present study intentionally included object labels that toddlers could reasonably be expected to know at 17 months of age. Therefore, a lack of group differences observed at 25 months of age likely reflects ceiling effects in task performance for both groups. Future work should test alternative iterations of that VAT that include more advanced vocabulary words in order to better differentiate group performance at later developmental ages.

An alternative interpretation of the observed difference in target identification between the groups at 17 months of age is that this finding may reflect a difference in object label processing speed, rather than an overall difference in object label knowledge. In the present task, onset of the visual and auditory stimulus occurred simultaneously. This prevented toddlers from visually exploring the object arrays prior to hearing the target object labeled. With an array size of eight items, there is potential for a high level of variability in the number of objects toddlers may fixate before first locating the target object, regardless of whether or not they recognize the object label. In this way, the early time course of gaze-data collected during this task approximates that from a visual search task. In order to better address whether or not differences in overall target fixation may reflect a difference in overall object label knowledge or object label processing speed, future work should vary the time between visual stimulus and auditory stimulus onset. Giving toddlers a period of time to scan the object array and become familiar with object locations prior to labeling the target object may allow experimenters to examine the time course of gaze data immediately following target object labeling. Such analyses would be similar to those commonly conducted when using the looking-while-listening method and may provide a direct means of testing for potential differences in processing speed.

This study was the first to examine the development of object category knowledge in a HR and LR sibling population. Research with older children and adults with ASD has suggested a general deficit in the domain of categorization (Gastgeb & Strauss, 2012) as well as a weakened link between receptive language abilities and category knowledge (Ungerer & Sigman, 1987). While a marginal finding suggested a lesser influence of the target label on attention distribution for the HR group compared to the LR group, this did not reach a level of significance. Therefore, data reported here do not support a deficit in the early knowledge of superordinate categories. A significant pattern observed was a weakened relation between the measure of receptive language (target identification) and measures of superordinate category understanding (fixation of other-category members and patterns of sequential looking) at 17 months of age. This suggests that differences in category knowledge may be a meaningful measure to consider when conceptualizing the receptive language deficits associated with HR populations. During language learning, typically developing children are sensitive to the common features of category members and use this information to generalize known object labels to newly encountered exemplars (Ferguson & Waxman, 2017). The inability to identify similarities between category members and/or utilize this information when learning new object referents may be a potential mechanism for the delays observed in the HR population. A lack of a similar finding at 25 months of age may simply be due to the limited vocabulary tested during this particular session. Future studies should further explore this emerging relation in children at risk for and with ASD.

The low behavioral demands placed on participants makes the VAT an ideal measure to use with atypically developing or otherwise impaired populations where task demands may be prohibitory to the successful completion of current standardized assessments. Data reported here validate the feasibly of using the VAT to measure receptive vocabulary and category knowledge in toddlers at familial risk for ASD. Without modification, the VAT can accommodate the testing of non-verbal to minimally verbal populations, as well as populations, such as children with anxiety, that may encounter many of the interaction-based hurdles similar to children with ASD. Thus, the VAT is a measure that can be administered uniformly across a broad spectrum of populations. Furthermore, the inherent flexibility of the stimulus arrays of the VAT allows experimenters the ability to test any number of lexical or category constructs, making it a highly customizable research paradigm.

Highlights.

• Successful application of the visual array task with infants at risk for ASD

• At risk toddlers demonstrated delays in object label knowledge at 17 months of age

• Receptive vocabulary related to category knowledge for at risk and control groups