A multilingual news corpus,
sentence-level and topic-coded.
MEDIATEXT is a research corpus of news articles segmented to the sentence and coded for CAP policy topic, sentiment, and named entities. The first release — MediaText CEE — covers four Visegrád media systems (Czechia, Hungary, Poland, Slovakia) at over 1.4 million articles. A parallel release — MediaText NYT — covers the New York Times from 1851 to the present. Compiled at poltextLAB and distributed under CC BY-NC 4.0.
Sentence-level. Multilingual. Documented.
MEDIATEXT is a sentence-level news corpus. Every article is segmented into sentences; each sentence carries a CAP (Comparative Agendas Project) policy topic, a sentiment score, and named-entity spans. Document-level aggregates roll up underneath. Two releases are documented in companion data papers: MediaText CEE (Czechia, Hungary, Poland, Slovakia — over 1.4 million articles) and MediaText NYT (the New York Times, 1851 to the present).
The same coding stack — the Babel Machine multilingual transformer paired with the perfAPI / gpt-oss:120b gateway — was applied to every sub-corpus, so a sentence-level row from the Hungarian press and one from the New York Times share the same label schema and codebook.
Public portal, deposited dataset.
A public reading portal at mediatext.poltextlab.com lets visitors browse overview tables, sentiment distributions, NER word clouds, and a per-row data explorer. The full programmatic release — vector embeddings on every text variable alongside metadata — is deposited on Harvard Dataverse and OSF for the closed period, under CC BY-NC 4.0.