Think about it! Improving defeasible reasoning by first modeling the question scenario

Defeasible reasoning is the mode of reasoning where conclusions can be\noverturned by taking into account new evidence. Existing cognitive science\nliterature on defeasible reasoning suggests that a person forms a mental model\nof the problem scenario before answering questions. Our research goal asks\nwhether neural models can similarly benefit from envisioning the question\nscenario before answering a defeasible query. Our approach is, given a\nquestion, to have a model first create a graph of relevant influences, and then\nleverage that graph as an additional input when answering the question. Our\nsystem, CURIOUS, achieves a new state-of-the-art on three different defeasible\nreasoning datasets. This result is significant as it illustrates that\nperformance can be improved by guiding a system to "think about" a question and\nexplicitly model the scenario, rather than answering reflexively. Code, data,\nand pre-trained models are located at https://github.com/madaan/thinkaboutit.\n

Paper

Similar papers

© 2026 NYSGPT2525 LLC