Supplementary material for our BEA 2023 paper, consisting of both the items generated by GPT-3 which passed the content review and were administered to Mturk workers (under /generated_items) and the prompts used to generate said items (under /prompts).

Prompts are text files marked as either simple or elaborated, depending on the prompt type. Note that these prompts were also used for GPT-4.

Generated items are csvs with the following columns:

1. item: the item content which consists of the premise and hypothesis, separated by a newline.
2. label: the item label, either entailment, contradiction, or neutral.
3. Prompt Type: Denotes which type of prompt was used to generate the item.
4. Top P: The setting of this parameter used when querying GPT-3.

Citation

If you find this work useful, please cite us:

Laverghetta Jr., A., & Licato, J. (2023). Generating Better Items for Cognitive Assessments Using Large Language Models. Proceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications.