Semantic Parsing in Task-Oriented Dialog with Recursive Insertion-Based Encoder

We introduce a Recursive INsertion-based Encoder (RINE), novel approach for semantic parsing in task-oriented dialog. Our model consists of an encoder network that incrementally builds the parse tree by predicting non-terminal label and its positions linearized tree. At generation time, constructs recursively inserting predicted labels at until termination. RINE achieves state-of-the-art exact match accuracy on low- high-resource versions conversational benchmark TOP, outperforming strong sequence-to-sequence models transition-based parsers. also show our design is applicable to nested named entity recognition task, where it performs par with designed task. Finally, we demonstrate 2-3.5 times faster than inference time.