1. Introduction
1.1. Multilingualism in ASD: Previous evidence for non-interactive foreign language acquisition
Autism Spectrum Disorder (ASD)Footnote 1 is a neurobiological disorder of early brain development that includes impairments in interaction and communication. Additionally, children with ASD exhibit patterns of activities, interests, and behaviours that are restricted and repetitive (American Psychiatric Association, 2013; World Health Organization (2019)). Like their non-autistic peers with typical language development (TLD), children with ASD can be raised in multilingual environments, exposed to two or more languages at home or in educational settings. Research on the effects of multilingualism in ASD is still scarce (for an overview, see Hantman et al., Reference Hantman, Choi, Hartwick, Nadler and Luk2023; Prévost & Tuller, Reference Prévost and Tuller2022; Yu, Reference Yu2018). Studies on foreign language acquisition through non-interactive exposure among autistic individuals have been notably limited, primarily focusing on individual cases (e.g., Kissine et al., Reference Kissine, Luffin, Aiad, Bourourou, Deliens and Gaddour2019; Riedel et al., Reference Riedel2020; Smith & Tsimpli, Reference Smith and Tsimpli1995; Vulchanova et al., Reference Vulchanova, Talcott, Vulchanova and Stankova2012; Zhukova et al., Reference Zhukova, Talantseva, An and Grigorenko2021). The current study aims to contribute to our understanding of the effects of naturalistic and non-interactive language acquisition in autism.
More than half of individuals with autism showcase distinct abilities, often termed ‘splinter skills,’ ‘islets of ability’, ‘savant skills’, or ‘talents’, which signify relative strengths compared to their typically developing counterparts (Itzchak et al., Reference Itzchak, Aviva and Zachor2013; Mottron et al., Reference Mottron, Ostrolenk and Gagnon2021; Zhukova et al., Reference Zhukova, Talantseva, An and Grigorenko2021). The Autism Diagnostic Interview-Revised (ADI-R) lists a range of ‘splinter skills’ commonly identified in individuals with ASD, including memory, visuospatial aptitude, reading, drawing, music, and mathematical calculation (Rutter et al., Reference Rutter, LeCouteur and Lord2003). Interestingly, the list does not explicitly include spontaneous foreign language acquisition via non-interactive exposure, despite several documented cases in the existing literature. However, it is reasonable to hypothesise that the global impact of COVID-19 and extensive lockdowns may have contributed to an increased prevalence of the phenomenon of non-interactive language learning worldwide, in that, the widespread availability of the Internet has empowered children to access content in various languages and choose a language as an area of special interest, potentially amplifying instances of spontaneous foreign language acquisition in children with ASD (Zhukova et al., Reference Zhukova, Talantseva, An and Grigorenko2021).
In contrast to multilingual children with and without ASD who acquire their languages through naturalistic interactions at home and school, this group of children achieves foreign language learning independently and spontaneously through non-interactive input. Interestingly, previous studies suggested that infants exposed to foreign languages through recordings, without interpersonal interactions, did not maintain the phonetic perception of the foreign language (Kuhl et al., Reference Kuhl, Tsao and Liu2003). Additionally, Roseberry et al. (Reference Roseberry, Hirsh-Pasek and Golinkoff2014) conducted a study in which children were taught new verbs in various conditions (live interactions, video chat, and non-contingent video training). Results showed that video training was not an effective tool for learning new words, suggesting that neurotypical children need social interaction to learn new words.
The topic of non-interactive language acquisition raises questions regarding the role of interaction in language acquisition (Kissine, Reference Kissine2021). Over the years, numerous reported cases have highlighted instances of spontaneous foreign language acquisition via non-interactive exposure among individuals with ASD. One of the most famous cases is that of Christopher, who, despite cognitive impairments, partially mastered at least 16 different foreign languages (Smith & Tsimpli, Reference Smith and Tsimpli1995). Similarly, EV, a ten-year-old girl from Bulgaria, independently acquired German by watching TV shows, demonstrating superior proficiency in both her native language (L1) and the acquired language (L2) across various aspects, including oral skills, vocabulary, and grammar (Vulchanova et al., Reference Vulchanova, Talcott, Vulchanova and Stankova2012). Kissine et al. (Reference Kissine, Luffin, Aiad, Bourourou, Deliens and Gaddour2019) described an unusual phenomenon in which five Tunisian boys with ASD used Modern Standard Arabic, a language generally limited to very formal, mostly written settings but also used on TV across the Arabic-speaking world. Similarly, more recently, Abd El-Raziq et al. (Reference Abd El-Raziq, Meir and Saiegh-Haddad2024b) reported the same phenomenon of Modern Standard Arabic use in several Palestinian-Arabic-speaking autistic children aged 4–11. The overuse of Modern Standard Arabic was observed in autistic children with intact structural language skills (4 out of 19) and in one child with a comorbid Developmental Language Disorder (1 out of 19). The authors proposed that the increased use of Modern Standard Arabic by autistic children might be accounted for by the fact that autistic individuals can acquire languages solely through non-interactive TV exposure. Modern Standard Arabic is used in non-interactive contexts such as storybook reading, religious texts like the Quran, and media (e.g., cartoons on TV and YouTube).
