ERASER: A Benchmark to Evaluate Rationalized NLP Models

By Jay DeYoung, Sarthak Jain, Nazneen Rajani, Eric P. Lehman, Caiming Xiong, Richard Socher, Byron C. Wallace. 2019

847 citations recorded by the source corpus.

State-of-the-art models in NLP are now predominantly based on deep neural networks that are opaque in terms of how they come to make predictions. This limitation has increased interest in designing more interpretable deep models for NLP that reveal the ‘reasoning’ behind model outputs. But work in this direction has been conducted on different datasets and tasks with correspondingly unique aims and metrics; this makes it difficult to track progress. We propose the Evaluating Rationales And Simple English Reasoning (ERASER a benchmark to advance research on interpretable models in NLP. This benchmark comprises multiple datasets and tasks for which human annotations of “rationales” (supporting evidence) have been collected. We propose several metrics that aim to capture how well the rationales provided by models align with human rationales, and also how faithful these rationales are (i.e., the degree to which provided rationales influenced the corresponding predictions). Our hope is that releasing this benchmark facilitates progress on designing more interpretable NLP systems. The benchmark, code, and documentation are available at https://www.eraserbenchmark.com/

Subfields: Artificial Intelligence, Natural Language Processing, Evaluation & Benchmarks

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 Natural Language Processing

In Evaluation & Benchmarks

Citations in the corpus

Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategies (2021, background): “, text snippets extracted from the context (DeYoung et al., 2020; Yang et al., 2018; Kwiatkowski et al., 2019; Khot et al., 2020a) as evidence for an answer.”
ProoFVer: Natural Logic Theorem Proving for Fact Verification (2021, background): “Plausibility: Rationales extracted based on attention scores are often used as means to highlight the reasoning involved in the decision making process of various models (DeYoung et al., 2020).”
SELFEXPLAIN: A Self-Explaining Architecture for Neural Text Classifiers (2021, background): “, 2018) and recently, Eraser benchmark (DeYoung et al., 2020).”
Measuring Association Between Labels and Free-Text Rationales (2020, background): “, 2020; Jacovi and Goldberg, 2020a) and evaluations (DeYoung et al., 2020) of explanatory methods with faithfulness as a goal focus on extractive rationales and rely on the suffi-”
WT5?! Training Text-to-Text Models to Explain their Predictions (2020, methodology): “Instead, we refer the interested reader to the surveys provided by (Doshi-Velez & Kim, 2017; Molnar, 2019; Guidotti et al., 2018). Most work on interpretability methods focuses on models for computer vision applications (e.g. Xu et al. (2015); Zhang & Zhu (2018)), whereas the interpretability of NLP”
HateXplain: A Benchmark Dataset for Explainable Hate Speech Detection (2020, methodology): “We follow the framework in the ERASER benchmark by DeYoung et al. (2020) to measure the explainability aspect of a model.”
Towards Faithfully Interpretable NLP Systems: How Should We Define and Evaluate Faithfulness? (2020, background): “Indeed, while attention mechanisms have been considered as “inherently in-terpretable” (Ghaeini et al., 2018; Lee et al., 2017), recent work cast doubt regarding their faithfulness (Serrano and Smith, 2019; Jain and Wallace, 2019; Wiegreffe and Pinter, 2019).”
Get Your Vitamin C! Robust Fact Verification with Contrastive Evidence (2021, background): “Current fact veriﬁcation datasets provide sentence-level rationales (DeYoung et al., 2020; Petroni et al., 2020) but do not enforce the model’s verdict to rely on them—leading to a potential discrepancy.”
Explaining Answers with Entailment Trees (2021, background): “While today’s explanation systems are good at providing a sentence or two of supporting evidence (“rationales”) for an answer (DeYoung et al., 2019), they rarely explain the chain of reasoning from what is known to the answer, i.e., how the answer follows, given the evidence – the goal of this work.”
NILE : Natural Language Inference with Faithful Natural Language Explanations (2020, background): “DeYoung et al. (2019) propose metrics to evaluate the faithfulness of such explanations.”
How Do Decisions Emerge across Layers in Neural Models? Interpretation with Differentiable Masking (2020, background): “Human-provided labels, such as human rationales (Camburu et al., 2018; DeYoung et al., 2020), will not help us in demonstrating this, as humans cannot judge if an interpretation is faithful (Jacovi and Goldberg, 2020).”
Formalizing Trust in Artificial Intelligence: Prerequisites, Causes and Goals of Human Trust in AI (2020, methodology): “We illustrate this point with the ERASER benchmark [13] as a case-study (Figure 3).”
Evaluating Explanations: How Much Do Explanations from the Teacher Aid Students? (2020, methodology): “Table 9: Comparing attribution methods as per the sufficiency (lower the better) and comprehensiveness metrics proposed in (DeYoung et al., 2020).”
Cross-Domain Transfer of Generative Explanations Using Text-to-Text Models (2021, methodology): “Another line of work for explainable NLP is based on rationalization pipelines that aim to produce extractive explanations by splitting the prediction process into two subsequent modules; a rationale extractor and a predictor [1,10,25,34].”
Fact or Fiction: Verifying Scientific Claims (2020, methodology): “We develop a baseline for scientiﬁc fact checking by adapting the “BERT-to-BERT” model for “hard” rationale selection presented in DeYoung et al. (2019) for a number of rationalized NLP tasks including F EVER ; this approach is also similar to the fact-checking model presented in Soleimani et al.…”
Prompting Contrastive Explanations for Commonsense Reasoning Tasks (2021, background): “Models that rationalize their decisions by extracting a contiguous subsequence of the input as an explanation (Lei et al., 2016; DeYoung et al., 2020; Paranjape et al., 2020) are inadequate in explaining 1Code is available at https://github.com/ bhargaviparanjape/RAG-X commonsense reasoning tasks…”