Learning to Embed Semantic Correspondence for Natural Language Understanding

Proceedings of the 22nd Conference on Computational Natural Language Learning (CoNLL 2018), pages 131–140 Brussels, Belgium, October 31 - November 1, 2018. c⃝2018 Association for Computational Linguistics 131 Learning to Embed Semantic Correspondence for Natural Language Understanding Sangkeun Jung1, Jinsik Lee2, and Jiwon Kim3 1,2,3T-Brain, AI Research Center, SK telecom 1Dept. of Computer Science and Engineering, Chungnam National University 1hugmanskj@gmail.com 2,3 {jinsik16.lee, jk}@sktbrain.com Abstract While learning embedding models has yielded fruitful results in several NLP subfields, most notably Word2Vec, embedding correspon- dence has relatively not been well explored es- pecially in the context of natural language un- derstanding (NLU), a task that typically ex- tracts structured semantic knowledge from a text. A NLU embedding model can facilitate analyzing and understanding relationships be- tween unstructured texts and their correspond- ing structured semantic knowledge, essential for both researchers and practitioners of NLU. Toward this end, we propose a framework that learns to embed semantic correspondence be- tween text and its extracted semantic knowl- edge, called semantic frame. One key con- tributed technique is semantic frame recon- struction used to derive a one-to-one mapping between embedded vectors and their corre- sponding semantic frames. Embedding into semantically meaningful vectors and comput- ing their distances in vector space provide a simple, but effective way to measure seman- tic similarities. With the proposed framework, we demonstrate three key areas where the em- bedding model can be effective: visualization, semantic search and re-ranking. 1 Introduction The goal of NLU is to extract meaning from a nat- ural language and infer the user intention. NLU typically involves two tasks: identifying user in- tent and extracting domain-specific entities, the second of which is often referred to as slot-filling (Mesnil et al., 2013; Jeong and Lee, 2006; Kim et al., 2016). Typically, the NLU task can be viewed as an extraction of structured text from a raw text. In NLU literature, the structured form of intent and filled slots is called a semantic frame. � ��: text ��: semantic frame ℛ� ℛ� � “Please list all flights from Dallas to Philadelphia on Monday.” ��� � intent : atis_flight fromloc. city_name Dallas toloc. city_name Philadelphia depart_date. day_name Monday � ∶ shared vector space � (a) Framework distance � distance � distance comparison �� �� �� visualization distance � � � � � (b) Applications Figure 1: Semantic vector learning framework and applications. We assume a pair of corresponding text and semantic frame (t, s), which has semanti- cally the same meaning in a raw text domain (χT ), and a semantic frame domain (χS) can be encoded to a vector v in a shared embedding vector space Z. RT and RS are two reader functions that en- code raw and structured text to a semantic vector. W is a writing function that decodes a semantic vector to a symbolic semantic frame. In this study, we aim to learn the meaningful distributed semantic representation, rather than fo- cusing on building the NLU system itself. Once we obtained a reliable and reasonable semantic …

Content truncated to 500 words