The human Faculty of Language allows the generation of an infinite set of structured expressions out of a finite set of words and rules. This property is assumed to result from a recursive process that embeds expressions within other expressions (1), creating complex hierarchical objects. The property of representing recursive hierarchical embedding (henceforth recursion) has been claimed in various domains (1–7). A crucial question remains whether recursion is domain-general, multiply-domain-specific (one recursive capacity per domain), or primarily linguistic and then used in other domains through linguistic computations (8–10).

In typical development cases and most populations, recursion emerges in syntax (1, 11–13) and at the level of linguistic discourse (14–16). This widespread and effortless use has led to the hypothesis that recursion is primarily linguistic, with its use in other domains dependent on the human Faculty of Language (17). Recursion has also been proposed in the domains of action planning (18–21), visuospatial processing (22), Theory of Mind (ToM) (23, 24), and music (25, 26). However, the prevalence, hierarchical depth, and automaticity of use in these domains seem limited compared to language.

Recursion has also been investigated in atypical development. Some studies have investigated the contribution of linguistic recursion to second-order ToM in autism (27, 28). Their results show that recursive syntax (sentential complementation) predicts second-order false beliefs (28) and that training with sentential complementation improves ToM, with most gains for children with mild autism (27). However, this relationship might be limited to explicit ToM and explained by the role of working memory in both domains (28–30).

In line with findings on specific autistic cognition, language development is heterogeneous in autism (31). Highly atypical developmental trajectories have been reported, including cases where missing early communication skills, such as joint attention, do not preclude the emergence of structural language (32). Similarly, literacy skills suggestive of implicit meaning-making have been observed in nonspeaking autistic individuals, offering a potential alternative route to language acquisition (33). As for syntax, syntactic deficits can occur even when vocabulary and sentence repetition are intact (34–36), and 20-30% of individuals with autism are non-verbal or minimally verbal (37). In this subgroup, linguistic recursion is absent, similar to cases in which children were not exposed to language in critical periods (see 38). In the latter, though, fluency and lexical abilities are preserved.

Linguistic comprehension in non-verbal and minimally verbal individuals with autism is vastly understudied. Slušná et al. (39) studied individuals with non- or minimally verbal autism and has found that expressive language (General level of no echolalic language, measured with the ADOS; see Table 1 for details) correlates with receptive language (or verbal mental age, measured with the Peabody Picture Vocabulary Test–III (PPVT), or with the Clinical Evaluation of Language Fundamentals (CELF) when PPVT score reached a verbal mental age (VMA) of 3 years) but not with non-verbal IQ. This study also shows a strong association between sense-making/ComFor (Forerunners in Communication) and Verbal Mental Age/Receptive Language.

In another study, Vicente et al. (40) investigated whether autistic adults with minimally receptive vocabulary could comprehend two-word phrases presented auditorily where a familiar noun is modified by a familiar adjective (e.g., red broom). They found that only one out of eight participants succeeded. Interestingly, the pattern was reversed with a visual version of the same task, as all but one participant could form simple symbolic compositions with iconic representations in all trials. This opens the question of whether visual-spatial recursion could also be available to minimally verbal populations with autism and, more broadly, whether language is necessary for recursion in vision.

Previous work on the relationship between language and visual recursion paints an ambiguous picture. On the one hand, verbal or motor interference does not affect the capacity to perform adequately in a visual recursion task (10). Moreover, when testing well-trained participants with motor, visual, and music recursion, fMRI experiments show domain-specific patterns of brain activity (5, 7, 41). Thus, each domain could implement similar recursive computations on independent biological systems (8, 9). On the other hand, when testing untrained participants, the behavioral and neural evidence is compatible with shared resources in acquiring recursive rules across domains (6, 42, 43). For instance, stroke patients with deficits in parsing sentences with two levels of embedding are also impaired in the processing of visual recursion, and visual recursion seems to activate a left-lateralized brain network, including the classic language brain areas (42, 43). However, these studies were performed in adults with fully developed language.

Perhaps recursion is domain-general at first, but acquires specializations depending on the degree of exposure and training with specific domains. The adult neural pattern associated with processing natural and artificial languages only emerges after extensive training and in specific development stages (44–47). More extensive exposure and training could explain why recursion is expressed first and more prevalently in language than in other domains. However, recursion might also become available later to other domains.

In our previous work in the visual domain, we found that children could master visual recursion around age 9 but not before (48). This is later than the acquisition of syntax. However, the age of acquisition of recursion may vary across linguistic forms, being available for simple adjectives and verbal compounds at age 2–3 and more complex sentence complements at age 5-7 (49). We also found that prior acquisition of a simpler iterative task was necessary to master recursion within the same developmental stage (48). This effect resembles the acquisition constraints in language, in which conjunctive constructions must be mastered before recursion (50, 51). In conjunction and iteration, items are added within a fixed hierarchical level without generating new levels (e.g., very, very, very big).

