Syntax Vector Learning Using Correspondence for Natural Language Understanding

Natural language understanding (NLU) is a core technique for implementing natural user interfaces. In this study, we propose neural network architecture to learn syntax vector representation by employing the correspondence between texts and syntactic structures. For representing structures of sentences, used three methods: dependency trees, phrase structure part speech tagging. A pretrained transformer text-to-vector syntax-to-vector projection approaches. The are projected onto common space, distances two vectors minimized according property representation. We conducted massive experiments verify effectiveness proposed methodology using Korean corpora, i.e., Weather, Navi, Rest, English ATIS, SNIPS, Simulated Dialogue-Movie, Dialogue-Restaurant, NLU-Evaluation datasets. Through experiments, concluded that our model quite effective in capturing learned representations useful.