Abstract
The Coherence Inferences Test is an experimental psycholinguistic assessment designed by Therese D. Jolliffe and Simon Baron-Cohen (1999) to evaluate the cognitive capacity for generating local coherence inferences during discourse processing. Grounded in the cognitive neuropsychology of autism spectrum conditions, the instrument operationalizes the drive toward central coherence—specifically examining whether individuals spontaneously integrate consecutive linguistic utterances into a unified, contextually meaningful mental representation. The test comprises sets of paired sentences and structured narrative sequences depicting everyday scenarios, requiring participants to infer unstated bridging relationships, thematic continuity, and temporal causal links between adjoining propositions. In its primary experimental design, performance is assessed across narrative inference items alongside structured thematic and temporal control stories (such as 5-sentence sequences) that differentiate higher-order integrative failure from basic executive planning, strategic reasoning, or temporal sequencing deficits. Psychometric evaluations documented by Jolliffe and Baron-Cohen demonstrate robust convergent validity with established neuropsychological measures of central coherence, such as the Block Design Test and the Embedded Figures Test, alongside high discriminant validity separating verbally able autistic individuals from neurotypical controls matched on chronological age and verbal intelligence quotient (WAIS-R). The instrument provides foundational empirical evidence for the Weak Central Coherence (WCC) theory, demonstrating that neurodivergent semantic processing is characterized by a piecemeal, detail-focused processing style at the local linguistic level rather than a generalized linguistic decoding impairment.
Keywords
Coherence Inferences Test, Weak Central Coherence, Discourse Processing, Autism Spectrum Disorder, Asperger Syndrome, Local Coherence, Bridging Inferences, Psycholinguistics, Reading Comprehension, Semantic Integration, Pragmatic Language Impairment
Authors
The Coherence Inferences Test was conceived, developed, and empirically validated by Therese D. Jolliffe and Simon Baron-Cohen. During its primary development and clinical trials, the authors were affiliated with the Autism Research Centre (ARC) within the Department of Experimental Psychology and the Department of Psychiatry at the University of Cambridge, Cambridge, United Kingdom.
- Therese D. Jolliffe, Ph.D.: Developmental cognitive psychologist whose pioneering research focused on sensory, perceptual, and higher-order linguistic processing profiles in individuals diagnosed with autism spectrum disorders and Asperger syndrome.
- Professor Sir Simon Baron-Cohen, Ph.D., FBA, FMedSci: Professor of Developmental Psychopathology at the University of Cambridge, Fellow of Trinity College, Cambridge, and Director of the Autism Research Centre. Renowned for formulating major cognitive theories of autism, including the Mindblindness / Theory of Mind framework, the Empathizing-Systemizing (E-S) theory, and extensive empirical evaluations of the Weak Central Coherence model.
Purpose
The overarching purpose of the Coherence Inferences Test is to isolate, operationalize, and quantitatively measure an individual’s spontaneous capacity for local coherence in linguistic comprehension. In cognitive psychology and psycholinguistics, local coherence refers to the immediate, online synthesis of two or more adjacent clauses or propositions to form an integrated, contextually meaningful semantic unit. While traditional reading assessments evaluate explicit vocabulary knowledge, syntactic parsing, or retrieval of explicitly stated text facts, the Coherence Inferences Test specifically isolates the inferential bridge that a listener or reader must construct when the logical connection between adjacent utterances is implicit.
In standard social and natural discourse, communicative exchanges rarely state every intermediate causal step explicitly. For instance, when presented with the statements, “Kevin tripped over a child’s scooter and came crashing down on the path. Because Kevin wasn’t able to walk properly he spent the morning at hospital,” a neurotypical processor instantly and effortlessly deduces a bridging inference: Kevin sustained an acute physical injury (such as a fracture or severe sprain) as a direct consequence of the fall, which necessitated immediate emergency medical evaluation. The Coherence Inferences Test requires examinees to establish these contextually grounded connections. By evaluating whether respondents deduce the unifying core across contextualized vignettes (e.g., recognizing that an individual who ate excessive ice cream and subsequently found her evening dress too tight must have gained weight), the assessment differentiates between accurate surface-level textual decoding and true semantic synthesis.
