A survey on improving NLP models with human explanations

Abstract

Training a model with access to human explanations can improve data efficiency and model performance on in-and out-of-domain data. Adding to these empirical findings, similarity with the process of human learning makes learning from explanations a promising way to establish a fruitful human-machine interaction. Several methods have been proposed for improving natural language processing (NLP) models with human explanations, that rely on different explanation types and mechanism for integrating these explanations into the learning process. These methods are rarely compared with each other, making it hard for practitioners to choose the best combination of explanation type and integration mechanism for a specific use-case. In this paper, we give an overview of different methods for learning from human explanations, and discuss different factors that can inform the decision of which method to choose for a specific usecase.

Teach Me How to Argue: A Survey on NLP Feedback Systems in Argumentation

“…Several surveys have been done in the field of argumentation (Ke and Ng, 2019;Habernal and Gurevych, 2016;Lawrence and Reed, 2020; and explainability (Danilevsky et al, 2020;Islam et al, 2021;Hartmann and Sonntag, 2022). As we would like to focus on how well a model can explain its results as a type of feedback for learners, we present here recent surveys related to feedback or explainability in argumentation.…”

XMD: An End-to-End Framework for Interactive Explanation-Based Debugging of NLP Models

“…Explanation-Based Model Debugging Many works have explored explanation-based debugging of NLP models, mainly differing in how model behavior is explained, how HITL feedback is provided, and how the model is updated (Lertvit-tayakumjorn and Toni, 2021;Hartmann and Sonntag, 2022;Balkir et al, 2022). Model behavior can be explained using instance (Idahl et al, 2021;Koh and Liang, 2017;Ribeiro et al, 2016) or task (Lertvittayakumjorn et al, 2020;Ribeiro et al, 2018) explanations, typically via feature importance scores.…”

“…On the other hand, explanation-based model debugging focuses on addressing spurious biases that actually influenced the given model's decisionmaking (Smith-Renner et al, 2020;Hartmann and Sonntag, 2022). In this paradigm, a human-in-the-loop (HITL) user is given explanations of the model's behavior (Sundararajan et al, 2017;Shrikumar et al, 2017) and asked to provide feedback about the behavior.…”