A Phrase Disambiguation Method of "Quanbu V de N" Based on SBERT Model and Syntactic Rule

A method combining syntactic rule and semantic preference in resolving ambiguous structures "(sic)V(sic)N (Quanbu V de N)" is proposed in this paper. First, a rule base is constructed, and the rules of ambiguity division are described in accordance with the syntactic structural constraints and semantic representation function. Secondly, the homomorphic and heterogeneous phrase is converted into two heteromorphic phrases "(sic)V(sic)N (Quanbu dou V de N)" and "(sic)V(sic)N (Quanbu de V de N)". We use the SBERT model to calculate the semantic similarity between the original sentence and the two candidate sentences so that the optimal phrase segmentation is obtained. The experiment was conducted on 646 texts with ambiguous phrases "Quanbu V de N" as the object for testing. The disambiguation accuracy and the F1 value are both over 95%, which proves that this method can effectively disambiguate this phrase in natural language.