Title
Follow along on your phone
aronvandepol.com/berlin

Looks matter

朝鮮의 新聞은 東京 大阪의 廣告로 사러 나아간다 …
… 그 新聞의 歷史만 보는 것이 아니라 그 新聞의 體裁를 보게 되는 …

“Korea’s newspapers live on advertising from Tokyo and Osaka. … Advertisers weigh not only a paper’s history but its format.”1

A self-ad of Paegyŏngsa 百榮社, the agency An probably ran, Tonga Ilbo, 1921

From plain to picture

That visual form, from plain type to the picture, is exactly what this study examines, and how widely it spread.

This talk shows three things
1The rise came from new firms arriving elaborate, not old ones changing.
2Japanese firms started elaborate about three times as often as Korean firms.
3The plain, words-only notice never fell below roughly 40% of any year’s ads.

The page grew more elaborate

The rise has been read three ways:

  • By count: Sin In-sŏp and Sŏ Pŏm-sŏk measured advertising space and saw steady growth.1
  • By close reading: Chun studied the most elaborate ads and found only a few firms looked modern.2
  • By business history: Eckert and McNamara studied the biggest firms, including Kyŏngbang.3

These and other readings go deep on particular cases. None maps the whole field, so we cannot see where the famous cases sit, or tell the exceptional from the ordinary.

Kyŏngbang, 1922 and May 1925

The pipeline

The problem

The source is three colonial dailies, scanned runs 1910 to early 1950: the Tonga Ilbo, the Chosŏn Ilbo, and the Maeil Sinbo.1

To map the whole field, three steps:

1Extract every advertisement, from every page.
2Read each one for its firm, products, and text.
3Score each one for visual elaboration.

Cutting out every advertisement

A YOLO26-L object detector cuts candidate advertisements out of the page.1 We train it to over-extract. It cuts out every block that could be an advertisement, real or not.

Trained on 600 hand-labelled pages. Then run over all 133,196 pages, producing 1,197,062 crops.

precision recall F1
0.81 0.84 0.82

Held-out on 120 unseen pages (1,246 ads), IoU 0.50 at confidence 0.30. F1 95% CI 0.76–0.87.

Chosŏn Ilbo, 1937. Every candidate block cut out. Red, the detector; green, the hand labels.

Reading each advertisement

A vision-language model, Qwen3-VL-235B, reads every crop and flags whether it is really an advertisement. It drops 546,748 crops that are not ads. After removing overlaps and failed reads, 623,303 advertisements remain. Each is transcribed into a structured record of about thirty fields. A few of them:

firm  Suzuki Shōten 鈴木商店
origin  Japanese category  food
product  味の素 (Ajinomoto)
headline  料理하시기에 苦心하실 것 없습니다
slogan  味の素
setting  vertical, illustrated

firm  Changmi Sanghoe 長美商會
origin  Korean category  retail
product  十八種一組 (18-piece set)
headline  十五週年記念特別大賣出
slogan  一組金額
place  京城府 (Kyŏngsŏng)

A rank from 0 to 1

  • A vision model (DINOv3) scores every advertisement from 0 to 1 for visual elaboration
  • It learns from pairwise comparisons, not absolute scores
  • The work that counts is illustration, design lettering, and layout, not typesetting
  • It agrees with human judgment 93 to 97 times in 100
  • Reprints of the same ad land 0.028 apart on the scale
  • Random pairs land 0.336 apart

Which took more work to make?

0.10 against 0.999. Anyone can call this pair

Which took more work to make?

0.998 against 0.996. The model learns the finer call

A thin middle

  • Grouped by setting, the field splits into a plain band (~42%), a thin middle (~24%), and an illustrated band (~34%)
  • Roughly 40% of advertisements stayed essentially text-only. The floor dipped under that line only once, to 39.8% in 1943
  • The middle was a place advertisements passed through, not a destination. The field stayed plain or moved to a separate photoengraving chain

Explore the corpus yourself

explorer.aronvandepol.com

Password WWC26

1

The rise came from new entrants, not firms changing.

