ACL_2023_Human_Study_Adversarial_Text.7z

This data is part of ACL 2023 publication "How do humans perceive adversarial text? A reality check on the validity and naturalness of word-based adversarial attacks". 

You will find:

1. all_text_adversarial_sentences.csv file that includes 25 283 successful adversarial texts that we generated using 9 attacks from TextFooler library over three sentiment analysis datasets ( IMDB movie reviews, Rotten Tomatoes movie reviews,  Yelp polarity service review)
2. questionnaire_answers folder with the survey answers from the participants for each part of the questionnaire.