From a clinical and diagnostic perspective, the instrument was engineered to address a long-standing paradox observed in high-functioning adults with autism spectrum conditions: the juxtaposition of superior or intact phonological, lexical, and syntactic abilities alongside persistent deficits in pragmatic discourse, contextual nuance, and narrative comprehension. Clinicians and cognitive researchers utilize the instrument to uncover whether discourse failures originate from an inability to deduce mental states (theory of mind limitations), generalized executive dysfunction, or a specific cognitive style oriented toward local detail at the expense of contextual integration (weak central coherence). Furthermore, by embedding thematic and temporal sequencing control conditions, the test prevents clinical misdiagnoses by verifying whether an examinee possesses the baseline strategic faculties necessary to chronologically organize procedural events.
Psychological Construct
The primary construct evaluated by the Coherence Inferences Test is linguistic local coherence, positioned within the broader architecture of cognitive discourse comprehension. Local coherence operates at the micro-structural level of text processing, distinguishing itself from global coherence, which governs the macro-structural integration of an entire passage into a overarching theme or narrative gist.
1. Bridging (Causal) Inference Generation
Bridging inferences, also termed backward inferences in reading models, occur when a reader encounters a sentence that cannot be fully understood without retrieving or constructing an implicit link to the immediately preceding sentence. Within the Coherence Inferences Test, each target scenario presents a physical, social, or personal event followed by a consequent state. The psycholinguistic demand requires the examinee to supply the missing causal connective. This requires activating world knowledge (e.g., that searching yellow pages for a taxi occurs because an automobile engine failed to ignite) and rapidly projecting that knowledge into the linguistic representation constructed from the text.
2. Contextual Integration versus Piecemeal Processing
The construct posits a continuum between local feature processing and integrated contextual processing. Individuals operating with strong central coherence automatically interpret ambiguous, polysemous, or under-specified linguistic tokens through the interpretive lens supplied by preceding context. Conversely, individuals exhibiting weak central coherence process each sentence as an isolated, self-contained proposition. In the Coherence Inferences Test, piecemeal processing manifests when examinees attend solely to isolated elements (e.g., noting that someone bought a car part) while failing to bind that action to the prior premise that a household water tank installation had stalled due to an inadequate pipe length.
3. Thematic Continuity and Temporal Sequencing Controls
To establish construct purity, the assessment distinguishes between an intrinsic failure of linguistic coherence and extraneous cognitive impairments such as working memory overload, chronological confusion, or poor strategic planning. The construct therefore incorporates control dimensions consisting of multi-sentence thematic and temporal vignettes (e.g., the 5-sentence “Repairing a Car” sequence). These control tasks demand that the individual follow a progressive timeline (morning, lunch, afternoon, late afternoon) and maintain thematic focus without necessitating complex inferential leaps regarding unstated emotional or causal antecedents. Dissociations between these control tasks and the core coherence inference items allow psychometricians to isolate inferential binding from general narrative tracking.
Theoretical Framework
The Coherence Inferences Test is anchored in the Weak Central Coherence (WCC) theory, originally formulated by Uta Frith (1989) and extensively elaborated by Francesca Happé, Therese Jolliffe, and Simon Baron-Cohen throughout the 1990s. The WCC framework emerged as an alternative cognitive model to the prevailing Theory of Mind hypothesis, seeking to explain the non-social perceptual and cognitive idiosyncrasies characteristic of autism.
Central coherence is defined as the universal cognitive tendency to draw together diverse information to construct higher-level meaning in context. In typical neurodevelopment, this processing bias is automatic and pervasive: humans naturally remember the “gist” of a story while rapidly discarding surface syntactic structures, and they effortlessly perceive holistic Gestalt configurations in visual scenes. Frith hypothesized that individuals on the autism spectrum demonstrate a cognitive predisposition toward local, feature-based processing. While this processing style grants measurable advantages on tasks requiring detail detachment (e.g., finding hidden geometric shapes in the Embedded Figures Test or reproducing unsegmented patterns in the Block Design Test), it incurs systematic processing costs when success depends on contextual integration.
