SUMMARY
Overall usage

• Google's translation service, which has been the most used translation service across all language pairs has been used for 71% of all all the published translations.
• MinT (Machine in Translation), a Wikimedia Foundation hosted open-source machine translation service, is the second most used translation service accounting for 16% of all the published translations.
• Compared to 2023 at the same time Google's service usage reduced by ~9%.
• No machine translation was used (scratch) for 6% of the published translations using the Content Translation tool.

YoY comparison

• On average, ~960 articles are translated and published daily using the Content Translation tool.
• Compared to 2023, the median number of daily translations:
  • using Google reduced from 669 to 626 (6% decrease).
  • using MinT increased from 48 to 118 (145% increase).
• for other services, the change is not more than 1-2%.

Language pairs where an optional service was used more or close to the default

• There are 44 language pairs where an optional service (or no service, i.e., scratch) was used more or close to the default.
• Among the language pairs where Google is the default:
  • For 20 language pairs, MinT was used more than Google.
  • For 10 language pairs, no service was used (i.e., scratch).
• Among the language pairs where MinT is the default, for 10 language pairs, Google was used more.

Usage of service at each target language

• MinT was used 100% for translating articles to nine languages.
• MinT was used for 90% of all translations for translating articles to 18 languages (within the respective language).
• MinT was used for the majority of the translations (>50% of the services) for translating articles to 18 languages.
• Google was used 100% for translating articles to nine languages.
• Google was used for 90% of all services for translating articles to 53 languages.
• Google was used for the majority of the translations (>50% of the services) to 49 languages.
• Only for translating articles to Aragonese, Apertium was used for 90% of the translations.
• Only for translating articles to Chuvash, Yandex was used for 100% of the translations.
• Only for translating articles to Basque (eu), Elia was the most used service (85% of all services).
• Even in languages where LingoCloud is supported, the usage has been quite low. For Chinese (zh), it was used for ~2% of 4000 translations, and for less than ~1% of 150 translations to Wu Chinese (wuu).

Percentage of MT content modified by the user

• The majority of translations across all MT services were modified by at least 10% at the time of publication.
• For machine translation suggestions from MinT, 32% were modified by less than 10%—the highest of all services.
• The percentage of translations with a human modification percentage between 10% and 50% for MinT and Google is 54%.
• The percentage of translations with a human modification percentage higher than 50% is the least for MinT and Elia at 13% and 11%, respectively.
• Apertium has the highest percentage of translations where the human modification percentage was more than 50%.

Deletion rates by MT service

• Articles translated using MinT were deleted the least: 2.23% of the 30,000 articles.
• Yandex and Google had the highest percentage of deleted articles, with more than 3%.
• 2.7% of the articles translated using Apertium were deleted.

Full report is available at https://analytics.wikimedia.org/published/reports/content_translation/mt_service_usage_analysis_2024_T370749.html

This is great! Thanks @KCVelaga_WMF.
The summary provides a very good and useful overview, looking forward to checking the full report in more detail.

Thank you @Pginer-WMF !

I just realized, I missed a sentence about YoY comparision.

Compared to last year, the year-over-year growth in usage of MinT translation service is ~8pp, while at the same time Google’s service usage reduced by ~9pp.

pp -> https://en.wikipedia.org/wiki/Percentage_point