Breaking NLI Systems with Sentences that Require Simple Lexical Inferences
This work addresses a critical limitation in NLI for AI and NLP researchers, though it is incremental as it focuses on exposing existing weaknesses rather than proposing a new solution.
The authors tackled the problem of natural language inference (NLI) systems failing to generalize to simple lexical inferences by creating a new test set that reveals deficiencies in state-of-the-art models, resulting in substantially worse performance compared to standard benchmarks like SNLI.
We create a new NLI test set that shows the deficiency of state-of-the-art models in inferences that require lexical and world knowledge. The new examples are simpler than the SNLI test set, containing sentences that differ by at most one word from sentences in the training set. Yet, the performance on the new test set is substantially worse across systems trained on SNLI, demonstrating that these systems are limited in their generalization ability, failing to capture many simple inferences.