Performance Analysis of a Serial Natural Language Processing Pipeline for Scaling Analytics of Academic Writing Process

Capturing and analyzing just the final submission of a writing assignment ignores a substantial amount of information, providing only a partial view of the writer’s effort and intent. Such a partial view of writing abilities limits opportunities for the generation of feedback to improve the final writing product as well as to aid in the development of effective writing techniques. Over-the-shoulder monitoring of the writing process for only a few individuals proves to be a challenge, while scaling specialized tutoring to as many writers as possible is simply impossible without leveraging technology. This research analyzes the computational requirements of a single-threaded writing analytics system for real-time monitoring and instructional intervention of writing processes. This chapter reports on the performance of this analytics system using the simulated writing processes of 391 compositions in higher education, a subset of the British Academic Written English (BAWE) corpus. It elaborates on computational requirements of analytics elements involving Natural Language Processing (NLP) and offers recommendations for building scalable big data NLP pipelines adapted to the analysis of academic writing process of learners.