Classification of visual and linguistic tasks using eye-movement features

Coco, Moreno I.; Keller, Frank · 2014 · OpenAlex-citations

DOI: 10.1167/14.3.11

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates whether eye-movement patterns can accurately classify the cognitive tasks being performed, addressing a gap left by Greene et al. (2012), who failed to classify visual tasks above chance levels. The authors hypothesize that task classification succeeds when tasks differ substantially in the cognitive domains involved, specifically comparing purely visual tasks with those requiring linguistic processing. The research aims to demonstrate that eye-movement features reflect the distinct goals of visual search, object naming, and scene description, thereby confirming the task-dependent allocation of visual attention. The researchers analyzed eye-tracking data from 74 participants performing three tasks: visual search (counting target objects), object naming (naming five objects), and scene description (generating spoken descriptions). Using photo-realistic indoor scenes, they extracted two sets of features: seven features previously used by Greene et al. (e.g., number of fixations, area covered) and fifteen additional features including initiation time, entropy of attention allocation, and mean saliency. They employed linear mixed-effects models to analyze feature differences across tasks and trained three classifier types—multinomial regression, least-angle regression, and support vector machines (SVM)—using ten-fold cross-validation to predict the task performed. The results demonstrated that all classifiers achieved accuracy well above chance, with SVMs reaching a maximum of 88% accuracy for visual search. Feature importance analysis revealed that initiation time alone was sufficient for above-chance classification (79% accuracy for object naming), ruling out the hypothesis that fixed termination times caused previous failures. The optimal classification performance utilized a set of seven features combining spatial information (e.g., entropy of attention) and temporal components (e.g., total fixation duration). Linear modeling confirmed that tasks elicited distinct eye-movement patterns, with search tasks showing different fixation distributions compared to naming and description tasks. These findings confirm that eye-movement responses are strongly modulated by task goals and the specific cognitive processes required, such as language integration. The study extends previous work by showing that accurate task classification is possible when tasks vary in their cognitive demands, rather than just their visual parameters. This supports the view that visual attention is actively controlled by task-specific goals, providing a causal link between eye-movement patterns and the underlying cognitive mechanisms of visual and linguistic processing.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• gaze based attention detection