This study utilizes the same dataset as Kato et al. (2024), which consists of a spoken language corpus containing speech samples from individuals with and without ASD. The dataset includes 64 individuals diagnosed with ASD (M = 18, SD = 3.48) and 71 non-ASD individuals (M = 19, SD = 2.77), all aged 14 and above. The selection of 14 years as the lower age bound was informed by a methodological rationale aimed at minimizing confounds arising from ongoing language development. In line with the position adopted in Kato et al. (2024), we draw on a limited, pro-critical period perspective—not as a theoretical commitment, but as a pragmatic framework for ensuring linguistic maturity in the sample. Specifically, our stance assumes that core morphosyntactic and pragmatic systems are typically stabilized by mid-adolescence, thereby providing a consistent developmental baseline for identifying group-level differences. This view aligns with foundational formulations of the Critical Period Hypothesis in first language acquisition (Lenneberg, 1967; Newport, 1990), while acknowledging the broader spectrum of positions in the literature, including counterevidence from adult learners, neuroplasticity research, and interactionist accounts. Although longitudinal evidence on Japanese is limited, sentence-final particle use and discourse structuring appear to be well established by early adolescence (Clancy, 1985), supporting our rationale for age selection.

This study analyzes spoken responses from ADOS-2, Modules 3 and 4, which assess verbal fluency, social cognition, and pragmatic language use in verbally fluent individuals. Fluency is defined by the ability to produce complex sentences, logical connectors (e.g., but, though), and event descriptions beyond the immediate context, with minor grammatical errors allowed. Module 3, used for adolescents, includes imaginative play, while Module 4, designed for older adolescents and adults, omits this task. Despite this difference, both modules share a consistent structure, enabling comparable data collection across age groups.

Participants with suspected ASD were audio-recorded while completing six to eight tasks in ADOS–2, Modules 3 and 4. These recordings were transcribed and annotated for lexicogrammar, and the resulting texts were stored in the spoken language corpus developed by Kato et al. (2022). The present study used these annotated texts for analysis.

From this corpus, semi-structured interview responses were selected. Topics included personal difficulties, social relationships (friendships, marriage, family), and hypothetical scenarios. The examiner used a conversational approach to encourage natural speech. While both questions and responses were transcribed, analysis focused solely on participants’ speech. No restrictions were placed on text length or word count, ensuring all available speech data was analyzed.

Kato et al. (2022) based their annotation scheme on SFL, which views language as a system of context-driven choices. In resource-selection mapping, speakers select from multiple lexicogrammatical options organized in a system network (Martin, 1992). Kato et al. (2024) developed four Japanese system networks: MOOD (clause types, modality), APPRAISAL (evaluative expressions), TRANSITIVITY (experiential representation), and LOGICAL (clausal relationships). These structured networks enable systematic linguistic analysis.

The annotation scheme, detailed in Supplementary Table S2, categorizes lexicogrammatical choices into 15 major categories with 140 tags, offering a comprehensive framework for analysis.

Kato et al. (2024) formulated two diagnostic models: the Tag Linear Model, based on annotated linguistic tags, and a deep learning-based approach that integrates these tags with textual analysis (Tag DNN, Text DNN, and Text+Tag DNN). Among these, the Text+Tag DNN model demonstrated the highest diagnostic effectiveness, achieving 80% accuracy, 82% precision, 73% sensitivity, and 87% specificity.

For the purpose of this study, the Tag Linear Model is the only one that allows direct identification of the discriminator. To ensure comparability across participants, the input to the Tag Linear Model was calculated by normalizing the frequency of each annotated tag by the total number of words produced by each speaker. This procedure allowed us to analyze proportional use of linguistic resources, independent of individual differences in speech quantity.

While deep learning models such as Tag DNN and Text+Tag DNN can capture nonlinear relationships and complex feature interactions—potentially enhancing predictive accuracy—they do not provide a direct explanation of how specific features influence classification. Even though these models were trained on the same annotated features, it cannot be assumed that they relied on the same linguistic cues; deep neural networks often transform and combine features in abstract ways that obscure their specific contributions.

Even so, the transparent evidence provided by the Tag Linear Model supports the existence of lexicogrammatical differences between ASD and non-ASD groups—offering interpretable linguistic insights, even if its statistical performance is slightly lower. This makes the linear model particularly valuable for research focused on identifying specific linguistic markers.

Due to the small sample size relative to the number of linguistic features, we employed the bootstrap method, which allows for numerical estimation of model parameters.

The bootstrap procedure involved 10,000 iterations, where a random sample was drawn with replacement, and a logistic regression model was trained on each resampled dataset. The coefficients were estimated using numerical optimization techniques. For each coefficient, we computed its mean and variance over the 10,000 iterations. Using these values, we derived the p-values to test the null hypothesis that the corresponding coefficient equals zero. These p-values indicate the statistical significance of each linguistic feature in distinguishing ASD from non-ASD speech.

Rankshifted clauses—including noun clauses, reported clauses, and adnominal clauses—along with LSR/projection-embedding and LSR/projection-idea, are significantly less used by individuals with ASD.

Rankshift in SFL refers to the downranking of a clause or phrase so that it functions within a lower-ranked unit, such as a nominal group (Halliday and Matthiessen, 2014). This process enhances meaning-making (semogenesis) by embedding clauses within other structures, enabling more complex and flexible sentence constructions. SFL organizes language hierarchically into clause, group/phrase, word, and morpheme, with rankshift occurring when a higher-ranked unit functions at a lower rank. Examples of rankshift are:

English

(i) nominalization (clause to noun phrase):

Here, that he is honest is a clause functioning as the object of the verb ‘believes.’ It is a rankshifted clause serving a nominal role.

