Model alignment methods like Direct Preference Optimization and Contrastive Preference Optimization have enhanced machine translation performance by leveraging preference data to enable models to reject suboptimal outputs. During preference data construction, previous approaches primarily rely on humans, strong models like GPT4 or model self-sampling. In this study, we first explain the shortcomings of this practice. Then, we propose Self-Supervised Preference Optimization (SSPO), a novel framework which efficiently constructs translation preference data for iterative DPO training. Applying SSPO to 14B parameters large language models (LLMs) achieves comparable or better performance than GPT-4o on FLORES and multi-domain test datasets. We release an augmented MQM dataset in https://github.com/sunny-sjtu/MQM-aug.

To facilitate the research on large language models (LLMs), this paper presents a comprehensive and unified library, LLMBox, to ease the development, use, and evaluation of LLMs. This library is featured with three main merits: (1) a unified data interface that supports the flexible implementation of various training strategies, (2) a comprehensive evaluation that covers extensive tasks, datasets, and models, and (3) more practical consideration, especially on user-friendliness and efficiency. With our library, users can easily reproduce existing methods, train new models, and conduct comprehensive performance comparisons. To rigorously test LLMBox, we conduct extensive experiments in a diverse coverage of evaluation settings, and experimental results demonstrate the effectiveness and efficiency of our library in supporting various implementations related to LLMs. The detailed introduction and usage guidance can be found at https://github.com/RUCAIBox/LLMBox.

We participate in the translation task on Spanish to Aragonese, Spanish to Aranese and Spanish to Asturian. Initially, we conduct preliminary experiments to assess the basic translation capabilities of various models and evaluate the impact of fine-tuning with different data types. We then choose to fine-tune the Qwen2-0.5B model using a forward synthesized pseudo-corpus from the Apertium translation system to replicate its fundamental performance. Building on this distillation model, we explore three optimization strategies across the three language directions: (1) Assembling the provided FLORES+ dev sets into a 5-shot format translation training dataset and performing few-shot fine-tuning to enhance model performance. (2) Utilizing the FLORES+ dev sets as training data and applying the Contrastive Preference Optimization (CPO) strategy for further refinement. (3) Retrieving the 20 most similar translation examples from the FLORES+ dev sets using the BM25 algorithm and performing 20-shot translations with the Claude 3.5-sonnet model. After evaluating these strategies, we select the best-performing approach for each language pair as our submission result.

This paper describes Shanghai Jiao Tong University (SJTU LoveFiction) Discourse-Level Literary Translation systems for the WMT24shared task. We participate in the literary translation task on Chinese → English, Chinese →German and Chinese → Russian with uncon-strained tack.Check our paper for detail.