LLMASP: Synthesising Answer Set Programs with LLMs
Abstract
Answer Set Programming (ASP) is a declarative programming framework. It is a powerful tool for solving complex logical and combinatorial problems, with many modern, real-world applications. However, its power is limited by a phenomenon known as the knowledge acquisition bottleneck, which is the difficulty in translating informal, domain-specific knowledge into the rigorous syntax required.
Meanwhile, Large Language Models (LLMs) have demonstrated efficacy in a wide range of natural language processing tasks, including generating code in imperative languages. Thus, we have identified an opportunity to utilise LLMs to mitigate the knowledge acquisition bottleneck, with the end goal of making ASP accessible to non-experts and accelerating the encoding process in general.
The LLMASP group has investigated solving this problem with a modular pipeline for translating a natural language problem into a corresponding Answer Set Program. The group investigates the efficacy of fine-tuning, chain-of-thought prompting, and few-shot prompting as LLM techniques, which have been divided between team members.
Videos 1
Watch presentations, demos, and related content
Like, comment, and subscribe on YouTube to support the creator!
Gallery 1
Explore the visual story of this exhibit
Title Image