Comparing Test Sets with Item Response Theory

TL;DR

This paper evaluates 29 NLP datasets using Item Response Theory to identify which datasets effectively distinguish among strong models and detect future improvements in natural language understanding.

Contribution

It introduces a uniform evaluation method using IRT on multiple datasets and identifies which datasets are most effective for benchmarking advanced models.

Findings

Quoref, HellaSwag, and MC-TACO are best for distinguishing strong models.

SNLI, MNLI, and CommitmentBank are saturated for current models.

Span selection tasks effectively differentiate model strengths.

Abstract

Recent years have seen numerous NLP datasets introduced to evaluate the performance of fine-tuned models on natural language understanding tasks. Recent results from large pretrained models, though, show that many of these datasets are largely saturated and unlikely to be able to detect further progress. What kind of datasets are still effective at discriminating among strong models, and what kind of datasets should we expect to be able to detect future improvements? To measure this uniformly across datasets, we draw on Item Response Theory and evaluate 29 datasets using predictions from 18 pretrained Transformer models on individual test examples. We find that Quoref, HellaSwag, and MC-TACO are best suited for distinguishing among state-of-the-art models, while SNLI, MNLI, and CommitmentBank seem to be saturated for current strong models. We also observe span selection task format, which is used for QA datasets like QAMR or SQuAD2.0, is effective in differentiating between strong and weak models.