Onto sentences Dataset is a collection of 1000 manually created sentences, supervised at word level according to a predefined ontology.
Data is split in 2 JSON files: train and test.
In each file it is indicated the number of sentences, how many distinct tokens are present in the corpus, and information about the ontology, where narrow and broad classes are considered separately. 
All the examples (JSON field "data") are collected in an array. Each element contains a list of words composing the sentence, and 2 arrays with narrow and broad targets respectively.
A target is a triple [l, b, e] where l is the label, while b (beginning) and e (end) delimit the interval of words to supervise with l.
