Summary of Beyond English-centric Llms: What Language Do Multilingual Language Models Think In?, by Chengzhi Zhong et al.
Beyond English-Centric LLMs: What Language Do Multilingual Language Models Think in?
by Chengzhi Zhong, Fei Cheng, Qianying Liu, Junfeng Jiang, Zhen Wan, Chenhui Chu, Yugo Murawaki, Sadao Kurohashi
First submitted to arxiv on: 20 Aug 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 A study investigates whether non-English-centric large language models (LLMs) operate in their respective dominant languages. Specifically, researchers examine how intermediate layer representations, when unembedded into the vocabulary space, exhibit higher probabilities for certain dominant languages during generation. This phenomenon is dubbed “internal latent languages.” The paper explores this concept using various LLMs and evaluation metrics. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models are super smart computers that can understand and generate human-like text. Some of these models were trained on non-English data, like Chinese or Spanish. Researchers wondered if these models think in their dominant language, which is the language they were trained on. They found that when these models generate new text, they tend to use words and phrases more often from their training language. This study looks at how this works and why it matters. |
