The name of each file corresponds to the name of each sample (set of instances) in the paper
"Behavior Analysis of NLI Models: Uncovering the Influence of Three Factors on Robustness"
used to analyze the robustness of the NLI models proposed.

Each file has the following structure:

gold_label [separation_marker] premise [separation_marker] hypothesis [separation_maker] word1 [separation_maker] word2

Where: 

	'gold_label' is the type of relationship between a premise sentence and a hypothesis sentence ('neutral', 'entailment', or 'contradiction')

	[separation_maker] is a tab separating two fields

	'premise' is a premise sentence

	'hypothesis' is a hypothesis sentence

	'word1' and 'word2' form a target word pair as described in the above paper
	(in the case of a control sample this word pair is in its original form, while in the case of a transformed sample this word pair is already swapped)


NOTE: two files have a slightly different file structure, namely 'I_TA2.txt' and 'I_TA3.txt', where the last two fields ('word1' and 'word2') are not included.
