No Noun Phrase Left Behind: Detecting and Typing Unlinkable Entities

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 893–903, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics No Noun Phrase Left Behind: Detecting and Typing Unlinkable Entities Thomas Lin, Mausam, Oren Etzioni Computer Science & Engineering University of Washington Seattle, WA 98195, USA {tlin, mausam, etzioni}@cs.washington.edu Abstract Entity linking systems link noun-phrase men- tions in text to their corresponding Wikipedia articles. However, NLP applications would gain from the ability to detect and type all entities mentioned in text, including the long tail of entities not prominent enough to have their own Wikipedia articles. In this paper we show that once the Wikipedia entities men- tioned in a corpus of textual assertions are linked, this can further enable the detection and fine-grained typing of the unlinkable en- tities. Our proposed method for detecting un- linkable entities achieves 24% greater accu- racy than a Named Entity Recognition base- line, and our method for fine-grained typing is able to propagate over 1,000 types from linked Wikipedia entities to unlinkable entities. De- tection and typing of unlinkable entities can increase yield for NLP applications such as typed question answering. 1 Introduction A key challenge in machine reading (Etzioni et al., 2006) is to identify the entities mentioned in text, and associate them with appropriate background in- formation such as their type. Consider the sentence “Some people think that pineapple juice is good for vitamin C.” To analyze this sentence, a machine should know that “pineapple juice” refers to a bev- erage, while “vitamin C” refers to a nutrient. Entity linking (Bunescu and Pas¸ca, 2006; Cucerzan, 2007) addresses this problem by link- ing noun phrases within the sentence to entries in a large, fixed entity catalog (almost always example noun phrases status “apple juice” “orange juice” present “prune juice” “wheatgrass juice” absent “radiation exposure” “workplace stress” present “asbestos exposure” “financial stress” absent “IJCAI” “OOPSLA” present “EMNLP” “ICAPS” absent Table 1: Wikipedia has entries for prominent entities, while missing tail and new entities of the same types. Wikipedia). Thus, entity linking has a limited and somewhat arbitrary range. In our example, systems by (Ferragina and Scaiella, 2010) and (Ratinov et al., 2011) both link “vitamin C” correctly, but link “pineapple juice” to “pineapple.” “Pineapple juice” is not entity linked as a beverage because it is not prominent enough to have its own Wikipedia entry. As Table 1 shows, Wikipedia often has prominent entities, while missing tail and new entities of the same types.1 (Wang et al., 2012) notes that there are more than 900 different active shoe brands, but only 82 exist in Wikipedia. In scenarios such as in- telligence analysis and local search, non-Wikipedia entities are often the most important. Hence, we introduce the unlinkable noun phrase problem: Given a noun phrase that does not link into Wikipedia, return whether it is an entity, as well its fine-grained semantic types. Deciding if a non- Wikipedia noun phrase is an entity is …

Content truncated to 500 words