Update README.md

README.md

@@ -19,7 +19,9 @@ This is a simplified version of SQuAD v1 dataset.
 
 <!-- Provide a longer summary of what this dataset is. -->
 The modifications made the original SQuAD v1 dataset:
+* Extract only the needed columns to train a seq2seq model (context, question and answer)
 * Preprocess the text to have only English letters and some basic punctuation marks.
 * Punctuation marks are separated from the adjacent words (i.e: hello! -> hello !)
 * Remove context length greater than 175 tokens (this removes around 20% of the original dataset)
-* Remove questions and answers longer than 30 tokens (this removes around 140 examples from the original dataset)
+* Remove questions and answers longer than 30 tokens (this removes around 140 examples from the original dataset)
+* Add three additional columns to show the word count for each context, question and answer