Translational NLP: A New Paradigm and General Principles for Natural Language Processing Research

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4125–4138 June 6–11, 2021. ©2021 Association for Computational Linguistics 4125 Translational NLP: A New Paradigm and General Principles for Natural Language Processing Research Denis Newman-Griffis1 Jill Fain Lehman2 Carolyn Rosé3 Harry Hochheiser1 1Department of Biomedical Informatics, University of Pittsburgh, USA 2Human-Computer Interaction Institute, Carnegie Mellon University, USA 3Language Technologies Institute, Carnegie Mellon University, USA {dnewmangriffis, harryh}@pitt.edu, {jfl, cprose}@cs.cmu.edu Abstract Natural language processing (NLP) research combines the study of universal principles, through basic science, with applied science tar- geting specific use cases and settings. How- ever, the process of exchange between ba- sic NLP and applications is often assumed to emerge naturally, resulting in many inno- vations going unapplied and many important questions left unstudied. We describe a new paradigm of Translational NLP, which aims to structure and facilitate the processes by which basic and applied NLP research inform one another. Translational NLP thus presents a third research paradigm, focused on under- standing the challenges posed by application needs and how these challenges can drive in- novation in basic science and technology de- sign. We show that many significant advances in NLP research have emerged from the in- tersection of basic principles with application needs, and present a conceptual framework outlining the stakeholders and key questions in translational research. Our framework pro- vides a roadmap for developing Translational NLP as a dedicated research area, and identi- fies general translational principles to facilitate exchange between basic and applied research. 1 Introduction Natural language processing (NLP) lies at the in- tersection of basic science and applied technolo- gies. However, translating innovations in basic NLP methods to successful applications remains a difficult task in which failure points often appear late in the development process, delaying or pre- venting potential impact in research and industry. Application challenges range widely, from changes in data distributions (Elsahar and Gallé, 2019) to computational bottlenecks (Desai et al., 2020) and integration with domain expertise (Rahman et al., 2020). When unanticipated, such challenges can be fatal to applications of new NLP methodologies, leaving exciting innovations with minimal practical Applications Linguistics Modeling Identifies new research questions Discovers underlying phenomena Guides model development Enables application goals Drive modeling needs and constraints Provides evidence for linguistic theory Translational NLP Figure 1: Interactions between linguistic theory, model development, and applications in NLP research. Solid lines indicate moving from basic research to applica- tions, and dashed lines indicate how applied research feeds back into basic study. Translational NLP devel- ops processes to realize this exchange. impact. Meanwhile, real-world applications may rely on regular expressions (Anzaldi et al., 2017) or unigram frequencies (Slater et al., 2017) when more sophisticated methods would yield deeper in- sight. When successful translations of basic NLP insights into practical applied technologies do oc- cur, the factors contributing to this success are rarely analyzed, limiting our ability to learn how to enable the next project and the next technology. We argue for …

Content truncated to 500 words