XLEnt: Mining a Large Cross-lingual Entity Dataset with Lexical-Semantic-Phonetic Word Alignment

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 10424–10430 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 10424 XLEnt: Mining a Large Cross-lingual Entity Dataset with Lexical-Semantic-Phonetic Word Alignment Ahmed El-Kishky1 Adithya Renduchintala2 James Cross2 Francisco Guzmán2 Philipp Koehn3 1Twitter Cortex, 2 Facebook AI 3Johns Hopkins University aelkishky@twitter.com, {adirendu,jcross,fguzman}@fb.com, phi@jhu.edu Abstract Cross-lingual named-entity lexica are an impor- tant resource to multilingual NLP tasks such as machine translation and cross-lingual wikifica- tion. While knowledge bases contain a large number of entities in high-resource languages such as English and French, corresponding en- tities for lower-resource languages are often missing. To address this, we propose Lexical- Semantic-Phonetic Align (LSP-Align), a tech- nique to automatically mine cross-lingual en- tity lexica from mined web data. We demon- strate LSP-Align outperforms baselines at ex- tracting cross-lingual entity pairs and mine 164 million entity pairs from 120 different lan- guages aligned with English. We release these cross-lingual entity pairs along with the mas- sively multilingual tagged named entity corpus as a resource to the NLP community. 1 Introduction Named entities are references in natural text to real-world objects such as persons, locations, or or- ganizations that can be denoted with a proper name. Recognizing and handling these named entities in many languages is a difficult, yet crucial, step to language-agnostic text understanding and multilin- gual natural language processing (NLP) (Sekine and Ranchhod, 2009). As such, cross-lingual named entity lexica can be invaluable resources towards making tasks such as entity linking, named entity recognition (Ren et al., 2016b,a), and information and knowledge base con- struction (Tao et al., 2014) inherently multilingual. However, the coverage of many such multilingual entity lexica (e.g., Wikipedia titles) is less complete for lower-resource languages and approaches to au- tomatically generate them under-perform due to the poor performance of low-resource taggers (Feng et al., 2018; Cotterell and Duh, 2017). To perform low-resource NER, previous efforts have applied word alignment techniques to project Figure 1: Identify entity pairs by projecting English enti- ties onto lower-resource languages via word-alignment. available labels to other languages. Kim et al. (2010) applies heuristic approaches with alignment correction using an alignment dictionary of entity mentions. Das and Petrov (2011) introduced a novel label propagation technique that creates a tag lexicon for the target language, while Wang and Manning (2014) instead projected model ex- pectation rather than labels thus transferring word boundary uncertainty. Additional work jointly per- forms word alignment while training bilingual name tagging (Wang et al., 2013); however this method assumes the availability of named entity taggers in both languages. Other methods have leveraged bilingual embeddings for projection (Ni et al., 2017; Xie et al., 2018). In this work, we propose using named-entity projection to automatically curate a large cross- lingual entity lexicon for many language pairs. As shown Figure 1, we construct this resource by performing NER in a higher-resource language, then projecting the entities onto text in a lower- resource language using word-alignment models. Our main contribution is the construction and release …

Content truncated to 500 words