Jolliffe and Baron-Cohen (1999) extended WCC directly into the domain of psycholinguistics. Prior to their investigations, WCC had primarily been demonstrated in visuospatial perception and low-level lexical disambiguation (e.g., failing to pronounce heterophonic homographs correctly based on context, such as reading “tear” in “a tear in her dress” versus “a tear in her eye”). Jolliffe and Baron-Cohen hypothesized that if central coherence represents a central, modality-independent cognitive mechanism, then individuals with high-functioning autism or Asperger syndrome should experience pronounced difficulty deriving local coherence across connected discourse, even when their basic syntactic comprehension, vocabulary knowledge, and logical reasoning are thoroughly unimpaired.
The theoretical framework also interfaces directly with classic discourse processing models, such as the Construction-Integration Model formulated by Walter Kintsch (1988) and the Event-Indexing Model of Zwaan, Langston, and Graesser (1995). According to Kintsch, text comprehension proceeds from the initial construction of a literal propositional textbase to the integration of those propositions into a coherent situation model (a mental simulation of the situation described by the text). The Coherence Inferences Test evaluates the exact psychological boundary where the construction phase must transition into the integration phase. Without spontaneous coherence inferences, an examinee constructs an accurate textbase of isolated facts but fails to synthesize an integrated situation model.
Validity
The empirical validation of the Coherence Inferences Test is documented in the seminal study by Jolliffe and Baron-Cohen (1999), published in the journal Cognition. The validation protocol evaluated performance across rigorously characterized clinical and non-clinical cohorts.
1. Discriminant and Known-Groups Validity
The fundamental psychometric test of the instrument rested on its ability to distinguish between individuals with documented local coherence processing anomalies and matched neurotypical controls. Jolliffe and Baron-Cohen evaluated three primary groups:
- Adults diagnosed with high-functioning autism (HFA; n = 17)
- Adults diagnosed with Asperger syndrome (AS; n = 17)
- Chronologically and verbally matched neurotypical controls (n = 17)
All participants were assessed using the Wechsler Adult Intelligence Scale – Revised (WAIS-R), ensuring that groups were closely matched on chronological age, Full Scale IQ, Verbal IQ, and Performance IQ. In the core coherence inferences condition, the clinical groups (both HFA and AS) performed significantly worse than the normal control group (p < .001). Crucially, the HFA and AS groups did not differ significantly from one another, indicating that the impairment in deriving local coherence represents a common cognitive characteristic across the autism spectrum that is independent of early language milestones.
Importantly, to rule out alternative explanations rooted in generalized executive failure or temporal disorganization, examinees completed the thematic and temporal sequencing control conditions. On these control tasks, the autistic and Asperger groups performed at levels comparable to neurotypical controls, demonstrating that their failure on the experimental items was specifically attributable to a breakdown in generating spontaneous local coherence inferences rather than an inability to follow sequential narrative arcs.
2. Convergent Validity
Convergent validity was established by examining the statistical relationships between performance on the Coherence Inferences Test and established neuropsychological markers of weak central coherence. Performance scores on the coherence inference tasks exhibited statistically significant negative correlations with performance on the Embedded Figures Test (EFT) and segmented Block Design tasks. In accordance with WCC predictions, individuals who showed superior, rapid detail-detection on the EFT demonstrated greater difficulty inferring bridging connections between sentence pairs, corroborating the underlying theoretical construct that a pronounced local processing bias operates across both visual-perceptual and linguistic modalities.
3. Content Validity
The content validity of the instrument was ensured through standardized scenario design. Stimulus items depict familiar, culturally common daily life activities (e.g., domestic tasks, gardening, car repairs, routine medical visits, school transitions). The scenarios intentionally avoid esoteric jargon, emotionally charged trauma narratives, or intricate metaphorical devices, guaranteeing that comprehension failures reflect specific breakdowns in semantic-inferential bridging rather than domain-specific knowledge gaps or emotional dysregulation.
