Summary of Prefix Text As a Yarn: Eliciting Non-english Alignment in Foundation Language Model, by Runzhe Zhan et al.
Prefix Text as a Yarn: Eliciting Non-English Alignment in Foundation Language Model
by Runzhe Zhan, Xinyi Yang, Derek F. Wong, Lidia S. Chao, Yue Zhang
First submitted to arxiv on: 25 Apr 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 a novel training-free alignment method named PreTTY to bridge the gap between large language models (LLMs) and specific preferences, specifically in cross-lingual generation tasks. The authors critique the hypothesis that supervised fine-tuning (SFT) is merely “superficial” and demonstrate that SFT may be constrained by its reliance on prior tokens. They achieve comparable performance to SFT without training by employing minimal task-related prior tokens. Experiments on machine translation and part-of-speech tagging across eight languages show the efficacy of PreTTY in cross-lingual settings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making language models work better for different languages. Right now, we use a way called supervised fine-tuning (SFT) to make these models do what we want. But some people think that SFT isn’t really making the model understand what we want it to do, just doing what we tell it to do. The authors of this paper looked into this and found that SFT might not be as good as we thought because it relies on words that come before the ones we’re trying to translate. They came up with a new way called PreTTY that doesn’t need training and can make language models work better for different languages. |
Keywords
» Artificial intelligence » Alignment » Fine tuning » Supervised » Translation
