An auditable rule-based baseline for speech-act annotation in Uzbek literary discourse: A source-derived pilot study
Annotatsiya
This study presents an auditable rule-based baseline for reproducing speech-act labels in low-resource Uzbek literary discourse. The evidence comprises an aggregate inventory of 74 speech-act instances from Sadriddin Ayni's The Death of the Usurer. Only 29 of these are available as item-level source-analysis fragments with text and class labels. The operational inventory contains five classes: representatives, directives, expressives, commissives and declaratives. A deterministic procedure normalises Uzbek apostrophe variants and applies ordered lexical and punctuation cues to suggest one label for scholarly review. On the source-derived calibration subset, 28 of 29 suggestions matched the inherited manual labels (accuracy 96.55%, macro-F1 0.982, Cohen's kappa 0.950). A majority-class baseline achieved 37.93% accuracy. A source-verification sensitivity check corrected three classifier inputs without changing any prediction. The sole mismatch was a rhetorical question labelled expressive but predicted as directive. The same examples informed rule formulation and scoring. The results therefore measure internal calibration fit, not unseen-text performance or inter-annotator reliability. The contribution is a transparent symbolic-AI baseline, a reproducible audit trail and a protocol for building a larger, independently annotated Uzbek corpus. Such a corpus could support future human-computer interaction and statistical modelling.
Hali tarjima qilinmagan