Explaining NLP Models via Minimal Contrastive Editing (MiCE)

By Alexis Ross, Ana Marasovic, Matthew E. Peters. 2020

142 citations recorded by the source corpus.

Humans have been shown to give contrastive explanations, which explain why an observed event happened rather than some other counterfactual event (the contrast case). Despite the influential role that contrastivity plays in how humans explain, this property is largely missing from current methods for explaining NLP models. We present Minimal Contrastive Editing (MiCE), a method for producing contrastive explanations of model predictions in the form of edits to inputs that change model outputs to the contrast case. Our experiments across three tasks--binary sentiment classification, topic classification, and multiple-choice question answering--show that MiCE is able to produce edits that are not only contrastive, but also minimal and fluent, consistent with human contrastive edits. We demonstrate how MiCE edits can be used for two use cases in NLP system development--debugging incorrect model outputs and uncovering dataset artifacts--and thereby illustrate that producing contrastive explanations is a promising research direction for model interpretability.

Subfields: Artificial Intelligence, Information Systems, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

In Natural Language Processing

Citations in the corpus

Tailor: Generating and Perturbing Text with Semantic Controls (2021, methodology): “They usually rely on application-speciﬁc labels (Ross et al., 2021; Madaan et al., 2020b; Sha et al., 2021; Akyürek et al., 2021) or require pairs of original and perturbed sentences (Wu et al., 2021), which are expensive to generalize.”
Polyjuice: Generating Counterfactuals for Explaining, Evaluating, and Improving Models (2021, methodology): “…semantics while changing model predictions (Ribeiro et al., 2018; Iyyer et al., 2018; Li et al., 2021), whereas concurrent work to our own (Madaan et al., 2021; Ross et al., 2020) automatically generates x̂ that change predictions for explanation or analysis, with no constraints on semantics.”
Get Your Vitamin C! Robust Fact Verification with Contrastive Evidence (2021, background): “This can both ensure consistency with external sources of truth, and improve the model’s faithfulness via connecting the verdict with a specific evidence (Jacovi and Goldberg, 2020; Ross et al., 2020).”
Teach Me to Explain: A Review of Datasets for Explainable NLP (2021, background): “For example, Ross et al. (2020) propose contrastive explanations as the minimal edits of the input that are needed to explain a different (contrast) label for a given input-label pair for a classiﬁcation task.”
Prompting Contrastive Explanations for Commonsense Reasoning Tasks (2021, methodology): “Ross et al. (2020) use this idea to generate contrastive explanations, while it has also been used for evaluation (Gardner et al., 2020) and training (Kaushik et al., 2019) with the aim of addressing model robustness.”
Explainability for Large Language Models: A Survey (2023, methodology): “Specifically, MiCE uses gradient-based attribution to select tokens to mask in the first stage and focuses on optimizing for minimal edits through binary search [Ross et al. 2021].”
Combining Feature and Instance Attribution to Detect Artifacts (2021, background): “…existing artifacts and their effects on models (Gururangan et al., 2018), using instance attribution methods to surface artifacts and reduce model bias (Han and Tsvetkov, 2021; Zylberajch et al., 2021), or use artifact detection as a metric to evaluate interpretability methods (Ross et al., 2021).”
Teach Me to Explain: A Review of Datasets for Explainable Natural Language Processing (2021, methodology): “[107] Alexis Ross, Ana Marasović, and Matthew Peters.”