(ii) relative clauses (clause within noun phrase):

The relative clause that you lent me modifies the noun book, embedding a clause within a noun phrase.

(iii) adjective phrases within noun phrases:

The prepositional phrase with the hat functions as a modifier of man, representing a rankshifted phrase embedded within a noun phrase.

Japanese

(i) nominalization using no and koto:

The clause kare-ga kuru (he is coming) is nominalized using no, allowing it to function as the object of the verb shitte-iru (know). Interlinear glosses follow the Leipzig Glossing Rules. A full list of grammatical abbreviations used in the glosses is provided in Supplementary Table S3.

(ii) relative clauses (pre-nominal modification):

The clause watashi-ga kinō mita (I watched yesterday) modifies the noun, eiga, (movie), embedding the clause within the noun phrase.

(iii) embedded questions:

The clause kare-ga nani-o itta ka (what he said) functions as the object of oboete-inai, (don’t remember), illustrating a rankshift.

Rankshift is a linguistic mechanism where a higher-rank unit functions at a lower rank within another structure, allowing for complex sentence constructions. For example, a clause can function within a noun group, as seen in relative clauses. Both English and Japanese utilize rankshift, though their grammatical structures and markers differ.

While words and morphemes are part of the rank scale, rankshift primarily involves embedding clauses within groups or phrases rather than focusing on smaller linguistic units. Understanding rankshift helps in analyzing sentence complexity and structural flexibility.

Integrating traditional Japanese grammar with SFL provides a comprehensive framework for annotating complex sentences. Traditional Japanese grammar emphasizes syntactic structures, categorizing clauses based on form, such as relative clauses, nominal clauses, and reported clauses. In contrast, SFL focuses on logico-semantic relations (LSR), which describe the logical and semantic relationships between clauses, including elaboration, extension, enhancement, and projection. Combining these approaches allows for an annotation scheme that captures both syntactic structure and semantic function in complex sentences.

One advantage of this hybrid framework is its ability to address overlapping categories. For instance, in the phrase kino yonda hon, (the book I read yesterday), traditional grammar classifies the clause as a modifying clause, focusing on its pre-nominal modification. SFL, however, analyzes it as embedding, where a rankshifted clause functions within a nominal group. By integrating both perspectives, this approach ensures a precise representation of both structural and functional aspects of sentence complexity.

Additionally, this combined framework enhances the classification of projection clauses. Traditional grammar identifies nominal clauses as noun-like structures and quotation clauses as reported speech or thought. SFL’s LSR framework categorizes these under projection, divided into projection: idea (thought representation) and projection: locution (speech representation). For example, the sentence kare-ga kuru-koto-wo shitte-iru, (I know that he is coming) is a nominal clause in traditional grammar but is analyzed as projection: idea in SFL, representing a mental process. Similarly, kare-wa asu iku-to itta, (he said, ‘I will go tomorrow’) is a reported clauses in traditional grammar but falls under projection: locution in SFL due to its reported speech nature. This combined framework ensures a clear distinction between thought and speech projections, enhancing linguistic precision.

Beyond Japanese-specific classifications, integrating SFL facilitates cross-linguistic comparisons. While traditional Japanese grammar focuses on language-specific distinctions, SFL’s functional categories align with universal syntactic and semantic principles, simplifying the comparison of complex sentence structures across languages like Japanese and English.

Practically, this dual-layered annotation scheme offers a flexible and comprehensive analysis. Traditional classifications, such as relative, nominal, and reported clauses, provide clear form-based distinctions, while SFL’s LSR categories, including embedding, projection-idea, and projection-locution, offer function-based classifications. This integration allows researchers to examine Japanese complex sentences from both structural and functional perspectives.

By distinguishing embedded versus ranking elements, projection types, and overlapping categories, this hybrid approach ensures a more accurate and detailed representation of complex syntactic structures. Notably, our annotation scheme incorporates 13 items from the LSR framework, with six identified as discriminators (see Supplementary Table S2). We proceed to analyze specific example sentences to illustrate these concepts.

(i) A reported clause is a rankshifted clause that expresses the content of speech, thought, commands, requests, or wishes. In Japanese, it is typically marked by to for direct and indirect speech and yō-ni for reported thoughts or intentions.

The clause minna-ga watashi-no-koto-wo hen-da (Everyone [was saying] I was weird) is rankshifted and functions as the object of itte-ta (was saying).

(ii) A noun clause is another type of rankshifted structure that functions as a noun within a sentence. It is commonly formed using formal nouns like koto and no, allowing the clause to serve as a subject, object, or complement.

The clause sensei-ga yasashii (The teacher is kind) is rankshifted and treated as a noun phrase within the larger sentence.

(iii) An adnominal clause is a rankshifted clause that functions as a premodifier of a head noun. In SFL, this is analyzed as downranking, where a clause is embedded within a noun phrase, increasing syntactic complexity. The more rankshifted elements a clause contains—such as adnominal clauses—the more structurally intricate it becomes. In Japanese, modifiers follow a regressive pattern, meaning they precede the head noun, whereas in English, relative clauses follow a progressive pattern, appearing after the noun.

In traditional grammar, adnominal clauses correspond to relative clauses, as they add descriptive information to the head noun:

The clause, konomae byōin-de hanashita (who I spoke to at the hospital the other day) modifies sensei (the doctor).

The embedded clause densha-de tonari-ni suwatte-ta (who was sitting next to me on the train) modifies hito (person).

These lexicogrammatical structures serve as one of the key discriminators between ASD and non-ASD individuals. We will now probe the cognitive reasoning behind this distinction.

