Modeling Semantics with Gated Graph Neural Networks for Knowledge Base Question Answering

Proceedings of the 27th International Conference on Computational Linguistics, pages 3306–3317 Santa Fe, New Mexico, USA, August 20-26, 2018. 3306 Modeling Semantics with Gated Graph Neural Networks for Knowledge Base Question Answering Daniil Sorokin and Iryna Gurevych Ubiquitous Knowledge Processing Lab (UKP) and Research Training Group AIPHES Department of Computer Science, Technische Universit¨at Darmstadt www.informatik.tu-darmstadt.de/ukp/ Abstract The most approaches to Knowledge Base Question Answering are based on semantic parsing. In this paper, we address the problem of learning vector representations for complex semantic parses that consist of multiple entities and relations. Previous work largely focused on selecting the correct semantic relations for a question and disregarded the structure of the semantic parse: the connections between entities and the directions of the relations. We propose to use Gated Graph Neural Networks to encode the graph structure of the semantic parse. We show on two data sets that the graph networks outperform all baseline models that do not explicitly model the structure. The error analysis confirms that our approach can successfully process complex semantic parses. 1 Introduction Knowledge base question answering (QA) is an important natural language processing problem. Given a natural language question, the task is to find a set of entities in a knowledge base (KB) that constitutes the answer. For example, for a question “What is Princess Leia’s home planet?” the answer, “Alderaan”, could be retrieved from a general-purpose KB, such as Wikidata1. A successful KB QA system would ultimately provide a universally accessible natural language interface to factual knowledge (Liang, 2016). QA requires precise modeling of the question semantics through the entities and relations available in the KB in order to retrieve the correct answer. Figure 1 shows how the above example question could be modeled using Wikidata. The depicted graph structure consists of entities and relations from the KB and the special q-node. Any entity in the KB that can take the place of the q-node will be a part of the answer. In this paper, we describe a semantic parsing approach to the problem of KB QA. That is, for each input question, we construct an explicit structural semantic parse (semantic graph), as in Figure 1. Semantic parses can be deterministically converted to a query to extract the answers from the KB. Similar graphical representations were used in previous work (Yih et al., 2015; Reddy et al., 2016; Bao et al., 2016). However, the modern semantic parsing approaches usually focus either on the syntactic analysis of the input question (Reddy et al., 2016) or on detecting individual KB relations (Yu et al., 2017), whereas the structure of the semantic parse is ignored or only approximately modeled. Reddy et al. (2016) use the syntactic structure of the question to build all possible semantic parses and then apply a linear model with manually defined features to choose the correct parse. A subset of their features encodes basic information about the graph structure of the semantic parse (e.g. number of nodes). The state-of-the-art approach of Yih et al. (2015) …

Content truncated to 500 words