Between Computation and Interpretation. On Automated Classification of the Serbian 2024–26 Student Protest Slogans
Between Computation and Interpretation. On Automated Classification of the Serbian 2024–26 Student Protest Slogans

Summary/Abstract: An earlier study on Serbian student protest slogans compared manual close reading annotation, keyword-based assignment, LDA topic modelling, and AI-generated labels as procedures for thematic classification. The present article argues that the comparison rested on a methodologically unstable foundation: the manually constructed categories conflate lexical-semantic domains with discourse functions and therefore cannot serve as a reliable standard against which automated procedures are measured. Setting manual annotation aside, the present article compares the two fully automated systems – LDA and AI labelling – directly. Both methods are sensitive to the same broad vocabulary fields but impose structurally incommensurable organisations on them; for 73% of slogans, their outputs are practically unrelated – a methodological instability that is the central finding.

• Details
• Contents