Spatial Language and Cognition in Autistic Preschoolers

The development of spatial language – the lexical domain describing the features, locations, and relations of and between objects in space – has garnered considerable interest in the neurotypical (NT) literature as an intersection point between linguistic and nonverbal visuospatial skills, providing valuable insight into the relationship between language and cognitive development. While spatial language and spatial cognition are closely linked in NT children, autistic individuals demonstrate an uneven profile where visuospatial skills have been proposed as an area of relative strength (Edgin & Pennington, 2005; Stevenson & Gernsbacher, 2013), but linguistic skills are much more variable (Anderson et al., 2007; Brignell et al., 2018; Ellis Weismer & Kover, 2015; Tager-Flusberg, 2016). This uneven profile has motivated a small number of recent studies to begin investigating spatial language in autistic adults, adolescents, and older children (Bochynska, Coventry, et al., 2020; Bochynska, Vulchanova, et al., 2020), which have yielded several insights including a surprising relationship between ASD symptom severity and spatial language. However, the early development of the relationship between spatial language and nonverbal spatial cognition in ASD has not yet been explored. Moreover, the contributions of child-based and environmental variables known to influence spatial language in NT children and older autistic individuals also warrant examination in a younger cohort of autistic children. Addressing this gap in the literature, the present study investigated the longitudinal relationship of autistic children’s spatial language and nonverbal spatial cognition in the context of parent-child interactions and examined two predictors of spatial language production identified in previous studies - ASD symptom severity and parent spatial language input.

Spatial Language in Typical Development

For most neurotypical (NT) children, spatial language acquisition begins early in childhood coinciding with conceptual development, such that more complex spatial relational terms are acquired later than simpler terms (Weist et al., 2000). Spatial language and visuospatial cognition are tightly yoked in NT children (see Wu et al., 2022 for a review). This connection has been cited extensively as evidence of the critical role of language in structuring conceptual development (e.g., Gentner et al., 2013; Loewenstein & Gentner, 2005; Majid et al., 2004; Pyers et al., 2010). Preschool children’s spatial language abilities have been linked to performance in a number of nonverbal spatial tasks (Hermer-Vazquez et al., 2001; Hund et al., 2021; Miller et al., 2020; Polinsky et al., 2017; Pruden et al., 2011; Simms & Gentner, 2019; Verdine et al., 2014). For example, 3- to 4-year-old children’s parent-reported spatial language has been implicated in spatial search task performance (Hund et al., 2021) and spatial assembly (Verdine et al., 2014), while 4-year-olds’ spatial word production during dyadic parent interaction predicted spatial reasoning abilities (as measured by time to complete puzzle; Polinsky et al., 2017). Preschoolers’ spatial word production was also predictive of nonverbal spatial problem-solving (Pruden et al., 2011). Task-relevant spatial language predicted mental rotation abilities in 4.5- to 6-year-old children (Miller et al., 2020), reorientation in 5- to 7-year-old children (Hermer-Vazquez et al., 2001) and ability to locate a midpoint between objects in 2.5- to 5.5-year-old children (Simms & Gentner, 2019).

Spatial Language and Cognition in ASD

Despite carrying important theoretical implications concerning the language-cognition interface in this population, considerably less is known about spatial language abilities in ASD. Two recent studies have examined spatial language in autistic adults, adolescents, and older children (ages 9 – 27) who had language and cognitive abilities in the average range (often referred to in the literature as “high-functioning,” Bochynska, Coventry, et al., 2020; Bochynska, Vulchanova, et al., 2020). Taking into account previous work suggesting areas of relative strength in spatial cognition and preliminary reports of difficulties with spatial language (Perkins et al., 2006), the first study examined the relationship between spatial language and non-linguistic spatial representations (Bochynska, Vulchanova, et al., 2020). Results indicated no differences in spatial representation between autistic participants and nonverbal cognition- and age-matched NT peers, though the ASD group demonstrated difficulties with spatial language over and above general language ability which the researchers proposed may reflect a delay in the development of spatial language. Overall linguistic abilities did not account for the group difference in spatial language, which was driven by the younger participants in the ASD group (Bochynska, Vulchanova, et al., 2020). Findings revealed subtle differences in spatial language such that the ASD group made more errors in processing direction words (i.e., right/left) and produced fewer proximal terms (i.e., near/far). Structurally, areas of difficulty in spatial language in the ASD group were reflected in their spatial representations, suggesting parallels between linguistic and non-linguistic spatial abilities in ASD similar to that of NT peers. While this study provided critical new insights into spatial language and nonverbal spatial cognition in ASD, it is not yet known whether this connection is as robust in early development. Given that spatial language appeared to improve with age in this previous study while the foundations of spatial cognition typically begin early in prelinguistic development (Casasola, 2018), spatial language development may lag behind nonverbal spatial cognition in this population. Therefore, a goal of the present study was to extend these findings by investigating how the relationship between spatial language and spatial cognition unfolds in the early development of autistic children.

