The use of argumentation in education has shown improvement in students' critical thinking skills, and computational models for argumentation have been developed to further assist this process. Although these models are useful for evaluating the quality of an argument, they often cannot explain why a particular argument score was predicted, i.e., why the argument is good or bad, which makes it difficult to provide constructive feedback to users, e.g., students, so that they can strengthen their critical thinking skills.
In this survey, we explore current NLP feedback systems by categorizing each into four important dimensions of feedback (Richness, Visualization, Interactivity and Personalization). We discuss limitations for each dimension and provide suggestions to enhance the power of feedback and explanations to ultimately improve user critical thinking skills.
Feel free to explore, on our website, the different references we found for each dimension. For more information please check the original paper published at the 10th Workshop on Argument Mining (ArgMining 2023) as well as the slides and the poster presented during the conference.
Level of feedback details given by a model, i.e., what is the error identified by the model and why it is an error.
Model's ability to present feedback, i.e., how the feedback is shown to the end user.
Model's ability to allow the user to communicate with other users or the model itself, i.e., with whom the user is talking.
Model's ability to adapt the feedback to the users' background, i.e., to whom the feedback is given.
The following figure shows possible feedback highlighting each dimension for a given argument consisting of two claims and one premise.
In this example:
In the Richness dimension, a faulty generalization in the argument is identified (cf. What) and explained (cf. Why).
In the Visualization dimension, symbols, highlights and other important visual elements are added to the feedback to make it more understandable.
In the Interactivity dimension, the user can ask for more explanations to the model.
In the Personalization dimension, the user is a child and receives appropriate feedback regarding their profile.
We hope our survey contributes to enriching the research community focused on argumentation with a comprehensive understanding of current perspectives in NLP systems for teaching how to argue.
As potential areas for improvement to enhance the quality of educational argumentative systems, we highlighted the following points:
(1) generate accurate, constructive feedback for a real-life input,
(2) tailor the output based on the user's background,
(3) evaluate and compare end-to-end systems more deeply,
(4) improve models' abilities to adapt to unknown topics,
(5) collaborate with pedagogical teams and actual students, and finally
(6) take into consideration ethical issues.
Please refer to Figure 2 and the original paper for more information.