Discourse frames — how the world press writes about AI
Derived from world_press data · 12 press excerpts on AI
Research question — With what frames does the world press narrate AI — and does the distribution of frames differ by zone (production, components, adoption, absence)?
Raw data sample_discourse_worldpress.csv · 12 rows
Derived from world_press data — the actual articles cited by the lab series (ep1), recomposed as excerpt units. Outlet and date are carried over unchanged.
| id | outlet | region | date | excerpt |
|---|---|---|---|---|
| P01 | CNBC | North America | 2026-06-29 | The Korean government announced a semiconductor ecosystem project worth 800 trillion won, with Samsung Electronics and SK Hynix taking part. The report described it as a long-term investment in which each company builds new fabrication plants, alongside AI data centers and advanced packaging. |
| P02 | Fortune | North America | 2026-06-28 | Citing Korea Economic Daily, it reported that the two groups’ ten-year investment could reach as much as 2,000 trillion won. The expansion of the semiconductor supply chain was narrated as a response to AI demand. |
| P03 | TradingKey | Market analysis | 2026-06 | SK Hynix is the largest supplier of the HBM that goes into Nvidia AI accelerators, and is said to supply more than 60% of Nvidia’s HBM volume. The narration is that a substantial share of the world’s AI computation hangs on one company’s component. |
| P04 | CNBC | North America | 2026-07-02 | It reported that a Nasdaq semiconductor sell-off carried over to the Korean market within a day, closing SK Hynix down 14.6% and Samsung Electronics down 9.1%. An analysis that the two stocks make up about half of the entire KOSPI market capitalization ran alongside it. |
| P05 | North American outlet | North America | 2026-06 | Coverage continued about the equity relationship between the US government and OpenAI. An agenda of state-company relations that only holds for whoever builds the most advanced models was treated as the issue of the page. |
The study, step by step — what was actually produced
① Splitting into units
One excerpt = one unit of analysis (row mode). 12 units, with outlet, region and date attached to each unit as metadata.
② Frame coding
With Entman’s four functions as the codebook, all 12 excerpts were coded by hand — problem definition 4 · causal diagnosis 2 · moral evaluation 2 · treatment recommendation 4.
Judgment memo — the investment announcements (P01, P02) convey facts, but the emphasis of the narration lies on “a response to AI demand,” so they were coded as treatment recommendation.
③ Crossing by zone
Taking the four zones of the ep1 canon (build · hold up · bring in · inside the unseen) as the meta axis, the frame distribution was crossed against it. The holding-up zone clusters on problem definition and treatment recommendation; inside the regions that were not seen, the clustering is on causal diagnosis.
④ Opposed vocabularies
The distribution of production verbs (announce · build · develop · invest) was contrasted with that of reaction verbs (catch up · protect · appeal · demand · define). In the bringing-in zone and inside the unseen regions, reaction verbs dominate.
⑤ Results
The frame structure table, the cross table and the quotation table are laid out as paper-ready tables. Every quote carried is a sentence that exists in the source excerpts.
Key results
Frame × zone cross table
Each cell counts excerpts. The zone names are the observational descriptions of the ep1 canon.
| Zone | Problem definition | Causal diagnosis | Moral evaluation | Treatment recommendation | Total |
|---|---|---|---|---|---|
| Countries that build | 1 | 0 | 0 | 0 | 1 |
| Countries that hold it up | 2 | 0 | 0 | 2 | 4 |
| Countries that bring it in | 1 | 0 | 1 | 1 | 3 |
| Inside the regions that were not seen | 0 | 2 | 1 | 1 | 4 |
| Total | 4 | 2 | 2 | 4 | 12 |
Opposed vocabularies — production verbs against reaction verbs
The demonstration of quantification the ep1 canon foreshadowed — counted by presence in an excerpt unit.
| Zone | Production verbs (announce · build · develop · invest) | Reaction verbs (catch up · protect · appeal · demand · define) |
|---|---|---|
| Countries that build | 0 | 0 |
| Countries that hold it up | 4 | 0 |
| Countries that bring it in | 1 | 3 |
| Inside the regions that were not seen | 1 | 3 |
Representative quotes
Every quote is a sentence that exists in the raw data (excerpt-existence check passed).
“A local expert pointed to electricity, not algorithms, as the primary constraint”
Causal diagnosis · P09 · Global Voices
“An op-ed ran arguing that the country should not try to build frontier AI”
Moral evaluation · P07 · Australian outlet (op-ed)
“a Nasdaq semiconductor sell-off carried over to the Korean market within a day”
Problem definition · P04 · CNBC
“Ghana defined AI as a sovereign capability in its national AI strategy”
Treatment recommendation · P11 · Al Jazeera
Paper-ready tables
Table 1. Theme structure
Categories, their definitions, and how often they appeared.
| Category | Definition | Frequency |
|---|---|---|
| Problem definition | Narration that defines what is at issue | 4 |
| Causal diagnosis | Narration that points to the cause of the issue | 2 |
| Moral evaluation | Narration carrying a normative judgment | 2 |
| Treatment recommendation | Narration carrying a response or prescription | 4 |
Table 2. Quotations by category
Participant words supporting each category.
| Category | Case | Quotation |
|---|---|---|
| Causal diagnosis | P09 | A local expert pointed to electricity, not algorithms, as the primary constraint |
| Moral evaluation | P07 | An op-ed ran arguing that the country should not try to build frontier AI |
| Treatment recommendation | P01 | announced a semiconductor ecosystem project worth 800 trillion won |
Copying opens an editable duplicate in your workbench — this showcase stays as it is.