TBMM Parliamentary Proceedings Corpus with Speaker-Turn Segmentation, Party Linkage, and Validated Metadata, 1950–2023

A speech-level corpus of Turkish Grand National Assembly (TBMM) parliamentary proceedings covering 4 January 1950 through 23 April 2023 — the full period of competitive multi-party parliamentary politics in Turkey. The corpus contains 2,418,313 speaker turns across five chambers (TBMM, Millet Meclisi, Senato, Danışma Meclisi, MGK), of which 702,503 are linked to specific members of parliament and parties via a Wikipedia-derived roster of 10,740 MP-term records with gender annotations. An analytical core subset of 136,118 high-confidence floor speeches (49.4 million words) is provided for quantitative political analysis. This subset covers 2,868 unique MPs across 26 political parties and is filtered to include only substantive speeches with exact name+province+decade linkage at HIGH or MEDIUM confidence. The corpus combines two sources. For 1950–2018, it extends the Turkronicles OCR archive (Yazar et al. 2025, CC BY 4.0) with additional speaker-turn segmentation and linkage layers. For the 27th Legislative Term (2018–2023), it adds 314,899 speaker turns extracted directly from 537 official Word-format session transcripts published by TBMM. Speaker boundaries are detected using a ten-pattern unified parser with decade-aware OCR normalization, covering chairman, president (REİS), MPs (including initialled names and continuation lines), ministers (including state ministers and the prime minister), group spokespersons, and committee roles. A targeted correction in this release recovers more than 6,000 multi-word party-group speeches (AK PARTİ GRUBU ADINA, İYİ PARTİ GRUBU ADINA, ANAVATAN PARTİSİ GRUBU ADINA) that earlier pipelines misclassified as individual MP turns. All legacy speaker patterns were preserved without regression. Known limitations — including approximately 525 residual OCR-noise leaks, uneven linkage coverage in the bicameral era, and unnormalized province surface forms — are documented in the README. The dataset is released under CC BY 4.0.