Furthermore, LP, a young man with Asperger Syndrome, independently learned multiple languages, some within a remarkably short period of three days, attributed to his unique synaesthetic abilities (Riedel et al., Reference Riedel2020). Another noteworthy case is Alex, an 11-year-old Russian boy born in Russia (Zhukova et al., Reference Zhukova, Talantseva, An and Grigorenko2021). Despite his linguistic environment, he learned English by watching cartoons and produced his first words in English. Unlike other cases, Alex’s English skills developed while his Russian skills lagged behind. This prompted his family to move to England to support his chosen language. Dumont et al. (Reference Dumont, Biston, Wintgens, Clin and Kissine2022) documented the profiles of three children with ASD who learned English via YouTube. The children were tested in both French and English and were found to have either balanced or differing abilities in the two languages. In terms of phonology, researchers found that vowel acquisition in English was achieved in all three children, while consonant acquisition was achieved only in one of them. Recently, a group of 12 French-speaking children with ASD were documented to show unexpected acquisition of English via screens (Dumont et al., Reference Dumont, Belenger, Eigsti and Kissine2024): this group of children showed enhanced pitch discriminations. In a similar vein, Kadiri and Anasse (Reference Kadiri and Anasse2023) documented the case of a six-year-old autistic child from Morocco who, once again, spontaneously acquired proficiency in English without any formal instruction. Through exposure to YouTube videos, the child demonstrated functional and communicative use of English. Notably, consistent with Alex’s case, the six-year-old’s use of his native Moroccan dialect exhibited a delay. Researchers intriguingly propose that unexpectedly acquired English might serve as the primary language (L1). More recently, Balčiūnienė and Kornev (Reference Balčiūnienė and Kornev2023a, 2023Reference Balčiūnienė and Kornevb) described a case study of a Lithuanian-speaking boy who independently began to speak English at the age of 4. The study included a battery of tests assessing grammar, receptive and expressive vocabulary, and discourse skills (storytelling and dialogue). The researchers found delayed and impaired grammatical abilities in the native (Lithuanian) language but not in English.
To conclude, it is possible to hypothesise that these cases demonstrate instances where language development in ASD unfolds along an alternative acquisition path that does not closely relate to the communicative function of language (for a detailed discussion, see Kissine et al., Reference Kissine, Saint-Denis and Morton2023; Kuhl et al., Reference Kuhl, Tsao and Liu2003; Roseberry et al., Reference Roseberry, Hirsh-Pasek and Golinkoff2014).
1.2. Morpho-syntax in monolingual and multilingual children with ASD
According to the DSM-5 (American Psychiatric Association, 2013), language is not included in the diagnosis of ASD. It is estimated that 15–20% of children with ASD remain non-verbal, while approximately 60–80% develop expressive and receptive structural language skills. Among children with ASD who develop fluent speech, there is a broad spectrum of language skills. Some demonstrate exceptional abilities (as outlined in the previous subsection), while others exhibit heterogeneous profiles (for an overview, see Schaeffer et al., Reference Schaeffer, Abd El-Raziq, Castroviejo, Durrleman, Ferré, Grama and Tuller2023). Some children with ASD show morpho-syntactic skills similar to their typically developing peers (ASD+NL), while others present with morpho-syntactic impairments (ASD+ LI) (e.g., Abd El-Raziq et al., Reference Abd El-Raziq, Meir and Saiegh-Haddad2024a; Durrleman & Delage, Reference Durrleman and Delage2016; Kjelgaard & Tager-Flusberg, Reference Kjelgaard and Tager-Flusberg2001; Meir & Novogrodsky, Reference Meir and Novogrodsky2020; Tager-Flusberg, Reference Tager-Flusberg2016).
Children with ASD+LI are reported to have difficulties comprehending and producing various syntactic structures, such as wh-questions, relative clauses, passives, and constructions involving argument movement (for an overview, see Schaeffer et al., Reference Schaeffer, Abd El-Raziq, Castroviejo, Durrleman, Ferré, Grama and Tuller2023, and studies cited therein). A study by Meir and Novogrodsky (Reference Meir and Novogrodsky2020) on the morpho-syntactic abilities of multilingual Russian-Hebrew-speaking children with ASD aged 4;6–9;0 provided evidence for two language profiles (ASD+NL and ASD+LI) in multilingual children with ASD. The authors reported lower syntactic abilities among children with ASD compared to children with TLD. However, some children with ASD (monolingual and multilingual) exhibited typical morpho-syntactic abilities (ASD+NL). Another group of children with ASD (ASD+LI) tended to show comorbid Developmental Language Disorder, previously referred to as Specific Language Impairment. Similarly, Andreou et al. (Reference Andreou, Tsimpli, Durrleman and Peristeri2020), who tested bilingual (Albanian-Greek) and monolingual (Greek-speaking) children with ASD, found that bilinguals scored higher than monolinguals in complex sentences, particularly in adverbials and relatives. Furthermore, recent studies confirm that errors observed in children with ASD+LI have a morpho-syntactic nature rather than a pragmatic one (see Abd El-Raziq et al., Reference Abd El-Raziq, Meir and Saiegh-Haddad2024a; Sukenik, Reference Sukenik2023).
To sum, monolingual and multilingual children with ASD are showing heterogeneity of morpho-syntactic profiles. Furthermore, it is known that multilingual children with TLD also show varying profiles, as their languages skills are unevenly distributed across the two languages (Kohnert, Reference Kohnert2010). Thus, it is important to assess morpho-syntactic skills of multilingual children with ASD in their two languages. This study will be the first to test English-Hebrew-speaking multilinguals with ASD extending previous research on multilingual children with ASD (see research by Meir & Novogrodsky, Reference Meir and Novogrodsky2020 on Russian-Hebrew speaking children with ASD and on multilingual Greek by Andreou et al., Reference Andreou, Tsimpli, Durrleman and Peristeri2020.
