Skip to content

CONCEPT Cited by 1 source

Cross-lingual knowledge transfer

Cross-lingual knowledge transfer is the LLM phenomenon of applying knowledge learned during pre-training in one language to question-answering or task performance in a different language. A model that has read about "the periodic table of elements" in English should — in a well-transferred-knowledge regime — be able to answer questions about it in Swahili, Bengali, or Tagalog without those answers ever appearing in its training set in those languages.

The phenomenon matters because:

  • Training corpora are heavily English-dominant. If knowledge doesn't transfer, low-resource-language deployment is bottlenecked on training-corpus availability in those languages.
  • Conversely, if knowledge does transfer well, low-resource- language deployment can ride on the much larger English- language pre-training corpus — a high-leverage outcome.

The 2026-05-28 Google Research I/O 2026 roundup post cites the ECLeKTic benchmark as Google Research's named instrument for measuring this — "a benchmark which shows how LLMs operate in different languages" (Source: sources/2026-05-28-google-a-new-era-of-innovation-google-research-at-io-2026).

This is a minimum-viable wiki page; the underlying paper / benchmark blog has the formal task definition and metrics.

Seen in

Last updated · 542 distilled / 1,571 read