Identification of Homonyms in Japanese Verb Phrases Using Syntactic and Semantic Attributes of Verbs

Abstract

Identifying homonyms is considered to be an important part of language understanding process. In Japanese, many verbs have homonyms. Some of auxiliary verbs and connective expressions also have multiple meanings. Therefore, Japanese understanding system must have the ability to select one correct meaning among them. It seem that top-down analysis based on contextual information such as "script" is indispensable to complete identification of homonyms in sentences. However, homonyms can often be identified within scope of a verb phrase when a verb is used with auxiliary verbs or connective expressions. For example, verb "SAKERU" has two meanings "avoid" and "split (intransitive) in two". When this verb is used with auxiliary verb "TAI (want)" like "SAKE-TAI" the meaning of "SAKERU" is "avoid". Because "TAI" requires volitional verbs. "SAKERU" as "avoid" is volitional, but "split in two" is not volitional. This paper discusses a method for identifying homonyms in Japanese verb phrases written in KANA. The method is based on several syntactic and semantic attributes of verbs and connection rules of auxiliary verbs and connective expressions. The attributes discussed here are volition, aspectual primitives, operativity, semantic categories based on the concept of "transfer and existence" and so forth. The connection rules are represented in term of these attributes. A connection rule, which is attached to each auxiliary verb or connective expression, checks whether it can be connected to a verb or a phrase by referring to its attribute values and it updates the values if necessary. The identification process proceeds by activating all rules associated with auxiliary verbs in a phrase in turn. On completion of activation of all the rules, permissible combinations of word meanings remain. This identification process is a part of Japanese understanding system called "MARION" now under development.