Do international news headlines lean on high-arousal language? An exploratory lexical pilot (n = 97)
To probe it, I built a small, zero-dependency Python script that reads live RSS feeds from major international broadcasters. It tokenises each headline, strips syntax, and scores it against a curated lexicon of high-arousal civic terms to produce an Affective Intensity Score (AIS). This is a simple lexical (bag-of-words) method, not a semantic model — a point that matters for how much weight the result can bear (see Section 3).
Across the three outlets sampled, average affective intensity was very low, and — the number that matters — 0.0% of headlines crossed the high-arousal threshold at any outlet. On this sample, then, these international broadcasters show no sign of optimising headlines for affective intensity. The small gaps between outlets below (BBC 1.4%, NYT 0.9%, Al Jazeera 0.3%) are noise at this scale, not a ranking of who is most inflammatory.
A core tenet of honest measurement is acknowledging the limitations of the tools employed. This pilot uses a lexical (bag-of-words) approach, which lacks semantic nuance and struggles with context and negation (e.g., telling a legitimate state “crisis” apart from sensationalised political “chaos”). The near-zero scores may reflect this bluntness as much as genuinely calm headlines.
Furthermore, this framework currently relies on an English-centric NLP lexicon. Applying this model to South Asian media ecosystems requires training a localized Large Language Model (LLM) on a Bengali linguistic corpus to capture the true epistemic texture of regional disinformation. This pilot proves the structural data pipeline; Phase 2 will demand deep-learning semantic integration.
Even the top-scoring headlines in the sample score low, and they are flagged only for ordinary hard-news words — “protest,” “attack,” “deadly,” “dead.” That is exactly the bag-of-words weakness from Section 3: the lexicon catches routine news vocabulary, not affective framing.
| Outlet | Affective AIS Score | Headline Text (Tokenized) |
|---|---|---|
| NYT World | 12.5% Hits: protest | Tractors Clog Copenhagen as Farmers Protest Fertilizer Restrictions |
| NYT World | 10.0% Hits: attack | Weeks After Failed Drone Attack, Germany Points Finger at Russia |
| BBC Asia | 9.1% Hits: deadly | Nepal intensifies relief and recovery efforts after deadly floods, UN says |
| Al Jazeera | 8.3% Hits: dead | 5 dead, dozens missing after migrant boat found after weeks at sea |
| NYT World | 8.3% Hits: intense | U.S. and Iran Survey the Damage After a Night of Intense Strikes |
The script is small and dependency-free, so it is easy to rerun and adapt. Because it reads live RSS feeds, rerunning it returns current headlines, not the 97 captured here — the stored dataset is the archived snapshot. As Section 3 notes, the English lexicon and bag-of-words scoring make this a starting scaffold, not a finished measure; a Bengali corpus and a semantic model would be the next step before any claim about South Asian media.
The code and the captured dataset are public on GitHub.
View Code & Dataset on GitHub →