Work

Trustworthy confidence for Kazakh morphological tagging codebase

Year: 2026
Type: book
Abstract: Code and measured results for the paper "Trustworthy Confidence for Kazakh Morphological Tagging: Calibration and Selective Decision-Making in a Low-Resource Text-Correction Pipeline". One code path measures four taggers — KazBERT, KazRoBERTa, XLM-R base and a feature-based CRF — on the canonical 754/162/162 split of UD_Kazakh-KTB (7,376 / 1,566 / 1,594 tokens, seed 42, decontaminated), with temperature fitted on dev and every calibration/selective-prediction metric evaluated on the held-out test split.
Language: English
Open Access status: green
OpenAlex ID: W7215923514
FWCI:
Cites:
Cited by:
Related to: 10