Generalized Quantifiers as a Source of Error in Multilingual NLU Benchmarks
This addresses errors in NLU benchmarks for researchers and developers, but it is incremental as it builds on existing theories and benchmarks.
The paper tackles the problem of natural language understanding (NLU) models struggling with quantifier semantics, finding that quantifiers are pervasive in benchmarks and cause performance drops, with multilingual models showing poor reasoning abilities but not necessarily worse for non-English languages.
Logical approaches to representing language have developed and evaluated computational models of quantifier words since the 19th century, but today's NLU models still struggle to capture their semantics. We rely on Generalized Quantifier Theory for language-independent representations of the semantics of quantifier words, to quantify their contribution to the errors of NLU models. We find that quantifiers are pervasive in NLU benchmarks, and their occurrence at test time is associated with performance drops. Multilingual models also exhibit unsatisfying quantifier reasoning abilities, but not necessarily worse for non-English languages. To facilitate directly-targeted probing, we present an adversarial generalized quantifier NLI task (GQNLI) and show that pre-trained language models have a clear lack of robustness in generalized quantifier reasoning.