Machine Translation Digest for Jul 21 2025
Here is today's selection of cs.CL papers exploring various advancements in machine translation and language modeling. The common theme revolves around enhancing language models, with a focus on diverse applications such as text detoxification, medical knowledge, and reinforcement learning. These studies highlight both the development of new benchmarks and frameworks, as well as improvements in existing datasets to better support language model evolution.
Evaluating Text Style Transfer: A Nine-Language Benchmark for Text Detoxification
Despite recent progress in large language models (LLMs), evaluation of text generation tasks such as text style transfer (TST) remains a significant challenge. Recent studies (Dementieva et al., 2024; Pauli et al., 2025) revealed a substantial gap between automatic metrics and human judgments. Moreover, most prior work focuses exclusively on English, leaving multilingual TST evaluation largely unexplored. In this paper, we perform the first comprehensive multilingual study on evaluation of text detoxification system across nine languages: English, Spanish, German, Chinese, Arabic, Hindi, Ukrainian, Russian, Amharic. Drawing inspiration from the machine translation, we assess the effectiveness of modern neural-based evaluation models alongside prompting-based LLM-as-a-judge approaches. Our findings provide a practical recipe for designing more reliable multilingual TST evaluation pipeline in the text detoxification case.
A Novel Self-Evolution Framework for Large Language Models
The capabilities of Large Language Models (LLMs) are limited to some extent by pre-training, so some researchers optimize LLMs through post-training. Existing post-training strategies, such as memory-based retrieval or preference optimization, improve user alignment yet fail to enhance the model's domain cognition. To bridge this gap, we propose a novel Dual-Phase Self-Evolution (DPSE) framework that jointly optimizes user preference adaptation and domain-specific competence. DPSE introduces a Censor module to extract multi-dimensional interaction signals and estimate satisfaction scores, which guide structured data expansion via topic-aware and preference-driven strategies. These expanded datasets support a two-stage fine-tuning pipeline: supervised domain grounding followed by frequency-aware preference optimization. Experiments across general NLP benchmarks and long-term dialogue tasks demonstrate that DPSE consistently outperforms Supervised Fine-Tuning, Preference Optimization, and Memory-Augmented baselines. Ablation studies validate the contribution of each module. In this way, our framework provides an autonomous path toward continual self-evolution of LLMs.
BEnchmarking LLMs for Ophthalmology (BELO) for Ophthalmological Knowledge and Reasoning
Current benchmarks evaluating large language models (LLMs) in ophthalmology are limited in scope and disproportionately prioritise accuracy. We introduce BELO (BEnchmarking LLMs for Ophthalmology), a standardized and comprehensive evaluation benchmark developed through multiple rounds of expert checking by 13 ophthalmologists. BELO assesses ophthalmology-related clinical accuracy and reasoning quality. Using keyword matching and a fine-tuned PubMedBERT model, we curated ophthalmology-specific multiple-choice-questions (MCQs) from diverse medical datasets (BCSC, MedMCQA, MedQA, BioASQ, and PubMedQA). The dataset underwent multiple rounds of expert checking. Duplicate and substandard questions were systematically removed. Ten ophthalmologists refined the explanations of each MCQ's correct answer. This was further adjudicated by three senior ophthalmologists. To illustrate BELO's utility, we evaluated six LLMs (OpenAI o1, o3-mini, GPT-4o, DeepSeek-R1, Llama-3-8B, and Gemini 1.5 Pro) using accuracy, macro-F1, and five text-generation metrics (ROUGE-L, BERTScore, BARTScore, METEOR, and AlignScore). In a further evaluation involving human experts, two ophthalmologists qualitatively reviewed 50 randomly selected outputs for accuracy, comprehensiveness, and completeness. BELO consists of 900 high-quality, expert-reviewed questions aggregated from five sources: BCSC (260), BioASQ (10), MedMCQA (572), MedQA (40), and PubMedQA (18). A public leaderboard has been established to promote transparent evaluation and reporting. Importantly, the BELO dataset will remain a hold-out, evaluation-only benchmark to ensure fair and reproducible comparisons of future models.
Off-Policy Corrected Reward Modeling for Reinforcement Learning from Human Feedback
Reinforcement Learning from Human Feedback (RLHF) allows us to train models, such as language models (LMs), to follow complex human preferences. In RLHF for LMs, we first train an LM using supervised fine-tuning, sample pairs of responses, obtain human feedback, and use the resulting data to train a reward model (RM). RL methods are then used to train the LM to maximize the reward given by the RM. As training progresses, the responses generated by the LM no longer resemble the responses seen by the RM during training, leading to the RM becoming inaccurate. The score given by the RM keeps increasing, but the learned behavior no longer matches the human preferences. This issue is known as overoptimization. We investigate overoptimization from the point of view of distribution shift and show that the shift results in an inconsistent estimate of the RM parameters, leading to an inconsistent estimate of the policy gradient. We propose Off-Policy Corrected Reward Modeling (OCRM), which iteratively off-policy corrects the RM using importance weighting, without requiring new labels or samples. This results in a more accurate RM, which empirically leads to an improved final policy. We validate our approach in experiments with summarization and chatbot datasets and show that it performs significantly better than standard RLHF methods and baselines. Our implementation is available at https://github.com/JohannesAck/OffPolicyCorrectedRewardModeling
ChiMed 2.0: Advancing Chinese Medical Dataset in Facilitating Large Language Modeling
Building high-quality data resources is crucial for advancing artificial intelligence research and applications in specific domains, particularly in the Chinese medical domain. Existing Chinese medical datasets are limited in size and narrow in domain coverage, falling short of the diverse corpora required for effective pre-training. Moreover, most datasets are designed solely for LLM fine-tuning and do not support pre-training and reinforcement learning from human feedback (RLHF). In this paper, we propose a Chinese medical dataset named ChiMed 2.0, which extends our previous work ChiMed, and covers data collected from Chinese medical online platforms and generated by LLMs. ChiMed 2.0 contains 204.4M Chinese characters covering both traditional Chinese medicine classics and modern general medical data, where there are 164.8K documents for pre-training, 351.6K question-answering pairs for supervised fine-tuning (SFT), and 41.7K preference data tuples for RLHF. To validate the effectiveness of our approach for training a Chinese medical LLM, we conduct further pre-training, SFT, and RLHF experiments on representative general domain LLMs and evaluate their performance on medical benchmark datasets. The results show performance gains across different model scales, validating the dataset's effectiveness and applicability.
| Curated by yukajii.com |