Improving Natural Language Inference Using External Knowledge in the Science Questions Domain

  • 2018-09-15 14:37:46
  • Xiaoyan Wang, Pavan Kapanipathi, Ryan Musa, Mo Yu, Kartik Talamadupula, Ibrahim Abdelaziz, Maria Chang, Achille Fokoue, Bassem Makni, Nicholas Mattei, Michael Witbrock
  • 4

Abstract

Natural Language Inference (NLI) is fundamental to many Natural LanguageProcessing (NLP) applications including semantic search and question answering.The NLI problem has gained significant attention thanks to the release of largescale, challenging datasets. Present approaches to the problem largely focus onlearning-based methods that use only textual information in order to classifywhether a given premise entails, contradicts, or is neutral with respect to agiven hypothesis. Surprisingly, the use of methods based on structuredknowledge -- a central topic in artificial intelligence -- has not receivedmuch attention vis-a-vis the NLI problem. While there are many open knowledgebases that contain various types of reasoning information, their use for NLIhas not been well explored. To address this, we present a combination oftechniques that harness knowledge graphs to improve performance on the NLIproblem in the science questions domain. We present the results of applying ourtechniques on text, graph, and text-to-graph based models, and discussimplications for the use of external knowledge in solving the NLI problem. Ourmodel achieves the new state-of-the-art performance on the NLI problem over theSciTail science questions dataset.

 

Quick Read (beta)

loading the full paper ...