LOGICPO: Efficient Translation of NL-based Logical Problems to FOL using LLMs and Preference Optimization
This addresses a bottleneck in logical reasoning for AI applications like question answering, offering an incremental improvement in representation accuracy.
The paper tackles the problem of LLMs incorrectly converting natural language reasoning problems to logical formulations, proposing a finetuning approach with preference optimization that improves logical correctness by 10% and reduces syntax errors by 14% compared to GPT-3.5-turbo.
Logical reasoning is a key task for artificial intelligence due to it's role in major downstream tasks such as Question Answering, Summarization. Recent methods in improving the reasoning ability of LLMs fall short in correctly converting a natural language reasoning problem to an equivalent logical formulation, which hinders the framework's overall ability to reason. Towards this, we propose to use finetuning on a preference optimization dataset to learn to parse and represent a natural language problem as a whole to a consistent logical program by 1) introducing a new supervised and preference optimization dataset LogicPO, and 2) adopting popular techniques such as Direct Preference Optimization (DPO), Kahneman-Tversky optimization (KTO) to finetune open-source LLMs. Our best model with Phi-3.5 consistently outperforms GPT-3.5-turbo's (8-shot) by producing 10% more logically correct and with 14% less syntax errors. Through the framework and our improved evaluation metrics, we offer a promising direction in improving the logical reasoning of LLMs by better representing them in their logical formulations.