ASD Symptom Severity as a Predictor of Spatial Language

In another study, Bochynska, Coventry and colleagues (2020) reported greater difficulties in production of select spatial terms and spatial description recall in autistic individuals than nonverbal cognition- and age-matched NT peers which were not explained by overall language ability. Taken together, the findings of these studies suggest the areas of deficit observed in spatial language abilities of the ASD group may be attributable to other individual or environmental factors. An analysis of potential contributors to task performance in Bochynska, Coventry et al. (2020) yielded a surprising finding - ASD symptom severity was predictive of performance in both spatial language production and spatial description recall, suggesting that spatial language difficulties may be attributable to core characteristics of ASD. The authors did not speculate as to why ASD symptoms might impair spatial language beyond the overall language impairments often associated with ASD, and it is unlikely that this finding could be attributed solely to developmental delay as chronological age was also included as a predictor in these models. One possibility is that overall language ability, which was not included as a predictor, confounded the observed relationship between ASD symptom severity and spatial language production and recall. To tease apart their respective contributions, the present study examined ASD symptom severity as a predictor of children’s spatial language production over the preschool years when accounting for overall language. While investigating child characteristics associated with spatial language in ASD yielded important insights, a more complete understanding of early spatial language development also requires consideration of environmental factors. In the NT literature, a robust predictor of child production is parent input (Cartmill et al., 2010; Pruden et al., 2011; Pruden & Levine, 2017).

The Role of Parent Input in Spatial Language Acquisition

While its effect on early spatial language acquisition in autistic children is not yet known, several studies in the NT literature have implicated parent input as a critical factor shaping child spatial language. Pruden and colleagues (2011) examined the relationship between parent input and child production of three types of spatial language: shape terms (e.g., circle, triangle), dimensional adjectives (e.g., big, little), and spatial features (e.g., bent, curvy) during dyadic interactions in their homes longitudinally over 14–46 months of age and administered nonverbal spatial tasks when children were 52 months. Results indicated considerable variability in parent spatial language production. Parent spatial language was predictive of child spatial language production when controlling for overall parent language input, and child spatial language production was in turn predictive of their performance on nonverbal spatial tasks at 52 months (Pruden et al., 2011). Using the same sample and coding scheme for spatial language, Pruden and colleagues also found that boys heard and produced more spatial words than girls (but not more of all other word types), and that parent spatial language input when children were 14–26 months mediated these sex differences in child spatial word production at 34–36 months (Pruden & Levine, 2017). In a third study, they investigated whether adding gesture to parent spatial language input would be more predictive of children’s later spatial language production than parent spatial language alone. Parent spatial utterances including gestures predicted child spatial language production when parent spatial utterances without gesture and non-spatial utterances were controlled (Cartmill et al., 2010). Together, these findings provide strong evidence in support of a link between parent and child spatial language production among NT children. However, whether parent input would have a similar effect on autistic children’s spatial language has not yet been determined. While parent language input is broadly similar between NT and ASD groups (Bang & Nadig, 2015), difficulties in aspects of social communication and language processing may limit the ability of autistic children to benefit from parent input (Arunachalam & Luyster, 2018). Therefore, the relationship between parent and child spatial word production during dyadic interaction may be attenuated compared to that found in studies of NT children. Conversely, proposed strengths in visuospatial skills in ASD (e.g., Edgin & Pennington, 2005) may enhance children’s interest in (and therefore attention to) spatial words in parent input, resulting in a robust relationship between parent and child spatial word production.

The Present Study

In this study, we addressed several gaps in the literature regarding the development of spatial language in young autistic children. First, prior work in this area has been less generalizable to the broader population given the narrow range of linguistic and cognitive abilities in their samples. Therefore, we recruited younger children with a more representative spread of language and cognitive ability than previous studies (Bochynska, Coventry, et al., 2020; Bochynska, Vulchanova, et al., 2020) to increase the generalizability of our findings. Second, previous studies have not considered change in spatial language and its associated factors over time, which will be critical to understanding how development in spatial language unfolds. Using standardized assessments and language samples collected during naturalistic parent-child interactions at multiple time points, we sought to investigate the interactive relationship between child spatial language production and nonverbal spatial cognition over time during the preschool years. Based on previous findings in NT preschoolers and older autistic individuals, we anticipated a significant association between these two domains which may grow in strength over time as linguistic abilities begin to catch up to nonverbal cognition for some children. However, if early language delays are pervasive enough to dissociate spatial language production from nonverbal spatial cognition in this stage of development, we may not observe such a relationship. Next, we examined potential child and environmental predictors of child spatial language production found in studies of older autistic individuals and NT children. We asked whether ASD symptom severity was associated with child spatial language production over the preschool years. If the relationship of ASD symptoms to spatial language in early development reflects that of older autistic individuals, then we would expect ASD symptom severity to predict child spatial language production even when accounting for overall language ability (Bochynska, Coventry, et al., 2020; Bochynska, Vulchanova, et al., 2020). If, however, the effect of ASD symptom severity is confounded by linguistic ability, then we would expect only overall language to predict spatial language production. Finally, we examined the effect of parent spatial language input on child production during the preschool years. If the relationship between parent and child spatial word production mirrors that of NT children, then we would expect a significant relationship between parent spatial input and child production even when accounting for overall parent and child language. Conversely, this association might not be observed if social communication and language processing difficulties associated with ASD impair children’s ability to benefit from parent input (i.e., Arunachalam & Luyster, 2018). In this paper, we answered the following primary research questions:

1. What is the longitudinal relationship between young autistic children’s spatial language production and nonverbal spatial cognition when accounting for overall child language during the preschool years?

