This paper proposes an efficient decision tree algorithm to identify the semantic content in an utterance, represented in a forma J language such as interlingua. Semantic content in utterance (concept) i8 represented as a certain combination of terms predefined. Possible combinations of those terms are also predefined. The proposed method employs a set of decision trees, each of which infer the existence of a certain term in the utterance, while examining the existence of other terms with its referential links on the nodes. Moreover, an algorithm is pro- posed to drive those trees under a dynamic ordering scheme. With this algorithm we obtained the optimal combination of terms for given utterance, for certain predefined combinations. Experimental results showed 11/100 reduction in error rate, compared to the traditional method which employs mutually independent decision trees.
展开▼
