Non-Autoregressive Semantic Parsing for Compositional Task-Oriented Dialog
This work addresses the problem of real-time conversational AI for developers by reducing latency in semantic parsing, though it is incremental as it builds on existing non-autoregressive and seq2seq methods.
The authors tackled the high latency and compute requirements of sequence-to-sequence models for semantic parsing in task-oriented dialog by proposing a non-autoregressive approach with convolutional neural networks, achieving up to an 81% reduction in latency on the TOP dataset while maintaining competitive performance.
Semantic parsing using sequence-to-sequence models allows parsing of deeper representations compared to traditional word tagging based models. In spite of these advantages, widespread adoption of these models for real-time conversational use cases has been stymied by higher compute requirements and thus higher latency. In this work, we propose a non-autoregressive approach to predict semantic parse trees with an efficient seq2seq model architecture. By combining non-autoregressive prediction with convolutional neural networks, we achieve significant latency gains and parameter size reduction compared to traditional RNN models. Our novel architecture achieves up to an 81% reduction in latency on TOP dataset and retains competitive performance to non-pretrained models on three different semantic parsing datasets. Our code is available at https://github.com/facebookresearch/pytext