NaturalLanguageProcesing4All
Arthur Hjorth · 2021
This paper presents a pilot study of NaturalLanguageProcessing4All (NLP4All), a Constructionist, low-threshold, XAI learning tool designed to bring Natural Language Processing methods into high school classrooms. Specifically, NLP4All is designed to let non-programmers explore different corpora of text through classification activities. Together with a high school Social Studies teacher, I developed a 2-week (6-hour) learning unit focusing on analyzing tweets from political parties to explore the differences and similarities between their policy views and communication styles. In the analysis, I find that text classification shows unexplored promise as a learning activity; that students were able to draw on their prior knowledge to classify tweets; that using NLP4All to collaboratively classify tweets led to productive classroom discussions; and that while students were able to build good machine learning models for classifying tweets, their rationales often focused on identifying one party, rather than distinguishing between parties. Finally, I discuss other educational contexts where NLP and ML can be productive for children, and future design features that may be worth exploring.