The Role of Mutual Information and Semantic Similarity in Sentence Processing: The Case of Dangling Topic Construction in Chinese

This study uses computational approaches to clarify the syntactic problems of dangling topic construction in Mandarin Chinese. Previous studies of topic construction in Mandarin Chinese have often syntactically analyzed the relationship between the topic and the rest of the construction. However, syntactic approaches have not clarified dangling topic construction, that is, constructions where the topic seems to dangle. That meant these previous studies had to implicitly make use of other approaches (such as pragmatic knowledge) to advance their arguments. The difficulty is that the concepts from pragmatics that these studies used were very vague and subjective. In order to tackle this problem, this paper explicitly computes the relation in dangling topic construction in Chinese using semantic similarity and pointwise mutual information. We propose three methods for computing the semantic similarity between topic and comment. We also collected experiential data using human ratings of the acceptance degree for a set of dangling topic constructions. The results demonstrate that pointwise mutual information and the three measures of semantic similarity can make good predictions about the data on the human rating of these dangling topic constructions. The event knowledge could be integrated with semantic similarity in order to strengthen the predictability of cognitive processes in understanding sentences. These measures play a similar role to that of semantic plausibility in sentence processing. This is the first time that PMI and sentence-based semantic similarity are employed to predict how humans comprehend sentences. The PMI and sentence similarity measures are likely to shed further light on the notion of topic construction and to help in seeing how Chinese speakers understand and process sentences. More importantly, this study creates a novel, effective and practical computational approach for predicting sentence comprehension/processing and syntactic analysis.