Diversify Your Datasets: Analyzing Generalization via Controlled Variance in Adversarial Datasets
This work addresses the need for more robust NLP models by improving challenge dataset design, though it is incremental as it builds on prior adversarial dataset methods.
The authors tackled the problem of models overfitting to specific adversarial datasets by controlling additional variance aspects, such as syntactic complexity, to test generalization on phenomena like dative alternation and numerical reasoning, showing that models can still fail on differently distributed challenge datasets.
Phenomenon-specific "adversarial" datasets have been recently designed to perform targeted stress-tests for particular inference types. Recent work (Liu et al., 2019a) proposed that such datasets can be utilized for training NLI and other types of models, often allowing to learn the phenomenon in focus and improve on the challenge dataset, indicating a "blind spot" in the original training data. Yet, although a model can improve in such a training process, it might still be vulnerable to other challenge datasets targeting the same phenomenon but drawn from a different distribution, such as having a different syntactic complexity level. In this work, we extend this method to drive conclusions about a model's ability to learn and generalize a target phenomenon rather than to "learn" a dataset, by controlling additional aspects in the adversarial datasets. We demonstrate our approach on two inference phenomena - dative alternation and numerical reasoning, elaborating, and in some cases contradicting, the results of Liu et al.. Our methodology enables building better challenge datasets for creating more robust models, and may yield better model understanding and subsequent overarching improvements.