URIEL+ expands linguistic database with writing systems and genealogical coverage for 26,000+ languages

Researchers have substantially expanded URIEL+, a foundational linguistic knowledge base that encodes languages through geographic, genetic, and typological information. The enhancement addresses a persistent challenge in cross-lingual AI: sparse data for thousands of languages, particularly those with few speakers or limited digital resources.

What's new in URIEL+#

The upgraded version introduces three major additions:

• Script vectors describing writing system properties now cover 7,488 languages, reducing feature sparsity by 14%
• Glottolog integration brings nearly 18,710 new languages into the database, expanding total coverage from around 3,000 to more than 19,000 languages (a 1,007% increase)
• Lineage imputation propagates typological and script features across language family trees for 26,449 languages, improving imputation quality metrics by up to 35%

Why this matters#

URIEL+ underpins multilingual research across machine translation, language modeling, and transfer learning. However, its original design left many languages underrepresented, with incomplete feature sets and sparse genealogical links. Low-resource languages—which account for the vast majority of the world's linguistic diversity—suffered most from these gaps. Richer linguistic metadata helps models understand structural relationships and transfer knowledge more effectively across language families.

Performance on real tasks#

When tested on cross-lingual transfer benchmarks focused on low-resource scenarios, the expanded URIEL+ showed mixed but sometimes promising results. Some tasks saw gains up to 6%, though performance was occasionally variable compared to the original version—suggesting the value of the new data may depend on how well it aligns with specific downstream applications.

The work, available on arXiv, was led by Mason Shipton and colleagues in computational linguistics.