HkAmsters at CMCL 2022 shared task : predicting eye-tracking data from a gradient boosting framework with linguistic features

Abstract

Eye movement data are used in psycholinguis-tic studies to infer information regarding cogni-tive processes during reading. In this paper, wedescribe our proposed method for the SharedTask of Cognitive Modeling and ComputationalLinguistics (CMCL) 2022 - Subtask 1, whichinvolves data from multiple datasets on 6 lan-guages. We compared different regression mod-els using features of the target word and itsprevious word, and target word surprisal as re-gression features. Our final system, using agradient boosting regressor, achieved the low-est mean absolute error (MAE), resulting in thebest system of the competition.