Discriminators of inferential ability, including modality-probability, conditional clauses, te-form conjunctive clauses-cause/reason, expansion-enhancement-cause-conditional, and evidentiality-appearance, are significantly less used by individuals with ASD, reflecting challenges in inferential reasoning. Two primary factors that hinder inferential ability can be considered: (i) context integration difficulties and (ii) weak central coherence (WCC):

(i) Context integration difficulties

Contextual integration is crucial for inferencing, allowing individuals to connect linguistic elements across discourse and apply background knowledge. While ASD individuals do not entirely lack inferential ability, they often struggle with making contextually appropriate inferences (Bodner et al., 2015). Even when they understand individual linguistic components, they may fail to integrate them into a coherent whole, affecting their processing of conditional, implicit causal, probable, and evidential relationships.

ASD individuals with strong verbal skills can generate inferences but often produce contextually inappropriate responses that do not align with broader discourse structures (Bodner et al., 2015). They may activate background knowledge but struggle to apply it within a given context, leading to fragmented reasoning and misinterpretation of inferential cues (McKenzie et al., 2010).

These difficulties are most evident in discourse requiring cohesive information linking. Typically developed (TD) individuals process cause-effect relationships across sentences, whereas ASD individuals interpret statements in isolation, leading to breakdowns in inferential comprehension, especially when information is implicit.

Since context integration deficits occur at the discourse level, they do not manifest in specific lexicogrammatical structures, such as conditionals or evidential markers. Instead, they affect the ability to integrate meaning across discourse, making the resulting deficits hard to demonstrate through isolated examples.

(ii) WCC and the role of sensory precision

WCC theory posits that individuals with ASD focus on local details rather than global meaning, leading to fragmented linguistic comprehension (Frith, 1989; Happé and Booth, 2008). Instead of forming broader representations of meaning, they process each linguistic unit independently, making it difficult to integrate elements into a coherent whole, particularly in inferential structures requiring contextual linking.

ASD individuals struggle to integrate prior knowledge with real-time sensory input, making it difficult to establish common knowledge (Hohwy and Palmer, 2014). Their model emphasizes that heightened reliance on sensory precision contributes to challenges in processing implicit social cues, affecting inferential discourse. This can also be explained in terms of causal reasoning, where individuals on the spectrum prioritize immediate perceptual details while neglecting broader contextual cues. As a result, this rigidity impairs their ability to infer others’ intentions, align mental states, and establish common knowledge, all of which are essential for understanding inferential discourse.

The following are examples of lexicogrammatical effects:

(i) conditional clauses (cause/reason, resultative condition)

[1] cause/reason conditionals

TD individuals flexibly infer cause-effect relationships, while ASD individuals may struggle when causality relies on pragmatic inference rather than direct linguistic cues. They process clauses separately, focusing on surface-level connections, which makes unstated causal links harder to recognize. This reliance on syntactic clarity creates challenges in processing implied causality.

[2] resultative conditionals

TD individuals integrate background knowledge to interpret causal relationships, while ASD individuals rely on rigid interpretations, making it difficult to apply contextual information flexibly (McKenzie et al., 2010). This difficulty in updating reasoning when presented with new information affects their ability to process hypothetical or resultative conditionals effectively.

Since these reasoning difficulties arise from multiple cognitive factors, several mechanisms explain why ASD individuals struggle with conditional clauses. First, inhibitory control deficits hinder their ability to suppress reality-based knowledge when considering hypothetical alternatives (Beck et al., 2009). For instance, when hearing If I had left earlier, I wouldn’t have been late, TD individuals recognize the implied alternative timeline, whereas ASD individuals may focus solely on the factual outcome of being late.

Second, reduced cognitive flexibility in ASD leads to rigid application of conditional logic, making it difficult to integrate counterexamples or exceptions (McKenzie et al., 2010). This rigidity impairs processing of conditionals requiring perspective shifts, especially in hypothetical reasoning. Contextual integration deficits further compound these challenges, as ASD individuals struggle to adjust conclusions when new information contradicts prior assumptions (Pijnacker et al., 2009).

(ii) te-form/conjunctive clauses-cause/reason

The te-form links multiple clauses, often indicating an inevitable cause-effect relationship. However, ASD individuals may interpret these clauses as separate events rather than a unified causal sequence.

While TD speakers infer inevitable cause-effect relationships, ASD individuals may struggle to establish this link, leading to fragmented interpretation. Even with explicit reason/cause markers such as kara (because), node (since), tame-ni (in order to), sei-de (due to), or okage-de (thanks to), they still face difficulties processing causality. This challenge intensifies when causality is implicit, as in te-form conjunctive-cause/reason. Without explicit causal markers, ASD individuals may fail to infer the reason-result relationship automatically, relying instead on overt cues, further demonstrating their difficulty in integrating unstated causal links.

(iii) modalization-probability

Modalization-probability expresses likelihood, possibility, or certainty. In Japanese, modal markers like darō (probably), kamoshirenai (might), ni-chigainai (must be), and hazuda (should be) indicate different degrees of probability.

Example:

TD speakers infer certainty levels using context and background knowledge, but ASD individuals often process probabilistic expressions rigidly, struggling with: (1) interpreting likelihood; (2) adjusting meaning by context; (3) recognizing implied probability without markers.

For instance, kamoshirenai (might/maybe) conveys uncertainty, yet ASD individuals may take it as absolute or dismiss it as vague. This challenge worsens when probability is implied rather than explicitly stated, leading to misinterpretations.

