Fundamental Reasoning Paradigms Induce Out-of-Domain Generalization in Language Models
This addresses the challenge of enhancing reasoning generalization in language models, which is incremental as it builds on existing research but introduces a systematic exploration of core paradigms.
The study tackled the problem of whether fundamental reasoning paradigms (deduction, induction, abduction) can improve out-of-domain generalization in large language models, finding that their approach led to substantial performance gains of up to 14.60 on realistic tasks.
Deduction, induction, and abduction are fundamental reasoning paradigms, core for human logical thinking. Although improving Large Language Model (LLM) reasoning has attracted significant research efforts, the extent to which the fundamental paradigms induce generalization has yet to be systematically explored. In this study, we shed light on how the interplay between these core paradigms influences LLMs' reasoning behavior. To this end, we first collect a new dataset of reasoning trajectories from symbolic tasks, each targeting one of the three fundamental paradigms, to abstract from concrete world knowledge. Then, we investigate effective ways for inducing these skills into LLMs. We experiment with a battery of methods including simple fine-tuning, and more complex approaches to increase model depth, or transform a dense model to a mixture-of-experts. We comprehensively evaluate induced models on realistic out-of-domain tasks, that are entirely formulated in natural language and contain real-world knowledge. Our results reveal that our approach yields strong generalizability with substantial performance gains (up to $14.60$) across realistic tasks.