Wikipedia's founder said he used ChatGPT in the review process for an article and thought it could be helpful. Editors replied to point out it was full of mistakes.
You can do that, that’s fine. As long as you can verify it is an accurate translation, so you need to know the subject matter and the target language.
But you could probably also have used Google translate and then just fine tune the output yourself. Anyone could have done that at any point in the last 10 years.
As long as you can verify it is an accurate translation
Unless the process has changed in the last decade, article translations are a multi-step process, which includes translators and proof-readers. It’s easier to get volunteer proof-readers than volunteer translators. Adding AI for the translation step, but keeping the proof-reading step should be a great help.
But you could probably also have used Google translate and then just fine tune the output yourself. Anyone could have done that at any point in the last 10 years.
Have you ever used Google translate? Putting an entire Wikipedia article through it and then “fine tuning” it would be more work than translating it from scratch. Absolutely no comparison between Google translate and AI translations.
Same in Hungarian, machine translation still often gives hilariously bad results. It’s especially prone to mixing up formal and informal ‘you’ within the same paragraph, something which humans never do. At least it’s easy to tell when a website is one of those ‘auto-translated to 30 languages’ content mill.
I recently have edited a small wiki page that was obviously written by someone that wasn’t proficient in English. I used AI to just reword what was already written and then I edited the output myself. It did a pretty good job. It was a page about some B-list Indonesian actress that I just stumbled upon and I didn’t want to put time and effort into it but the page really needed work done.
Wikipedia’s translation tool for porting articles between languages currently uses google translate so I could see an LLM being an improvement but LLMs are also way way costlier than normal translation models like google translate. Would it be worth it? And also would the better LLM translations make editors less likely to reword the translation to make it’s tone better?
Honestly, translating the good articles from other languages would improve Wikipedia immensely.
For example, the Nanjing dialect article is pretty bare in English and very detailed in Mandarin
This is the goal of Abstract Wikipedia. https://meta.wikimedia.org/wiki/Special:MyLanguage/Abstract_Wikipedia
You can do that, that’s fine. As long as you can verify it is an accurate translation, so you need to know the subject matter and the target language.
But you could probably also have used Google translate and then just fine tune the output yourself. Anyone could have done that at any point in the last 10 years.
Unless the process has changed in the last decade, article translations are a multi-step process, which includes translators and proof-readers. It’s easier to get volunteer proof-readers than volunteer translators. Adding AI for the translation step, but keeping the proof-reading step should be a great help.
Have you ever used Google translate? Putting an entire Wikipedia article through it and then “fine tuning” it would be more work than translating it from scratch. Absolutely no comparison between Google translate and AI translations.
That depends on if you are capable of translating the language if you don’t know the language then the translator will give you a good start.
Google translate is horrendously bad at Korean, especially with slang and accidental typos. Like nonsense bad.
Same in Hungarian, machine translation still often gives hilariously bad results. It’s especially prone to mixing up formal and informal ‘you’ within the same paragraph, something which humans never do. At least it’s easy to tell when a website is one of those ‘auto-translated to 30 languages’ content mill.
I recently have edited a small wiki page that was obviously written by someone that wasn’t proficient in English. I used AI to just reword what was already written and then I edited the output myself. It did a pretty good job. It was a page about some B-list Indonesian actress that I just stumbled upon and I didn’t want to put time and effort into it but the page really needed work done.
Wikipedia’s translation tool for porting articles between languages currently uses google translate so I could see an LLM being an improvement but LLMs are also way way costlier than normal translation models like google translate. Would it be worth it? And also would the better LLM translations make editors less likely to reword the translation to make it’s tone better?
You can use an LLM to reword the translation to make the tone better. It’s literally what LLMs are designed to do