2. Does ASD symptom severity predict children’s spatial word production over and above the effect of overall child language?

3. What is the effect of parent spatial language input on autistic preschool children’s spatial word production when accounting for overall parent and child language?

Methods

Participants

Data utilized in this study were initially collected as part of a large longitudinal project investigating predictors and outcomes of language growth in autistic preschool children, which was approved by the Education and Behavioral/Social Sciences Institutional Review Board at the University of Wisconsin-Madison (e.g., Davidson & Ellis Weismer, 2014; Ellis Weismer & Kover, 2015; Ray-Subramanian & Ellis Weismer, 2012; Venker et al., 2014). One hundred twenty-nine autistic preschool children and their parents participated in the larger project, recruited from the local community, clinics, and early intervention programs. Written consent was provided by all parents prior to study enrollment. Through parent report, participants were screened before participating in the study based on the following exclusionary criteria: speaking language(s) other than English in the home, history of seizure disorder, prematurity, multiple birth, cerebral palsy, or chromosomal abnormalities (see Table 1 for participant characteristics at Visit 1). Upon entering the study, participants either had an existing ASD diagnosis or had been flagged for concern due to behaviors consistent with ASD. As part of the study procedure, all participants had ASD diagnoses confirmed by our study team before data were collected. All participants included in the study met DSM IV-TR (American Psychiatric Association, 2000) diagnostic criteria for ASD at each visit (the current diagnostic standard at the initiation of the longitudinal project). An experienced clinical psychologist determined ASD diagnoses based on administration of the Autism Diagnostic Observation Schedule or Autism Diagnostic Observation Schedule Toddler Module (ADOS or ADOS-T; Lord et al., 2002; Luyster et al., 2009), the Autism Diagnostic Interview – Revised (ADI-R) Toddler Module (Kim et al., 2013), and best estimate clinical diagnosis. Participants’ race/ethnicity composition was: 86% White, 2% Black, 3% Hispanic, 2% Native American, and 8% Other.

Table 1.

Participant characteristics at Visit 1.

Measure	M	SD	Range
Chronological Age (in Months)	31.17	4.08	23–39
ADOS Calibrated Severity Score	7.48	1.81	3–10
Mullen Nonverbal Ratio IQ	78.29	13.93	44–115
PLS-4 Total Language Raw Score	135.64	21.57	100–204
PLS-4 Total Language Standard Score	64.56	11.77	50–102

Procedure

For the larger longitudinal project, children participated in 3–4 laboratory visits approximately 12 months apart. At each visit, participants were administered a battery of assessments measuring language, cognition, and autism diagnosis. All assessment measures were the most current version at the time of data collection. Natural language samples were also collected during dyadic parent-child play sessions at Visits 1, 2, and 4. Mean chronological ages at each visit for the participants included in analyses for the current study were (in months): 31.2 (SD=4.1, Range=23–39; Visit 1), 44.3 (SD=4.0, Range=37–53; Visit 2), and 66.5 (SD=5.0, Range=57–79; Visit 4). Because the standardized assessments utilized in this study were not collected for the entire sample at Visit 3, only Visit 1, 2, and 4 data were included in the present analyses. Due to participants not completing language samples, coding errors, or attrition between visits, spatial language data were ultimately available for 93 participants at Visit 1, 111 participants at Visit 2, and 101 participants at Visit 4. The relatively smaller sample size at Visit 1 reflects the fact that a subset of participants was not producing enough spoken language at that point in development to obtain a valid spontaneous language sample.

Standardized Measures

To assess structural language ability for the larger project, participants were administered the Preschool Language Scales – Fourth Edition (PLS-4; Zimmerman et al., 2002) by a certified speech-language pathologist. The PLS-4 is a standardized language assessment for preschool-age children with two subscales: Auditory Comprehension (AC) measuring receptive language, and Expressive Communication (EC) measuring expressive language. The subscales are added together to acquire a composite measure of overall language ability, the Total Language (TL) score. While not used in analyses for the present study, PLS-4 TL raw and standard scores at Visit 1 are included for descriptive purposes among sample characteristics in Table 1.

