Home › Categories › Technology› Retrieval Augmentation Reduces Hallucination In Conversation

Retrieval Augmentation Reduces Hallucination In Conversation.pdf

2021.findings-emnlp.320.pdf

Preview of Retrieval Augmentation Reduces Hallucination in Conversation

🔗 Source: aclanthology.org

📊 Size: 482 KB

👤 Author: Kurt Shuster ; Spencer Poff ; Moya Chen ; Douwe Kiela ; Jason Weston

⬇️ Downloads: 153

Summary

Researchers from Facebook AI Research explored the use of neural-retrieval-in-the-loop architectures for knowledge-grounded dialogue, a task that requires querying based on complex multi-turn dialogue context and generating conversationally coherent responses. They studied various types of architectures with multiple components, including retrievers, rankers, and encoder-decoders, to maximize knowledgeability while retaining conversational ability. The best models obtained state-of-the-art performance on two knowledge-grounded conversational tasks, substantially reducing the problem of knowledge hallucination in state-of-the-art chatbots by over 60%. The approach was shown to be effective in reducing hallucinated responses, especially on out-of-distribution topics and test data, with knowledgeability metric gains of 70% for in-distribution data and 85% for out-of-distribution data.

Download File

Description

Researchers from Facebook AI Research explored the use of neural-retrieval-in-the-loop architectures for knowledge-grounded dialogue, a task that requires...

Technical Information

File Format: PDF
File Size: 482 KB
Pages: 20
Language: EN
Author: Kurt Shuster ; Spencer Poff ; Moya Chen ; Douwe Kiela ; Jason Weston
Total Downloads: 153
Last Updated: 4 hours ago

Document Overview

This PDF document about Retrieval Augmentation Reduces Hallucination in Conversation provides comprehensive information and guidance. Whether you're a beginner or advanced user, this resource offers valuable insights into Retrieval Augmentation Reduces Hallucination in Conversation.