Summary of Cross-lingual Conversational Speech Summarization with Large Language Models, by Max Nelson et al.
Cross-Lingual Conversational Speech Summarization with Large Language Models
by Max Nelson, Shannon Wotherspoon, Francis Keith, William Hartmann, Matthew Snover
First submitted to arxiv on: 12 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 new approach to cross-lingual conversational speech summarization is proposed, which builds upon existing resources such as the Fisher and Callhome Spanish-English Speech Translation corpus. The summaries are generated using GPT-4 from reference translations and treated as ground truth. A baseline cascade-based system is built using open-source speech recognition and machine translation models. LLMs are tested for summarization, and the impact of transcription and translation errors is analyzed. The adapted Mistral-7B model outperforms off-the-shelf models and matches GPT-4’s performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Cross-lingual conversational speech summarization is a big challenge because we don’t have enough good resources to help us learn. There are some transcriptions of conversations in different languages, but there aren’t any summaries. We took an existing dataset and added summaries to it. These summaries were made using GPT-4 from the original translations, and they’re considered correct. The goal is to make similar summaries even when there are errors in the transcription or translation. We built a basic system that uses free speech recognition and machine translation tools. Then, we tested different large language models (LLMs) to see how well they do at summarizing. When we adapted one of these models, called Mistral-7B, it did much better than the other models and was just as good as GPT-4. |
Keywords
» Artificial intelligence » Gpt » Summarization » Translation
