How to Use Theory to Implement Natural Language Processing for Peer-Feedback
Martin Greisel, Elisabeth Bauer, Ilia Kuznetsov, Markus Berndt, Markus Dresel, Martin Rudolf Fischer, Ingo Kollar, Frank Fischer · Computer-supported collaborative learning/The Computer-Supported Collaborative Learning Conference · 2023
Whenever learners produce text, natural language processing (NLP) has great potential to improve learning.Theories from learning sciences should guide the implementation of NLP into concrete learning scenarios.However, theoretical concepts are much more abstract than the targets and inputs NLP can work with.Therefore, a process is needed which translates theory into NLP tasks.As such a process is missing, we propose a terminological and procedural scheme which researchers and practitioners can employ to develop NLP-based adaptive support measures for learning processes.It defines a sequence of leverage points, support measures, adaptation targets, automation goals, data, prediction targets, input, intrinsic metrics, NLP model, and extrinsic metrics.To illustrate it, we apply it to peer-feedback as a use case. Problem statementRecent developments in artificial intelligence (AI) promise to foster various learning processes (Dawson et al., 2019).Yet, to do so, theories of learning are indispensable (Wise & Shaffer, 2015).For example, peer-feedback is assumed to boost learning because, among other reasons, it provides feedback which can be easier to understand than teacher feedback and constitutes an additional learning opportunity (Li et al., 2020).However, for this purpose, learners should provide high-quality feedback (Patchan et al., 2016), but not all learners have the necessary prior knowledge and skills.Consequently, AI might support the learners when composing their feedback.As feedback often is written text, natural language processing (NLP) is the most relevant field of AI for this task.Yet, applying NLP to peer-feedback is complex: NLP and learning sciences have their own terminologies and approaches to conceptualize phenomena.Hence, mapping the theoretical learning process to NLP becomes a complex endeavor of synchronizing terminology, goals, and procedures.A guiding framework that helps researchers and practitioners through this development process is missing.For this reason, in this conceptual paper, we propose a terminological and procedural scheme to guide the development of NLP support measures and exemplify it for peer-feedback.