FOLLIE (First-Order Logic for Language Inference and Entailment) is the first dataset for French with natural language sentences and their corresponding first-order logic (FOL) formulas. The natural language sentences in the dataset were drawn from all the existing French Natural Language Inference (NLI) datasets, namely DACCORD, FraCaS-FR, GQNLI-FR, RTE3-FR (dev and test), SICK-FR, and XNLI (dev and test). The resulting FOLLIE dataset not only provides logical representations of the sentences but also preserves the original sentence-pair structure and the gold label (entailment, neutral, or contradiction) for each pair. It can therefore be used either for natural language–to–FOL translation tasks or for NLI/RTE tasks using neurosymbolic approaches (i.e., solving NLI by reasoning over FOL formulas). The logical formulas in the dataset are fully compatible with the NLTK Logic Parser.

Creator(s)

Maximos Skandalis

Distributor(s)

Maximos Skandalis

Right Holder(s)

Maximos Skandalis