Both of our submissions to the International Conference on Natural Language and Speech Processing have been accepted for oral presentation: FrIdéo and Framing by Wording, Framing by Selection.
FrIdéo builds a continuous, auditable ideology scale for French national news. Rather than resting on any single indicator, it fuses nine independent families of evidence, and it separates partisan endorsement from critical reporting — the distinction that lexical approaches usually collapse, and the reason an outlet that quotes a position gets read as holding it. The scores for thirty outlets are published as open data alongside the paper.
Framing by Wording, Framing by Selection audits French news headlines from 2022 to 2025 along two dimensions at once: how a story is worded, and which stories get covered at all. Treating those separately matters, because a newsroom can shape coverage without writing a single slanted sentence. The paper releases a 10,000-headline supervision set and a recalibration protocol other researchers can reuse.
Both papers are listed on the publications page with their abstracts, results and citations. The datasets behind them are open and free to download.
Read both papers, with their results, methods and citations, on the publications page.