SynthWorlds: Controlled Parallel Worlds for Disentangling Reasoning and Knowledge in Language Models
This addresses the challenge of accurately evaluating reasoning in language models for AI researchers, though it is incremental as it builds on existing disentanglement methods.
The paper tackles the problem of disentangling reasoning ability from factual knowledge in language models by introducing SynthWorlds, a framework that constructs parallel corpora with identical structure but different knowledge relevance, and finds a persistent performance gap due to memorized knowledge in experiments on tasks like multi-hop question answering.
Evaluating the reasoning ability of language models (LMs) is complicated by their extensive parametric world knowledge, where benchmark performance often reflects factual recall rather than genuine reasoning. Existing datasets and approaches (e.g., temporal filtering, paraphrasing, adversarial substitution) cannot cleanly separate the two. We present SynthWorlds, a framework that disentangles task reasoning complexity from factual knowledge. In SynthWorlds, we construct parallel corpora representing two worlds with identical interconnected structure: a real-mapped world, where models may exploit parametric knowledge, and a synthetic-mapped world, where such knowledge is meaningless. On top of these corpora, we design two mirrored tasks as case studies: multi-hop question answering and page navigation, which maintain equal reasoning difficulty across worlds. Experiments in parametric-only (e.g., closed-book QA) and knowledge-augmented (e.g., retrieval-augmented) LM settings reveal a persistent knowledge advantage gap, defined as the performance boost models gain from memorized parametric world knowledge. Knowledge acquisition and integration mechanisms reduce but do not eliminate this gap, highlighting opportunities for system improvements. Fully automatic and scalable, SynthWorlds provides a controlled environment for evaluating LMs in ways that were previously challenging, enabling precise and testable comparisons of reasoning and memorization.