Demographic information for the full sample is presented in Table 1. Seventy-six percent of the autistic children in our population were minimally verbal or non-verbal, according to the ADOS module used for their assessment (Toddler module or Module 1), and 37% of children in the clinical group. Of note, given the age range of our sample (24–84 months), limited oral language would meet developmental expectations for the children at the lower end of the age range (see Figure S1 in Additional File 4 for the age distribution by group and ADOS module).

The intensity of the interest in letters and numbers, respectively, was rated for each child using an ordinal scale (None < Moderate < Intense < Exclusive) according to the clinician reports. Despite the absence of functional communicative oral language in the majority of autistic children, their interest in letters was generally higher than that of the clinical group. Twenty percent of the reports of autistic children mentioned an intense or exclusive interest in letters, when it was the case for only 3% of the clinical group. For autistic children, the odds of having a greater level of interest in letters were 2.78 times higher than for children in the clinical group (OR = 2.78, 95% CI [1.55–5.17], p = 8.42e−4). Their interest in numbers was also higher, 17% of autistic children had an intense or exclusive interest in numbers, while it was the case for only 2% of the clinical group. Autistic children had 3.49 times higher odds of having a greater level of interest in numbers than children in the clinical group (OR = 3.49, 95% CI [1.85–6.96], p = 2.07e−04). The analyses were adjusted for age and for the clinicians who performed the autism assessment (see Table S2 in Additional file 5).

Demographic information for the sub-sample used in Study 2 is presented in Table 2.

Language development Language level was documented through the choice of ADOS module at the time of assessment and by section C of the QIMET. Eighty-four percent of the autistic children were non-verbal or minimally verbal according to the ADOS module used for their assessment (Toddler module or Module 1), 27% of the clinical group, and none in the typically developing group (no developmental delay in this group). Figure S2 in Additional File 4 shows the age distribution by group and ADOS module. On the QIMET, thirty-three autistic children, 2 in the clinical group, and none in the TD group were reported as completely non-verbal by their caregiver (see Table 3 for detailed results on language ability extracted from items C3, C4, and C5 of the QIMET). Of the 33 autistic children reported as fully non-verbal, the QIMET revealed that 8 could sing the alphabet song, 9 could name letters, 9 could count numbers, 1 could spell some words, 3 did some pretend-reading, and 2 could read some words.

Presence and intensity of the interest in letters and numbers Most autistic, clinical, and TD children in our sample had developed at least a moderate interest in letters (autistic: 82%; clinical: 87%; TD: 95%) and numbers (autistic: 83%; clinical: 85%; TD: 97%) at the time of the study. Thirty-seven percent of autistic children had an intense or exclusive interest in letters, 23% of the clinical group, and 25% of the TD group, according to their parents. Thirty-six percent of autistic children had an intense or exclusive interest in numbers, 23% of the clinical group, and 24% of the TD group.

Age of emergence of the interest in letters and numbers The median age of emergence of the interest in letters, written words and reading in autistic children was 30 months. Compared to autistic children, the emergence of this interest was delayed in the clinical group (median: 36 months; HR = 0.73, 95% CI [0.55–0.97], p = 0.029), but the TD children were no different (median: 28.5 months; HR = 1.17, 95% CI [0.87–1.58], p = 0.294) (Fig. 1a). The median age of emergence of the interest in numbers and mathematical symbols in autistic children was 30 months. Compared to autistic children, the emergence of this interest was delayed in the clinical group (median: 36 months; HR = 0.63, 95% CI [0.47–0.84], p = 0.002), but the TD children were no different (medians: 30 months; HR = 1.30, 95% CI [0.97–1.75], p = 0.077) (Fig. 1b). The age of emergence of both these interests in autistic children did not differ according to the child’s communicative language ability, i.e., between autistic children evaluated with ADOS module 2 and up (grouped as communicative language users) and others evaluated with Toddler module and module 1 (grouped as minimally or non-verbal) (Letters: HR = 0.88, 95% CI[0.51–1.51], p = 0.634; Numbers: HR = 0.85, 95% CI[0.49–1.47], p = 0.559).