(iv) evidentiality-appearance

Evidentiality refers to linguistic markers that indicate the source of information, whether directly observed, inferred, or reported (Teruya, 2007). It is categorized into three main types: hearsay, appearance, and reasoning, as shown below. Among these, appearance-based evidentiality was found to be used significantly less frequently.

[1] hearsay (reported evidence): refers to cases where the speaker conveys information obtained from others rather than direct experience. The marker sōda (I hear) is commonly used for this function.

In this sentence, the speaker relays information from a source rather than making a direct observation.

[2] appearance (sensory-based inference): involves judgments based on visual or sensory perception. The speaker makes an inference from observed sensory input, rather than reporting or reasoning from background knowledge.

Here, the heat of the soup is inferred from visible cues such as steam or bubbling liquid rather than from direct contact.

[3] reasoning (cognitive inference): refers to cases where the speaker makes a conclusion based on logical deduction or contextual knowledge rather than sensory perception. This is commonly expressed in Japanese with darō (I guess) or yōda (seems).

The speaker does not directly witness the event but infers it based on expected timing and general knowledge of flight schedules.

Among the three types of evidentiality, the reduced use of appearance-based evidentiality in ASD may stem from atypical sensory perception. Some individuals experience hypersensitivity, making them distrust sensory input and avoid markers like sōda (looks/seems). Others, with reduced sensitivity, may miss relevant cues, hindering sensory-based inferences. These variations in sensory perception likely contribute to the observed patterns in evidentiality use, leading ASD individuals to rely less on appearance-based evidentiality.

Although both probability and reasoning involve uncertainty, they function on fundamentally different principles.

(i) probability as an internal judgment: refers to the speaker’s internal assessment of likelihood, based on their own reasoning or inference. It reflects a subjective evaluation rather than an externally sourced fact.

The phrase hazuda (must) expresses a high degree of certainty but remains an internal assumption based on contextual knowledge rather than external evidence.

(ii) evidentiality as an external source of knowledge: indicates the information source—whether directly observed, inferred from sensory input, or reported by others (Teruya, 2007). Unlike probability, it concerns how the speaker knows something rather than their level of certainty.

The darō (I guess) here functions as both a probability and an evidential marker, implying that the speaker is making an informed guess, likely based on general knowledge rather than direct observation.

In SFL, Kadooka et al. (2016) categorizes evidentiality as a subtype of modality, whereas Teruya (2007) treats it as an independent system related to propositional validity, which aligns with Kamio (1990), who argues that absolute evidence is unattainable in conversation, meaning that all statements rely on some form of inference—whether subjective (probability) or based on external sources (evidentiality).

ASD individuals struggle with inferential language due to deficits in common knowledge, social inferencing, sensory processing, and limited exposure to inferential discourse. These challenges affect both comprehension and lexicogrammatical choices.

The discussion so far has focused on input processing, but in ASD, deficits in inferential comprehension also impact language output. Difficulty interpreting implicit meanings and complex relationships leads to a preference for simpler, more explicit structures. Lacking a strong grasp of syntactic complexity and inferential cues, ASD speakers rely on straightforward constructions. In essence, what one cannot infer, one cannot effectively produce.

Fauconnier (1994) introduced mental space theory to explain how language constructs cognitive representations beyond direct real-world reference. Instead of assuming that linguistic expressions correspond directly to external reality, mental space theory posits that meaning is structured through mental spaces—temporary cognitive domains that allow individuals to interpret different perspectives, hypothetical scenarios, and belief states. Mental space theory integrates pragmatics (context-based meaning) and semantics (linguistically encoded meaning) to explain how we comprehend language beyond surface-level interpretation.

A classic example is metonymy, where an author’s name represents their literary works:

While metonymy relies on direct associations, metaphor processing—a well-documented challenge for individuals with ASD (Happé, 1993; Norbury, 2005; Kalandadze et al., 2021)—requires more complex conceptual mapping across mental spaces. Understanding metaphors involves linking distinct conceptual domains, allowing individuals to interpret abstract meanings beyond literal word definitions. For example, interpreting Knowledge is light—a metaphor that maps illumination onto understanding—requires mental space construction.

Mental space theory further explains belief spaces, demonstrating how mental spaces allow us to process different perspectives:

Here, believes triggers a shift from the speaker’s own reality to Len’s subjective belief, creating a mental space distinct from objective reality (Figure 3). This showcases how conceptual mappings guide meaning interpretation.

In discourse, mental space builders (e.g., if A, then B; Max believes X; in Len’s mind) establish and structure mental spaces, facilitating shifts between different perspectives and realities. These linguistic elements are essential for processing counterfactual reasoning, hypothetical scenarios, and belief attribution.

This study found that individuals with ASD exhibit significantly less frequent use of mental space builders, suggesting difficulty in constructing and managing mental spaces. This may contribute to challenges in understanding hypothetical reasoning, interpreting figurative language, and shifting between different perspectives in discourse (Figure 4).

Individuals with ASD, who tend to process language literally, may struggle to establish such conceptual mappings, leading to difficulty in understanding non-literal expressions. This study identifies the following discriminators in ASD and non-ASD:

(i) projection-idea; represents subjective thoughts and beliefs, often marked by to omou (I think) in Japanese. This functions as an mental space builder by shifting discourse from an objective statement to the speaker’s internal cognition or belief state.

Here, to-omou (I think) creates a mental space distinct from objective reality, framing the content as the speaker’s subjective belief rather than an asserted fact.

(ii) Modalization-probability; expresses uncertainty, possibility, or hypothetical reasoning, often using kamoshirenai (might) in Japanese. As an mental space builder, it shifts the meaning into a hypothetical space, marking the statement as non-definitive.

