Diacritic Restoration in the Serbo-Croatian Macro-Language: diacritic-stripped writing (šišana) as a written register and an on-device task

Preprint 2 of a study of diacritic restoration (šišana/dešišavanje) in the Bosnian–Croatian–Serbian (BCS) Latin standards. Comparing three independently compiled standard lexicons, it shows that the diacritization of a stripped form is standard-independent for more than 99. 5% of the lexicon: standard differences surface as different stripped forms, not as different diacritizations of the same form. The result is corroborated by the minimal foreign-language contamination of the shared lexicon and by the structural asymmetry of detecting the standard from text, and is supported by an independent Universal Dependencies benchmark against the corpus-trained REDI tool, on which a compact (~12. 8 MB) offline dictionary restorer reaches near-ceiling accuracy at the lowest false-positive rate — a different point on the accuracy/footprint trade-off rather than a claim of superior accuracy. A native-speaker validation study is outlined. This deposit contains the article in two language versions: English (preprint2deposit. pdf, with full data appendices) and a neutral Serbo-Croatian/BCS version (preprint2depositₕbs. pdf). Full derived data lists are reproducible from open frequency sources via the included pipeline and are available from the author on request.