A Context-aware Framework for Translation-mediated Conversations

José Pombal, Sweta Agrawal, Emmanouil Zaranis, Patrick Fernandes, André F. T. Martins · 2024

Read A Context-aware Framework for Translation-mediated Conversations by José Pombal, Sweta Agrawal, Emmanouil Zaranis, Patrick Fernandes, André F. T. Martins (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-04205.txt.

Description

José Pombal and co-authors improve machine translation for live chats between people who share no language in 2024. Current tools can err and break a talk down. A key cause is lost context, which leaves vague or dropped details. Their frame feeds chat context into a large language model at training and at use.

Tests cover two task domains: customer chat and user help chats. Their system, called TowerChat, beat top known systems on many language pairs. Gains held across a set of auto quality scores. Checks show the model uses context in the planned way. Translations stay more consistent with what the speaker meant. The work points to context as a fix for chat translation flaws.

Catalog record

CREATORJosé Pombal, Sweta Agrawal, Emmanouil Zaranis, Patrick Fernandes, André F. T. Martins
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE58 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-04205

Browse the full library · About the Tower of Babel