Here, kamoshirenai (may/might) creates a mental space of uncertainty, where the event is not confirmed but considered possible. ASD individuals, who often prefer explicit and concrete statements, may struggle with this probabilistic reasoning.

(iii) Conditional clauses; establish hierarchical relationships between clauses, structuring causal, adversative, or hypothetical relationships. In mental space theory, these mental space builders introduce dependency between mental spaces, signaling how one event relates to another.

[1] Cause/reason conditional

[2] Resultative Conditional (Causal Relationship: If X Happens, Y Follows)

Here, to (if) builds a conditional mental space, where the speaker’s emotional response (I am happy) is dependent on the condition (everybody laughs).

The reduced ability to construct mental spaces in ASD may explain challenges in metaphor comprehension, pointing to the role of mental space builders in conceptual thought and discourse navigation. In sum, mental space theory provides a framework for understanding how language constructs meaning beyond direct reference. Reduced mental space builder use in ASD suggests difficulties in mental space construction, contributing to challenges in hypothetical reasoning, belief attribution, and metaphor comprehension. To explain this, we adopt the decoupling perspective.

Decoupling is the cognitive ability to separate immediate perceptions from alternative representations, supporting imagination, flexible thinking, and abstract reasoning. Leslie (1987) introduced this concept in relation to pretend play, where children distinguish reality from pretense. For example, when a child pretends a banana is a telephone, they must decouple its real function as food from its imagined use, avoiding confusion.

Decoupling is a key mechanism of meta-representation—the ability to form representations of representations (Leslie, 1987). His model views pretend play as an early example of this ability, involving three processes: the Expression Raiser, which copies and disconnects a primary representation from reality; the Manipulator, which integrates the decoupled representation into a pretend scenario; and the Interpreter, which reconnects it to real-world interactions (Leslie and Roth, 1993).

Difficulties in decoupling have been linked to ASD, where individuals struggle with pretend play, flexible thinking, and understanding others’ beliefs, thoughts, and emotions (Leslie and Roth, 1993; Kinoshita, 1992; Frith, 1989). Reduced spontaneous pretend play in ASD suggests decoupling deficits, contributing to social and communication challenges (Wing and Gould, 1979). Beyond pretend play, decoupling is crucial for understanding non-literal language, including metaphors, irony, and figurative expressions. Impairments in this area lead to overly literal interpretations and communication difficulties.

While Leslie’s model has been highly influential, Perner (1988, 1991) argued that pretend play relies on a simpler as if representation rather than meta-representation, allowing children to engage in pretense without understanding mental states. Other researchers (Wellman, 1990; Astington, 1996) challenge Leslie’s emphasis on innate cognitive mechanisms, instead emphasizing the role of environmental and cultural factors in cognitive development.

Despite these debates, decoupling remains central to cognitive flexibility, imagination, and social cognition. Its impairment in ASD helps explain difficulties in communication, perspective-taking, and abstract thinking.

Even in ASD individuals with strong linguistic skills, difficulties in symbolic reasoning and communication persist (Leslie and Roth, 1993). The significantly lower use of mental space builders in ASD suggests cognitive differences in:

While ASD individuals retain some decoupling ability, its efficacy varies across the spectrum. Impairments in meta-representational abilities result from a malfunction in the decoupling mechanism (Leslie and Roth, 1993). This dysfunction is an innate cognitive impairment rather than a general developmental issue (Leslie, 1991).

This study, focusing on individuals aged 14 and older, does not address whether cognitive abilities are innate or acquired but examines how decoupling dysfunction affects linguistic expression. The significantly lower use of mental space builders in the ASD group suggests distinct cognitive traits, indicating difficulties in distinguishing reality from mental representations—key to constructing mental spaces, similar to pretend play (Kato, 2025, in press).

A broader theoretical issue arises: if decoupling deficits explain the reduced use of mental space builders, does this suggest mental space construction is not an independent cognitive function but a secondary effect of decoupling ability? This question leads to a fundamental issue regarding mental space theory’s cognitive status and its relationship with decoupling (Kato, 2025, in press).

Takubo (1992) views mental space construction as a cognitive construct that actively organizes semantic and pragmatic information in real-time processing. Under this view, mental spaces function as cognitive tools for counterfactual reasoning, belief attribution, and conceptual organization, suggesting mental space construction is an intrinsic cognitive mechanism. In contrast, Sakai (2014) argues that mental space construction is not an independent cognitive function but a symbolic system modeling how language encodes conceptual structures. Mental space construction, in this view, does not govern real-time cognition but serves as a linguistic mapping tool within discourse. Thus, while Takubo sees mental space construction as a core cognitive mechanism, Sakai considers it a framework shaped by other cognitive abilities.

This study aligns with Sakai’s perspective, suggesting that ASD individuals’ reduced use of mental space builders stems from decoupling deficits rather than mental space construction impairments. Decoupling, as described by Leslie (1987), allows individuals to separate real-world knowledge from hypothetical scenarios, a crucial ability for constructing mental spaces (Kato, 2025, in press). However, Fauconnier (1994) did not explore mental space theory’s cognitive basis, leaving open whether mental space construction relies on specialized neural mechanisms or broader cognitive functions. If mental space construction is only a symbolic mapping system, its role in language comprehension must be reconsidered in relation to foundational cognitive abilities.

Future research must clarify whether mental space construction is: (i) an independent cognitive construct or (ii) a system shaped by other cognitive mechanisms. Resolving this distinction has significant implications for understanding language processing differences between TD and atypical populations.