1.3. Input/exposure in multilingual children with and without ASD
The linguistic development of young multilingual children is profoundly influenced by both the quantity and quality of input/exposure in each language (for an overview, see Armon-Lotem & Meir, Reference Armon-Lotem, Meir, De Houwer and Ortega2019; Paradis, Reference Paradis2011, Reference Paradis2023). Crude quantitative exposure measures may include the Age of Onset of multilingualism (AoO), naturalistic and/or non-interactive (e.g., via TV, YouTube, cartoons), Length of Exposure (LoE) to the SL, cumulative exposure, and relative exposure to each language, among many others (see Armon-Lotem & Meir, Reference Armon-Lotem, Meir, De Houwer and Ortega2019; Unsworth, Reference Unsworth, Grüter and Paradis2014). Qualitative aspects encompass the caretaker’s proficiency in providing native or non-native input, as well as high or low levels of code-mixed input, language use in different settings (e.g., home vs. school), and family language policy. Additionally, children receive input from multilingual and bidialectal speakers, which may diverge from monolingual norms.
Research on the effects of input/exposure in multilingual children with ASD remains limited and presents conflicting evidence. One line of research suggests that input/exposure shapes outcomes in children with ASD. Gonzalez-Barrero and Nadig (Reference Gonzalez-Barrero and Nadig2018) demonstrated that language exposure significantly predicted vocabulary and morphological skills in multilingual English-French children with both ASD and TLD, who varied in their exposure to French. Similarly, Cohen et al. (Reference Cohen, Guerra, Miguel, Bottema-Beutel and Oliveira2023) documented code-switching patterns (e.g., “Papi quién is this? [Daddy who is this?]”) in a Spanish-English child with ASD, aligned with code-switching patterns observed in family members, thereby suggesting that daily language interactions predict child outcomes.
In contrast, the evidence does not uniformly support a straightforward link between input/exposure and linguistic abilities in children with ASD. For instance, Armon-Lotem and Meir (Reference Armon-Lotem and Meir2022) examined the relationship between language exposure and linguistic performance using the LITMUS SRep tasks in Russian-Hebrew-speaking children with TLD, Developmental Language Disorder (DLD), and ASD. They found that increased exposure benefited children with TLD and DLD, but for bilinguals with ASD, increased exposure did not consistently translate into improved intake and subsequent output. Similarly, Beauchamp et al. (Reference Beauchamp, Rezzonico, Bennett, Duku, Georgiades, Kerns and Elsabbagh2023) compared monolingual and bilingual children with ASD and found that language exposure did not significantly impact their performance on macrostructure or microstructure measures. Likewise, Hambly and Fombonne (Reference Hambly and Fombonne2012), who investigated bilingual children with ASD and compared the timing of language exposure (infancy versus post-infancy period), suggested that sequentially exposed children did not differ from their simultaneously exposed peers in terms of dominant language skills.
1.4. The current study
The current study was devised to innovate by exploring the distinction between naturalistic and non-interactive foreign language acquisition in children with ASD (in our study, English) comparing them to their multilingual peers with TLD. Notably, this research is the first of its kind to investigate the phenomenon of spontaneous foreign non-interactive language acquisition (i.e., English) in autism, departing from the historical practice of relying solely on case studies, comparing this unique group of children to those with and without ASD acquiring both of their languages naturalistically (in our study English and Hebrew). Furthermore, the study examined language abilities among participating children in both English and Hebrew.
Before presenting the specific research questions of the study, it is important to describe the sociolinguistic situation in Israel, where the data were collected for children with and without ASD. Hebrew is the societal language in Israel, while English functions as a minority heritage language. However, English plays a unique role in Israel as a global language and lingua franca (see Gordon & Meir, Reference Gordon and Meir2024; Fridman & Meir, Reference Fridman and Meir2023; Karpava et al., Reference Karpava, Zabrodskaja, Ritter, Meir and Ringblom2024; Rose et al., Reference Rose, Armon-Lotem and Altman2023). English is ubiquitous in Israel – it is often heard (not dubbed) in movies, television, and on the Internet; it is prestigious and assumed to facilitate the advancement of its speakers, thus it is encouraged; and it is formally taught in schools from a young age.
The three research questions addressed in the study were as follows.
Research Question 1: Are there group differences between multilingual children with ASD who show unexpected spontaneous English acquisition via non-interactive sources (ASD-NI), those with ASD acquiring languages naturalistically (ASD-Nat), and typically developing multilinguals (TLD-Nat)?
Previous research indicates that multilingual children with ASD generally exhibit a broad range of syntactic abilities. However, as a group, we anticipate that children with ASD will demonstrate diminished morpho-syntactic abilities compared to children with TLD (see subsection 1.2). Hence, we expect that children in the ASD groups will exhibit lower morpho-syntactic abilities than children in the TLD group. At the individual level, we anticipate that some children with ASD will exhibit typical morpho-syntactic abilities on par with multilinguals without ASD, while others will face challenges in both of their languages.
Research Question 2: Do multilingual children in the three groups show individual differences within the two languages, i.e., English and Hebrew?
