Update README.md

README.md

@@ -92,7 +92,7 @@ The sentences in the dataset are primarily curated to benchmark the performance
 
 The sentences in the dataset are collected from  publicly available documents. After collection, they are proofread and lightly edited to remove ambiguity and ensure they align with the specific rule being tested.
 
-The dataset contains 60 sentence pairs for each rule. Thus, it is representative of the rules rather than of language in actual use.
+The dataset contains 60 sentence pairs for each rule. Thus, it is representative of the rules rather than of language in actual use. The unacceptable version of the sentences are made by manually removing or adding commas.
 
 ## Generalizability