Adaptive Gamified Interface for Identifying Reading Comprehension Barriers in Autism

Abstract

Adaptive Gamified Interface for Identifying Reading Comprehension Barriers in Autism

Reading comprehension is multifaceted and complex as it needs parallel processing of varied cognitive, lexico-semantic, and pragmatic faculties; it is important as it is intricately needed to acquire educational success, broader socio-economic, neurocognitive, and occupational competencies (Kritsotakis, G., & Morfidi, E. (2025)). When these core processing faculties are compromised, then children are more likely to have educational underachievement, which is common among children who require special attention including those with autism spectrum disorder (ASD) (Davidson, M. M. (2021))  Within this population, reading for understanding is especially challenging stemming from deficits in communication, social-interaction and cognitive processing which current readability indices like Flesch-Kincaid fail to capture. This study aims to identify distinct cognitive boundaries by evaluating how structured multi-modal scaffolding affects comprehension and mental fatigue, while simultaneously documenting how specific linguistic boundaries inhibit reading fluency. To address this unresolved area, we have developed Text Detective, a hierarchically scaffolded assessment tool employing a selection-based response format on a personal computer framework. Introducing the interface as a “game” served as an essential behavioural catalyst, eliciting immediate attentional capture and intrinsic motivation. The system architecture features a three-tiered structure to methodically extract how multi-modal stimuli affect cognitive load while evaluating metrics across four salient linguistic barriers: lexical semantics (literal vs. figurative mapping), pragmatics (including irony and semantic anomalies), discourse processing (using anaphora resolution and temporal & causal sequencing), and affective linguistic & narrative perspective (including behavioural cue decoding, perspective-taking, emotional nuance calibration):

• Level 1 (Visual Scaffolding): Questions were paired with visual hints such as when evaluating figurative language; a sad image accompanies the target idiom (‘feeling blue’) with multiple-choice options (Sad, Happy, Annoyed).
• Level 2 (Auditory Reinforcement): Questions and options were supplemented by a slow, neutral AI-generated TTS audio narration in an everyday Indian English accent.
• Level 3 (Independent Text): Questions were presented as text alone with options, removing external stimuli.

The subjects were selected following an ABLLS–R framework to verify basic decoding skills. The experiment is under active data collection with an initial cohort of seven subjects. To mitigate unfamiliarity and random selection, participants completed a comprehensive interface orientation, were kept under supervision while playing the game, and asked to retake the level if arbitrary option selection was observed. The results were obtained anonymously in the form of a single Excel sheet for each subject and a master Excel sheet containing all the subjects’ compiled data. Using preliminary data, we found that 4 out of 7 subjects performed best on Level 3 countering standard educational assumptions. This strongly suggests that multi-modal compounding offered severe sensory distractions. Conversely, the remaining subjects performed well with the audio reinforcement in Level 2, while none demonstrated peak performance with the visual images in Level 1. Syntactically, it was identified that sentences containing more than 2 to 3 verbs caused cognitive drift making them remember the final clause. Finally, a significant pragmatic discrepancy was observed: while subjects were fully capable of identifying why a statement was ‘funny’ when prompted, six out of seven subjects flatly responded ‘no’ when simply asked if the statement was funny or not. In the future, we aim to build a customized readability index based on more data across all identifiable linguistic feature sets, and finally develop an automated text simplification system using AI tools like ML and DL to ensure classic literature is fully accessible to autistic learners.

References

Kritsotakis, G., & Morfidi, E. (2025).

Davidson, M. M. (2021).