A Neural Question Answering Model Based on Semi-Structured Tables

Proceedings of the 27th International Conference on Computational Linguistics, pages 1941–1951 Santa Fe, New Mexico, USA, August 20-26, 2018. 1941 A Neural Question Answering Model Based on Semi-Structured Tables Hao Wang1, Xiaodong Zhang1, Shuming Ma1, Xu Sun1, Houfeng Wang1,2, Mengxiang Wang3 1 MOE Key Lab of Computational Linguistics, Peking University, Beijing, 100871, China 2 Collaborative Innovation Center for Language Ability, Xuzhou, Jiangsu, 221009, China 3 Teachers' College of Beijing Union University, Beijing, 100011, China {hhhwang, zxdcs, shumingma, xusun, wanghf}@pku.edu.cn 41326498@qq.com Abstract Most question answering (QA) systems are based on raw text and structured knowledge graph. However, raw text corpora are hard for QA system to understand, and structured knowledge graph needs intensive manual work, while it is relatively easy to obtain semi-structured tables from many sources directly, or build them automatically. In this paper, we build an end-to-end system to answer multiple choice questions with semi-structured tables as its knowledge. Our system answers queries by two steps. First, it finds the most similar tables. Then the system measures the relevance between each question and candidate table cells, and choose the most related cell as the source of answer. The system is evaluated with TabMCQ dataset, and gets a huge improvement compared to the state of the art. 1 Introduction Question answering is a practical task, and there are many related challenges. Due to the complexity of natural language, those challenges are hard to solve. A QA model requires knowledge, which con- tains facts, and fetches the answer from the knowledge. For most QA systems, raw text and structured knowledge graph are used as their knowledge. However, raw text corpora are hard to understand for machine. Besides, most knowledge graphs contain too much noise and still need manual intervention. In contrast, the semi-structured data are more flexible to be comprehended than raw text corpora, and much easier to build from text automatically than knowledge graphs. Besides, there are many documents which can be easily converted to semi-structured tables, such as announcements published by institutions and knowledge in science books, etc. We would like to utilize these kinds of knowledge in QA systems. There are many works on QA based on other knowledge. Some of them are based on raw text corpora, such as Miller et al. (2016), Min et al. (2017), Yin et al. (2016). Besides, QA models based on knowledge graphs are improving rapidly. Freebase (Bollacker et al., 2008) and Wikidata (Vrandecic and Kr¨otzsch, 2014) are well-known structured knowledge bases, which are built almost manually by their users. Al- though there are some studies on automatic knowledge graph construction, like Sateli and Witte (2015), it is not good enough. There are also many studies on those structured data, like Yih et al. (2015), Yao and Durme (2014). Some of them focused on text description of each item, while others, such as Zhang et al. (2016), try to obtain embeddings of the items. Although there is limited work on semi-structured tables, there has already been research of QA based on simple …

Content truncated to 500 words