Retrieve-and-Fill for Scenario-based Task-Oriented Semantic Parsing

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 430–447 May 2-6, 2023 ©2023 Association for Computational Linguistics Retrieve-and-Fill for Scenario-based Task-Oriented Semantic Parsing Akshat Shrivastava Shrey Desai Anchit Gupta Ali Elkahky Aleksandr Livshits Alexander Zotov Ahmed Aly Meta {akshats, shreyd, alielk, anchit, alll, alexzotov, ahhegazy}@meta.com Abstract Task-oriented semantic parsing models have achieved strong results in recent years, but un- fortunately do not strike an appealing balance between model size, runtime latency, and cross- domain generalizability. We tackle this prob- lem by introducing scenario-based semantic parsing: a variant of the original task which sep- arates disambiguating an utterance’s “scenario” (an intent-slot template with variable leaf spans) and generating its frame, complete with on- tology and utterance tokens. This formula- tion closely ties to the data collection process where the scenarios are first designed followed by crowd sourced utterance annotation. Con- cretely, we create a Retrieve-and-Fill (RAF) ar- chitecture comprised of (1) a retrieval module which ranks the best scenario given an utter- ance and (2) a filling module which imputes spans into the scenario to create the frame. Our model is modular, differentiable, interpretable, and allows us to garner extra supervision from scenarios. RAF achieves strong results in high- resource, low-resource, and multilingual set- tings, outperforming recent approaches despite, using base pre-trained encoders and efficient decoding. 1 Introduction Task-oriented conversational assistants typically first use semantic parsers to map textual utterances into structured frames for language understand- ing (Hemphill et al., 1990; Coucke et al., 2018; Gupta et al., 2018; Rongali et al., 2020; Agha- janyan et al., 2020). While these parsers achieve strong performance with rich supervision, they of- ten face obstacles adapting to novel settings, es- pecially ones with distinct semantics and scarce data. Recent approaches address this by improv- ing parsers’ data efficiency, such as by using pre- trained representations (Aghajanyan et al., 2020; Rongali et al., 2020), optimizing loss functions (Chen et al., 2020b), and supplying natural lan- guage prompts (Desai et al., 2021). However, these approaches typically rely on larger models and longer contexts, impeding their applicability in real-world conversational assistants. Even though non-autoregressive models can alleviate some con- cerns (Babu et al., 2021; Shrivastava et al., 2021; Zhu et al., 2020), to the best of our knowledge, there exists no approach which strikes an appealing balance between model size, runtime latency, and cross-domain generalizability. We begin tackling this problem by introducing scenario-based task-oriented semantic parsing which is more closely tied to how new task do- mains are developed. A slight variation on orig- inal semantic parsing, we are given access to all supported scenarios apriori and have to parse the utterance given this scenario bank. Here, a sce- nario is akin to a incomplete frame; it is, precisely, an intent-slot template with variables as leaf spans (e.g., IN:GET_WEATHER [SL:LOCATION x1 ] ]), indicating it maps to a family of linguistically sim- ilar utterances. As domain development and data- collection usually starts out with designing the set of supported …

Content truncated to 500 words