TriggerNER: Learning with Entity Triggers as Explanations for Named Entity Recognition

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8503–8511 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 8503 TriggerNER: Learning with Entity Triggers as Explanations for Named Entity Recognition Bill Yuchen Lin†∗, Dong-Ho Lee†∗, Ming Shen†, Ryan Moreno†, Xiao Huang†, Prashant Shiralkar‡, Xiang Ren† †University of Southern California ‡ Amazon {yuchen.lin,dongho.lee,shenming,morenor}@usc.edu, {huan183,xiangren}@usc.edu, shiralp@amazon.com Abstract Training neural models for named entity recognition (NER) in a new domain often re- quires additional human annotations that are usually expensive and time-consuming to col- lect. Thus, a crucial research question is how to obtain supervision in a cost-effective way. In this paper, we introduce “entity triggers,” an effective proxy of human explanations for fa- cilitating label-efficient learning of NER mod- els. An entity trigger is defined as a group of words in a sentence that helps to explain why humans would recognize an entity in the sentence. We crowd-sourced 14k entity trig- gers for two well-studied NER datasets1. Our proposed model, Trigger Matching Network, jointly learns trigger representations and soft matching module with self-attention such that can generalize to unseen sentences easily for tagging. The framework is significantly more cost-effective than the traditional frameworks. 1 Introduction Named entity recognition (NER) is a fundamental information extraction task that focuses on extract- ing entities from a given text and classifying them using pre-defined categories (e.g., persons, loca- tions, organizations) (Nadeau and Sekine, 2007). Recent advances in NER have primarily focused on training neural network models with an abun- dance of human annotations, yielding state-of-the- art results (Lample et al., 2016). However, collect- ing human annotations for NER is expensive and time-consuming, especially in social media mes- sages (Lin et al., 2017a) and technical domains such as biomedical publications, financial docu- ments, legal reports, etc. As we seek to advance NER into more domains with less human effort, ∗The first two authors contributed equally. 1The code, data, and a longer version of the paper are at http://github.com/INK-USC/TriggerNER 𝑡! = 2,5,6 → 7 𝑡" = 11,12,13 → 7 2 5 6 7 8 We had a fantastic lunch at Rumble Fish yesterday , I-RES B-RES 11 12 13 where the food is my favorite . Figure 1: We show two individual entity triggers: t1 (“had ... lunch at”) and t2 (“where the food”). Both are associated to the same entity mention “Rumble Fish” (starting from 7th token) typed as restaurant (RES). how to learn neural models for NER in a cost- effective way becomes a crucial research problem. The standard protocol for obtaining an anno- tated NER dataset involves an annotator select- ing token spans in a sentence as mentions of en- tities, and labeling them with an entity type. How- ever, such annotation process provides limited su- pervision per example. Consequently, one would need large amount of annotations in order to train high-performing models for a broad range of en- tity types, which can clearly be cost-prohibitive. The key question is then how can we learn an ef- fective NER …

Content truncated to 500 words