The dataset is a collection of stories, where each story mostly talks about an entity, also called as main entity. Sentences express triple-like facts of kind (E,R,E), 
i.e Entity Relation Entity.

Stories are collected into a single .txt file properly formatted to gather sentences, supervisions, story identifier and other meta information about a sentence in the story.
Each line contains a sentence, its story identifier, the id of the fact expressed in the text, and the textual form (paired with its supervision id) of the two entities and 
the relation. Each token is separated by the tab.

Here we show an example of a sentence present in the dataset:

HEADER:	        story_id  fact_id  main_entity_id    entity_1/id	  relation/id	       entity_2/id	       sentence
SENTENCE:	    1	    349          76	  [Fresh fruits/76]	{are eaten by/8}	[Dan/29]	Fresh fruits are eaten by Dan.


This is a sentence that belongs to story "1", it states a fact (fact id is "349") which involves entities "76" and "29" linked by relation "8". "76" is also the main entity of story "1". 
"Fresh fruits" is the surface form of entity "76", "are eaten by" is the textual representation of relation "8", "Dan" refers instead to entity instance "29".