We expect to find differences between the two languages in morpho-syntactic abilities. It has been established that language skills in multilinguals are unevenly distributed across their two languages (Kohnert, Reference Kohnert2010). Additionally, we anticipate that some children in the ASD-NI group will perform better in English due to it being their favourite language, or language of choice.
Research Question 3: Are there effects of exposure variables on children’s morpho-syntactic abilities in English and Hebrew within the three groups?
Previous research conflicts regarding the influence of language exposure on language skills in multilingual children with ASD (Armon-Lotem & Meir, Reference Armon-Lotem and Meir2022; Beauchamp et al., Reference Beauchamp, Rezzonico, Bennett, Duku, Georgiades, Kerns and Elsabbagh2023; Gonzalez-Barrero & Nadig, Reference Gonzalez-Barrero and Nadig2018; Hambly & Fombonne, Reference Hambly and Fombonne2012). Therefore, the current study aims to provide further insights into the intricate interplay between exposure and language development in ASD. We expect to find a relationship between exposure and linguistic abilities in the TLD group, yet no clear relationship in the ASD group. Based on parental reports, the following indices were derived: Length of Exposure to English/Hebrew, Current Exposure to English/Hebrew, the child’s preferred language, and the language spoken by the child with the mother.
2. Method
All research materials, data, and the analysis codes are available from the OSF repository: https://osf.io/4abnd/?view_only=33f560e404174e3583c1e5a6b5bcf29c.
2.1. Participants
A total of 46 multilingual English-Hebrew-speaking children with and without ASD, aged 4;10 to 12;0, participated in the study. Autistic children were diagnosed before the study and attended kindergartens and schools in various educational settings (i.e., classes for autistic children, Special Education, or mainstream schools). Most children came from mid-to-high SES backgrounds, as measured by their mothers’ education levels.
Participants were divided into three groups: multilingual autistic children who acquired English via non-interactive exposure (ASD-NI), multilingual autistic children who acquired English naturally via interactive exposure (ASD-Nat), and non-autistic typically developing multilingual children (TLD-Nat) (see Table 1).
Table 1. Background information of the participants per group
In the ASD-NI group, where multilinguals acquired English non-interactively, all children were born to non-English speaking families: Hebrew was the home language for most children (n=8), while four children were raised in homes where a language other than Hebrew was spoken (Persian (n=1), Russian (n=2), and Spanish (n=1)). Only one child, who spoke Russian, maintained his home language and continued to use it with his mother. The remaining multilingual children reported that they stopped using their home language and switched to English and Hebrew.
In the ASD-Nat group, most children were born to families who moved to Israel from English-speaking countries; one child was exposed to English due to relocation to an English-speaking country.
In the TLD-Nat group, most children were born to immigrants who moved to Israel from English-speaking countries. However, two children were exposed to English due to their families relocating out of and then returning to Israel. Thus, in the ASD-Nat and TLD-Nat, children with and without ASD were exposed to naturalistic input in English and in Hebrew, while children in the ASD-NI group were exposed to English via non-interactive sources as reported by the parents (YouTube, TV, computer games).
All children in the ASD groups underwent the Autism Diagnostic Observation Schedule, Second Edition [ADOS-2] (Lord et al., Reference Lord, Rutter, Risi, Gotham and Bishop2012) or the Autism Diagnostic Interview-Revised [ADI-R] (Rutter et al., Reference Rutter, LeCouteur and Lord2003) to confirm their autism diagnosis and to assess autism severity. Non-verbal IQ was measured in all children (with and without autism) using the Raven Colored Progressive Matrices (Raven et al., Reference Raven, Court and Raven1998). There were no significant differences in non-verbal IQ scores among the three groups (Table 1). Similarly, there were no differences observed between the two autism groups in terms of autism severity scores (Table 1).
The three groups did not differ in terms of chronological age, mothers’ education, or number of siblings per family. However, the groups did differ in their length of exposure to Hebrew/English and parental ratings of the child’s proficiency in English and Hebrew (Table 1). As expected, children who acquired English naturalistically (ASD-Nat and TLD-Nat) had longer exposure to English compared to those who acquired it non-interactively (ASD-NI). This difference arose because children in the ASD-NI group began learning English later, possibly due to exposure through screens starting around 24–36 months of age. Parental ratings of proficiency in English and Hebrew were higher for the ASD-NI group compared to the ASD-Nat group. The groups with autism were predominantly male, consistent with the typical gender ratio among autistic children.
2.2. Tasks
2.2.1. Background tasks
Background information was collected using a shortened and adapted version of the Bilingual Parents’ Questionnaire (BIPAQ) and through interviews (Abutbul-Oz & Armon-Lotem, Reference Abutbul-Oz and Armon-Lotem2022). The BIPAQ gathered information on age, gender, length of exposure, and age of onset of Hebrew/English. Parents were also asked about other languages spoken at home. Additionally, the questionnaire provided information on whether children had any health or developmental concerns (e.g., Otitis Media, neurological or social disorders). Parents of children in the ASD-NI group were interviewed regarding their child’s English acquisition, including when they started speaking English, the number of hours spent using media, language preference, and possible reasons for English use. Furthermore, the children were interviewed about their English language acquisition, including reasons and methods of English acquisition, language preference, etc. The parental questionnaire included the following measures of language exposure: the Age of Onset to English/Hebrew (in years), the Length of Exposure to English/Hebrew (calculated based on the child’s chronological age and the Age of Onset, measured in years), Current Exposure to English/Hebrew in percentages (e.g., 0% English – 100% Hebrew; 25% English – 75% Hebrew; 50% English – 50% Hebrew, etc., converted to a 1–5 scale), the child’s preferred language, and the language spoken by the child to the mother (only Hebrew, only English, both English and Hebrew, or other).
