README.md

metadata

pretty_name: Political Parliamentary Corpus (PPC)
language:
  - de
  - en
  - it
  - pl
  - tr
multilinguality:
  - multilingual
license: other
task_categories:
  - text-generation
  - fill-mask
tags:
  - politics
  - parliament
  - social-science
size_categories:
  - 10M<n<100M
configs:
  - config_name: de
    data_files:
      - split: train
        path: data/de/*.parquet
  - config_name: en
    data_files:
      - split: train
        path: data/en/*.parquet
  - config_name: it
    data_files:
      - split: train
        path: data/it/*.parquet
  - config_name: pl
    data_files:
      - split: train
        path: data/pl/*.parquet
  - config_name: tr
    data_files:
      - split: train
        path: data/tr/*.parquet
default_config_name: de

Political Parliamentary Corpus (PPC)

A multilingual corpus of parliamentary speech, party manifestos and (for German) historical newspapers, exposed with one config per language. Every record follows a single unified schema, so the languages are directly comparable.

• 44,978,179 documents (~16.5B tokens, chars/4 estimate)
• 5 languages: de, en, it, pl, tr
• Coverage 1803–2026
• 22 sources, unified schema

Languages / configs

Config	Language	Documents	~Tokens	Years	Sources
de	German	4,446,524	3.3B	1867–2026	bundestag_official, ddb_newspapers, germaparl, manifesto_de, parlamint_at, parlamint_de_beta, reichstag_bsb
en	English	17,782,842	6.7B	1803–2026	hansard_historic, manifesto_uk, parlamint_gb, theyworkforyou
it	Italian	5,151,530	3.5B	1848–2022	italian_parliament, itaparl, manifesto_it, parlamint_it
pl	Polish	15,607,398	2.5B	1919–2025	manifesto_pl, parlamint_pl, ppc_pl, sejm_pl
tr	Turkish	1,989,885	621.5M	1950–2023	manifesto_tr, parlamint_tr, tbmm_speeches_v1
total		44,978,179	16.5B

Document types

Type	Documents
parliamentary_speech	38,426,749
parliamentary_session	6,541,163
news	9,963
manifesto	304

Usage

from datasets import load_dataset

# load a single language (config)
ds = load_dataset("<your-username>/ppc", "de", split="train")
print(ds[0]["text"])

# stream instead of downloading everything
ds = load_dataset("<your-username>/ppc", "de", split="train", streaming=True)

Fields

Field	Type	Description
id	string	the document identifier
source	string	originating source dataset
country	string	ISO country code
lang	string	ISO 639-1 language code
doc_type	string	manifesto / news / parliamentary_session / parliamentary_speech
date	string	document date (string, may be partial e.g. YYYY-MM)
year	int32	document year (int)
title	string	title where available
text	string	the document text
speaker	string	speaker name where available
speaker_role	string	speaker role where available
party	string	party where available
house	string	chamber / house where available
granularity	string	document / sentence / session / speech
url	string	source URL where available
meta	string	JSON string of source-specific extra fields

meta is stored as a JSON string so the Parquet schema stays identical across every source; parse it with json.loads(row["meta"]) when present.

Sources

Source	Lang	Documents	Years
ppc_pl	pl	15,326,265	1919–2025
hansard_historic	en	9,462,457	1803–2004
theyworkforyou	en	7,449,341	1945–2026
itaparl	it	4,905,544	1948–2022
germaparl	de	1,334,367	1949–2021
parlamint_de_beta	de	1,318,629	1949–2025
tbmm_speeches_v1	tr	1,226,139	1950–2023
bundestag_official	de	1,089,317	2025–2026
parlamint_gb	en	870,973	2015–2022
parlamint_tr	tr	763,728	2011–2022
parlamint_at	de	358,048	1996–2022
reichstag_bsb	de	336,115	1867–1939
parlamint_pl	pl	280,947	2015–2022
parlamint_it	it	210,789	2013–2022
italian_parliament	it	35,105	1848–2022
ddb_newspapers	de	9,963	1945–2024
sejm_pl	pl	148	2015–2023
manifesto_it	it	92	1963–2018
manifesto_de	de	85	1949–2025
manifesto_uk	en	71	1964–2024
manifesto_pl	pl	38	1991–2019
manifesto_tr	tr	18	1954–2018