Oculex: Taxonomia Oculomotora Orientada A Comportamento Para Suporte Clínico Na Avaliação da Afasia Baseada Em Machine Learning

Clinical assessment of aphasia relies predominantly on accuracy/error measures and expert judgment, which may mask relevant processing differences. Individuals with similar performance can adopt distinct visual and cognitive strategies during linguistic tasks. Eye tracking enables the recording of attentional allocation dynamics between target and distractors, latency to target convergence, and revisit patterns. The Test for Reception of Grammar (TROG-2) and its Brazilian adaptation (TROG2-Br) provide a consolidated psychometric framework for measuring morphosyntactic comprehension across increasing levels of complexity. Despite advances in machine learning (ML) approaches applied to oculomotor signals, the literature remains fragmented, lacking reusable structures that articulate metrics, behavioral interpretation, and linguistic-clinical hypotheses in tasks analogous to the TROG-2. This paper proposes OcuLex, a behavior-oriented oculomotor taxonomy conceived as an intermediate layer between raw eye-tracking signals and ML pipelines aimed at clinical decision support. The proposal is supported by a systematic mapping that revealed the absence of studies simultaneously integrating aphasia, eye tracking, TROG-2, ML, and formal taxonomy frameworks. The solution organizes the domain into three dimensions—processing level, behavioral profile, and linguistic-clinical correlate—augmented by a cross-cutting annotation layer and a minimal ontological core. A preliminary validation based on aggregated statistics from a Brazilian digital TROG-2 dataset with eye tracking indicates a predominance of the Exploratory-Search and Late-Search profiles in the aphasic group. The taxonomy demonstrated utility in translating statistical differences into behaviorally interpretable and clinically meaningful profiles.