LAB-002 — CASE K
Topic: Multilingual file naming, ASCII transliteration, contextual verification and human correction.
Date: November 2, 2026
Method: Qualitative observation of a single human–AI-assisted publication workflow.

OBSERVED EVENT

During preparation of Turkish and English PDFs for the Degismisism DGS-Y1.0 Zenodo record, the Turkish file was initially named “Degismisism_DGS-Y1.0_TR_DOI.pdf”. Although its content was Turkish, the filename used “Degismisism”, the English name of the concept. The user identified the mismatch and proposed “Degismisizm_DGS-Y1.0_TR_DOI.pdf”. The filename was corrected, and the Turkish PDF was checked: its main heading was “Değişmişizm”, while English occurrences appeared in appropriate bilingual title or abstract contexts.

THE LINGUISTIC DISTINCTION

Turkish concept: Değişmişizm
ASCII-safe Turkish filename token: Degismisizm
English concept: Degismisism
English filename token: Degismisism

Transliterating Turkish-specific characters into ASCII-compatible forms (ğ→g, ş→s, ı→i) is not the same as translating or replacing a concept's designated language form. The distinction between “Degismisizm” and “Degismisism” is not solely an ASCII issue: the final “z” versus “s” reflects the different language-specific names.

INTERPRETATION

The observable mismatch was between the TR language indicator and the English token in the filename. The record does not establish that the model is incapable of ASCII conversion. A plausible failure mode is insufficient separation between transliteration and cross-language naming. A single case cannot establish a system-wide error rate.

PROPOSED CORRECTION CHECKLIST

1. Fix the output language (TR or EN) before selecting the filename.
2. Select the concept name authorized for that language: TR=Değişmişizm; EN=Degismisism.
3. Apply character conversion for filesystem compatibility only after selecting the name.
4. Cross-check the filename, cover language, in-document title, DOI and version code.
5. Independently verify both the filename and the document content; do not rely on visual resemblance.
6. Retain human review prior to publication and log the correction with its rationale.

CONCLUSION AND LIMITATIONS

This qualitative case documents how human review identified and corrected a naming inconsistency in a bilingual publication workflow. It is not a controlled experiment, a metric of automatic spelling performance or generalizable evidence about all language models.

Related DOI: https://doi.org/10.5281/zenodo.23877575
Conceptual context: reasoned correction and meta-correction in Degismisism; corrigibility and human oversight in Istisarelism.