Tasvir Bankası: A Turkish Literary Scene-State-Description Dataset and Annotation Pipeline

Tasvir Bankası v0.3.1 is a Turkish literary scene-state-description dataset and reproducible annotation pipeline release prepared by Furkan Yaşar. Its central contribution is the transformation of Turkish literary prose into structured records that distinguish scene-bearing, state-bearing, dialogue-bearing, tense-bearing, and descriptive-evidence-bearing textual units. This Zenodo upload is the publication manifest archive for the release. It preserves citation metadata, license policy, source-rights documentation, validation summaries, and release manifest/checksum references. It does not contain the full gated Hugging Face dataset payload and does not bypass the intended non-commercial gated access workflow. The full Hugging Face dataset package described by this manifest contains 2,172 train records, a 92-record label-level curated gold-eval subset, and a 98-record human curation ledger. The public-safe GitHub pipeline package is tracked separately as `v0.1.0-public-preview`; the dataset release package is tracked as `v0.3.1`, with final publication-polish packaging applied before upload. The project addresses a gap in Turkish literary NLP by providing a carefully engineered pilot resource for scene-level, state-level, and descriptive-evidence-level analysis of literary prose. Although the current implementation is Turkish-specific, the staged pipeline architecture is portable to other languages if language-specific morphology, tense, lexical matching, and prompt assumptions are rewritten. The release is methodologically adjacent to literary NLP datasets that structure literary texts for computational analysis, such as LitBank-style annotation resources, quotation-attribution corpora, and literary coreference datasets. Its distinctive focus is the combined treatment of scene, state, and description as structured annotation targets for Turkish prose. Source texts are documented as rights-reviewed public-domain candidates under the Turkish death+70 rule. Where available, source-edition metadata such as publisher, edition year, printing number, and ISBN are documented in `SOURCE_TEXT_INVENTORY.md` and `SOURCE_RIGHTS_REVIEW_TABLE.md`. Modern edition-specific paratext, cover material, forewords, notes, and publisher-added materials are excluded from the intended source boundary. This release does not constitute a legal determination regarding edition-specific rights; users are responsible for verifying source and jurisdictional status before reuse. The release is intended for research, education, digital humanities, computational narratology, Turkish literary NLP, reproducibility review, and non-commercial evaluation. Commercial use, resale, integration into paid text-generation products, commercial model training, or redistribution as part of a commercial dataset package is not permitted without explicit written permission from the author. Related resources: GitHub public-safe annotation pipeline package:https://github.com/Kon-tiki-ship/tasvir-bankasi-turkish-literary-annotation-pipeline Hugging Face gated dataset release:https://huggingface.co/datasets/Kon-tiki-ship/tasvir-bankasi-turkish-literary-scene-state-description-dataset Zenodo publication DOI:https://doi.org/10.5281/zenodo.20579958 Recommended citation: Yaşar, F. (2026). Tasvir Bankası v0.3.1: A Turkish Literary Scene-State-Description Dataset and Annotation Pipeline. Zenodo. https://doi.org/10.5281/zenodo.20579958 Last updated for final upload package: `2026-06-07T10:50:54Z`.