Natural Language Inference (NLI), also known as Recognizing Textual Entailment (RTE)

  • the task of determining the inference relation between two (short, ordered) texts: entailment, contradiction, or neutral (MacCartney and Manning 2008)

Natural Language Inference - Papers with Code

Benchmarking Datasets

Hugging Face đŸ€— Datasets > Natural Language Inference

  • The Stanford Natural Language Inference (SNLI) Corpus
    • 570k human-written English sentence pairs
    • manually labeled for balanced classification
    • labels entailment, contradiction, and neutral
    • benchmark for evaluating representational systems for text (especially including those induced by representation-learning methods) and resource for developing NLP models of any kind
  • MultiNLI
    • Multi-Genre Natural Language Inference (MultiNLI)
    • crowd-sourced
    • 433k sentence pairs
    • annotated with textual entailment information
    • modeled on the SNLI corpus
      • 
differs in that covers a range of genres of spoken and written text
      • 
and supports a distinctive cross-genre generalization evaluation
    • served as basis for the shared task of the RepEval 2017 Workshop at EMNLP in Copenhagen