The Raven’s Colored Progressive Matrices non-verbal IQ test (Raven et al., Reference Raven, Court and Raven1998) was administered to control for non-verbal IQ. Descriptive statistics and group differences for background measures are presented in Table 1.
2.2.2. Morpho-syntactic abilities
Previous research has established a robust link between sentence repetition (SRep) skills and overall syntactic proficiency. Researchers and clinicians widely employ SRep tasks across diverse typical and atypical populations to gauge morpho-syntactic skills (Marinis & Armon-Lotem, Reference Marinis, Armon-Lotem, Armon-Lotem, de Jong and Meir2015; Andreou et al., Reference Andreou, Tsimpli, Durrleman and Peristeri2020; Meir & Novogrodsky, Reference Meir and Novogrodsky2020; Meir et al., Reference Meir, Walters and Armon-Lotem2016; Sukenik & Friedmann, Reference Sukenik and Friedmann2018). The LITMUS SRep tasks, developed within the COST Action IS0804 (Marinis & Armon-Lotem, Reference Marinis, Armon-Lotem, Armon-Lotem, de Jong and Meir2015), were utilised. SRep is a complex task as it involves sentence comprehension, storage in short-term memory, and regeneration using long-term memory along with recently activated lexical items (for more details, see Potter & Lombardi, Reference Potter and Lombardi1998).
The sentences were pre-recorded by native speakers of the respective languages to ensure consistency in presentation. The English LITMUS SRep (Marinis & Armon-Lotem, Reference Marinis, Armon-Lotem, Armon-Lotem, de Jong and Meir2015) and the Hebrew LITMUS SRep (based on the Hebrew version by Meir et al., Reference Meir, Walters and Armon-Lotem2016) included 30 sentences of various lengths and syntactic complexities. The English LITMUS SRep comprised the following 10 structures: Subject-Verb-Object (SVO) with auxiliaries/modals, passives, object questions (who, what, which), sentential adjuncts, and object relative clauses. The Hebrew LITMUS SRep task included SVO sentences with obligatory and optional prepositions, biclausal sentences with coordination and subordination, object wh-questions, oblique wh-questions, object relatives, VSO, conditionals, and biclausal sentences with conjunctional phrases.
The tasks were presented using PowerPoint 2016 on a laptop/ipad, either in a quiet room or via Zoom. A child received a score of 1 if the morpho-syntactic structure was correctly repeated and 0 if it was not preserved. Lexical substitutions were considered correct (e.g., uncle/man, soup/food, cooked/made). This scoring method is advantageous for evaluating syntactic abilities without penalising vocabulary errors (Marinis & Armon-Lotem, Reference Marinis, Armon-Lotem, Armon-Lotem, de Jong and Meir2015).
2.3. Procedure
All procedures in the current study were conducted in compliance with relevant guidelines and regulations, following a protocol approved by the Institutional Review Board (IRB) of Bar Ilan University and the Chief Scientist of the Israeli Ministry of Education. Written informed parental consent was obtained from each child’s parents prior to their participation. Additionally, oral assent was obtained from the children before each testing session.
Children were recruited through advertisements on social media and professional networks. Testing commenced in 2019 and continued through the years 2020–2023. Before each meeting, parents received a brief verbal explanation of the study along with a detailed written explanation provided in both English and Hebrew.
Due to the COVID-19 pandemic, some children were assessed remotely via Zoom. We ensured an equal number of participants were tested via Zoom in both clinical groups (autism and non-autism). Previous studies have shown a correlation between telehealth and face-to-face assessment scores (Sutherland et al., Reference Sutherland, Trembath, Hodge, Rose and Roberts2019; Hao et al., Reference Hao, Franco, Sundarrajan and Chen2021). Face-to-face meetings were primarily conducted in the children’s homes or educational settings. We made concerted efforts to foster cooperation and obtain accurate scores reflecting each child’s abilities.
Each child participated in three to five sessions in a quiet room, with structured settings used as needed. Simple rewards were provided to all participants. Verbal tasks were administered twice, once in English and once in Hebrew, during different sessions. The tests were administered by the first author, a certified and experienced speech and language pathologist proficient in both English and Hebrew. The language of the initial assessment session was counterbalanced among the participants, with some starting in English and others in Hebrew. Each session lasted between 20 and 40 minutes, depending on the child’s cooperation.
3. Results
All statistical analyses were conducted in R version 4.0.5. (R Core Team, 2022). The binomial models for SRep were built by adding the random and fixed variables in a step-by-step procedure, starting with an intercept-only model as a baseline. The null models included both by-subject random intercepts and by-stimulus random intercepts. The variables and/or the interactions of the variables were kept in the model only if they significantly improved the fit of the model and resulted in a reduced AIC-value. In the Results section, we report the minimally adequate models that performed significantly better than the intercept-only baseline model.
