#small language model
Open access
CountryCode-Decide: messy country names to ISO 3166 codes, with calibrated abstention
Oct 2026 · Zenodo (CERN European Organization for Nuclear Research)
Data Quality and Management
Topic Modeling
Abstract
A small open-weight model (mmBERT-small, 8-bit ONNX, CPU) that maps a country or US state written as free text in company registers and forms (any language or script, misspelled, abbreviated, historical) to ISO 3166-1 alpha-2 and ISO 3166-2:US codes with calibrated probabilities, and abstains with a short candidate list when a certified error-rate target cannot be met. Training combines Unicode CLDR and Wikidata names, real register strings and generated variants that passed quality gates and a human label audit; answer thresholds are certified by fixed-sequence testing on held-out real register strings; evaluation uses fresh registers read once after a frozen pre-registration.
