Counterfactually-Augmented SNLI Training Data Does Not Yield Better Generalization Than Unaugmented Data

By William Huang, Haokun Liu, Samuel R. Bowman. 2020

41 citations recorded by the source corpus.

A growing body of work shows that models exploit annotation artifacts to achieve state-of-the-art performance on standard crowdsourced benchmarks—datasets collected from crowdworkers to create an evaluation task—while still failing on out-of-domain examples for the same task. Recent work has explored the use of counterfactually-augmented data—data built by minimally editing a set of seed examples to yield counterfactual labels—to augment training data associated with these benchmarks and build more robust classifiers that generalize better. However, Khashabi et al. (2020) find that this type of augmentation yields little benefit on reading comprehension tasks when controlling for dataset size and cost of collection. We build upon this work by using English natural language inference data to test model generalization and robustness and find that models trained on a counterfactually-augmented SNLI dataset do not generalize better than unaugmented datasets of similar size and that counterfactual augmentation can hurt performance, yielding models that are less robust to challenge examples. Counterfactual augmentation of natural language understanding data through standard crowdsourcing techniques does not appear to be an effective way of collecting training data and further innovation is required to make this general line of work viable.

Subfields: Artificial Intelligence, Computer Science Applications, 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 Computer Science Applications

In Natural Language Processing

In Evaluation & Benchmarks

Citations in the corpus

Polyjuice: Generating Counterfactuals for Explaining, Evaluating, and Improving Models (2021, background): “, 2018), whereas Polyjuice can miss more complex or context-dependent changes, and could benefit from target perturbations that compensate for its lack of domain knowledge (targeted guidance is also helpful for human annotators (Huang et al., 2020)).”
Dynabench: Rethinking Benchmarking in NLP (2021, background): “necessarily lead to better generalization (Huang et al., 2020).”
Analyzing Dynamic Adversarial Training Data in the Limit (2021, background): “, 2021b), encouraging counterfactual examples (Kaushik et al., 2020; Huang et al., 2020; Gardner et al., 2020), or providing prompts that workers can edit (Bowman et al.”
Competency Problems: On Finding and Removing Artifacts in Language Data (2021, background): “This suggests that care must be taken when performing edit-based data augmentation, as large edited training datasets are not likely to be artifact-free (cf. Tafjord et al., 2019; Huang et al., 2020).”
Analyzing Stereotypes in Generative Text Inference Tasks (2021, background): “There has simultaneously been a range of work aimed to mitigate problems of stereotyping in NLP systems, including many in the space of text generation (Sheng et al., 2020; He et al., 2019; Clark et al., 2019; Huang et al., 2020).”