Negotiating particles, found to be used significantly less by ASD individuals, serve as a discriminator in JA. Negotiating particles are essential in Japanese interpersonal communication. Particles like ne and yo appear at clause endings to facilitate interaction. For example, …shimasu (do) becomes an interpersonal statement with ne (…shimasu-ne (yo)), engaging the listener. English lacks direct equivalents, instead relying on intonation, modal verbs, or discourse markers (right?, you know, isn’t it?). However, these elements do not function as fixed grammatical particles like their Japanese counterparts.

These particles are lexicogrammatical elements that convey the speaker’s attitudinal stance toward a proposition or proposal (Teruya, 2007). They regulate turn-taking, attention management, and stance negotiation in discourse. In particular, ne and yo frequently serve interpersonal functions. Ne is associated with confirmation and shared cognition, fostering social alignment. Yo conveys insistence, assertion, or emphasis, marking statements as informative or directive (Nihon-kijutsubunpo-kenkyukai, 2008). Yone combines yo and ne, indicating that the speaker believes their cognition is acceptable to the hearer. Kane blends the interrogative ka with ne, introducing a questioning function while softening inquiries.

Maynard (1997) analyzed 60-min daily conversations among 20 pairs of subjects and found that sentence-ending particles accounted for 35% of all sentence-final expressions, with ne and yo being the most frequently used. In Japanese socio-cultural contexts, these particles reflect the speaker’s intentions regarding interpersonal distance and relationship-building with the interlocutor.

(i) Ownership of information and negotiating particle selection

The use of ne, yo, and yone is influenced by how the speaker perceives the ownership of information. A reduced use of these particles in spoken Japanese among ASD individuals is considered an indicator of pragmatic impairment, as these structures are essential for engagement and shared cognition. The selection of negotiating particles depends on the relative degree of information ownership between the speaker (S) and hearer (H) (Table 2).

The information ownership framework demonstrates how these particles mediate epistemic relationships, signaling the speaker’s assumption about what the listener already knows. Yo is used when the speaker possesses more information than the listener, but it may be softened with yone when the hearer is expected to confirm the statement (Maynard, 1993). Yo often signals new information (Kitagawa, 1984), draws the hearer’s attention, or prompts action (Shirakawa, 1992; Izuhara, 2001). Compared to ne, which seeks confirmation and attunement, yo carries a stronger directive force and asserts the speaker’s viewpoint more forcefully (Imamura, 2011; Tokieda, 1951). While both particles regulate interpersonal interaction, yo has a stronger attention-calling function (Izuhara, 2001).

(ii) Pragmatic functions of ne from a calling-attention perspective

The particle ne plays multiple roles in managing attention and interaction in discourse (Maynard, 1997) and also serves politeness-related functions (Usami, 1997; Table 3).

In spoken Japanese, ne and yo are crucial for managing attention and interpersonal alignment. Their role in drawing the hearer’s focus and fostering joint understanding suggests that reduced use among ASD individuals reflects pragmatic impairment. This study links their diminished use to deficits in JA, a key ability in social cognition and communication development.

The cognitive function of ne in calling for attention directly connects it to JA, a key mechanism in early social-cognitive development (Mundy et al., 1990). ASD individuals frequently exhibit difficulties in gaze-following, pointing, and object-sharing, which are essential for coordinating social interactions (Leekam, 2005). Since JA is linked to social motivation (Mundy and Sigman, 2006), children who are more socially engaged tend to develop stronger communication skills. However, ASD individuals often lack social motivation, process emotions differently, and exhibit reduced orientation to faces (Dawson et al., 2005; McPartland et al., 2011), limiting their opportunities for language socialization. JA consists of two components; Responding to JA and Initiating JA. Responding to JA and initiating JA are supported by distinct but interrelated neural networks, with initiating JA being more closely tied to social motivation (Schilbach et al., 2010; Schietecatte et al., 2012). Since ne and yo serve as verbal markers of JA, their reduced use in ASD individuals suggests weak social motivation and impaired initiating JA.

In sum, ne, yo, yone, and kane serve as key markers of social alignment, information ownership, and attention management. Their reduced use in ASD individuals reflects pragmatic impairment, weak social motivation, and JA deficits. As these particles facilitate interpersonal engagement, their absence suggests broader communication challenges, reflecting cognitive-linguistic difficulties in ASD social interaction (Kato et al., 2022).

ASD individuals use the existential process significantly more, making it a discriminator in self-other differentiation and agency awareness. In SFL, the process system refers to how language represents different types of experiences through the transitivity system. The clause serves as the primary unit for expressing these experiences, and the process system organizes human experiences into six distinct Process types (Table 4; see Supplementary Figure S1).

The transitivity system in SFL structures human experiences into linguistic categories, allowing language to not only describe reality but also construct it. By distinguishing between different types of processes, language provides a systematic way of organizing how we perceive actions, cognition, relationships, behaviors, speech, and existence.

This study reveals significant differences in existential process use between ASD and non-ASD individuals, with ASD individuals using existential processes more frequently. Existential processes indicate existence or presence without implying action or emotion, distinguishing between animate and inanimate subjects: iru (be) for animate entities and aru (be) for inanimate ones.

In Japanese, iru (animate) and aru (inanimate) express both existence and possession, unlike English, which separates BE-type (existence) and HAVE-type (possession) verbs (Takezawa, 2003). These verbs indicate: (1) inherent ownership; (2) human relationships (e.g., kinship); (3) part-whole relationships (e.g., body parts); (4) spatial relationships (Chappell and McGregor, 1996). While (1)-(3) reflect possession (X has Y), (4) conveys existence (Y is at X), aligning with transitive (possession) and intransitive (existence) structures (Kishimoto, 2003). English have can imply existence (We have much snow), showing overlap between HAVE and BE verbs (Yamaguchi, 2003), but in Japanese, aru and iru are more natural for existence.

