On the Effectiveness of Sentence Encoding for Intent Detection Meta-Learning
21127Authors
nan
Abstract
AbstractRecent studies on few-shot intent detection have attempted to formulate the task as a meta-learning problem, where a meta-learning model is trained with a certain capability to quickly adapt to newly specified few-shot tasks with potentially unseen intent categories. Prototypical networks have been commonly used in this setting, with the hope that good prototypical representations could be learned to capture the semantic similarity between the query and a few labeled instances. This intuition naturally leaves a question of whether or not a good sentence representation scheme could suffice for the task without further domain-specific adaptation. In this paper, we conduct empirical studies on a number of general-purpose sentence embedding schemes, showing that good sentence embeddings without any fine-tuning on intent detection data could produce a non-trivially strong performance. Inspired by the results from our qualitative analysis, we propose a frustratingly easy modification, which leads to consistent improvements over all sentence encoding schemes, including those from the state-of-the-art prototypical network variants with task-specific fine-tuning.
Keywords
Full Content
Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 348 - 374 July 10-15, 2022 ©2022 Association for Computational Linguistics The Why and The How: A Survey on Natural Language Interaction in Visualization Henrik Voigt1, Özge Alacam2, Monique Meuschke3, Kai Lawonn1 and Sina Zarrieß2 1University of Jena 2University of Bielefeld 3University of Magdeburg 1first.last@uni-jena.de 2first.last@uni-bielefeld.de 3last@isg.cs.uni-magdeburg.de Abstract Natural language as a modality of interaction is becoming increasingly popular in the field of visualization. In addition to the popular query interfaces, other language-based interactions such as annotations, recommendations, expla- nations, or documentation experience growing interest. In this survey, we provide an overview of natural language-based interaction in the research area of visualization. We discuss a renowned taxonomy of visualization tasks and classify 119 related works to illustrate the state- of-the-art of how current natural language in- terfaces support their performance. We exam- ine applied NLP methods and discuss human- machine dialogue structures with a focus on initiative, duration, and communicative func- tions in recent visualization-oriented dialogue interfaces. Based on this overview, we point out interesting areas for the future application of NLP methods in the field of visualization. 1 Introduction Natural language as a modality for interacting with visual models enjoys increasing popularity in human-computer interface research in the fields of Human-Computer-Interaction (HCI) and Visu- alization (VIS) (Yu and Silva, 2020; Srinivasan et al., 2020a; Liu et al., 2021; Narechania et al., 2021; Kim et al., 2021c). At the same time, interest in tasks involving the visual modality has grown strongly in NLP research in recent years (Suhr et al., 2017; Hudson and Manning, 2019; Acharya et al., 2019). While there are common interests and paral- lel trends in VIS and NLP, research in these fields often adopts different perspectives on what inter- action is and how it should be modeled. Broadly speaking, in the VIS community, a lot of research aims to understand why users need to interact with a visualization and what users’ intents are when they interact with a visual model (Dimara and Perin, 2020). Therefore, Brehmer and Munzner (2013) categorize users’ data-related intents into visual- ization tasks and introduce a taxonomy to describe them in general terms and compare them among applications. Recent contributions show that dif- ferent forms of natural language-based interaction prove suitable to support users in accomplishing various visualization tasks. This concerns not only the popular query interfaces, but also, on a broader scale, the provision of recommendations, annota- tions, explanations, documentations, or for support in analytical conversation. However, the variety of existing visualization tasks benefiting from natural language interaction beyond simple query inter- faces has not yet received much attention in the NLP community. On the other hand, modern NLP methods offer enormous potential for modeling multi-modal dialogues in the visualization tasks. In this survey, we aim to complement the why- oriented perspective of classifying visualization tasks by intent in VIS with the how-oriented di- alogue modeling perspective in NLP for works …
Content truncated to 500 words
Labeled Text
AbstractRecent studies on few-shot intent detection have attempted to formulate the task as a meta-learning problem, where a meta-learning model is trained with a certain capability to quickly adapt to newly specified few-shot tasks with potentially unseen intent categories. Prototypical networks have been commonly used in this setting, with the hope that good prototypical representations could be learned to capture the semantic similarity between the query and a few labeled instances. This intuition naturally leaves a question of whether or not a good sentence representation scheme could suffice for the task without further domain-specific adaptation. In this paper, we conduct empirical studies on a number of general-purpose sentence embedding schemes, showing that good sentence embeddings without any fine-tuning on intent detection data could produce a non-trivially strong performance. Inspired by the results from our qualitative analysis, we propose a frustratingly easy modification, which leads to consistent improvements over all sentence encoding schemes, including those from the state-of-the-art prototypical network variants with task-specific fine-tuning.
Publication Details
- Paper ID:
- 21127
- Year:
- 2022
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan