Summary of Where Does In-context Translation Happen in Large Language Models, by Suzanna Sia et al.
Where does In-context Translation Happen in Large Language Models
by Suzanna Sia, David Mueller, Kevin Duh
First submitted to arxiv on: 7 Mar 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 investigates the self-supervised learning abilities of large language models in performing Machine Translation (MT) tasks. Specifically, it aims to identify the region where these models transition from being in-context learners to actual translation models. To achieve this, the authors conduct layer-wise context-masking experiments on four different models: GPTNeo2.7B, Bloom3B, Llama7b, and Llama7b-chat. The results show a “task recognition” point where the MT task is encoded into input representations, making attention to context unnecessary. This discovery has significant implications for computational efficiency, with a 45% reduction in computational resources required when prompting with five examples. The study also highlights the importance of layer-wise fine-tuning, particularly in the layers critical to task recognition. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how big language models can learn to translate languages without being specifically trained for that job. They found a special point where the model figures out what it’s supposed to do and starts doing the translation itself. To find this point, they tested different parts of the model with different tasks. The results show that once the model knows what to do, it doesn’t need as much information from its surroundings. This discovery can help make the models more efficient by reducing the amount of computer power needed for translations. |
Keywords
» Artificial intelligence » Attention » Fine tuning » Prompting » Self supervised » Translation