The Autism Diagnostic Observation Schedule or Autism Diagnostic Observation Schedule Toddler Module (ADOS or ADOS-T; Gotham et al., 2006; Luyster et al., 2009) was administered as a part of the comprehensive autism diagnostic assessment provided by a psychologist on our research team. In addition to assisting with diagnosis, this assessment also provided our measure of ASD symptom severity, the Calibrated Severity Score (CSS). The CSS is a standardized measure of ASD symptom severity on a scale of 1–10, where lower severity scores represent lower levels of impairment (Gotham et al., 2009).

Our measure of nonverbal spatial cognition was derived from the Mullen Scales of Early Learning (MSEL; Mullen, 1995), which was administered by the trained psychologist on our research team as part of the larger longitudinal project. The MSEL is a standardized measure for children ages birth-68 months assessing development across five domains: fine motor, gross motor, visual reception, expressive language, and receptive language and is considered suitable for assessing autistic children (Akshoomoff, 2006). Each of the five scales yields a T-score, percentile rank, and age equivalent, and together as a Composite measure yields a standard score and percentile rank. T-scores from the Visual Reception scale (mean of 50, standard deviation of 10) were utilized as our proxy measure of nonverbal spatial cognition. This scale has been used previously as a measure of spatial cognition in autistic preschoolers (Hellendoorn et al., 2015). Items in this scale probe skills across the visuospatial domain, tapping into underlying conceptual knowledge of spatial terms and relations and demonstration of spatial awareness.

Natural Language Samples

At each lab visit, a fifteen-minute parent-child play-based language sample was collected. Parents were instructed to “play with your child as you would at home.” At Visits 1 and 2, Parent-child dyads were provided with a Fisher-Price Little People Animal Sounds Farm (with batteries removed) and several Playskool Mr. Potato Head sets. At Visit 4, dyads were offered a Playmobil camping and playground set. Language samples were recorded using a Sony DCR-SR80 Handycam digital video camera and uploaded to a computer using Nero 8 ShowTime software. Transcription of language samples was completed by trained undergraduate and graduate students using Systematic Analysis of Language Transcript (SALT) software (Miller & Iglesias, 2012). Transcribers were required to attain 90% or greater agreement on morpheme and segmentation criterion with an experienced transcriber on three consecutive transcripts prior to beginning independent transcription, and reliability was conducted on 10% of samples. 45 transcripts were ultimately reviewed, achieving morpheme agreement of 93% and segmentation agreement of 96%.

For the purposes of this study, we define spatial language at the lexical level, i.e., total spatial words produced. To acquire our measures of parent and child spatial language, we coded the transcripts following similar procedures to previous studies in the NT literature examining spatial language in parent-child interactions (Cartmill et al., 2010; Pruden et al., 2011; Pruden & Levine, 2017). Like these studies, we employed the System for Analyzing Children’s Language About Space (Cannon et al., 2007) to identify spatial words produced by parents and children in the transcripts. Following Cannon and colleagues’ (2007) guide, categories included in our parent and child spatial language variables were:

Spatial Dimensions – describing size of objects, people, spaces (i.e., big, little, short)

Shapes – describing enclosed two- and three-dimensional objects and spaces (i.e., circle, triangle, square)

Locations and Directions - describing relative positions in space (i.e., in, from, under)

Orientation and Transformation – describing relative orientation or transformation in space (i.e., upside down, flip, rotate)

Continuous Amount – describing relative amount of continuous quantities (i.e., part, some, enough)

Deictics – words that rely on context to understand their referent (i.e., here, there, where)

Spatial Features and Properties – describing features and properties of 2D and 3D objects, people, spaces, and properties of their features (i.e., flat, round, straight)

Pattern – indicating talking about a spatial pattern (i.e., design, sequence, order)

Analytical Strategy

To calculate our measures of child and parent spatial language, cumulative totals of words across all categories were summed for each transcript for each child and parent. Total spatial words produced from each category by children and parents at each visit are described in Table 2. “Other” (non-spatial) words produced by parents and children were also calculated by subtracting total spatial words from the number of total words produced throughout the sample and were included in analyses to account for the effects of overall parent and child language production. In consideration of the wide natural variability in language in ASD, we aimed to retain as much data as possible and therefore set our threshold for removal of outliers at the 99.99 percentile. Multiple outlier diagnostics consistently identified one data point for child total spatial words (116 words) and one data point for parent total spatial words (293 words) as outliers, both of which were above the 99.99 percentile for the sample. To avoid undue influence from these outliers on analysis results, these points were removed prior to analysis.

Table 2.

Number of words produced by children in each spatial category from Cannon et al. (2007) and total spatial words for parents and children at Visit 1 (M = 31. months), Visit 2 (44 months), and Visit 4 (66 months).

