Update README.md

README.md

@@ -50,8 +50,8 @@ More information needed
 
 ## Training and evaluation data
 
-Subset of nlpaueb/finer-139 split (roughly 16% of full train split) and and full validation split.
-Given the original proportion of 'O' labels (~80%), we reduced the train split records to match the original majority class proportion given the reduced set of labels.
+Subset of nlpaueb/finer-139 split and and full validation split.
+Given the original proportion of 'O' labels (approx. 80%), we reduced the train split records to match the original majority class proportion given the reduced set of labels.
 Original train split size: 900384 records
 Subset train split size: 142513 records (~16% of original amount of records)