3.1. Morpho-syntactic abilities in English and in Hebrew: SRep performance in the three groups
The results of the SRep performance in Hebrew and English are presented in Figure 1. The figures reveal a large variability across all groups, with a notably higher dispersion of scores in English (ASD-Nat: M=0.47, SD=0.49, Min-Max: 0–1; ASD-NI: M=0.39, SD=0.49, Min-Max: 0–1; TLD-Nat: M=0.55, SD=0.50, Min-Max: 0–1) compared to Hebrew (ASD-Nat: M=0.82, SD=0.39, Min-Max: 0–1; ASD-NI: M=0.67, SD=0.47, Min-Max: 0–1; TLD-Nat: M=0.90, SD=0.39, Min-Max: 0–1). Table 2 depicts the results of a binomial mixed-effects logistic regression model that analysed the data in both Hebrew and English. The model included Group (TLD-Nat, ASD-NI, ASD-Nat) and Language (English, Hebrew) as fixed effects, as well as their interactions. The model revealed main effects of Group and Language, as well as a significant Group*Language interaction. Follow-up pairwise analyses with Bonferroni corrections for multiple comparisons on the significant interaction showed no group differences in the English language (TLD-Nat vs. ASD-Nat: p=1.00; TLD-Nat vs. ASD-NI: p=.082; ASD-Nat vs. ASD-NI: p=.41). However, group differences emerged in the Hebrew language (TLD-Nat vs. ASD-Nat: p=.56; TLD-Nat vs. ASD-NI: p=.02; ASD-Nat vs. ASD-NI: p=.51). Thus, differences were observed only in Hebrew between the TLD-Nat and ASD-NI groups, with the latter performing significantly lower.
Figure 1. Individual differences in the SRep task per group per language.
Note: Thick black lines inside yellow diamonds indicate mean values. Violin plots show score distributions. Individual points and lines represent participant trends in English and Hebrew.
Table 2. Parameters for the binomial mixed-effects logistic regression model results for SRep performance in both languages
Subsequently, we analysed the SRep task separately in English and Hebrew to evaluate the effect of the syntactic structure. The results are provided in Figure 2A for English and Figure 2B for Hebrew.
Figure 2. The performance in the SRep tasks per structure per group.
The results for the binomial mixed-effects logistic regression model for the English SRep data (see Supplementary Materials S1) included Group (TLD-Nat, ASD-NI, ASD-Nat) and Structure (SVO, biCl, Obl-Q, Obj-REL, VSO, ADV-CONJ, COND) as fixed effects, as well as their interactions. The model showed significant effects of Group, Structure, and a Group*Structure interaction. Follow-up pairwise analyses with Bonferroni corrections for multiple comparisons on the significant Group*Structures interaction showed that differences between the TLD-Nat and ASD-Nat group emerged only for the bi-clausal sentences (ASD-NI < TLD-Nat, p=.04); group differences for other structures and other languages were not significant (see Supplementary Materials S2).
The outcomes for the corresponding model of the SRep data in Hebrew can be found in Supplementary Materials (S3). The model included Group (TLD-Nat, ASD-NI, ASD-Nat) and Structure (SVO, SVO2AUX, BiCl, Sh-PAS, L-Pass, S-REL, WH-Q, Obj-REL, COND) as fixed effects, as well as their interactions. Like the English data, the model showed significant effects of Group, Structure, and Group*Structure interactions. Follow-up pairwise analyses revealed differences between the TLD-Nat and ASD-NI groups only for bi-clausal sentences, object questions, and conditionals, with the ASD-NI showing lower performance (p=.009, p=.005, p=.02, respectively) (see Supplementary Materials S4). Differences for other structures were not significant. Furthermore, differences between TLD-Nat and ASD-Nat were not significant. Finally, no group differences were observed between the two autistic groups.
3.2. Language exposure variables and morpho-syntactic abilities in English and in Hebrew: Correlational analyses
The correlational analysis of the exposure to English/Hebrew and morpho-syntactic abilities in the two languages across autistic and non-autistic groups is presented in Table 3. The results indicate significant correlations between the length of exposure and morpho-syntactic abilities in Hebrew for both the ASD and TLD groups. However, no correlations were observed between the length of exposure and morpho-syntactic abilities in English for either the ASD or TLD groups. It is also noteworthy that no correlations were found between parental ratings in English or Hebrew among the ASD and TLD groups.
Table 3. Spearman method with pairwise-deletion analysis for TLD-Nat and ASD groups
4. Discussion
The present study investigated language acquisition outcomes in multilingual children with ASD who acquire English through non-interactive means, compared to their autistic and non-autistic multilingual peers who acquire languages naturalistically through exposure at home and in educational settings. The study assessed the morpho-syntactic skills of all participants in both Hebrew and English using SRep tasks. Existing research on unexpected non-interactive language acquisition among autistic children is limited, primarily consisting of individual case studies. In this study, we present data from 14 autistic children demonstrating non-interactive acquisition of English and 12 autistic individuals for whom English is their home language, compared to 20 naturalistic multilinguals. Addressing our first research question, we investigated potential group differences in morpho-syntactic skills. Our second research question aimed to explore performance discrepancies within the two languages among multilingual individuals. The third research question examined the relationship between input/exposure and children’s morpho-syntactic abilities in both their languages.
4.1. The children`s morpho-syntactic abilities: Group comparisons (ASD-Nat, ASD-NI, and TLD-Nat)
At the group level, the analysis revealed differences only in Hebrew, while in English, the three groups were statistically indistinguishable. Significant individual differences should be acknowledged in both languages across all three groups. Specifically, in Hebrew, differences were observed only between the TLD-Nat and the ASD-NI groups, whereas the TLD-Nat and the ASD-Nat groups showed similar performance.
