This repository contains my code and data for my contributions in AI2's Natural Instructions Expansion repository.
I have contributed the following datasets so far:
- Detoxifying Language Models Risks Marginalizing Minority Voices
- Given a prompt and two completions from a language model, select the completion that is less toxic, more fluent, more topical, and more stylistically similar with respect to continuing the prompt.
- SCAN
- Given a command in a limited form of natural language, provide the correct sequence of actions that executes the command to thus navigate an agent in its environment.
- Odd-Man-Out
- Given a category and a set of five words, select the word from the set that does not belong (i.e. is the least relevant) with the other words in the category.
- ToM QA
- Given a story, answer the question regarding a character's beliefs about reality or another character's beliefs.
- CODAH
- Given a prompt and four completions, select the completion that is the most plausible in continuing or answering the prompt.
- StereoSet
- Given two sentences regarding the stereotypes of either gender, profession, race, or religion, determine whether the second sentence is either a stereotype, anti-stereotype, or is unrelated with respect to the first sentence and the target.
- Jeopardy!
- Given a category and a trivia clue, generate the best answer that belongs in the category and is described by the clue.
With these 7 datasets, I have contributed 26 tasks, totaling to 111,008 instances. Each task is a variation of the description in each indented bulletpoint above. See the .json files in each directory for more information on each task.