Given this distinction, aru and iru play a central role in existential processes, serving as core BE-type verbs that express existence (exist, happen, arise) or circumstance (sit, stand, grow), with 74% of existential process expressions in this study involving aru and iru.

ASD individuals showed significantly higher existential process use, especially for spatial relationships. This section examines differences in existential process use between ASD and non-ASD groups, analyzing them in terms of self-other awareness, JA, and sense of agency.

Neisser (1988) identified 5 types of self-knowledge, with this study focusing on two: (1) ecological self—the most basic self-awareness, emerging from sensory experiences and physical interactions. (2) interpersonal self—self-awareness formed through social interactions and emotional exchanges.

The ecological self reflects how individuals perceive themselves via direct environmental engagement, like navigating a dark room using tactile and auditory sensations. In contrast, the interpersonal self emerges through early social interactions, such as infant-caregiver reciprocity, which forms the basis for understanding social relationships (Honda, 2005).

Infants initially form binary relationships, focusing on either their environment (e.g., object attachment) or people (e.g., bonding with caregivers). By 9–12 months, these relationships merge into a ternary relation, enabling infants to perceive objects within shared social contexts. This marks the emergence of JA, a key milestone in social and cognitive development (Mundy et al., 1990).

The significantly higher frequency of existential process constructions in the ASD group, especially those involving inanimate subjects (e.g. hon-ga aru, ‘there is a book’; isu-ga aru, ‘there is a chair’), reflects not just a stylistic preference but a fundamentally different mode of world construal. Existential processes serve to assert the presence of entities or conditions without specifying agentive or experiential participation. In systemic functional terms, this process type encodes existence rather than action, perception, or mental engagement. The prominence of existential processes in the ASD group suggests a representational stance centered on environmental registration rather than interactional positioning. Unlike mental or material processes that presuppose relational dynamics between participants (e.g., Actor and Goal, Senser and Phenomenon), existential processes foreground isolated states of being. This aligns with the idea that individuals with ASD may remain anchored in what Neisser (1988) terms the ecological self—defined by one’s perception of the immediate environment as opposed to intersubjective awareness. Previous studies suggest this ecological orientation persists in ASD beyond infancy, with delays or deficits in the emergence of the interpersonal self (Honda, 2005). The ASD group’s reliance on existential processes reflects this persistence in linguistic form: a grammar of observation rather than engagement. Although Japanese allows for subject omission and implicit agency (Ikegami, 2006), the distributional skew observed in our corpus suggests a cognitive-linguistic divergence not explained by typological norms alone.

The transition from ecological to interpersonal self-awareness is typically mediated by the emergence of JA, which allows individuals to coordinate attention with others toward shared referents. This ability supports the development of shared intentional frameworks and the recognition of others as intentional agents. Research has consistently found that JA is impaired in ASD (Mundy et al., 1990; Leekam, 2005), with downstream effects on social cognition and language development. These impairments appear in reduced frequency and complexity of gaze-following, pointing, and protodeclarative gestures—behaviors that scaffold shared meaning. In linguistic terms, the ability to construe another as a co-present participant underlies the selection of animate-subject existential processes (e.g., neko-ga iru “There is a cat”). These constructions index not only presence but co-awareness—an acknowledgment of an animate being as a potential object of shared attention. The ASD group’s low usage of such constructions suggests a diminished capacity or motivation to encode this interpersonal alignment. As with JA behaviors, the grammatical data reflect the absence of a ternary relation (self–other–object) and a retreat into simpler dyadic structures (self–object). This interpretation is consistent with neurocognitive findings. Chiu et al. (2008) found reduced activation in the middle cingulate cortex when individuals with ASD were asked to attribute actions to themselves, indicating a deficit in self-monitoring during interaction. Lombardo et al. (2009) similarly reported reduced functional connectivity between the ventromedial prefrontal cortex and somatosensory cortices—regions involved in self-other mapping. existential process usage, then, offers a linguistic marker of the developmental plateauing of the interpersonal self.

The linguistic patterns observed in our data can also be interpreted in relation to the embodied sense of agency. Agency entails more than motor control; it involves recognizing oneself as the initiator of action in a socially meaningful context. In individuals with ASD, the subjective experience of agency is often less differentiated, particularly in interpersonal or affectively complex situations (Williams and Happé, 2009). This is not merely a behavioral matter but grounded in the neurophysiological organization of self-related cognition. Existential processes provide a linguistic alternative to transitive or volitional constructions. By focusing on what exists or is located, without marking an agent or initiator, existential processes enable speakers to describe the world while avoiding grammatical commitments to causality or volition. This strategy may appeal to speakers who experience uncertainty in attributing or interpreting actions with respect to themselves or others. These tendencies mirror known disruptions in body schema and sensorimotor awareness in ASD. Neuroimaging studies reveal atypical activation in the right temporoparietal junction and posterior parietal cortex—areas essential for spatial integration and proprioception (Devinsky, 2000; Mesulam, 2000; Tsakiris et al., 2008). Farrer et al. (2003) showed these same regions are critical for distinguishing self-generated from externally caused actions. The avoidance of agency in existential process constructions may thus reflect broader disruptions in embodied experience. Our data suggest that existential processes function not only as existential statements but also as compensatory forms that reduce the representational burden of interpersonal causality (Kato, 2024b).