The findings related to English are noteworthy, considering that for the ASD-NI group, Hebrew was the ambient language in their home environment, while exposure to English occurred later in their childhood through non-interactive sources. Despite this delayed exposure and the non-interactive nature of their input, children in the ASD-NI group achieved comparable mastery in morpho-syntactic skills to their peers, both autistic and non-autistic, who were exposed to English naturalistically at home. One possible explanation for these findings in English could be that children in the ASD-NI group had a higher motivation to learn the new language (English) because it aligned with their special interests, whereas for children in the ASD-Nat group, English was just another language in their environment. Additionally, the elimination of social pressure may have been beneficial for children with ASD who experience difficulties with face-to-face interaction.
Regarding differences in performance across various morpho-syntactic structures, we found that most structures posed similar difficulty levels for all groups. In Hebrew, bi-clausal sentences with advanced conjunctions were notably more challenging, while in English, bi-clausal sentences, object questions, and conditionals presented greater difficulties. There were no statistically significant differences in group performance across the remaining structures.
In conclusion, multilingual children in all three groups demonstrated a diverse range of skills in both languages, with no observed group differences in English. However, differences were noted between the TLD-Nat and ASD-NI groups in Hebrew. It is important to note that most children in the ASD-NI group were born to Hebrew-speaking parents, and Hebrew was their primary language at home, yet despite ample exposure to Hebrew at home children in the ASD-NI group transitioned to English.
4.2. The children`s morpho-syntactic skills in the two languages: English vs. Hebrew
Regarding language differences, nearly all children in our study exhibited notable heterogeneity in morpho-syntactic skills across their languages: English and Hebrew. These results align with previous studies suggesting that linguistic abilities among bilinguals are unevenly distributed between their two languages (Kohnert, Reference Kohnert2010). Our findings reveal a considerably higher dispersion of scores in English compared to Hebrew. This observation aligns with our expectations, considering that the children in the study currently reside in Israel, where Hebrew is the dominant environmental language. Therefore, the English task may have posed greater difficulties to all children growing up in Israel, where English is ubiquitous but not the dominant societal language.
Secondly, the design of the SRep task in English and Hebrew may also contribute to the asymmetry across the languages, as the tasks in English and Hebrew may not be equally difficult. Future studies should include monolingual English- and Hebrew-speaking controls to determine the comparability of the tasks under monolingual language acquisition. Currently, there are no published norms available to assess English-Hebrew-speaking children on SRep tasks. Sarig (Reference Sarig2014) conducted a study on English-Hebrew multilinguals with and without DLD and found that the mean accuracy rate for Hebrew-English children with TLD aged 5;9 to 6;3 on SRep was around 42.5% in English and 65% in Hebrew. Similarly to Sarig (Reference Sarig2014), our study showed lower scores across all groups on the English SRep task. Previously, there has been conflicting evidence regarding the equivalence of difficulty of LITMUS tasks. For instance, Meir et al. (Reference Meir, Walters and Armon-Lotem2016) demonstrated that LITMUS SRep tasks in Russian and Hebrew are of parallel difficulty. On the other hand, Van Wonderen and Unsworth (Reference Van Wonderen and Unsworth2021) showed differences in the performance of age-matched Spanish-speaking and Dutch-speaking monolingual children in the LITMUS Cross-linguistic Lexical Tasks, suggesting caution in using LITMUS tasks to compare the performance of multilingual children across languages. Therefore, based on the performance of multilinguals in the current study, we cannot rule out the possibility that the task in one language was more challenging than its parallel counterpart in the other language.
4.3. Language exposure and morpho-syntactic skills
Previous studies examining the relationship between language exposure/input and linguistic performance have suggested that increased exposure benefits children with TLD. However, for multilingual children with ASD, this relationship is often unclear, as heightened exposure does not consistently translate into improved linguistic performance. Some researchers argue that increased exposure does not enhance performance among multilingual children with ASD (Armon-Lotem & Meir, Reference Armon-Lotem and Meir2022; Beauchamp et al., Reference Beauchamp, Rezzonico, Bennett, Duku, Georgiades, Kerns and Elsabbagh2023; Hambly and Fombonne, Reference Hambly and Fombonne2012), while others posit the opposite (Gonzalez-Barrero & Nadig, Reference Gonzalez-Barrero and Nadig2018). The results of our study are mixed, showing a positive correlation between language exposure and linguistic ability in Hebrew. However, no such correlation was found in English. Several explanations are posited: First, as mentioned earlier, English in Israel serves as a lingua franca and is widely accessible, potentially leading to heightened exposure among children. Second, regarding the children in the ASD-NI group, it is plausible that parents were unable to accurately estimate the Age of Onset (AoO) for their children, as much of the exposure occurred without parental oversight. It is likely that parents were unaware of when their children began acquiring English. Third, the level of exposure may also be intertwined with the special interests of these children. Therefore, it might not be solely the quantity of exposure to English that is related to morpho-syntactic acquisition, but also their specific interests and the necessity for using the English language. Finally, it is plausible to suggest that the indices used in the current study, based on the parental questionnaire, were not sufficiently sensitive to capture group differences. In addition to the language history questionnaires employed in this study, direct observation of linguistic environments – such as the evaluation of input amount and type – should be considered. One potential tool is the Electronically Activated Recorder (EAR), an audio recording device also available via smartphone apps, which can assess real-world linguistic input available to children. The EAR provides a “ground truth” estimate of day-to-day linguistic environments, which cannot be accurately captured through questionnaires alone (see Cychosz et al., Reference Cychosz, Villanueva and Weisleder2021). In conclusion, future research should further investigate both the quantity and quality of input/exposure, as well as the motivation to acquire the language, in children with and without ASD.