The typical ad climbed

  • The typical ad’s elaboration score climbed from roughly 0.4 to 0.63 between the 1910s and 1939
  • The plainest tenth barely moved off the floor
  • The top quarter’s share of ads more than doubled over the same years

Firms do not climb

  • About 60% of firms stay within ±0.10 of their entry level

Nor over their lifetime

  • Re-based to its entry level, the median firm shows zero net change across its life
  • The cohort started at 10,617 firms, thinning with age

It was the newcomers

  • Roughly three in four late-1930s ads, at every level of elaboration, came from new firms
  • So the newcomers carried the field’s whole rise in sophistication

2

How a firm entered depended on its origin.

The gap was set at entry

  • Japanese firms launched with a richly illustrated ad 35% of the time, Korean firms just 11%
  • About six in ten Korean firms started with a plain, words-only ad
  • Even in the late 1930s, Korean firms still started a step plainer than Japanese firms

Trade and origin both shaped the look

  • The more modern the trade, the more elaborate its advertising
  • Within nearly every trade, Korean firms ranked plainest, then Japanese, then Western

Every trade spanned plain to elaborate

Sorted at the start, then frozen

  • Japanese firms started in the top, most-illustrated tier three times as often as Korean firms
  • About two in three firms of either origin stayed in the tier they entered
  • The colonial gap was a starting line, not a different path

The picture cost a plate

  • A plain notice cost no plate and a worked headline could still come from type. The picture had to be hand-drawn and cut as a plate
  • That apparatus was thin and dear. In the 1930s Korean ads carried a picture about half as often as Japanese ones
  • The local supply was real but thin. In 1927 Korea’s whole design and illustration stock was one freelancer

Famous for design, plain in print

  • Cinema advertising is credited with driving design lettering forward
  • Its own ads were mostly plain listings. Only about one in a hundred reached the elaborate top, like this Ben-Hur
  • So the cinema trade’s design fame outran its plain record

Where the famous cases sit

  • Yongkeun Chun reads the most elaborate ads and calls their world a displayed modernity few firms could sustain
  • Placed on the whole scale, the firms he and Sin In-sŏp single out scatter apart
  • Every canonical Korean case sits below the Japanese imperial brands, Ajinomoto, Morinaga, Club and Lion, at the top of the field

Climbing was rare, and rarely kept

  • Of more than five thousand steady Korean firms, about eight climbed from plain to elaborate and held it
  • About one climb in three reversed, up then back down
  • Most well-known Korean advertisers never climbed at all

Tokyo and Osaka filled the top

  • Korean firms’ share of the showiest ads fell from over half to about one fifth by the 1930s
  • The room at the top grew. Korean firms did not climb into it
  • Dentsū, Japan’s largest advertising agency, had run a Seoul branch since 1907. It alone handled more than half of all advertisement production

3

The plain ad never went away.

The floor never gave way

  • The plain notice never fell far below 40% of a year’s advertising
  • Korean advertising stayed plainer than Japanese throughout. In the early 1940s 63% of Korean ads were words-only
  • The floor was the look of an under-capitalised market. No Korean firm was large enough to sustain a campaign

After liberation: a plain shock, then recovery

  • Liberation brought the plainest page since the 1910s. Through 1945 and 1946, more than three in four ads were words-only
  • By 1949 elaboration was back near its early-1940s level, largely without the Japanese firms
  • Enough of the craft had taken local root to revive it. The record ends on the edge of the 1950 war

A words-only pharmacy notice, Chosŏn Ilbo, March 1946.

The modern look came in with new firms, not out of old ones.
How modern a firm looked was set at the start, by its origin.

And beneath it, the plain notice never gave way.

References

Arnold, Taylor, and Lauren Tilton. Distant Viewing: Computational Exploration of Digital Images. 1st ed. MIT Press, 2023.
Chun, Yongkeun. “Displayed Modernity: Advertising and Commercial Art in Colonial Korea, 1920-1940.” Thesis, Royal College of Art, 2020.
Eckert, Carter J. Offspring of Empire: The Koch’ang Kims and the Colonial Origins of Korean Capitalism, 1876-1945. Korean Studies of the Henry M. Jackson School of International Studies. Univ. of Washington Press, 2014.
McNamara, Dennis L. The Colonial Origins of Korean Enterprise, 1910-1945. Cambridge University Press, 1990. https://doi.org/10.1017/CBO9780511528101.
Sin, In-sŏp, and Pŏm-sŏk Sŏ. Han’guk Kwanggosa 한국 광고사 [The History of Korean Advertising]. Revised 3rd Edition. Nanam 나남, 2011.