Natural Language Processing (NLP) is one of the most captivating applications of Deep Learning. In this survey, we consider how the Data Augmentation training strategy can aid in its development. We begin with the major motifs of Data Augmentation summarized into strengthening local decision boundaries, brute force training, causality and counterfactual examples, and the distinction between meaning and form. We follow these motifs with a concrete list of augmentation frameworks that have been developed for text data. Deep Learning generally struggles with the measurement of generalization and characterization of overfitting. We highlight studies that cover how augmentations can construct test sets for generalization. NLP is at an early stage in applying Data Augmentation compared to Computer Vision. We highlight the key differences and promising ideas that have yet to be tested in NLP. For the sake of practical implementation, we describe tools that facilitate Data Augmentation such as the use of consistency regularization, controllers, and offline and online augmentation pipelines, to preview a few. Finally, we discuss interesting topics around Data Augmentation in NLP such as task-specific augmentations, the use of prior knowledge in self-supervised learning versus Data Augmentation, intersections with transfer and multi-task learning, and ideas for AI-GAs (AI-Generating Algorithms). We hope this paper inspires further research interest in Text Data Augmentation.
- Award ID(s):
- 2105329
- NSF-PAR ID:
- 10482432
- Publisher / Repository:
- Association for Computational Linguistics
- Date Published:
- Journal Name:
- Findings of the Association for Computational Linguistics: EMNLP 2023
- Page Range / eLocation ID:
- 11792 to 11806
- Format(s):
- Medium: X
- Location:
- Singapore
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
Abstract -
more » « less
Abstract Background Real‐world engineering problems are ill‐defined and complex, and solving them may arouse negative epistemic affect (feelings experienced within problem‐solving). These feelings fall into sequenced patterns (affective pathways). Over time, these patterns can alter students' attitudes toward engineering. Meta‐affect (affect or cognition about affect) can shape or reframe affective pathways, changing a student's problem‐solving experience.
Purpose/Hypothesis(es) This paper examines epistemic affect and meta‐affect in undergraduate students solving ill‐defined problems called open‐ended modeling problems (OEMPs), addressing two research questions: What epistemic affect and transitions between different affective states do students report? And, how does meta‐affect shape students' affective experiences?
Design/Method We examined 11 retrospective interviews with nine students performed across two semesters in which students completed OEMPs. Using inductive and deductive coding with discourse analysis, we systematically searched for expressions conveying epistemic affect and for transitions in affect; we performed additional deductive coding of the transcripts for meta‐affect and synthesized these results to formulate narratives related to affect and meta‐affect.
Results Together, the expressions, transitions, and meta‐affect suggest different types of student experiences. Depending on their meta‐affect, students either recounted experiences dominated by positive or negative affect, or else they experienced negative emotions as productive.
Conclusions Ill‐defined complex problems elicit a wide range of positive and negative emotions and provide opportunities to practice affective regulation and productive meta‐affect. Viewing the OEMPs as authentic disciplinary experiences and/or the ability to view negative emotions as productive can enable overall positive experiences. Our results provide insight into how instructors can foster positive affective pathways through problem‐scaffolding or their interactions with students.
-
more » « less
The escalating global energy predicament implores for a revolutionary resolution—one that converts sunlight into electricity—holding the key to supreme conversion efficiency. This comprehensive review embarks on the exploration of the principle of generating multiple excitons per absorbed photon, a captivating concept that possesses the potential to redefine the fundamental confines of conversion efficiency, albeit its application remains limited in photovoltaic devices. At the nucleus of this phenomenon are two principal processes: multiple exciton generation (MEG) within quantum-confined environments, and singlet fission (SF) inside molecular crystals. The process of SF, characterized by the cleavage of a single photogenerated singlet exciton into two triplet excitons, holds promise to potentially amplify photon-to-electron conversion efficiency twofold, thereby laying the groundwork to challenge the detailed balance limit of solar cell efficiency. Our discourse primarily dissects the complex nature of SF in crystalline organic semiconductors, laying special emphasis on the anisotropic behavior of SF and the diffusion of the subsequent triplet excitons in single-crystalline polyacene organic semiconductors. We initiate this journey of discovery by elucidating the principles of MEG and SF, tracing their historical genesis, and scrutinizing the anisotropy of SF and the impact of quantum decoherence within the purview of functional mode electron transfer theory. We present an overview of prominent techniques deployed in investigating anisotropic SF in organic semiconductors, including femtosecond transient absorption microscopy and imaging as well as stimulated Raman scattering microscopies, and highlight recent breakthroughs linked with the anisotropic dimensions of Davydov splitting, Herzberg–Teller effects, SF, and triplet transport operations in single-crystalline polyacenes. Through this comprehensive analysis, our objective is to interweave the fundamental principles of anisotropic SF and triplet transport with the current frontiers of scientific discovery, providing inspiration and facilitating future ventures to harness the anisotropic attributes of organic semiconductor crystals in the design of pioneering photovoltaic and photonic devices.
-
Community organizers build grassroots power and collective voice in communities that are structurally marginalized in representative democracy, particularly in minoritized communities. Our project explores how self-identified community organizers use the narrative potentials of data to navigate the promises of data activism and the simultaneous risks posed to working-class communities of color by data-intensive technologies. Our nine respondents consistently named the material, financial, intellectual, and affective demands of data work, as well as the provisional, tenuous possibility of accomplishing movement work via narratives bolstered by data. Our early results identified two important factors in community organizers’ assessment of the efficacy and political potential of narratives built with data: audience and legitimacy.more » « less
-
Vincenot, Christian (Ed.)Effectively communicating risk is critical to reducing conflict in human-wildlife interactions. Using a survey experiment fielded in the midst of contentious public debate over flying fox management in urban and suburban areas of Australia, we find that stories with characters (i.e., narratives) are more effective than descriptive information at mobilizing support for different forms of bat management, including legal protection, relocation, and habitat restoration. We use conditional process analysis to show that narratives, particularly with accompanying images, are effective because they cause emotional reactions that influence risk perception, which in turn drives public opinion about strategies for risk mitigation. We find that prior attitudes towards bats matter in how narrative messages are received, in particular in how strongly they generate shifts in affective response, risk perception, and public opinion. Our results suggest that those with warm prior attitudes towards bats report greater support for bat dispersal when they perceive impacts from bats to be more likely, while those with cool priors report greater support for bat protection when they perceive impacts from bats to be more positive, revealing 1) potential opportunities for targeted messaging to boost public buy-in of proposals to manage risks associated with human-wildlife interactions, and 2) potential vulnerabilities to disinformation regarding risk.more » « less
