Abstract:

Background:
Science discourse, central to students’ science learning, involves asking questions, sharing experiences, and generating evidence-based explanations to understand the natural world (Bae et al., 2021; Lemke, 2004). However, opportunities for rich science discourse are often lacking in middle school classrooms, especially for students historically underrepresented in science (Emdin, 2020). Drawing from sociocultural theory this study focuses on "third spaces" in classrooms that combine students' life experiences with academic discourse, fostering inclusive science learning environments (Moje, 2004; Moll et al., 1992; Vygotsky, 1957).

Purpose of the Study:
The purpose of this explanatory sequential mixed methods study is to apply Natural Language Processing (NLP) techniques to analyze science classroom discourse, with a focus on discursive resources and the distribution of agency among participants to understand how science discourse can be made more equitable.

Methodology:
Seven science classroom videos from a U.S. middle school science partnership project were analyzed. A comprehensive set of NLP techniques (Liu & Cohen, 2021), including LIWC for examining language patterns and BERTopic for topic modeling was used to quantitatively identify patterns in discourse. Qualitative analyses of classroom videos provide contextual insights.

Findings:
Findings showed that students generally used more analytic language than teachers, indicating a high level of engagement in scientific discourse. There was also notable alignment between teacher and student communication styles, as evidenced by moderate to high language coordination scores. Topic modeling showed a clear distinction between academic and everyday talk, with teacher talk characterized by a higher number academic topics while students engaged in a mix of both academic and everyday language. In summary, these various discourse indices demonstrate evidence of hybrid science discourse, where the teacher and students are drawing from scientific and out-of-school resources to make sense of science ideas.

In terms of agency however, analysis of classroom dynamics demonstrated traditional teacher-led discourse patterns, evidenced by a higher frequency of teacher versus student spoken words. Additionally, the majority of questions posed in classrooms were closed-ended, focused on reinforcing knowledge recall rather than student-driven exploration of science ideas. Clout analyses also suggested that teachers typically adopt a more authoritative stance in communication.

Altogether, findings highlight the complexity of discourse patterns in science classrooms. While diverse discursive resources were present, findings also underscore the need for approaches that encourage student agency. That is, talk opportunities in which students take ownership of their science learning and have more influence over what and how they are talking about science.

Discussion:
This study highlights the potential of NLP techniques to provide scalable insights into classroom discourse, particularly in terms of the discursive resources present and the distribution of agency between teacher and students. The findings show various ways to integrate students' lived experiences into classroom discourse, promoting a balance between academic and everyday language to foster engagement in science. Findings also point to the need to create more opportunities for student agency in science learning.

Keywords:

STEM education, discourse, classrooms, natural language processing, students, equity.