Summary of Multi-document Grounded Multi-turn Synthetic Dialog Generation, by Young-suk Lee et al.
Multi-Document Grounded Multi-Turn Synthetic Dialog Generation
by Young-Suk Lee, Chulaka Gunasekara, Danish Contractor, Ramón Fernandez Astudillo, Radu Florian
First submitted to arxiv on: 17 Sep 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 introduces a technique for generating synthetic dialogues that incorporates three main ideas. It uses taxonomy-driven user queries generated with Chain-of-Thought (CoT) prompting to control the overall dialog flow, and mimics real-world use of retrievers to update grounding documents after every user-turn. The technique also applies LLM-as-a-Judge to filter out queries with incorrect answers. Human evaluation suggests that the synthetic dialog data is diverse, coherent, and includes mostly correct answers. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper creates a new way to generate conversations by controlling what people say next using taxonomy-driven prompts and updating documents after each turn. It also filters out wrong answers using large language models as judges. The generated dialogue data is found to be good quality, with humans rating it high on being coherent and correct. |
Keywords
» Artificial intelligence » Grounding » Prompting
