How Parsing Operator Chunking predicts Reading Times in Sentence Processing

Abstract

Elaborating on earlier theories in the ?field of sentence processing and parsing strategies, Hale (2014) proposed that the Chunking Theory of Learning (CTL) might be considered a good potential for relating a concrete mechanism to the sentence complexity metric Surprisal (Hale, 2001), which provides a mathematical specification of the probability of the next word in the sentence. Applying CTL to sentence processing, Hale assumed that parsing operators can be fused together into a quicker executing macro-operator if used more often, resulting in faster parsing for more familiar sentence structures, at the same time reducing surprisal effects. The present study provides an examination of Hale's theory on parsing action chunking, testing its predictions on the Natural Stories Corpus (Futrell et al., 2018). The results show a correlation between cohesion degree of parsing operator trigrams and average reading times, supporting the idea that parsing action tuples can be learned to be a chunk. We will conclude that the presented results are in line with Hale's predictions, and that further research should give insight into possible internal or external effects at play.