Source-primed Multi-turn Conversation Helps Large Language Models Translate Documents

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hu, Hanxu, Vamvas, Jannis, Sennrich, Rico
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917956151148544
author Hu, Hanxu
Vamvas, Jannis
Sennrich, Rico
author_facet Hu, Hanxu
Vamvas, Jannis
Sennrich, Rico
contents LLMs have paved the way for truly simple document-level machine translation, but challenges such as omission errors remain. In this paper, we study a simple method for handling document-level machine translation, by leveraging previous contexts in a multi-turn conversational manner. Specifically, by decomposing documents into segments and iteratively translating them while maintaining previous turns, this method ensures coherent translations without additional training, and can fully re-use the KV cache of previous turns thus minimizing computational overhead. We further propose a `source-primed' method that first provides the whole source document before multi-turn translation. We empirically show this multi-turn method outperforms both translating entire documents in a single turn and translating each segment independently according to multiple automatic metrics in representative LLMs, establishing a strong baseline for document-level translation using LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2503_10494
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Source-primed Multi-turn Conversation Helps Large Language Models Translate Documents
Hu, Hanxu
Vamvas, Jannis
Sennrich, Rico
Computation and Language
LLMs have paved the way for truly simple document-level machine translation, but challenges such as omission errors remain. In this paper, we study a simple method for handling document-level machine translation, by leveraging previous contexts in a multi-turn conversational manner. Specifically, by decomposing documents into segments and iteratively translating them while maintaining previous turns, this method ensures coherent translations without additional training, and can fully re-use the KV cache of previous turns thus minimizing computational overhead. We further propose a `source-primed' method that first provides the whole source document before multi-turn translation. We empirically show this multi-turn method outperforms both translating entire documents in a single turn and translating each segment independently according to multiple automatic metrics in representative LLMs, establishing a strong baseline for document-level translation using LLMs.
title Source-primed Multi-turn Conversation Helps Large Language Models Translate Documents
topic Computation and Language
url https://arxiv.org/abs/2503.10494