Quantifying parsing complexity as a function of grammar complexity


Recent research has used measures of syntactic complexity generated by computational models to predict human processing effort in behavioral and neurophysiological experiments. These models often rely on simplifying grammatical assumptions and the dominant view in the literature has been that such simplifications are adequate approximations for processing accounts. However, the effect of chosen grammar on processing complexity has not been empirically investigated in a natural text. We compared the impact of grammar on estimated processing complexity for a 12 minute English narrative. One grammar was based on the Penn Treebank 2 schema, and the second was based in the Minimalist Grammar framework and represented as an X-bar schema. The impact of grammar was contrasted against the impact of parsing strategy using two stack-based strategies and complexity was measured in terms of the number of nodes built and the stack depth word-by-word. The impact of grammar choice matched or exceeded the impact of parsing strategy. Thus even for a single, fixed corpus and a fixed parsing strategy, the grammar matters.

Connectedness: Papers by and for Sarah VanWagenen