Summary of Paying More Attention to Source Context: Mitigating Unfaithful Translations From Large Language Model, by Hongbin Zhang et al.
Paying More Attention to Source Context: Mitigating Unfaithful Translations from Large Language Model
by Hongbin Zhang, Kehai Chen, Xuefeng Bai, Yang Xiang, Min Zhang
First submitted to arxiv on: 11 Jun 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 paper proposes methods to improve the multilingual machine translation capabilities of large language models (LLMs). Unlike traditional encoder-decoder models, decoder-only LLMs lack explicit alignment between source and target contexts. Analyzing contribution scores during generation reveals that LLMs can be biased towards previously generated tokens over corresponding source tokens, leading to unfaithful translations. To address this issue, the authors propose adjusting source context attention weights, suppressing irrelevant target prefix influence, and avoiding over-reliance on the target prefix in instruction tuning. Experimental results demonstrate the effectiveness of these methods across multiple language pairs using human-collected test sets. Additionally, human evaluation shows that the proposed method reduces hallucinatory translations and facilitates faithful translation generation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps large language models (LLMs) translate languages better. Right now, LLMs can translate well between many languages, but sometimes they don’t get it exactly right. To fix this, the authors came up with new ways to make LLMs pay attention to what’s being translated from. They tested these methods and found that they work across different language pairs. The results also show that these methods help reduce mistakes and give more accurate translations. |
Keywords
» Artificial intelligence » Alignment » Attention » Decoder » Encoder decoder » Instruction tuning » Translation
