🤖 AI Summary
Dialogue state tracking (DST) suffers from catastrophic forgetting under continual service addition, as task distribution shifts degrade performance on previously seen services. To address this, we propose a continual-learning-oriented DST reformulation: DST is recast as a fine-grained, context-aware, example-guided question-answering task, effectively mitigating cross-service distribution shift. We introduce, for the first time, a retrieval-augmented in-context learning mechanism that dynamically retrieves relevant historical state examples—thereby avoiding over-reliance on service-specific parameter memorization. Further, we integrate dialogue-level memory replay with joint training of a lightweight 60M-parameter model, eliminating the need for parameter expansion or strong regularization. Evaluated on standard continual DST benchmarks, our method achieves state-of-the-art performance, demonstrating exceptional stability and generalization—particularly in long-term incremental settings.
📝 Abstract
Dialogue systems are frequently updated to accommodate new services, but naively updating them by continually training with data for new services in diminishing performance on previously learnt services. Motivated by the insight that dialogue state tracking (DST), a crucial component of dialogue systems that estimates the user's goal as a conversation proceeds, is a simple natural language understanding task, we propose reformulating it as a bundle of granular example-guided question answering tasks to minimize the task shift between services and thus benefit continual learning. Our approach alleviates service-specific memorization and teaches a model to contextualize the given question and example to extract the necessary information from the conversation. We find that a model with just 60M parameters can achieve a significant boost by learning to learn from in-context examples retrieved by a retriever trained to identify turns with similar dialogue state changes. Combining our method with dialogue-level memory replay, our approach attains state of the art performance on DST continual learning metrics without relying on any complex regularization or parameter expansion methods.