Emergence of the interest in letters and in numbers by diagnostic group. Proportion of autistic, non-autistic clinical, and typically developing children by age of onset of their interest (in months). a. Proportion showing an interest in letters, written words and reading by age. b. Proportion showing an interest in numbers and mathematical symbols by age

Behaviors relevant to the interest in letters and numbers A detailed questionnaire was used to assess the presence or absence of specific behaviors of interest and competence relating to letters and numbers. Behaviors of interest and competence were grouped under index scores. The prevalence of each behavior of interest and competence within index scores in clinical and TD children, shown in Fig. 2 (see Table S3 for detailed results), were compared to autistic children. The autistic and TD groups share an equivalent age distribution, while the clinical group is slightly older, resulting in a conservative bias in these analyses, if the autistic group shows a higher frequency of these behaviors than the clinical group in spite of being younger on average. Autistic children stand out from clinical children by the fact that letters were “special” for 51% of them (p = 0.002). Four out of five behaviors of interest in written material related to a form of social reciprocity were less represented in autistic than clinical and TD children (i.e., pretends to read; brings book to be read by adult; likes reading for others; often accepts to read with adult). Autistic children showed lower levels of competence than clinical and TD children in recognizing some written words, counting out loud, counting objects of the same category and being able to count. Some behaviors of interest and competence were more represented in autistic than TD children, such as interest in letters on screens, the tendency to place letters in alphabetical order, and interest in games with numbers or math. However, these behaviors were no different in frequency from those observed in the clinical group. Similarly, some behaviors of interest and competence were less represented in autistic than clinical children (pronounces the sounds that letters make, writes numbers or mathematical symbols) and TD children (pretends to write, understands that we read from left to right). Some of these differences may be attributed to age, which was not controlled for in these comparisons. The significance level was set at p = 0.05 after Bonferroni correction.

Comparative prevalence of letter- and number-related behaviors of interest and competence grouped under index scores: a. non-social interest in letters; b. social interest in letters; c. interest in letters not specified; d. competence with letters; e. interest in numbers; and f. competence with numbers. Statistical differences are indicated next to the clinical and TD groups showing significant differences compared to the autistic group, with significance levels marked as * (p < .05), ** (p < .01), and *** (p < .001) after Bonferroni correction for multiple comparisons

Index scores of interest and competence When controlling for age, not being autistic was associated with a higher index score for social interest in letters, a particularly large difference in the case of TD children; (clinical: ß = 0.46, 95% CI [0.23–0.70], p = 1.0e−04; TD: ß = 1.30, 95% CI [1.05–1.54], p = 2.2 e−22). There was an opposite trend for non-social interest in letters, where the clinical group tended to score lower than the autistic group, and the TD group showed no difference (clinical: ß = − 0.41, 95% CI [− 0.68–− 0.14], p = 0.003; TD: ß = − 0.16, 95% CI [− 0.43–0.12], p = 0.27). Being TD was associated with a higher competence score with letters (clinical: ß = − 0.004, 95% CI [− 0.23–0.22], p = 0.98; TD: ß = 0.24, 95% CI [0.01–0.48], p = 0.04). Not being autistic was associated with a lower level of interest in numbers (clinical: ß = − 0.384, 95% CI [− 0.64–− 0.12], p = 0.004; TD: ß = − 0.537. 95% CI [− 0.81–− 0.27], p = 1.2 e−04), despite a generally higher level of competence (clinical: ß = 0.329, 95% CI [0.09–0.57], p = 0.007; TD: ß = 0.774. 95% CI [0.52–1.02], p = 3.1 e−09). Index scores were standardized within the sample to simplify effect sizes interpretation. Analyses were adjusted for age (See Table 4 and Table S4 in Additional file 5).

Notable additional information: Very few children across groups could read full sentences, but the three participants who could in the autistic group had read their first word at 24, 30 and 36 months, about two years before the 14 participants in the clinical group (mean: 56.5 months; minimum age: 30 months) and the four in the TD group (mean: 51 months; minimum age: 36 months). Several of the autistic children interested in written material displayed “unexpected bilingualism” [23, 24], i.e., they had reached various levels of self-taught skills in a language not spoken in their environment, sometimes to the point of refusing to speak the language most present in their environment, and insisting on speaking English even when their parents could not understand. There were multiple mentions by caregivers of a strong interest in foreign languages and alphabets (e.g., changing the language of video subtitles). The caregivers mentioned the use of tablets, videogames, TV, and YouTube as potential explanations for their child’s foreign language acquisition.