Original title: Measuring What Is Missing: A Natural Language Processing Approach to Quantifying Cultural Representation in K-12 STEM Curriculum
Avyay Bommaraju · August 2026
When students do not see their cultures in STEM lessons, they can struggle to feel like they belong in science. Researchers wanted to measure this problem across school materials, but no scalable tool existed to do it. To fix this, they built a computer program that automatically scans text and gives it a score out of 100 based on six areas, such as mentions of diverse scientists and cultural ideas. They tested the tool on 44 math and science texts. They found that every single text scored below 40, putting all of them in the lowest representation tier. Across all texts, the average score was just 9.99. Even worse, 38 of the 44 texts failed to name a single scientist from an underrepresented background. The researchers showed that these cultural gaps are systematic, and the platform that serves the most under-resourced students had the lowest scores.
Imagine planning a class party where every student brings a dish from their home culture. If you only hear from a few students, the menu ends up representing only a few cultures, and the rest of the class feels invisible. You want to measure how complete the menu is, but checking every dish by hand takes forever. In this paper, the textbooks are the menu, and the tool automatically counts how many cultural perspectives are represented, so schools can see the gap without reading every page.
Students from underrepresented backgrounds are less likely to see themselves in science when their cultures are missing from the curriculum. This study shows the absence is measurable and systematic, and that the free platform serving the most under-resourced students has the lowest representation. That gives schools and publishers a concrete way to detect and fix the gap.
The abstract doesn't say whether the CRI's weights are optimal or whether the 44-text sample represents all K-12 STEM curriculum. It also doesn't establish a causal link between low representation and student outcomes. The human validation covered only 20 of the 44 texts, so the full corpus was not checked for reliability.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Measuring What Is Missing: A Natural Language Processing Approach to Quantifying Cultural Representation in K-12 STEM Curriculum — Avyay Bommaraju — EdArXiv — doi:10.35542/osf.io/ejnmf_v4
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.