Using Natural Language Processing to Support K–12 Students’ Textual Interpretation

• In Event: Integrating AI Education in K–12 Humanities Curricula: Opportunities, Challenges, Design Considerations, and Promising Solutions

Abstract

Objectives or purposes: We share ways to use Artificial Intelligence (AI), and specifically Natural Language Processing (NLP), to build language-based visualizations that will help secondary school students see underlying textual patterns and relationships, and help English Language Arts teachers teach textual interpretation. These language visualizations include word clouds that help students make predictions about a story, line graphs that track conflicts across a narrative, or interactive charts that help students manipulate character descriptions and develop judgments about characters. In our paper, we explore the benefits and challenges of using language visualizations to help teachers and students address common literacy challenges such as schema building, comprehension, tracking literary patterns, distinguishing narrator from author, and building thematic interpretations, as well as newer challenges, such as the need for data literacy.

Conceptual framework: NLP and data visualizations arise from disciplines such as linguistics, data science, and the digital humanities (the latter discipline uses computational analyses of language to help reveal texts’ underlying structures). We build on premises from these disciplines: First, that NLP and related textual visualizations are a useful medium for teachers’ and students’ textual explorations; and second, that data literacy— knowing how to observe, predict, and interrogate data— is increasingly important in a world where students are already immersed in social media’s claims, online quizzes, and other forms of data representation.

Methods and Data Sources: While digital humanities is a growing discipline at the post-secondary level (e.g., Craig et al., 2021), it is rarely part of K-12 curricula. With the advent of more accessible NLP tools, younger students can now take advantage of this medium. Our university-based design team partnered with middle school English Language Arts (ELA) teachers and district coordinators to co-design activities at the intersection of ELA instruction and data literacy. Our data include artifacts and notes from design sessions.

Results: Our presentation will share examples of NLP data visualizations that address four of the ELA literacy needs articulated by our teacher partners - comprehension, prediction, character analysis, and authorial choice. For instance, we will discuss how the integration of NLP tasks (e.g., named entity recognition, part-of-speech tagging), can assist in the identification of verbs or adjectives most closely associated with characters in a narrative. A representation of those associations can serve as a basis for story prediction or character judgments. In addition, when students use NLP-generated sentiment analysis, they can systematically study how people/groups are represented within a text, thereby facilitating reflection on issues of power, race, and gender as these relate to textual effects or authorial choices. Further, students can compare their own interpretations with NLP data, and learn to challenge AI-generated representations.

Significance: Broadly, we hope to illuminate the potential of NLP as a vehicle for K-12 students’ exploration of texts, and to inspire further research at the intersection of AI and language arts.

Authors

• Raquel Coelho, University of Pittsburgh

• Dora Demszky, Stanford University

• Sarah Levine, Stanford University

• Victor R. Lee, Stanford University

• Dorna Abdi, Stanford University

• Christine Bywater, Stanford University