Stress distribution based on dependency parsing of Chinese discourse

Abstract

Dependency parsing is known as a syntactic or a shallow semantic analysis in NLP (Natural Language Processing). This paper conducts an interface study between syntax (dependency) and prosody (stress). The stress distribution patterns are statistically analyzed across 24 dependency relations based on HIT (Harbin Institute of Technology) dependency scheme [1] for a Chinese spoken discourse corpus. It shows that the intonation stress is more likely to appear at dependency relations of SBV, ATT, ADV and VOB. Besides, the stress distribution pattern within each relation is analyzed. The rule that the modifier is more likely to be stressed has been proved to some extent. The results demonstrate that there is an intrinsic association between stress distribution and dependency relation.