Data for ETS and Persuade Corpora used in Assessing Writing paper by Deane and Hoang, forthcoming in 2025

This dataset contains data derived from two corpora: A proprietary ETS corpus and the PERSUADE corpus.One of the files contained in data.tar contains essays and holistic human scores from the PERSUADE 2.0 corpus, enriched with automated writing traits calculated based on the model described in:Deane, Paul, Duanli Yan, Katherine Castellano, Yigal Attali, Michelle Lamar, Mo Zhang, Ian Blood et al. "Modeling writing traits in a formative essay corpus." <i>ETS Research Report Series</i> 2024, no. 1 (2024): 1-62.There are also files derived from an ETS proprietary corpus consisting of human-scored essays written to standard prompts associated with ETS' former digital writing service, Criterion. Due to privacy and contractual concerns, the original essays could not be shared. However, the factor scores for these essays are also provided.In addition, for each essay, document embeddings are provided for each layer (layers 0 to 12) of the base microsoft/deberta-v3-xsmall model in Huggingface.