Soft Gazetteers for Low-Resource Named Entity Recognition

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8118–8123 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 8118 Soft Gazetteers for Low-Resource Named Entity Recognition Shruti Rijhwani, Shuyan Zhou, Graham Neubig, Jaime Carbonell Language Technologies Institute Carnegie Mellon University {srijhwan,shuyanzh,gneubig,jgc}@cs.cmu.edu Abstract Traditional named entity recognition models use gazetteers (lists of entities) as features to improve performance. Although modern neu- ral network models do not require such hand- crafted features for strong performance, recent work (Wu et al., 2018) has demonstrated their utility for named entity recognition on En- glish data. However, designing such features for low-resource languages is challenging, be- cause exhaustive entity gazetteers do not ex- ist in these languages. To address this prob- lem, we propose a method of “soft gazetteers” that incorporates ubiquitously available infor- mation from English knowledge bases, such as Wikipedia, into neural named entity recogni- tion models through cross-lingual entity link- ing. Our experiments on four low-resource languages show an average improvement of 4 points in F1 score.1 1 Introduction Before the widespread adoption of neural networks for natural language processing tasks, named en- tity recognition (NER) systems used linguistic fea- tures based on lexical and syntactic knowledge to improve performance (Ratinov and Roth, 2009). With the introduction of the neural LSTM-CRF model (Huang et al., 2015; Lample et al., 2016), the need to develop hand-crafted features to train strong NER models diminished. However, Wu et al. (2018) have recently demonstrated that inte- grating linguistic features based on part-of-speech tags, word shapes, and manually created lists of entities called gazetteers into neural models leads to better NER on English data. Of particular inter- est to this paper are the gazetteer-based features – binary-valued features determined by whether or not an entity is present in the gazetteer. 1Code and data are available at https://github. com/neulab/soft-gazetteers. Although neural NER models have been ap- plied to low-resource settings (Cotterell and Duh, 2017; Huang et al., 2019), directly integrating gazetteer features into these models is difficult be- cause gazetteers in these languages are either lim- ited in coverage or completely absent. Expanding them is time-consuming and expensive, due to the lack of available annotators for low-resource lan- guages (Strassel and Tracey, 2016). As an alternative, we introduce “soft gazetteers”, a method to create continuous-valued gazetteer fea- tures based on readily available data from high- resource languages and large English knowledge bases (e.g., Wikipedia). More specifically, we use entity linking methods to extract information from these resources and integrate it into the commonly- used CNN-LSTM-CRF NER model (Ma and Hovy, 2016) using a carefully designed feature set. We use entity linking methods designed for low-resource languages, which require far fewer resources than traditional gazetteer features (Upad- hyay et al., 2018; Zhou et al., 2020). Our experiments demonstrate the effectiveness of our proposed soft gazetteer features, with an av- erage improvement of 4 F1 points over the baseline, across four low-resource languages: Kinyarwanda, Oromo, Sinhala, and Tigrinya. 2 Background Named Entity Recognition NER …

Content truncated to 500 words