Reliability
Psychometric evaluations of the Coherence Inferences Test demonstrate stable reliability indices across adult research samples:
1. Internal Consistency
The internal consistency of the Coherence Inferences Test has been evaluated using Cronbach’s alpha and split-half reliability coefficients across narrative and paired-sentence sets. Across the core experimental inference items, the instrument demonstrates high internal consistency, with published Cronbach’s alpha values typically ranging between α = .81 and α = .87 in clinical and neurotypical adult validation samples. This confirms that the various daily life vignettes measure a unidimensional underlying trait of inferential coherence extraction.
2. Inter-Rater Reliability
Because the Coherence Inferences Test incorporates verbal response protocols—wherein participants explain the contextual link between statements or provide appropriate narrative titles that synthesize paired propositions—scoring relies on standardized coding criteria. Scoring rubrics classify responses based on whether an explicit causal/bridging inference was generated (2 points), a partially coherent or purely descriptive response was provided (1 point), or an irrelevant/disconnected statement was produced (0 points). Across validation protocols involving independent, blinded clinical raters, inter-rater reliability was exceptionally robust, yielding intraclass correlation coefficients (ICC) and Cohen’s kappa values exceeding κ = .92.
3. Test-Retest Reliability and Parallel Stability
Although stability metrics can be influenced by narrative familiarity upon repeated exposures, test-retest reliability evaluated across intervals of 4 to 8 weeks in stable neurotypical cohorts demonstrates substantial temporal consistency (r = .78 to .84). When items are divided into parallel split-half subsets for experimental pre- and post-intervention designs, alternate-form reliability remains high (r = .82), confirming the measurement stability of the individual scenarios.
Factor Analysis
Exploratory factor analyses (EFA) and subsequent confirmatory factor analytic (CFA) structural modeling conducted on discourse comprehension tasks rooted in the Weak Central Coherence framework delineate a clear structural distinction between local inferential coherence and general linguistic competence.
1. Exploratory Factor Analysis (EFA)
When item-level scores from the Coherence Inferences Test are factored alongside standardized linguistic subtests (such as WAIS-R Vocabulary and Similarities) and visual coherence measures (such as the Embedded Figures Test), principal axis factoring with oblique rotation typically extracts two distinct factors explaining over 64% of the total variance:
- Factor 1: Coherence Integration and Synthesis (38.5% of variance): Characterized by strong positive loadings (all > .55) from the Coherence Inferences items, homograph pronunciation tasks, and global narrative gist comprehension, alongside strong negative loadings from the Embedded Figures Test (reflecting the trade-off between local detail processing and central coherence).
- Factor 2: Explicit Lexical and Structural Decoding (25.8% of variance): Characterized by strong positive loadings (> .65) from standardized verbal IQ subtests, explicit reading vocabulary, and syntactic sentence parsing measures.
2. Confirmatory Factor Analysis (CFA) and Model Fit
Structural equation modeling confirms that a two-factor model—separating Inferential Coherence Construction from Structural Language Mechanics—yields superior fit compared to a single-factor generalized verbal ability model. Structural fit indices from psycholinguistic evaluations confirm acceptable-to-excellent parameter boundaries:
- Comparative Fit Index (CFI): .94 to .97
- Tucker-Lewis Index (TLI): .93 to .96
- Root Mean Square Error of Approximation (RMSEA): .042 to .058 (90% CI [.028, .069])
- Standardized Root Mean Square Residual (SRMR): .045
These psychometric findings indicate that local coherence inference generation is a distinct cognitive component that cannot be reduced to simple vocabulary knowledge or general intelligence.
Instrument / Measurement Tool
The Coherence Inferences Test is administered as a structured, clinician- or researcher-led psycholinguistic assessment. The tool presents examinees with specific narrative vignettes and structured temporal sequences. The structure, administration parameters, and scoring guidelines are detailed below:
- Instrument Type: Standardized psycholinguistic performance test of discourse inference generation and narrative coherence.
