Update README.md

README.md

@@ -176,5 +176,5 @@ Despite originally being intended for Natural Language Inference (NLI), this dat
 ```python
 {'anchor': 'A person on a horse jumps over a broken down airplane.', 'positive': 'A person is outdoors, on a horse.', 'negative': 'A person is at a diner, ordering an omelette.'}
 ```
-* Collection strategy: Reading the SNLI & MultiNLI datasets, for each "premise" making a list of entailing and contradictory sentences using the dataset labels. Then, considering all possible triplets out of these entailing and contradictory lists. The reverse ("entailment" as "anchor" and "premise" as "positive") is also included.
+* Collection strategy: Reading the SNLI & MultiNLI datasets, for each "premise" making a list of entailing and contradictory sentences using the dataset labels. Then, considering all possible triplets out of these entailing and contradictory lists. The reverse ("entailment" as "anchor" and "premise" as "positive") is not included.
 * Deduplified: Yes