Summary of Pomp: Probability-driven Meta-graph Prompter For Llms in Low-resource Unsupervised Neural Machine Translation, by Shilong Pan et al.
POMP: Probability-driven Meta-graph Prompter for LLMs in Low-resource Unsupervised Neural Machine Translation
by Shilong Pan, Zhiliang Tian, Liang Ding, Zhen Huang, Zhihua Wen, Dongsheng Li
First submitted to arxiv on: 11 Jan 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 This research paper focuses on improving unsupervised neural machine translation (UNMT) for low-resource languages (LRLs). UNMT methods like back-translation and transfer learning have limitations due to synthetic data noise, language bias, and error propagation. Large Language Models (LLMs) can enhance NMT with in-context learning (ICL) and supervised fine-tuning, but they require sufficient training data, which is a challenge for LRLs. The authors propose the Probability-driven Meta-graph Prompter (POMP), an innovative approach that employs multiple auxiliary languages to mitigate linguistic noise and improve translations during training. POMP constructs a dynamic graph of auxiliary languages and samples paths to prompt LLMs. Rewards are estimated by scores and used to update probabilities, demonstrating significant improvements in translation quality for three LRLs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research paper is about making it easier to translate texts from low-resource languages without having any training data. Currently, this process can be noisy and biased towards certain languages. The authors propose a new method called POMP that uses multiple auxiliary languages to help machines learn and improve translations. POMP helps reduce noise and bias by sampling paths through these auxiliary languages. The results show significant improvements in translation quality for three low-resource languages. |
Keywords
» Artificial intelligence » Fine tuning » Probability » Prompt » Supervised » Synthetic data » Transfer learning » Translation » Unsupervised
