Awardees: Stuart Shieber and Allen Schmaltz (SEAS)

Summary: Awardees explored the potential impact of a writing instruction method that emphasizes the editorial and revision process, which has the potential advantages of scalability and skill transferability.

Professor Stuart Shieber and doctoral candidate Allen Schmaltz, both of SEAS, explored the impact of devising a method of writing instruction emphasizing the editorial and revision process. Toward this end, they explored the feasibility of designing a software tool that can facilitate the re-writing process and at the same time, collect natural language data to further enhance the tool and by extension, student learning. Initial work focused, in particular, on machine learning algorithms for the task of grammatical error identification and correction, which is a component task of the larger endeavor of (semi-)automated (re-)writing assistance.

They applied their initial algorithms to the Automated Evaluation of Scientific Writing Shared Task 2016, a standardized test environment for evaluating approaches for identifying errors in academic text. Their system was the highest performing on the binary classification task.

Next steps include incorporating the initial learning approaches into a software tool that can be deployed in a college writing course. Writing feedback is useful for student learning, but is relatively costly to provide. Extending their initial research, they plan to explore whether a machine learning based tool can, in practice, effectively streamline and automate aspects of instructor feedback, while simultaneously improving the individual learning of student editors who are also incorporated into the editing feedback loop. This work aims to further research in natural language correction–via new algorithms and generating additional training data for the research community–and at the same time, build a practical tool of pedagogical utility for students at Harvard and the broader public.

See more:

Allen Schmaltz, Yoon Kim, Alexander M. Rush, and Stuart Shieber. 2016. Sentence-level grammatical error identification as sequence-to-sequence correction. In Proceedings of the 11th Workshop on Innovative Use of NLP for Building Educational Applications, pages 242–251, San Diego, CA, June. Association for Computational Linguistics. http://aclweb.org/anthology/W/W16/W16-0528.pdf