Abstract
Large multilingual language models typically rely on a single vocabulary shared across 100+ languages. As these models have increased in parameter count and depth, vocabulary size has remained largely unchanged. This vocabulary bottleneck limits the representational capabilities of multilingual models like XLM-R. In this paper, we introduce a new approach for scaling to very large multilingual vocabularies by de-emphasizing token sharing between languages with little lexical overlap and assigning vocabulary capacity to achieve sufficient coverage for each individual language. Tokenizations using our vocabulary are typically more semantically meaningful and shorter compared to XLM-R. Leveraging this improved vocabulary, we train XLM-V, a multilingual language model with a one million token vocabulary. XLM-V outperforms XLM-R on every task we tested on ranging from natural language inference (XNLI), question answering (MLQA, XQuAD, TyDiQA), to named entity recognition (WikiAnn). XLM-V is particularly effective on low-resource language tasks and outperforms XLM-R by 11.2% and 5.8% absolute on MasakhaNER and Americas NLI, respectively.
| Original language | English |
|---|---|
| Title of host publication | EMNLP 2023 - 2023 Conference on Empirical Methods in Natural Language Processing, Proceedings |
| Editors | Houda Bouamor, Juan Pino, Kalika Bali |
| Publisher | Association for Computational Linguistics (ACL) |
| Pages | 13142-13152 |
| Number of pages | 11 |
| ISBN (Electronic) | 9798891760608 |
| DOIs | |
| State | Published - 2023 |
| Externally published | Yes |
| Event | 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023 - Hybrid, Singapore, Singapore Duration: 6 Dec 2023 → 10 Dec 2023 |
Publication series
| Name | EMNLP 2023 - 2023 Conference on Empirical Methods in Natural Language Processing, Proceedings |
|---|
Conference
| Conference | 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023 |
|---|---|
| Country/Territory | Singapore |
| City | Hybrid, Singapore |
| Period | 6/12/23 → 10/12/23 |
Bibliographical note
Publisher Copyright:©2023 Association for Computational Linguistics.
Fingerprint
Dive into the research topics of 'XLM-V: Overcoming the Vocabulary Bottleneck in Multilingual Masked Language Models'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver