PulseAugur
EN
LIVE 08:49:15

New RLHF framework improves Vietnamese historical manuscript translation

By PulseAugur Editorial · [2 sources] ·

Researchers have developed a new multimodal framework using Reinforcement Learning from Human Feedback (RLHF) to translate degraded Han-Nom manuscripts into modern Vietnamese. The system integrates visual features from CLIP ViT-L/14@336, Han-Nom representations from bert-base-chinese, and Vietnamese representations from vinai/phobert-base, alongside T5-small encoder states. Experiments comparing Proximal Policy Optimization (PPO), Direct Preference Optimization (DPO), and KTO showed that DPO achieved the best results across several metrics, significantly improving lexical and semantic quality for this low-resource historical translation task. AI

IMPACT This research advances multimodal translation techniques, potentially improving the accessibility of historical texts.

RANK_REASON The cluster contains an academic paper detailing a new research methodology and experimental results.

Read on arXiv cs.CL →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

New RLHF framework improves Vietnamese historical manuscript translation

COVERAGE [2]

  1. arXiv cs.CL TIER_1 English(EN) · Thi Kim Trang Vo, Nghia Hieu Nguyen, Ha Minh Tan ·

    Direct Image-to-Modern Vietnamese Translation of Han-Nom Manuscripts via Multimodal RLHF Preference Alignment

    arXiv:2607.11434v1 Announce Type: new Abstract: Translating Han-Nom manuscripts into modern Vietnamese is challenging because historical pages are often degraded, the script contains rare logographic characters, and parallel supervision is limited. We propose a multimodal RLHF pr…

  2. arXiv cs.CL TIER_1 English(EN) · Ha Minh Tan ·

    Direct Image-to-Modern Vietnamese Translation of Han-Nom Manuscripts via Multimodal RLHF Preference Alignment

    Translating Han-Nom manuscripts into modern Vietnamese is challenging because historical pages are often degraded, the script contains rare logographic characters, and parallel supervision is limited. We propose a multimodal RLHF preference-alignment framework that conditions Vie…

RELATED ENTITIES

RELATED TOPICS