Changes between Initial Version and Version 1 of Ticket #24067, comment 41


Ignore:
Timestamp:
2025-02-09T00:22:53+01:00 (12 months ago)
Author:
1ec5

Legend:

Unmodified
Added
Removed
Modified

  • Ticket #24067, comment 41

    initial v1  
    77Then there’s the workload. [https://josm.openstreetmap.de/browser/josm/trunk/resources/data/en.lang?rev=19306 en.lang] has nearly 55,000 words across 14,378 strings. By contrast, [https://app.transifex.com/openstreetmap/id-editor/core/ the core iD project] on Transifex has only 13,692 words across 1,862 strings, excluding the strings for id-tagging-schema, editor-layer-index, and osm-community-index. This is entirely reasonable, given JOSM’s sheer range of capabilities. Unfortunately, the Launchpad project lacks any distinction between more and less important strings, not even an option to focus on one dialog or another. Coming in as a new translator is rather like visiting your hometown in OpenHistoricalMap for the first time and feeling overwhelmed by the vast, undifferentiated blankness. The motivation isn’t there for a typical translator who commits to doing everything in one sitting until they’ve gotten enough dopamine from a steadily changing progress bar. Other TMSes integrate translation memory and machine translation (Google Translate, DeepL, etc.) as on-demand suggestions, which let translators focus on the more difficult strings. Launchpad has some translation memory from other FOSS projects, but in my experience they haven’t been as helpful as the ones from other TMSes.
    88
    9 Those who decry JOSM’s most recent foray into automation may not realize that JOSM shipped with machine-generated slop for years. Someone once laid waste to the Vietnamese localization using a machine translator, and it took some arm-twisting to get it removed: #21720. JOSM’s reputation in the Vietnamese community took a hit from that episode. LLM-powered machine translation could yield somewhat better results, but translating short strings is difficult without context, no matter who or what does the translating. OSM has its own jargon that would make the task more complicated, and JOSM also seems to suffer from some string reuse and “Lego brick” [https://learn.microsoft.com/en-us/globalization/internationalization/concatenation string concatenation]. Who knows if “[https://josm.openstreetmap.de/browser/josm/trunk/resources/data/en.lang?rev=19306#L2733 For]” can be translated accurately? A human translator would be quite puzzled by “[https://josm.openstreetmap.de/browser/josm/trunk/resources/data/en.lang?rev=19306#L2739 For the body]”, especially with the explanation “group ‘For the body’ combo combo ‘Second hand’”. In the hands of a machine… that would be a rather surreal prompt.
     9Those who decry JOSM’s most recent foray into automation may not realize that JOSM shipped with machine-generated slop for years. Someone once laid waste to the Vietnamese localization using a machine translator, and it took some arm-twisting to get it removed: #21720. JOSM’s reputation in the Vietnamese community took a hit from that episode. LLM-powered machine translation could yield somewhat better results, but translating short strings is difficult without context, no matter who or what does the translating. OSM has its own jargon that would make the task more complicated, and JOSM also seems to suffer from some string reuse and “Lego brick” [https://learn.microsoft.com/en-us/globalization/internationalization/concatenation string concatenation]. Who knows if “[https://josm.openstreetmap.de/browser/josm/trunk/resources/data/en.lang?rev=19306#L2733 For]” can be translated accurately? A human translator would be quite puzzled by “[https://josm.openstreetmap.de/browser/josm/trunk/resources/data/en.lang?rev=19306#L2739 For the body]”, especially with the explanation “group ‘For the body’ combo combo ‘Second hand’”. I can just see the MOTD image that comes out of that prompt…
    1010
    1111This isn’t just sniping from the sidelines. I’m trying to pitch in, but it sure feels like a drop in the bucket. Translators will respond better to a plea for help that comes with a good contributor experience, including moral support. We’ve have been competing against the machines for a long time now and know how to find places where we’re valued. A thoughtful approach to soliciting contributions will come in handy the next time the project desperately needs more translators, designers, or maintainers.