A Framework for Interactive Natural Language Debugging
Published in Advances in Cognitive Systems, 2022
Natural language systems that use hand-curated linguistic resources have advantages over Machine Learning systems in that their behavior can be examined and incrementally corrected. However, maintaining these systems is a challenge due to the amount of expertise required and the complexity of the debugging process. To address this challenge, we propose Interactive Natural Language Debugging (INLD), a framework for locating and correcting errors in a system’s linguistic resources by interacting with a user in natural language. As part of ongoing research, we present the INLD pipeline, a taxonomy of error types, and a formulation of INLD as a model-based diagnosis problem.
Recommended citation: Nakos, C., Kuthalam, M., & Forbus, K. D. (2022). A Framework for Interactive Natural Language Debugging. In Proceedings of the Tenth Annual Conference on Advances in Cognitive Systems. https://advancesincognitivesystems.github.io/acs2022/data/acs22_paper-7803.pdf