Association Rule Mining of Anaphora Based on ParCorFull Corpus

Based on the Apriori algorithm, the association rules of the anaphora in ParCorFull corpus are mined, and the similarities and differences between the written and spoken language represented by speech and discussion are compared. The rules of the whole mining show that the most important anaphora types in news, speech and discussion styles are noun phrases and pronouns respectively. In news style, pronoun is easier to realize the function of antecedent, while in speech and discussion style, noun phrase is more commonly used as anaphora. The rules of local mining show that in journalism, verb phrases are generally not omitted, noun phrases are more directly appeared without modification, and pronouns often point to singular person entity objects with the subject. The rules of speech and discussion are more flexible. Subordinate clauses are all used as anaphoric objects in anaphoric without anaphoric functions.