Online Resource 3: data and scripts for "Don't Worry, the Stars Know: Negation and Hedging in Turkish Digital Horoscope Discourse"

Data
- negation_annotation.tsv               One row per candidate (972) with the final, adjudicated codes: status, formal type,
                                        functions, primary function and metadata. The reported_speech and expectation_source
                                        columns hold the model's codes for these two fields; the author's codes, used in the
                                        article, are in the two files below.
- author_blind_codes.xlsx               The author's coding of all 972 candidates before adjudication (sheet 'Kodlama').
- model_annotation_preadjudication.tsv  The model's coding of the same candidates before adjudication.
- adjudication.xlsx                     The author's decisions on the 331 disagreements.
- reported_speech_author.csv           The author's reported (R) / voiced (V) / own-voice (0) codes for all 92 control
                                        instances, the 51 horoscope instances marked by either coding and a random sample of
                                        40 unmarked horoscope instances.
- expectation_source_author.csv         The author's expectation-source codes for the 103 horoscope instances with a
                                        reassuring or denial function (C, W, N, '-').
- wordcounts.json                       Cleaned word count per text (output of clean.py).
- producers.txt                         Producer per Instagram and YouTube text.
- per_text_rates.csv                    Per-text rates used in the text-level tests (output of analyse.py).

Scripts (Python 3.10+, pandas, numpy, scipy, openpyxl)
- analyse.py             All frequencies, tables and tests in Sections 4.1, 4.2, 4.4, 4.5 and 4.6 (Mann-Whitney U,
                         rank-biserial r, Fisher, Kruskal-Wallis, Holm, chi-square, robustness checks).
- reliability.py         Agreement before adjudication (Table 3): identification kappa, fine and macro kappa,
                         Krippendorff's alpha, mean set overlap, alpha with MASI distance.
- sensitivity.py         The absence/ascription sensitivity check in Section 4.2.
- reported_speech.py     The reported- and voiced-speech counts in Section 4.1.
- expectation_source.py  The expectation-source counts in Section 4.3 and the author-model agreement on that field.
- clean.py               Documentation of the cleaning and word-counting procedure (raw texts are not redistributed).

Outputs as run on 13 October 2026
- analysis_output.txt, reliability_output.txt, disagreements_full.tsv, sensitivity_output.txt,
  expectation_source_output.txt, reported_speech_output.txt

To reproduce, run each script in this folder, e.g.
    python analyse.py
    python reliability.py > reliability_output.txt
    python sensitivity.py
    python expectation_source.py
    python reported_speech.py
