Fact · Languages and accents
The SeamlessM4T paper reports a language as zero-shot for speech translation into English when it lacks the primary training data counted in its table.
Checked against its source on October 17, 2026
In the source's own words
“Absent such data, we report the language as zero-shot (when evaluating S2TT into English).”
SeamlessM4T: Massively Multilingual and Multimodal Machine Translation (full text), Seamless Communication, Meta AI (arXiv), 2023, Notes to the language table (resource level).
Open the original (DOI 10.48550/arXiv.2308.11596)- What it measures
- S2TT is the paper's abbreviation for speech-to-text translation. The resource level in the table is set by the amount of primary data, which the paper defines as open-source S2TT data and pseudo-labeled ASR data.
- What it does not show
- Zero-shot describes the training data, not the quality of the output. A language can be labeled zero-shot and still produce usable text, but the paper does not give a measured score for it.