Entity Linking in 100 Languages

TL;DR

This paper introduces a multilingual entity linking model that connects mentions across 100+ languages to a unified knowledge base, achieving state-of-the-art results and highlighting challenges in low-resource scenarios.

Contribution

It presents a new formulation and a dual encoder model for large-scale multilingual entity linking, covering over 20 million entities across 100+ languages, with improved training techniques.

Findings

Outperforms previous state-of-the-art in cross-lingual linking

Provides a new large-scale multilingual dataset, Mewsli-9

Highlights challenges in low-resource and rare entity linking

Abstract

We propose a new formulation for multilingual entity linking, where language-specific mentions resolve to a language-agnostic Knowledge Base. We train a dual encoder in this new setting, building on prior work with improved feature representation, negative mining, and an auxiliary entity-pairing task, to obtain a single entity retrieval model that covers 100+ languages and 20 million entities. The model outperforms state-of-the-art results from a far more limited cross-lingual linking task. Rare entities and low-resource languages pose challenges at this large-scale, so we advocate for an increased focus on zero- and few-shot evaluation. To this end, we provide Mewsli-9, a large new multilingual dataset (http://goo.gle/mewsli-dataset) matched to our setting, and show how frequency-based analysis provided key insights for our model and training enhancements.