Reproduce “Improving RAG Retrieval via Propositional Content Extraction: a Speech Act Theory Approach” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
João Alberto de Oliveira Lima
A method leveraging GPT-4 and Speech Act Theory to extract propositional content from user queries enhances Retrieval-Augmented Generation (RAG) system performance. By transforming queries into simplified, assertive statements, the approach boosts semantic similarity with knowledge base documents, leading to improvements in maximum cosine similarity by up to 7.5% for commissive and 6.9% for expressive speech acts.
Key insights
- Classical linguistic theories, specifically Speech Act Theory, provide a valuable theoretical framework for practical query reformulation, bridging the gap between philosophical understandings of language and modern neural information retrieval.
- Systematically simplifying existing user queries by removing pragmatic markers (rather than generating new text) is an efficient strategy to enhance query-document alignment without the computational overhead or potential for hallucination associated with generative LLMs.
- The efficacy of propositional content extraction varies across speech act types, with the most substantial improvements observed in queries rich in illocutionary or pragmatic markers like commissive, expressive, and indirect speech acts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.