4.4. Limitations
This study is the first of its kind to investigate non-interactive foreign language acquisition among autistic children in a group comparison rather than through individual case studies. We acknowledge that autistic children often exhibit heterogeneous language profiles. Additionally, all children in this study were multilingual, which typically results in uneven linguistic skills. These intertwined factors make it challenging to distinguish between their respective influences. Future studies should consider further subdividing participants based on their linguistic profiles to conduct more nuanced analyses of performance across subsets. Naturally, increasing the number of participants would provide a more robust understanding.
Another limitation of this study is the reliance on parental interviews and questionnaires to gather data on language exposure. It is important to note that digital screen media often does not require or involve caregiver engagement, which may be particularly relevant for parents of autistic children who might face challenges in engaging in mutual activities with their child. Consequently, the child might primarily interact with digital screen media independently. As a result, parental questionnaires may not fully capture the accuracy of language exposure, as parents were often uncertain about when their child first began non-interactive exposure to English. Additionally, we found a poor correlation between parental ratings of English proficiency and the child’s actual performance in English. Therefore, caution is advised when interpreting parental reports. As mentioned earlier, in addition to parental reports, direct measures of language input should be incorporated in future studies. Future research should also examine the role of motivation in language learning (in multilinguals with non-interactive language exposure) and language maintenance (in multilinguals acquiring both languages naturalistically). More detailed questionnaires on screen-type exposure and input should be utilised in multilingual research, similar to the approach used by Sun et al. (Reference Sun, Lim, Low and Kee2022), which assesses the duration, frequency, content, design, and use of bilingual children’s digital screen media exposure at home.
5. Conclusions and clinical implications: natural and unexpected acquisition of foreign languages in ASD
The phenomenon of unexpected foreign language acquisition via non-interactive sources among autistic children presents a dilemma for educators, parents, and other professionals involved in assessing, teaching, and providing therapy to autistic children. The results of our study suggest diverse paths to language acquisition: children may acquire their two languages through interactive naturalistic input at home (as observed in individuals with and without autism) or via non-interactive media (as seen in autistic children acquiring English through screens). Interestingly, significant group differences were found only in Hebrew, the societal and home communication language for most autistic children who acquired English non-interactively, with these individuals showing lower performance compared to typically developing multilinguals. Importantly, in English, the three groups were statistically indistinguishable.
Our study raises several new questions. First, does non-interactive foreign language acquisition occur exclusively among children with ASD, or does it also manifest in children with TLD? Second, is the incidental acquisition of foreign languages through media exposure becoming more common? Third, what are the underlying reasons for this unexpected acquisition of a non-ambient language among individuals with autism? Fourth, based on our findings and anecdotal data, what recommendations should parents and caregivers consider when they encounter this phenomenon? In our study, some children mentioned that they autonomously learned English to navigate the Internet and explore specific interests (e.g., animals, ancient languages, warriors). Conversely, some parents reported an emotional event triggering their child’s language shift. Some children in our study were proficient in both Hebrew and English, while others did not fully acquire Hebrew and switched to English when their Hebrew language skills were not fully developed. This poses dilemmas for school staff and parents regarding whether to encourage or discourage the use of a non-ambient language. Some children expressed reluctance to speak Hebrew and preferred communicating solely in English. Furthermore, some staff members noted language decline in Hebrew compared to English, with children showing word retrieval difficulties, accent changes, and morphological errors in Hebrew as the child progressed in English. Future longitudinal studies should examine language development and loss patterns in children with ASD who exhibit natural and unexpected multilingualism in both languages. Such studies could elucidate whether these children cease communication in the societal language and exclusively adopt English, or if their English skills serve as a bridge to enhance and master Hebrew. Moreover, such longitudinal studies should document changes in language input/exposure patterns (both interactive and non-interactive) in order to better understand the mechanisms driving language skill development and/or decline in autism.
Based on our findings, we recommend assessing multilingual children in both their languages, as evaluating only one language may not provide a comprehensive view of their abilities. Our findings corroborate previous studies indicating that language skills in multilingual children are unevenly distributed across their two languages (Kohnert, Reference Kohnert2010). The importance of dual-language assessment and treatment for multilingual children, with and without autism, should be underscored for speech-language pathologists and educators. Assessing both languages of multilingual children is expected to optimise treatment effectiveness based on the child’s profile, regardless of whether they acquire languages naturalistically through interactive exposure or unexpectedly through the Internet, television, and YouTube. Further analysis of individual cases and groups will deepen our understanding of this remarkable phenomenon.
Data availability
The datasets generated and analysed in the current study are on OSF.
Acknowledgements
The authors express their heartfelt gratitude to the children and their parents for their invaluable participation in this study. They also wish to thank the speech and language pathologists for their vital support in facilitating the recruitment of the children.
Funding statement
This research was partially funded by the National Institute for Psychobiology in Israel (NIPI) under the grant titled, “Exploring the Use of Dual Language Assessment for Bilingual Children with ASD: Implications for Diagnosis and Treatment.”
Open access
