Building Mathematical Voice: Representational Fluency and Task Progression in AI-Assisted LaTeX Authoring Among Secondary Students

Mathematical communication requires students to move fluently between natural language, symbolic notation, and visual representations. This study examines how 45 secondary mathematics students (aged 16-18) developed representational fluency through a scaffolded sequence of AI-assisted LaTeX authoring activities using Overleaf. Drawing on Duval's theory of representational registers and Sfard's commognitive framework, we investigate three interrelated dimensions: (a) the internal structure of students' acceptance of the tool as a medium for mathematical communication, (b) temporal progression across three increasingly complex authoring tasks, and (c) the qualitative themes students themselves identified as salient in their authoring experience. Acceptance items showed strong inter-correlations, with clarity of symbols and collaboration emerging as a tightly coupled pair (rs = .63, p_adj < .001), suggesting that notational precision and social mathematical discourse are perceived as mutually reinforcing. Task completion times followed a non-linear trajectory (Friedman chi2 = 60.3, p < .0001, Kendall's W = .69): algebraic derivation was fastest (M = 10.0 min), worksheet adaptation slowest (M = 17.0 min), and figure/table insertion returned to baseline (M = 9.8 min), consistent with a learning curve that peaks at intermediate complexity. Thematic analysis of 31 open-ended responses revealed eight themes, with learning and effort (65 coded references), AI assistance (62), and error handling (59) most prominent. These findings suggest that AI-assisted LaTeX authoring can serve as a vehicle for developing mathematical voice, enabling students to engage simultaneously with the precision of formal notation and the collaborative construction of mathematical meaning.