Summary of Word Alignment As Preference For Machine Translation, by Qiyu Wu et al.
Word Alignment as Preference for Machine Translation
by Qiyu Wu, Masaaki Nagata, Zhongtao Miao, Yoshimasa Tsuruoka
First submitted to arxiv on: 15 May 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed solution aims to address the long-standing issue of hallucination and omission in machine translation (MT), particularly when using large language models (LLMs). To mitigate this problem, the authors guide the LLM-based MT model to better word alignment. The study begins by investigating the correlation between word alignment and hallucination/omission phenomena in MT. Next, the authors propose utilizing word alignment as a preference signal to optimize the LLM-based model. This is achieved by constructing preference data from chosen and rejected translations from multiple MT tools. Direct preference optimization is then used to optimize the LLM-based model towards this preference signal. To evaluate the performance of the models in mitigating hallucination and omission, the authors propose selecting hard instances and utilizing GPT-4 for direct evaluation. The study verifies the rationality of these designed evaluation methods through experiments and presents extensive results demonstrating the effectiveness of word alignment-based preference optimization to mitigate hallucination and omission. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Machine learning educators can improve machine translation (MT) by reducing hallucination and omission, which occurs when large language models (LLMs) are used. Researchers investigated how LLMs align words in different languages and found that this process is connected to hallucination and omission problems. To solve this issue, they developed a method called word alignment-based preference optimization. This approach uses information from multiple MT tools to guide the LLM to make better translations. The study’s findings show that this approach can effectively reduce hallucination and omission in MT. |
Keywords
» Artificial intelligence » Alignment » Gpt » Hallucination » Machine learning » Optimization » Translation