Visit	Spatial Dimensions		Shapes		Location & Direction		Orientation & Transformation
	M(SD)	Range	M(SD)	Range	M(SD)	Range	M(SD)	Range
1	0.01 (0.10)	0 – 1	0.28 (2.40)	0 – 23	1.77 (5.41)	0 – 41	0.03 (0.18)	0 –1
2	0.29 (0.90)	0 – 6	0.12 (0.82)	0 – 7	6.52 (8.35)	0 – 35	0.14 (0.57)	0 – 4
4	1.44 (4.73)	0 – 43	0.08 (0.39)	0 – 3	14.32 (12.94)	0 – 51	0.28 (0.78)	0 – 5
	Continuous Amount		Deictics		Spatial Features & Properties		Patterns
	M(SD)	Range	M(SD)	Range	M(SD)	Range	M(SD)	Range
1	0.18 (0.92)	0 – 8	1.14 (3.08)	0 – 17	0.00 (0.00)	0 – 0	0.00 (0.00)	0 – 0
2	0.73 (1.84)	0 – 11	4.64 (8.29)	0 – 57	0.08 (0.45)	0 – 4	0.12 (0.44)	0 – 3
4	3.16 (4.90)	0 – 29	6.86 (7.44)	0 – 34	0.12 (0.74)	0 – 7	0.28 (0.71)	0 – 5
	Child Total Number of Spatial Words		Parent Total Number of Spatial Words
	M(SD)	Range	M(SD)	Range
1	3.69 (9.27)	0 – 67	81.31 (34.73)	19 – 195
2	12.37 (16.28)	0 – 84	78.11 (36.73)	15 – 234
4	26.32 (24.19)	0 – 116	88.99 (41.32)	14 – 293

To investigate the longitudinal relationship between spatial language production and nonverbal spatial cognition and assess the impact of ASD symptom severity and parent input on children’s spatial language production over time, linear mixed effects models were constructed in RStudio, version 4.0.3 (RStudio Team, 2020) using the lme4 package (Bates et al., 2015). To handle missing data, lme4 performed a complete records analysis. Because the data missing from this study are missing completely at random (MCAR) – that is, the missingness of the data does not depend on the specific research questions answered by the analysis – a complete records analysis is considered an appropriate approach, providing a valid and unbiased estimate of model parameters (Carpenter & Smuk, 2021). The random effects specified in each model were the maximal random effects structure that successfully converged (Barr et al., 2013) and avoided a singular fit. In each model, visit was included as an interacting fixed effect with the predictor of interest and as a random effect to account for non-independence in the data resulting from multiple observations from each participant (3 total, one at each visit), and allow us to investigate change in the relationship between variables over time.

To determine the relationship between children’s spatial language production and nonverbal spatial cognition, we fit a model regressing Mullen Visual Reception T-Scores on the two-way interaction between visit (contrast coded: −.5, 0, .5) and total child spatial words (mean-centered), including all lower order terms and “other” child words (the covariate) as fixed effects with a by-subject random slope for visit as a random effect.

To examine the effect of ASD symptom severity over the preschool years, a model was fit regressing children’s total number of spatial words on the two-way interaction between visit (contrast coded: −.5, 0, .5) and ASD symptom severity (i.e., ADOS CSS; mean centered), including all lower order terms and “other” child words (the covariate) as fixed effects and a by-subject random slope for visit as a random effect.

Finally, to examine the role of parental spatial language input, a model was fit regressing children’s total number of spatial words on the two-way interaction between visit (contrast coded: −.5, 0, .5), and parent’s total number of spatial words (mean centered), including all lower order terms. To account for parent and child overall language, parents’ and children’s total number of “other” (non-spatial) words were included as covariates. A by-subject random slope for visit was also included.

Results

Our first research question concerned the longitudinal relationship between child spatial language production and nonverbal spatial cognition (as measured by MSEL Visual Reception T-scores) during the preschool years. Model results yielded a significant main effect of total child spatial words produced (B = 0.19, SE = 0.08, t = 2.45, p = .015) when accounting for overall language and all other model terms (Figure 1). Change over time in the relationship between spatial language and spatial cognition was also observed, as the interaction between total child words produced and visit was significant, (B = −0.36, SE = 0.11, t = −3.18, p = .002) The covariate, “other” words produced (B = 0.02, SE = 0.01, t = 2.08, p = .039) was also significant, indicating that overall language contributed unique variance to nonverbal spatial cognition as well. The intercept (B = 41.21, SE = 1.12, t = 36.81, p < .001) indicated that the predicted MSEL Visual Reception T-score for children in this sample at the average value of all other predictors in the model was 41.21. The main effect of visit (p = .248) was non-significant.

Fig. 1.

