2026-10-06T16:00:10Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33941566 2026-09-22T16:16:28Z category_28864 category_29398 category_28849 item_type_3 month_year_09_2026

The Objective Projection Corpus: A Bilingual Turkish-English Resource for Six-Feature Craft Annotation in Narrative Text, with a Published Reliability Audit Levent BULUT (23152441) Artificial intelligence not elsewhere classified Computational linguistics Natural language processing dataset documentation narrative annotation inter-rater reliability computational narratology show-don't-tell bilingual corpus Objective Projection <p dir="ltr">We describe the Objective Projection Corpus, a bilingual Turkish–English resource built to support research on a specific narrative-craft distinction: whether an emotional or informational state is declared on the surface of a text (told) or must be reconstructed by the reader from physical, non-evaluative detail (shown). The corpus centers on 500 annotated scene pairs (300 Turkish, 200 English) in which each scene is scored against six craft features by a deterministic, bilingual,rule-based detector with no external dependencies alongside the data. Two of the six features are prohibitions (no explicit emotion naming, no simile); four are positive techniques (materialized metaphor, micro-focus, temporal anchor, atmosphere contradiction). Beyond the annotated core, the corpus includes a 60-scene single-variable ablation set, a 30-scene compliance benchmark, a hard-negatives collection targeting five specific ways a text can appear compliant while violating the scheme, and an evaluation directory containing every file needed to independently reproduce the corpus's own reliability statistics: a blind in-sample study (n=120) and an out-of-sample study (n=100) scoring the detector and four large language models against two independent human references. We report those reliability figures here in summary form and are explicit about what they show: the two prohibition-type features are detected reliably; the four positive-technique features range from unreliable to, in the case of materialized metaphor, statistically indistinguishable from chance. We describe the corpus's construction, its annotation scheme, its known limitations — including that all scenes are purpose-written under one methodology rather than sampled from naturally occurring text, and that the annotation layer is decisively not yet validated as ground truth for its four inferential features — and the specific gaps we consider most valuable for other researchers to close.</p> 2026-09-22T16:16:28Z Dataset Dataset 10.6084/m9.figshare.33941566.v1 https://figshare.com/articles/dataset/The_Objective_Projection_Corpus_A_Bilingual_Turkish-English_Resource_for_Six-Feature_Craft_Annotation_in_Narrative_Text_with_a_Published_Reliability_Audit/33941566 CC BY 4.0