Existential processes must also be considered within the broader context of self-referential language development. Atypical pronoun use—including pronoun reversal and third-person self-reference—is well documented in ASD (Kanner, 1943; Hobson, 1990; Tager-Flusberg, 1996), and reflects difficulties in adopting the deictic perspective needed to anchor the self in discourse (Kircher and David, 2003). These issues extend into the domain of inner speech. Individuals with ASD often rely less on internal verbalization and more on visual-spatial or motoric representations (Hurlburt et al., 1994; Whitehouse et al., 2006). Kennedy et al. (2006) found they report fewer autobiographical daydreams involving self or others, reinforcing the hypothesis that internal narrative frameworks—crucial for perspective-taking—are underdeveloped or qualitatively distinct. Kana et al. (2006) also report reduced connectivity between language and visual areas in ASD, consistent with less integrated verbal processing. existential processes, in this light, represent a linguistic manifestation of this cognitive profile. They allow reference without explicit speaker positioning. While consistent with Japanese norms of ellipsis and subject suppression (Ikegami, 2006), the frequency and distribution of existential processes in ASD speakers go beyond stylistic choice. As in Mach’s (1971) self-portrait—where the visual field is intact but the observer is absent— existential processes grammatically preserve the structure of perception while minimizing perspectival anchoring. They encode a mode of cognition that favors stability and perceptual focus over interpersonal dynamism.

In sum, ASD cognition is marked by a strong reliance on ecological self-awareness, with reduced integration of social and relational perspectives. This is linguistically reflected in a preference for existential processes, which emphasize existence over interaction (Kato, 2024b).

ASD, as defined in the DSM-5 (American Psychiatric Association, 2013), is characterized by RRBs, which appear in both motor and linguistic patterns. These include motor stereotypies (e.g., hand-flapping, rocking), repetitive object use (e.g., lining up toys), and linguistic repetition (e.g., echolalia, rigid phrase repetition). In sentence structure, ASD individuals often list multiple examples in fixed syntactic patterns, reinforcing speech predictability and reducing cognitive effort. Parallel clauses function as a linguistic form of RRBs, promoting structured repetition. Instead of integrating ideas into a single cohesive thought, ASD individuals list actions sequentially, reflecting behavioral rigidity, where repetition provides structure and minimizes the need for linguistic flexibility. The following compares ASD and TD speakers:

While the TD speaker generalizes, the ASD speaker lists experiences explicitly, avoiding abstraction and reinforcing repetitive linguistic patterns. This structured listing reflects perseveration, where monotropic attention makes shifting focus or reframing information difficult (Murray et al., 2005). It may also serve as a cognitive strategy to manage attention, helping ASD individuals maintain discourse fluency without frequent reorganization.

WCC theory (Frith, 1989; Happé and Frith, 2006) suggests that individuals with ASD exhibit a strong preference for local detail processing rather than global meaning integration. Unlike TD individuals who synthesize multiple pieces of information into a broader, cohesive whole, ASD individuals focus on each component individually, treating them as discrete and independent entities rather than interrelated concepts.

This cognitive bias is directly reflected in language processing. While TD speakers often compress multiple ideas into a single summarized statement, ASD individuals list each detail explicitly, avoiding higher-level abstraction. ex., in describing weekend activities:

The TD speaker abstracts experiences into a single category (e.g., relaxing), while the ASD speaker lists events individually, reflecting difficulty in synthesizing details into a broader theme. ASD speech prioritizes enumeration over integration, reinforcing structured repetition.

ASD individuals also struggle with implicit meaning, leading them to explicitly state details rather than rely on listener inference. This lack of inferential processing supports their use of parallel clauses, allowing sequential presentation of details without requiring abstraction or interpretation.

Executive function deficits, particularly in cognitive flexibility and higher-order reasoning, are frequently observed in ASD (Lopez et al., 2005). These deficits contribute to rigid speech structures, leading ASD individuals to rely on familiar syntactic patterns, such as parallel clauses, which provide a stable linguistic framework.

(i) Cognitive rigidity and difficulty shifting discourse

ASD individuals struggle with shifting discourse patterns, leading to perseverative speech. Instead of summarizing efficiently, they list details repetitively within a fixed syntactic structure (e.g., tari, toka, ka constructions). Once established, they persist in using these patterns rather than adapting to new contexts, reinforcing rigid and repetitive speech. ex.: describing their interests:

The ASD speaker’s reliance on enumeration over categorization reinforces perseverative speech and reflects rigid cognitive processing.

(ii) Inhibitory control deficits and overproduction of examples

ASD individuals struggle to suppress unnecessary details, leading to verbose, repetitive speech even after conveying the main idea. ex., when asked about favorite foods:

The ASD speaker’s continued listing reflects inhibition difficulties, making it hard to stop providing examples. This reinforces their reliance on parallel clauses as a structured, repetitive linguistic tool.

(iii) Predictive processing deficits and speech patterns in ASD

From the predictive processing theory (Van de Cruys et al., 2014) perspective, ASD individuals struggle with anticipating conversational information. This study suggests that prediction difficulties lead to overly explicit, detail-focused speech. Instead of filtering redundancy, they rely on bottom-up processing, making speech excessively precise or fragmented. Reduced sensitivity to implicit meaning and conversational cues further reinforces a rigid, sequential discourse style.

Rather than anticipating the listener’s inference, the ASD speaker lists each detail explicitly, favoring enumeration over summarization.

In summary, ASD speech patterns reflect RRBs, marked by structured listing, perseveration, and reduced syntactic flexibility. Their reliance on parallel clauses, excessive examples, and difficulty shifting discourse indicates cognitive rigidity, inhibitory control deficits, and predictive processing challenges. Instead of integrating information fluidly, they prioritize explicit enumeration, resulting in overly detailed, fragmented, and inflexible speech.