Surprising Parser Actions and Reading Difficulty

An incremental dependency parser’s probability model is entered as a predictor in a linear mixed-effects model of German readers’ eye-fixation durations. This dependencybased predictor improves a baseline that takes into account word length, n-gram probability, and Cloze predictability that are typically applied in models of human reading. This improvement obtains even when the dependency parser explores a tiny fraction of its search space, as suggested by narrow-beam accounts of human sentence processing such as Garden Path theory.