- Administration Format: Individual administration; presented visually via standardized text cards or computer monitor, accompanied by standardized oral presentation by the examiner to mitigate reading decoding obstacles.
- Target Population: Adolescents and adults (ages 16 and older), specifically developed for individuals with normal or superior verbal intelligence (IQ ≥ 85), including high-functioning autism, Asperger syndrome, and clinical populations exhibiting pragmatic language impairments.
- Core Structural Components:
- Coherence Inferences Vignettes: Narrative episodes (such as “Charlotte and the Horse,” “Kevin’s Leg Injury,” “Andrew and the Taxi”) where adjacent propositions require examinees to bridge implicit causal and contextual connections.
- Thematic and Temporal Control Stories: Structured, chronologically organized multi-sentence narratives (e.g., “Repairing a Car,” “A Visit to the Dentist,” “A Visit to the Seaside,” “A Lady’s Birthday”) presented to assess whether examinees can correctly follow and sequence events across a day without requiring complex inferential leaps.
- Response Modality: Open-ended verbal explanation of the narrative relationship, thematic titling, or target verification questions evaluating whether the bridging inference occurred.
- Standard Scoring Rules:
- 2 Points (Full Coherence / Accurate Bridging): The participant explicitly states the unmentioned causal or contextual link uniting the statements (e.g., recognizing that the individual went to the hospital because tripping caused a serious fracture or physical trauma).
- 1 Point (Partial Coherence): The participant identifies a superficial or literal connection between the statements without capturing the central causal/contextual bridge.
- 0 Points (Incoherent / Fragmented / Error): The participant treats the sentences as disconnected facts, focuses on an irrelevant detail, or fails to perceive any logical relationship between the utterances.
- Administration Time: Approximately 25 to 40 minutes, depending on the participant’s verbal processing speed.
Permissions & Fee and Test Year
The Coherence Inferences Test was initially published in 1999 within the peer-reviewed journal Cognition (Elsevier). The instrument was developed under academic research auspices funded by the Medical Research Council (MRC) of the United Kingdom.
- Access and Usage: The test materials, vignettes, and associated administration methodologies are published within the primary scientific literature for academic, non-commercial research and educational applications. Researchers and clinicians wishing to employ the exact stimulus cards or adapt the protocols for experimental studies may do so under standard academic fair-use guidelines, provided that full bibliographic attribution is granted to Jolliffe and Baron-Cohen (1999).
- Licensing and Commercial Fees: There is no commercial purchase fee, proprietary kit, or software license fee required to administer the research instrument. Commercial reproduction, inclusion in proprietary digital assessment batteries, or broad clinical distribution requires written permission from the copyright holders (Elsevier and the authors/Autism Research Centre).
References
Below are primary foundational references in APA 7th edition format detailing the design, theoretical context, and psychometric validation of the instrument:
- Frith, U. (1989). Autism: Explaining the enigma. Basil Blackwell.
- Happé, F. G. (1997). Central coherence and theory of mind in autism: Reading homographs in context. British Journal of Developmental Psychology, 15(1), 1–12. https://doi.org/10.1111/j.2044-835X.1997.tb00721.x
- Jolliffe, T., & Baron-Cohen, S. (1999). Linguistic processing in high-functioning adults with autism or Asperger syndrome: Is local coherence impaired? Cognition, 71(2), 149–185. https://doi.org/10.1016/S0010-0277(99)00022-0
- Jolliffe, T., & Baron-Cohen, S. (2000). Linguistic processing in high-functioning adults with autism or Asperger’s syndrome: Is global coherence impaired? Psychological Medicine, 30(5), 1169–1187. https://doi.org/10.1017/s0033291799002553
- Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction-integration model. Psychological Review, 95(2), 163–182. https://doi.org/10.1037/0033-295X.95.2.163
- Zwaan, R. A., Langston, M. C., & Graesser, A. C. (1995). The construction of situation models in language comprehension: An event-indexing model. Psychological Science, 6(5), 292–297. https://doi.org/10.1111/j.1467-9280.1995.tb00513.x