Parsing and reading times: Testing Hale’s Theory of Chunking in Human Language Parsing

Abstract

In the field of human language comprehension, Hale (2014) introduced a novel theory, which incorporates the chunking theory of learning into human language parsing. According to the chunking theory of learning, oft-repeated subtasks in cognitive processes are sped up by being combined into a single operation. Given that parsing language is an oft-practiced skill, it would make sense that the chunking theory of learning also applies to language parsing. To test this theory, Hale implements a left-corner parser, which previous research claims suitable for parsing natural language. He then creates a metric called cohesion, the log-likelihood ratio of parser-operations. Hale found a significant effect of cohesion on reading time. Chunks that appear often are read faster than chunks that are rarer. In this thesis, Hale's experiment has been repeated, with some improvements. The found results are similar to Hale's results, providing additional evidence for the relevance of the chunking theory of learning in human language parsing.