Model predictions for the relationship between nonverbal spatial cognition (as measured by MSEL Visual Reception T-scores) and total child spatial words at visit 1 (30 months), visit 2 (44 months) and visit 4 (66 months), adjusted for overall child language

Our second research question sought to investigate whether ASD symptom severity was related to autistic children’s spatial language production during the preschool years. The effect of ASD symptom severity was not significant (p = .33) when accounting for overall child language (as measured by “other” words produced) and other model terms. Visit (p = .63) and the interactive effect of ASD symptom severity with visit were likewise non-significant (p = .13). Only the covariate, “other” words produced (B = 0.10, SE = 0.00, t = 23.58, p < .001) significantly predicted child spatial words in this model, suggesting that spatial language was more strongly associated with overall child language than ASD symptom severity. The intercept (B = 13.88, SE = 0.53, t = 26.31, p < .001) indicated that the predicted number of spatial words produced by children at the average value of all other predictors in the model was 13.88. To investigate the discrepancy between our findings from this analysis and that of (Bochynska, Coventry, et al., 2020), we conducted a post-hoc analysis removing “other” child words (our measure of overall language production) from the mixed-effects model regressing spatial words on the interaction of visit (contrast coded: −.5, 0, .5) and ASD symptom severity (mean-centered) as fixed effects with a by-subject random intercept. As suspected, the main effect of ASD symptom severity did reach significance in this alternative model when overall child language was not accounted for (B = −2.01, SE = 0.54, t = −3.71, p < .001).

The final research question examined the effect of parent spatial input on child spatial word production over the preschool years. Model results indicated a significant main effect of total parent spatial words (B = 0.09, SE = 0.02, t = 3.68, p < .001) when accounting for “other” parent words, “other” child words, and all other model terms (Figure 2). The effect of “other” child words (B = 0.10, SE = 0.00, t = 26.33, p < .001) was also significant, indicating that children’s overall language also contributed to spatial language production when accounting for parent input and all other model terms. The interaction between visit and parent input approached but did not reach significance, (B = 0.07, SE = 0.04, t = 1.91, p = .057) suggesting that the strength of the relationship between parent and child spatial word production did not change significantly between visits but trended in a positive direction. The intercept (B = 13.96, SE = 0.51, t = 27.45, p < .001) indicated that the predicted number of spatial words produced by children at the average value of all other model terms was 13.96. Visit (p = .73) and “other” parent words (p = .07) were not significant. To explore the possibility that child factors might moderate the relationship between parent and child spatial language production, we conducted a post-hoc analysis regressing total child spatial words on the three-way interaction between visit (contrast coded: −0.5, 0, 0.5), nonverbal spatial cognition (mean-centered), and parent input (mean-centered), including parent and child “other” words as covariates and a by-subject random slope for visit. Model results revealed a significant two-way interaction between nonverbal spatial cognition and total parent spatial words even when accounting for “other” parent and child words and all other model terms (B = 0.05, SE = 0.02, t = 2.63, p = .009).

Fig. 2.

Model predictions for the relationship between total parent spatial words and total child spatial words at visit 1 (30 months), visit 2 (44 months) and visit 4 (66 months), adjusted for overall parent and child language

Discussion

The purpose of this study was to investigate the early development of spatial language and cognition in autistic children and specify child and environmental factors contributing to that development. Using language samples collected during naturalistic parent-child interactions over multiple time points during the preschool years, we examined the longitudinal relationship between children’s production of spatial words and nonverbal spatial cognition. We then analyzed the effects of ASD symptom severity and parent input on children’s spatial language production as these factors were previously implicated in studies with older autistic individuals or NT children. Results revealed a significant, positive relationship between spatial word production and nonverbal spatial cognition which decreased in magnitude over time. Contrary to a previous finding (Bochynska, Coventry, et al., 2020), ASD severity was not a predictor of spatial language production when accounting for overall child language. Parent spatial input, on the other hand, was significantly, positively related to child spatial word production even when accounting for overall parent and child language.

Our first aim was to examine the association between children’s spatial language production and nonverbal spatial cognition (as measured by MSEL Visual Reception T-scores; Mullen, 1995) and determine whether that relationship changed over the course of early development. Given ample evidence suggesting yoking between spatial language and spatial cognition in NT preschool children and older autistic individuals (e.g., Bochynska, Coventry, et al., 2020; Bochynska, Vulchanova, et al., 2020; Hund et al., 2021; Polinsky et al., 2017; Simms & Gentner, 2019; Verdine et al., 2014), we predicted a positive relation between spatial word production and spatial cognition even when accounting for overall language. Study findings supported this hypothesis; the main effect of spatial cognition positively predicted children’s spatial word production over and above the effect of “other” words produced (Figure 1). These results provide evidence in support of an association between spatial language and nonverbal spatial cognition despite the uneven profile generally reported between visuospatial skills and language ability in ASD. Moreover, given that both nonverbal spatial cognition and overall language contributed unique variance to spatial language production, these findings may lend support to an account of spatial language as a reflection of the intersection between nonverbal cognition and overall linguistic abilities. It should be noted, however, that these results cannot prove directionality in this relationship. Much of the NT literature in this area is based in a theoretical framework wherein language structures cognition (i.e., Gentner et al., 2013; Loewenstein & Gentner, 2005; Majid et al., 2004; Pyers et al., 2010), but many autistic children experience language delays during the preschool years (Eigsti et al., 2011). Therefore, it is also conceivable that during this stage of development in ASD, earlier growth in spatial cognition (i.e., Casasola, 2018) could shape spatial language, or that bidirectional effects between these domains could interact over time. A previous study examining the relationships among fine motor skills, visuospatial cognition, and expressive and receptive language in young autistic children concluded that visuospatial cognition mediated the relationship between fine motor development and later expressive and receptive language, suggesting cognition-language directionality for autistic children at this age (Hellendoorn et al., 2015). Further research will be necessary to establish the nature of the relationship between the linguistic and cognitive spatial domains in autistic preschoolers.

