![[ Visit Client Website ]](images/banner.gif)
Schedule:
Saturday, January 17, 2026
Monument, ML 4 (Marriott Marquis Washington DC)
* noted as presenting author
Background: Open-source language data offers higher education researchers and administrators new opportunities to examine communication and intent in real-world settings, including expressions of power, comprehension, hierarchy, and engagement. Using text-as-data methods, computational social scientists can empirically assess psychological well-being, emotional tone, and cognitive engagement. Language is a reliable predictor of future behavior, including academic success (e.g., through application essays), and linguistic assessment can be applied to internal sources (e.g., course papers, recorded discussions). This proof-of-concept (PoC) study demonstrates how publicly available text can be leveraged for ethical, scalable, and replicable assessment—offering a path forward for faculty and administrators seeking data-driven admissions and student support strategies. Our interdisciplinary project analyzes conversational dynamics in higher education discourse, showcasing possibilities for broader applications across admissions, student development, and pedagogical research.
Methods: Textual data were sourced from the publicly available “UNLV Public Debate: Should AI Be Used in Child Welfare Cases?” published on YouTube. The full transcript was analyzed using Linguistic Inquiry and Word Count (LIWC-22) to extract linguistic and psychological features, including analytical thinking, clout, authenticity, cognitive processes, and emotional tone. Extracted data were imported into Stata for statistical analysis and visualization.
Results: The analysis revealed moderate analytical thinking (M = 44.18) and elevated clout (M = 63.06), suggesting a confident but conversational communication style. Authenticity was relatively low (M = 29.39), and emotional tone skewed negative (M = 28.24). Cognitive process words (M = 22.43) and emotional expressions (positive = 2.53%; negative = 1.93%) appeared at low rates, reflecting a structured but emotionally restrained debate. A detailed comparison showed the for-AI speaker used more analytical, self-referential, and quantitatively framed language, indicating a logical argumentative style, while the against-AI speaker used more pronouns, function words, family references, and past-focused language—suggesting a socially grounded and emotionally contextualized approach. Compared to student speakers, the expert used significantly more technical, professional, and culturally focused language, with less reliance on function words, pronouns, and social references—delivering a more formal, content-heavy message with minimal interpersonal framing.
Implications: This PoC project demonstrates that natural language processing tools can rapidly and efficiently assess sentiment, tone, and cognitive engagement in communication. The approach holds promise for evaluating admissions materials (e.g., essays and application videos) and monitoring student development. The text-as-data method provides a scalable, replicable model for future research and assessment initiatives, with the potential to standardize evaluations and reduce subjectivity—supporting greater equity in higher education admissions and student success monitoring. This analysis establishes the viability of using text-as-data tools to extract meaningful signals from public discourse in education.
Methods: Textual data were sourced from the publicly available “UNLV Public Debate: Should AI Be Used in Child Welfare Cases?” published on YouTube. The full transcript was analyzed using Linguistic Inquiry and Word Count (LIWC-22) to extract linguistic and psychological features, including analytical thinking, clout, authenticity, cognitive processes, and emotional tone. Extracted data were imported into Stata for statistical analysis and visualization.
Results: The analysis revealed moderate analytical thinking (M = 44.18) and elevated clout (M = 63.06), suggesting a confident but conversational communication style. Authenticity was relatively low (M = 29.39), and emotional tone skewed negative (M = 28.24). Cognitive process words (M = 22.43) and emotional expressions (positive = 2.53%; negative = 1.93%) appeared at low rates, reflecting a structured but emotionally restrained debate. A detailed comparison showed the for-AI speaker used more analytical, self-referential, and quantitatively framed language, indicating a logical argumentative style, while the against-AI speaker used more pronouns, function words, family references, and past-focused language—suggesting a socially grounded and emotionally contextualized approach. Compared to student speakers, the expert used significantly more technical, professional, and culturally focused language, with less reliance on function words, pronouns, and social references—delivering a more formal, content-heavy message with minimal interpersonal framing.
Implications: This PoC project demonstrates that natural language processing tools can rapidly and efficiently assess sentiment, tone, and cognitive engagement in communication. The approach holds promise for evaluating admissions materials (e.g., essays and application videos) and monitoring student development. The text-as-data method provides a scalable, replicable model for future research and assessment initiatives, with the potential to standardize evaluations and reduce subjectivity—supporting greater equity in higher education admissions and student success monitoring. This analysis establishes the viability of using text-as-data tools to extract meaningful signals from public discourse in education.