I’m beginning to see all too often that the “AI” provided translation of a word in the iOS app is unreliable, and I’d like to turn it off - especially since it comes up top of the list. Here’s an example reading a Russian text: “счете” (actually счёте, but the use of umlaut is optional). The first definition of “frustration” is erroneous. The next word in context has a connection to frustration: “фрустрировано” and fully in context: “[в] счете фрустрировано его стремление к смыслу” on account of his frustration in his search for meaning.
So AI is going astray, and I want to turn AI off. I was happy with translation help previously. Not sure how AI helps at this time.
I think it depends on the language and on the perception. I have the opposite experience.
In the last 14 days I have read 85k words, and created 465 lingqs. Mixed iOS and web-app.
I can definitely say that I don’t recall completely wrong translations more than 10 times. If we are talking only about single wrong translations in context.
Perhaps. I have learned to watch closely. I’m feeling like there’s only a 50% chance it’s going to be right. When it does get it right, so did the usual dictionary. So far, I’m not seeing the point in Russian.
Here’s another example from this morning’s reading. I highlighted the person’s last name, but AI gave me a translation of the next three words: “From New Zealand.”
50% chance is quite high, I don’t have that percentage, not even close.
What you describe above is possibly an error of the prompt. Something the team can correct by improving the prompt. What I want to express is that it doesn’t depend on the AI but on how the team is using the AI with their system. For some reason, in that circumstance, the AI is returning something else because the commands given are not correct.
Give them a bit of time and the team will figure it out.
It’s always disconcerting when users are involuntarily or unknowingly made to be beta testers. If their use of AI is this rough around then edges it should be marked as such. They could easily tag this as “beta version, use at your own risk,” and I’d be fine with it.
My experience studying Czech and Bulgarian is that the definition I want keeps popping up and then being pushed off by the ai translation, which is a decent definition for the word in the context but NOT a good definition of the word in general. It’s very frustrating. Instead of clicking once to select the word and once to choose the right definition, I have to click twice to creat the Lingq and then 3 more times to reopen the Lingq, open the definition list, and select the right definition.
The more complicated the texts your reading become and the further you are moving away from the most basic and obvious language use, the more unreliable the translations become. The translation “retrieved from” is given to a big chunk of unknown words. Translations of sentences or part of sentences become nonsensical. Lingq is really just suitable for the most independent and alert learners who are able to detect errors, which of course is not always easy in a foreign language. If you cannot do that, you will unwittingly absorb a lot of bunk.
Sometimes perhaps, but the cure is to realize that context is everything. A dictionary only gives clues - so Lingq is no different. Consume massive amounts of input and gain your own experience with the language.
I do see it getting better, but I’m not a fan of AI translating a phase I have not selected. Here’s an example. The Russian text: “у Стивена Кинга” (English: belonging to Stephen King). Clicking on “Стивена” should translate to Stephen. It does, but the AI translation - the first thing in the list – translates to “Stephen King.” It is including additional words I have yet to click on.
I know this isn’t perhaps what you’re hoping for, but I must say eventually AI translations grew on me. Especially since the lingqs on Afrikaans are so lacking. One work around I would suggest might be to switch the AI model. Some are better in some languages than others.
