Yes, that setting will be coming soon.
PLEASE allow the option to turn off the AI suggested translations as soon as humanly possible.
@mark following this thread, please fix that when one blue word is clicked, AI would return the definition only for that word and not for all the surrounding words as well.
However, when more words are selected, AI is returning very good definitions so far.
Didnāt realize this! Thanks for pointing it out! Iāve now tried it and it works amazingly well! Solves all of my translations issues!
Hello
- The default translation engine for phrase and sentence translation was switched from Deepl to ChatGPT. Why? Is ChatGPT any better in translation issues? Cause Deepl also uses AI for its work and it is an engine specially designed for translation issues while ChatGPT is a more general neuronal network.
- The AI first suggestion for word translation comes from where? From ChatGPT? If I change the default translation engine in the settings, does the source of the word translation change too?
Youāre right that DeepL is a more typical neural network (I believe itās based on convolutional neural network technology), and in theory I also expected it to outshine LLMs, but in practice weāve found it doesnāt necessarily do a much better job. The large training corpora used for larger LLMs seems to put them on equal and sometimes superior footing to DeepL. There are certainly some exceptions and edge cases, but for shorter text, individual words or phrases that are partially cut off from the main text, the performance of DeepL drops off notably. In a sense, itās more ābrittleā than LLMs. DeepLās advantage over most commercial LLMs today is in translating long text between less resourced language pairs, such as translating Estonian to Lithuanian.
The AI suggestion for word translation is currently coming from finetuned GPT models that have been trained on datasets weāve curated here at LingQ. If you change the default translation engine in settings, that wonāt impact the contextual hint translations. However, weāre currently in the process of continuing to upgrade this system, so stay tuned for more enhancements!
I donāt know whether it is me who has learned how to use and not use the AI suggestions, or whether itās the AI that is learning, but Iād like to report that on my end, I have noticed significant improvements in the AI suggestions across different languages over about the past week.
I still occasionally get a phrase translation for a single word, but itās much less frequent now. Conversely, I sometimes get translations that include an explanation in keywords or suggest alternate definitions now!
I am guessing that both are trueāI am getting better at spotting the useful AI translations and ignoring the bad ones, plus the AI is learning what kinds of definitions are appropriate for use in LingQ.
The feature still needs refinement; for instance I am still unhappy with how it kicks out the second āpopular meaningā and would prefer if it were displayed in addition, rather than instead. But I do feel itās moving in the right, useful direction!
I find it annoying when it translates the meaning of two or three words together instead of giving me the single word, and then when I select those same words to create a lingq for that collocation, it gives me a completely wrong translation instead of the same one it gave me when I clicked on a single word.
But overall, I agree that it is becoming a good help in speeding things up and getting more translations. It is definitely worth spending more time improving this functionality. Itās going to be a killer feature. IMHO.
Oh yes, itās very annoying when it happens. Sometimes, it actually can be a helpful hint for understanding the word in context, but thatās not the same as being suitable to save as a definition (and sometimes it can be hard to figure out whether it did or did not translate multiple words). That said, it does seem to be moving in the right direction.
The AI is extremely annoying and we really, really need to have the option to turn it off. It keeps suggesting entire phrases for one word, or a meaning specific to that particular context, when the word has multiple meanings. This makes it impossible to review later. Most of the time now I just look the word up in a regular dictionary because I canāt trust either the AI or crowdsourced definitions. Itās a lot more work.
Great new additions! What about adding colors to mark different words? For example, when I study a language I first try to identify the verbs, because everything revolves around them. What about having a unique color to mark verbs (or nouns or adjectives) and be able to see them at a first glance?
Thank you so very, very, very much
ŠŃŠ¶Šµ Š“ŃŠŗŃŃ
Lingq provides quick links to some online dictionaries, and thatās all I use. The AI generated material is too unreliable, and the crowd sourced material is too random.
Thanks for fixing the phrase lingQ problem for Danish. Now I have a problem with German. Many times I get no translation for a phrase in that language and so Have to use the Google translate window. This involves several extra clicks and so is very slow.
āBelieve it or not, itās not possible to account for the way everyone uses the app.ā
Very professional.
Iām with you on this. I would prefer it if it would be possible to just show the dictionary entries upon clicking on a word. This would speed up things tremendously.
Especially the ai translations are not only errorsome but even seem to degrade over time. The longer the system is in place the more nonsense it seems to generate.
Thatās strange. Are you able to provide examples of the phrases that arenāt translating? Links to the lesson and the phrases causing problems. You can send these to support if you donāt want to post them here.
That isnāt possible but you can toggle the dictionary quickly using the keyboard shortcut D. We will refine the prompts to get the meanings showing more consistently. It seems there are certain words that provide phrase meanings instead of word meanings. We are working on fixing that.
Thank you, but I am already aware of that. It isnāt the opening of the dictionary but more the manual typing to copy the translations from the dictionary. Just clicking on a suggestion is much faster, but if the translations, either ai or user generated, arenāt relyable and I have to countercheck it with the dictionary (not everytime, but often enough) it really slows things down.
Maybe it would be worth to consider the possibility to sort of ādownvoteā suggestions? So a symbol next to each suggestion we can click on if we consider it to be wrong (or redundant, as there are quiet a lot of duplicates). If enough users have done this for a suggestion, it could be stopped to be shown. Something like that.
Premium users do have the ability to report or flag bad translations. The AI context based meaning is another option for you to compare with the existing popular meanings. That is all we can do. If you still donāt like any of the meanings, you have to check the dictionary and manually input a meaning that works for you. That is how LingQ works. It would be great if we could automate more from dictionaries but that is not possible.