While the main effect of nonverbal spatial cognition was positively associated with spatial language as hypothesized, an unexpected result of our analysis was a significant but negative two-way interaction between nonverbal spatial cognition and visit such that the magnitude of the relationship between nonverbal spatial cognition and total spatial words produced decreased over time as children reached the age of school entry. Given that previous work suggested an association between linguistic and non-linguistic spatial abilities in ASD continuing into adulthood and improvements in spatial language abilities with age (Bochynska, Coventry, et al., 2020; Bochynska, Vulchanova, et al., 2020), we expected that the relationship between spatial language and cognition would strengthen as many children who were initially preverbal or minimally verbal began to acquire more expressive language (Ellis Weismer & Kover, 2015). One possible explanation for this surprising finding is that spatial language development may have grown at a faster rate than spatial cognition during this period. The average number of total spatial words children produced rose from 3.69 at Visit 1 to 26.32 by Visit 4 - a nearly seven-fold increase (Table 2). MSEL Visual Reception T-scores, on the other hand, increased from an average of 36.84 at Visit 1 to 44.35 at Visit 4. This discrepancy may be due to our use of a raw score (total words produced) as a measure of spatial language but standardized scores as a measure of spatial cognition. As age-level standards for performance on the MSEL increased at each visit, growth in T-scores may have begun to plateau if children’s spatial cognitive ability grew at a slower rate than expected for same-age NT peers. Future work will be needed to determine whether this result would be replicated using a raw measure of spatial cognitive ability.

The second aim of this study was to investigate predictors of spatial language production in autistic preschoolers. We extended previous work in older autistic children, adolescents, and adults (Bochynska, Coventry, et al., 2020; Bochynska, Vulchanova, et al., 2020) by examining whether ASD symptom severity predicted spatial language production in a younger cohort with a more representative range of language and cognitive ability. We predicted that our pattern of results would align with those prior findings such that more severe ASD would be associated with lower spatial language production. However, study results did not support this hypothesis. When accounting for “other” child words, ASD symptom severity was not associated with spatial language production. One possibility for this discrepancy could be lower task demands associated with our spatial language data collection method (naturalistic play) compared to the experimental tasks in Bochynska, Coventry et al. (2020) which could have attenuated the effect of ASD symptom severity in our study. This incongruity in findings may also be attributed to the decision to account for overall child language in our study, whereas Bochynska, Coventry et al. (2020) did not. Their analysis included a continuous measure of ASD symptom severity across both NT and ASD groups. These groups differed significantly on expressive language, yet it was not included as a covariate in their analysis. Therefore, expressive language may be a hidden confound in the reported relationship between ASD severity and spatial language performance. To investigate this possibility further, we conducted a post-hoc analysis removing overall language production as a covariate. Indeed, the results of this model aligned with previous findings and indicated that ASD symptom severity was a predictor of spatial language. The notion that spatial language is more attributable to overall language than ASD severity also aligns more closely with an account of spatial language as a reflection of the interface of language and cognition and not an autism-specific area of difficulty. For example, spatial language deficits are also prominently observed in individuals with Williams syndrome, who as a group tend to demonstrate relative strengths in expressive language but difficulties in visuospatial cognition (Landau & Hoffman, 2005). Together, these findings suggest yoking between the linguistic and cognitive spatial domains such that perturbations in either may influence performance in the other regardless of diagnostic status.

Moreover, the overlap in variance contributed by overall language production and ASD symptom severity evident in this study may raise the broader question of which specific cognitive or behavioral constructs underlie measures of autism severity. Structural language impairment is not a core characteristic of ASD per DSM-5 criteria (American Psychiatric Association, 2013), but ASD severity as measured by ADOS CSS does take structural language level into account (Janvier et al., 2022; Gotham et al., 2009; Shumway et al., 2012). While beyond the scope of the present study, further investigation of the role of language in defining autism severity will be informative for future theoretical frameworks and diagnostic assessment of ASD.

