ETD

• natural language inference
• natural language processing
• predicate argument structure
• probing task
• transformers

The thesis describes the creation of an English dataset built to analyze the inference relation between pairs of sentences. Specific linguistic phenomena related to the predicate-argument structures of sentences play a key role in defining the relation. The dataset has been used to address Natural Language Inference probing tasks whose main goal is to analyze what kind of information some state-of-the-art Transformers models are able to grasp.