Improving Question Answering over Incomplete KBs with Knowledge-Aware Reader

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4258–4264 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 4258 Improving Question Answering over Incomplete KBs with Knowledge-Aware Reader Wenhan Xiong†, Mo Yu∗, Shiyu Chang∗, Xiaoxiao Guo∗, William Yang Wang† † University of California, Santa Barbara ∗ IBM Research {xwhan, william}@cs.ucsb.edu, yum@us.ibm.com, {shiyu.chang, xiaoxiao.guo}@ibm.com Abstract We propose a new end-to-end question an- swering model, which learns to aggregate an- swer evidence from an incomplete knowledge base (KB) and a set of retrieved text snip- pets. Under the assumptions that the structured KB is easier to query and the acquired knowl- edge can help the understanding of unstruc- tured text, our model first accumulates knowl- edge of entities from a question-related KB subgraph; then reformulates the question in the latent space and reads the texts with the accu- mulated entity knowledge at hand. The evi- dence from KB and texts are finally aggregated to predict answers. On the widely-used KBQA benchmark WebQSP, our model achieves con- sistent improvements across settings with dif- ferent extents of KB incompleteness.1 1 Introduction Knowledge bases (KBs) are considered as an es- sential resource for answering factoid questions. However, accurately constructing KB with a well- designed and complicated schema requires lots of human efforts, which inevitably limits the cover- age of KBs (Min et al., 2013). As a matter of fact, KBs are often incomplete and insufficient to cover full evidence required by open-domain questions. On the other hand, the vast amount of unstruc- tured text on the Internet can easily cover a wide range of evolving knowledge, which is commonly used for open-domain question answering (Chen et al., 2017; Wang et al., 2018). Therefore, to im- prove the coverage of KBs, it is straightforward to augment KB with text data. Recently, text-based QA models along (Seo et al., 2016; Xiong et al., 2017; Yu et al., 2018) have achieved remarkable performance when dealing with a single passage that is guaranteed to include the answer. However, they are still insufficient when multiple documents 1https://github.com/xwhan/Knowledge-Aware-Reader. Question: Who did Cam Newton sign with? Document Retrieval Cam Newton plays football Cam Newton career_starts 2011 Cam Newton date_of_birth 19890511 … KB SubGraph Answer: Carolina Panthers Cameron Newton (born May 11, 1989) plays for the Carolina Panthers of the National Football League (NFL) … Background knowledge: Cam Newton is a football player Textual evidence: Cam Newton plays for Panthers Figure 1: A real example from WebQSP. Here the an- swer cannot be directly found in the KB. But the knowl- edge provided by the KB, i.e., Cam Newton is a foot- ball player, indicates he signed with the team he plays for. This knowledge can be essential for recognizing the relevant text piece. are presented. We hypothesize this is partially due to the lack of background knowledge while distin- guishing relevant information from irrelevant ones (see Figure 1 for a real example). To better utilize textual evidence for im- proving QA over incomplete …

Content truncated to 500 words