Rethinking Multilingual Vision-Language Translation: Dataset, Evaluation, and Adaptation

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wang, Xintong, Pan, Jingheng, Liu, Yixiao, Zhao, Xiaohu, Lyu, Chenyang, Wu, Minghao, Biemann, Chris, Wang, Longyue, Xu, Linlong, Luo, Weihua, Zhang, Kaifu
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908407289610240
author Wang, Xintong
Pan, Jingheng
Liu, Yixiao
Zhao, Xiaohu
Lyu, Chenyang
Wu, Minghao
Biemann, Chris
Wang, Longyue
Xu, Linlong
Luo, Weihua
Zhang, Kaifu
author_facet Wang, Xintong
Pan, Jingheng
Liu, Yixiao
Zhao, Xiaohu
Lyu, Chenyang
Wu, Minghao
Biemann, Chris
Wang, Longyue
Xu, Linlong
Luo, Weihua
Zhang, Kaifu
contents Vision-Language Translation (VLT) is a challenging task that requires accurately recognizing multilingual text embedded in images and translating it into the target language with the support of visual context. While recent Large Vision-Language Models (LVLMs) have demonstrated strong multilingual and visual understanding capabilities, there is a lack of systematic evaluation and understanding of their performance on VLT. In this work, we present a comprehensive study of VLT from three key perspectives: data quality, model architecture, and evaluation metrics. (1) We identify critical limitations in existing datasets, particularly in semantic and cultural fidelity, and introduce AibTrans -- a multilingual, parallel, human-verified dataset with OCR-corrected annotations. (2) We benchmark 11 commercial LVLMs/LLMs and 6 state-of-the-art open-source models across end-to-end and cascaded architectures, revealing their OCR dependency and contrasting generation versus reasoning behaviors. (3) We propose Density-Aware Evaluation to address metric reliability issues under varying contextual complexity, introducing the DA Score as a more robust measure of translation quality. Building upon these findings, we establish a new evaluation benchmark for VLT. Notably, we observe that fine-tuning LVLMs on high-resource language pairs degrades cross-lingual performance, and we propose a balanced multilingual fine-tuning strategy that effectively adapts LVLMs to VLT without sacrificing their generalization ability.
format Preprint
id arxiv_https___arxiv_org_abs_2506_11820
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Rethinking Multilingual Vision-Language Translation: Dataset, Evaluation, and Adaptation
Wang, Xintong
Pan, Jingheng
Liu, Yixiao
Zhao, Xiaohu
Lyu, Chenyang
Wu, Minghao
Biemann, Chris
Wang, Longyue
Xu, Linlong
Luo, Weihua
Zhang, Kaifu
Computer Vision and Pattern Recognition
Computation and Language
Vision-Language Translation (VLT) is a challenging task that requires accurately recognizing multilingual text embedded in images and translating it into the target language with the support of visual context. While recent Large Vision-Language Models (LVLMs) have demonstrated strong multilingual and visual understanding capabilities, there is a lack of systematic evaluation and understanding of their performance on VLT. In this work, we present a comprehensive study of VLT from three key perspectives: data quality, model architecture, and evaluation metrics. (1) We identify critical limitations in existing datasets, particularly in semantic and cultural fidelity, and introduce AibTrans -- a multilingual, parallel, human-verified dataset with OCR-corrected annotations. (2) We benchmark 11 commercial LVLMs/LLMs and 6 state-of-the-art open-source models across end-to-end and cascaded architectures, revealing their OCR dependency and contrasting generation versus reasoning behaviors. (3) We propose Density-Aware Evaluation to address metric reliability issues under varying contextual complexity, introducing the DA Score as a more robust measure of translation quality. Building upon these findings, we establish a new evaluation benchmark for VLT. Notably, we observe that fine-tuning LVLMs on high-resource language pairs degrades cross-lingual performance, and we propose a balanced multilingual fine-tuning strategy that effectively adapts LVLMs to VLT without sacrificing their generalization ability.
title Rethinking Multilingual Vision-Language Translation: Dataset, Evaluation, and Adaptation
topic Computer Vision and Pattern Recognition
Computation and Language
url https://arxiv.org/abs/2506.11820