This directory contains transformations that are part of the Natural Language Augmenter (NL-Augmenter). Each subdirectory contains a single transformation (or a filter for a contrast set). A summary table of these transformation follows.
| Transformation | Description |
|---|---|
| back_translation | Converts an English sentence to German and back to English |
| butter_fingers_perturbation | Adds noise to all types of text sources (sentence, paragraph, etc.) proportional to noise emanating from keyboard typos making common spelling errors. |
| change_person_named_entities | Changes person named entities: Alex travels to the city everyday! --> Jacob travels to the city everyday! |
| change_two_way_ne | Changes the named entity in the source sentence and reflects the same change in the target sentence. Benefits Machine Translation tasks. |
| longer_names_ner | Elongates person names: Russel Peters is a comedian. --> Russel J. M. Peters is a comedian. |
| mixed_language_perturbation | This perturbation translates randomly picked words in the text from English to other languages (e.g., German). It can be used to test the robustness of a model in a multilingual setting. |
| without_punctuation | Hello Sam, how are you? --> hello sam how are you |
| longer_names_ner | Elongates person names: Russel Peters is a comedian. --> Russel J. M. Peters is a comedian. |
| mixed_language_perturbation | This perturbation translates randomly picked words in the text from English to other languages (e.g., German). It can be used to test the robustness of a model in a multilingual setting. |
| without_punctuation | Welcome to New York city! --> Welcome to New York city |
| replace_numerical_values | Changes numerical values. Jason's 3 sisters want to move back to India --> Jason's 6 sisters want to move back to India |
| redundant_context_for_qa | Duplicates the context in a QA setting. (context, question, short-answer) --> (context+context, question, short-answer) |
| lexical_counterfactual_generator | Generates a counterfactual example. (sentence1, sentence2, label) --> (sentence1, sentence2, label) |