EssayJudge is a dataset of over 1,000 student argumentative essays describing math or science charts with human-annotated, ground-truth scores for lexical quality. It is a multimodal dataset containing images of charts, essay responses, and ground-truth scores for ten lexical traits of essays. The traits are categorized into three levels of granularity: 1) fine-grained lexical features, 2) broader sentence-level features, and 3) discourse-level characteristics. All 10 traits were assessed for each essay by two experienced experts in English education. Essays in this dataset were written by English language learners.

This dataset is useful for training LLMs to disect lexical and argument quality from math and science essay responses.

Permissions and licensing

License

Apache 2.0

Key restrictions

No limitations.

How to access the data

This dataset is available to anyone. To access the data, head to the Data download tab and follow the instructions provided.