Finding the Cracks: Improving LLMs Reasoning with Paraphrastic Probing and Consistency Verification
This work addresses a key limitation in LLM reasoning for complex tasks, offering a method to reduce errors and hallucinations, though it is incremental as it builds on prior critical token concepts.
The paper tackles the problem of large language models' declining reasoning performance on complex tasks due to hallucinations and error accumulation by proposing the Paraphrastic Probing and Consistency Verification (PPCV) framework, which identifies critical tokens and verifies consistency to improve reasoning, resulting in substantial performance enhancements across multiple benchmarks.
Large language models have demonstrated impressive performance across a variety of reasoning tasks. However, their problem-solving ability often declines on more complex tasks due to hallucinations and the accumulation of errors within these intermediate steps. Recent work has introduced the notion of critical tokens--tokens in the reasoning process that exert significant influence on subsequent steps. Prior studies suggest that replacing critical tokens can refine reasoning trajectories. Nonetheless, reliably identifying and exploiting critical tokens remains challenging. To address this, we propose the Paraphrastic Probing and Consistency Verification~(PPCV) framework. PPCV operates in two stages. In the first stage, we roll out an initial reasoning path from the original question and then concatenate paraphrased versions of the question with this reasoning path. And we identify critical tokens based on mismatches between the predicted top-1 token and the expected token in the reasoning path. A criterion is employed to confirm the final critical token. In the second stage, we substitute critical tokens with candidate alternatives and roll out new reasoning paths for both the original and paraphrased questions. The final answer is determined by checking the consistency of outputs across these parallel reasoning processes. We evaluate PPCV on mainstream LLMs across multiple benchmarks. Extensive experiments demonstrate PPCV substantially enhances the reasoning performance of LLMs compared to baselines.