AutoTriggER: Named Entity Recognition with Auxiliary Trigger Extraction

While deep neural models for named entity recognition (NER) have shown impressive results, these models often require additional human annotation that is expensive and time-consuming to generate, especially for new or low-resource domains. Some success has been shown replacing conventional human annotation with distant supervision or other meta-level information (e.g. explanations). However, the costs of generating this additional information can still be prohibitive, especially in domains where existing resources (e.g. databases to be used for distant supervision) may not exist. In this paper, we present a novel two-stage framework (AutoTriggER) to improve NER performance by automatically generating and leveraging “entity triggers” for named entity recognition. These triggers—essentially human-readable “clues” in the text that can help guide the model to better decisions—are first identified automatically using a sampling and occlusion algorithm. Next, we propose a trigger interpolation network to leverage these triggers in a transformer-based NER model. By combining these stages, AutoTriggER is able to both create and leverage auxiliary supervision by itself. Through experiments on three well-studied NER datasets, we show that our automatically extracted triggers are well-matched to human triggers, and AutoTriggER improves performance over a standard RoBERTa-CRF architecture by nearly 0.5 F1 points on average and much more in a low resource setting.1 ACM Reference Format: Dong-Ho Lee1,3*, Ravi Kiran Selvam1*, Sheikh Muhammad Sarwar2, Bill Yuchen Lin1,, Fred Morstatter3, Jay Pujara1,3, Elizabeth Boschee3, James Allan2, Xiang Ren1,3. 2020. AutoTriggER: Named Entity Recognition with Auxiliary Trigger Extraction. In Proceedings of the 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’20), August 23– 27, 2020, Virtual Event, CA, USA. ACM, New York, NY, USA, 10 pages. https://doi.org/10.1145/3394486.3403153