Plausibility based comprehension in a neural network model of sentence processing

Anonymous

Keywords: Event processing, plausibility based comprehension, neural networks

Abstract: Psycholinguistic evidence has shown that human language comprehension does not always proceed in accordance with syntactic rules. Instead, these rules can be overridden by semantic plausibility, challenging classic linguistic theories and models. Here we show that the phenomenon of plausibility based comprehension naturally emerges in the comprehension performance of the Sentence Gestalt model, a neural network model trained on mapping sentences to event description based on a large scale corpus without any explicit syntactic training.