Interestingly, we also found that grammar comprehension (controlling for IQ) equally predicted accuracy in visual recursion and iteration. However, these grammar comprehension tasks included sentences with a maximum of one embedded clause. Unlike our study with stroke patients mentioned above, they were not designed to test for the similarities between linguistic and visual recursion (42).

This manuscript explores the relationship between visual and syntactic recursion from a developmental perspective by presenting a case report from an 11-year-old child (A.) with autism and hyperlexia (mechanical reading), whose linguistic skills do not exceed those of minimally verbal individuals despite possessing a visually-based, predominantly nominal vocabulary acquired by reading. We investigated whether visual recursion can dissociate from linguistic syntax by testing A. with a visual recursion task and a control visual iteration task. We compared his accuracy levels with 26 Typical Development (TD) children attending the 4th grade. If linguistic recursion is a prerequisite for visual recursion, we would expect A. to fail in the former. However, he could still succeed in the control iteration task due to his preserved visual cognitive performance. Conversely, if recursion in vision is not dependent on language, we expect A. to pass both recursion and iteration tasks.

We presented a case report of an 11 y.o. minimally verbal but hyperlexic autistic child who can represent visual recursion despite showing no evidence of this capacity in linguistic production or comprehension. His accuracy was above chance and was not significantly different from that of 9–10 y.o. children with TD. This suggests that visual recursion is not dependent on a primary linguistic recursive capacity but could be domain-general or multiply domain-specific.

These findings resonate with prior research on autistic cognition, which shows marked dissociations between verbal and nonverbal intelligence (64). Many minimally verbal autistic individuals demonstrate significantly higher scores on nonverbal reasoning tasks such as Raven’s Progressive Matrices despite low scores on verbal subtests of traditional IQ measures. A.’s case similarly suggests that the capacity for recursion may be preserved in visuospatial domains while remaining inaccessible in language.

Within the generative framework (65), language recursion is the core combinatorial capacity that builds an “infinite array of expressions” that are interpreted at two interfaces, a conceptual-intentional (C-I) system connecting the same mental expressions to reasoning and planning among “other activities of the internalized mental world” and a sensorimotor (S-M) system connecting with the external world via language production and perception (externalization). We surmise that A.’s impairment in linguistic recursion might result from deficits in both the C-I and the S-M systems required for language, which may prevent recursion to manifest linguistically.

Visual recursion is likely to hinge on different machinery. First, it does not involve an intentional component. Second, it has a sensory component but does not involve a structured motor output like linguistic utterances. Thus, while language recursion requires intact loops involving core, conceptual-intentional, and sensorimotor systems, visual recursion can operate with core and visual sensory and conceptual systems without intentional and motor components. As reviewed above, visual conceptual and combinatorial properties might be preserved in minimally verbal autism (40).

This interpretation remains agnostic regarding whether recursion is domain-general or multiply domain-specific (10). However, it hypothetically (but not necessarily) allows the same core capacity to interface with specialized sensorimotor/sensory and conceptual-intentional/conceptual systems. In this regard, A.’s hyperlexia suggests a strong visual sensory system and the ability to associate visual patterns with vocal labels but without linguistic semantics. Interestingly, although his visual IQ was low in most subtasks in the Leiter-3, he had a medium-range score in the subtest Classification and Analogies, which explicitly assesses the capacity for general pattern analysis and prediction of “what goes next”. The latter can indicate a capacity for recognizing abstract visual patterns and rules (i.e., a visual conceptual system).

In sum, specialization of the sensory and conceptual abilities to the visual domain could afford the interaction of these systems with a core combinatorial capacity and the detection of visual recursion. Conversely, a faulty integration of the subcomponents making the C-I and S-M systems would impede expressive and receptive recursion in language.

This study has two significant limitations. First, broader generalizations are impossible, as this case report presents a single subject with a rare profile: he is a minimally verbal autistic child with hyperlexia in 3 languages. Second, we did not directly test A. with a linguistic recursion comprehension task. As such, we do not know whether A. could demonstrate recursion in a reading or writing task, should he be taught to read or write. Recent evidence suggests that nonspeaking autistic adolescents and adults may display implicit knowledge of English orthography in response time-based spelling tasks, suggesting foundational literacy skills (33). However, it is unclear whether this implicit capacity for orthography translates into semantic and, more importantly, syntactic understanding (66). Here, we infer limited syntactic recursion from A.’s limited expressive compositionality (i.e., two-word phrase requests at most; e.g., pink milkshake) and his low (and likely overestimated) score on the vocabulary test PPVT. Since previous research has shown an association between receptive and expressive language (39, 52) and very limited phrase comprehension in minimally verbal autism (40), we adduce that A. was also limited in linguistic recursion.

Finally, it is important to acknowledge that a lack of functional language might stem from impairments at either abstract computational abilities, such as recursion, or from an absence of verbs and/or grammar, which are fundamental and specific to language. Our results show that recursion is present, at least in vision, albeit possibly made available by simpler domain-specific machinery. However, it is also possible that a domain-general system is not available to language due to the latter set of limitations.