Given positive findings regarding its significance in NT preschool children’s spatial language production in past studies, the second predictor we investigated in early spatial language development in ASD was parent input. We anticipated that if autistic children showed a similar pattern of results to previous research in NT preschoolers, parent input would have a positive effect on autistic preschoolers’ spatial language production. Indeed, parent spatial word production was positively associated with children’s spatial word production, even when accounting for “other” parent and child words (Figure 2). This finding aligns with evidence from NT children (Cartmill et al., 2010; Pruden et al., 2011; Pruden & Levine, 2017) and suggests that autistic children as a group demonstrate the ability to pick up on spatial language used by their parents and incorporate it into their own production during dyadic interactions. This finding provides promising evidence that parent input could support autistic children’s spatial language development. However, future work will be needed to better specify this relationship, including potential mediators and moderators. If spatial language use between parents and autistic children follows a transactional model (e.g., Sameroff & Fiese, 2000), then parents’ spatial language input could also depend on child characteristics influencing children’s input to parents. Possible moderators of the parent-child spatial language relationship could include parent awareness of children’s existing spatial language knowledge, or children’s attention to spatial features and concepts in the input they receive. Alternatively, strengths in nonverbal spatial cognition could facilitate children’s ability to associate spatial language from parent input with more robust cognitive representations. To explore this question further, we conducted a post-hoc analysis examining whether spatial cognition moderated the relationship between parent and child spatial language (accounting for parent and child “other” words). Results indicated that spatial cognition was a significant moderator such that the spatial language use of children with higher spatial cognition was more strongly, positively associated with parent spatial input than that of children with lower spatial cognition. While preliminary, this finding suggests a need to consider reciprocal interactions among child and parent variables in the development of spatial language in ASD, including more domain-general mechanisms such as selective attention (Miller & Simmering, 2018) which are beyond the scope of the current study but could underlie several findings related to spatial language.

Limitations and Future Directions

While adding several novel contributions to the existing literature, the present findings are also limited by relatively low power due to small sample size, the lack of data at the third time point, and relatively coarse measures of key variables from standardized assessments. A contribution of the present study was the use of a measure of spatial cognition general enough to ensure that its relationship to spatial language was not proximal or task specific. However, it was not precise enough to allow a more detailed analysis of parallels between linguistic and cognitive representations. In addition to investigating possible explanatory mechanisms and determining the directionality of the reported relationships, future research should take a more fine-grained approach to identify specific areas of strength and difficulty in the spatial language of young autistic children and determine the extent to which they reflect the structure of emerging nonverbal spatial representations. Considering the discrepancy between receptive and expressive language abilities in autistic preschoolers (Ellis Weismer & Kover, 2015) and evidence suggesting that autistic children have difficulty with abstraction and concept formation (Kuschner et al., 2007), children’s understanding of spatial words may not be as robust relative to their use of those words. Measuring receptive spatial language was beyond the scope of the present study, so further work is also needed to determine whether children’s understanding of spatial words is on par with their production. Relatedly, a final limitation of this study was in narrowly defining spatial language at the lexical level, whereas language comprises far more complex integration of grammatical, semantic, and pragmatic contextual grounding. Therefore, a future direction for this line of research will be to examine child use of spatial words at the discourse level. Children’s use of spatial words in context could provide insight into the nature of their spatial representations beyond what can be gleaned from word counts alone. Finally, there remains a need to compare autistic children with NT peers to determine whether autistic children experience developmental delays in spatial language and to examine whether effects of nonverbal spatial cognition and parent input differ depending on diagnostic status.

Clinical Implications

Investigating spatial language in autistic children and their parents will help inform future caregiver-mediated interventions, particularly in identifying potential intervention targets. While not often directly targeted in language interventions, increasing spatial language may not only grow children’s expressive and receptive vocabulary, but also emphasize relational concepts supporting more complex language and ideas. Targeting spatial language in intervention could also translate to improved academic achievement by bringing children’s attention to spatial features and concepts foundational to learning in science, technology, engineering, and mathematics (STEM) (Ferrara et al., 2011; Pritulsky et al., 2020; Uttal et al., 2013). It will be critical to examine whether parent training can increase spatial language abilities in autistic children and determine if those language skills can in turn support aspects of spatial cognition which might be impaired as well as build upon existing strengths. Other modifications to the learning environment which have been found to support spatial language in NT children, including playing with blocks (Borriello & Liben, 2018; Ferrara et al., 2011) and adding gesture (Cartmill et al., 2010) should also be considered in future intervention studies with autistic children.

Conclusions

Like NT children and children with other developmental disabilities (i.e., Williams syndrome), results of the present study suggest that spatial language and spatial cognition are associated in ASD. Moreover, overall child language and parent input also appear to contribute to spatial language, but ASD symptom severity does not. Taken together, these findings lend support to an account of spatial language as a reflection of the intersection of cognition and language regardless of diagnostic status. Further research is needed to determine the directionality of the spatial language-spatial cognition interface as well as to elucidate underlying mechanisms supporting spatial language acquisition. Autistic preschoolers overall appeared to benefit from parent input during play interactions, highlighting the potential for future parent-mediated interventions targeting spatial language. However, future work will need to better specify this relationship, including identification of potential moderators and mediators. The insights gained by examining this unique domain of lexical knowledge will not only inform theoretical accounts of the cognitive-linguistic profile in ASD, but also guide future clinical interventions - especially those targeting complex language and functional, academically relevant vocabulary and concepts.