Latent Entities Extraction: How to Extract Entities that Do Not Appear in the Text?

Proceedings of the 22nd Conference on Computational Natural Language Learning (CoNLL 2018), pages 200–210 Brussels, Belgium, October 31 - November 1, 2018. c⃝2018 Association for Computational Linguistics 200 Latent Entities Extraction: How to Extract Entities that Do Not Appear in the Text? Eylon Shoshan1, Kira Radinsky1,2 {eylonsho999,kirar}@cs.technion.ac.il 1Department of Computer Science, Technion - Israel Institute of Technology, Haifa, Israel 2eBay Research, Israel Abstract Named-entity Recognition (NER) is an impor- tant task in the NLP field , and is widely used to solve many challenges. However, in many scenarios, not all of the entities are explicitly mentioned in the text. Sometimes they could be inferred from the context or from other in- dicative words. Consider the following sen- tence: “CMA can easily hydrolyze into free acetic acid.” Although water is not mentioned explicitly, one can infer that H2O is an en- tity involved in the process. In this work, we present the problem of Latent Entities Extrac- tion (LEE). We present several methods for determining whether entities are discussed in a text, even though, potentially, they are not explicitly written. Specifically, we design a neural model that handles extraction of multi- ple entities jointly. We show that our model, along with multi-task learning approach and a novel task grouping algorithm, reaches high performance in identifying latent entities. Our experiments are conducted on a large biolog- ical dataset from the biochemical field. The dataset contains text descriptions of biological processes, and for each process, all of the in- volved entities in the process are labeled, in- cluding implicitly mentioned ones. We believe LEE is a task that will significantly improve many NER and subsequent applications and improve text understanding and inference. 1 Introduction Named entity recognition (NER) is an impor- tant building block in many natural-language- processing algorithms and applications. For ex- ample, representing texts as a knowledge graph, where nodes are extracted entities, has been proved to be effective for question answering (Be- rant and Clark, 2014) as well as for summarization tasks (Ganesan et al., 2010). Other applications, such as semantic annotation (Marrero et al., 2013) require recognition of entities in the text as well. Babych and Hartley (2003) have also shown that identifying named entities correctly, has an effect both on the global syntactic and lexical structure, additionally to the local and immediate context. NER today focuses on extracting existing enti- ties in the text. However, many texts, contain “hid- den” entities, which are not mentioned explicitly in the text, but might be inferred from the context. For example, special verbs could help a human reader infer the discussed entity implicitly. Con- sider the following textual passage of a biochemi- cal reaction: “At the plasma membrane, phos- phatidylcholine is hydrolyzed, removing one of its acyl groups, to 1-acyl lysophosphatidylcholine by membrane- associated phospholipase b1. ” The words water or H2O are not mentioned. Nonetheless, one could easily infer that water is involved in the process, since the word hydrolyzed refers to water. Therefore, water is a latent entity in …

Content truncated to 500 words