@davideroccato Mine works like this.
I just clicked the “refresh” button and it changed the translation.
The first one was wrong, the second one was right.
3 Likes
@D.lfzM I can definitely see the change. I’ve tried right now and there was no reaction at all. I will try in the next days with another source. I have Google Translator as main one.
Yes, it won’t change it if it’s already correct. Or, at least re-generating just generates the same translation. It is there for when the translation has been generated or split incorrectly for some reason and is wrong. It allows you to generate it again.
2 Likes
@Ginkgo58 We’ll look into that. What language combination are you using?
@ed_shin They are meant to be context specific and you can always save multiple meanings. Now, multiple meanings are combined when shown in the minimized popup so you can see all your meanings together. The combining of meanings can be turned off in the settings.
2 Likes
I cannot see this on the web app. I assume it’s not widely available yet?
1 Like
Sorry! It should be there later today.
2 Likes
@mark For me it often is quiet the opposite. Single words are given translation suggestions that correspond to the phrase they are part of. My language combination is Korean-German (learning the first, beeing the latter).
2 Likes
Yes, that is my observation, as well, in various languages. The translation for a whole phrase will be suggested as a definition for a single word (which, of course, will not fit the next time i run into that word, as per LingQ’s lingQing concept).
Perhaps what’s needed is AI-assisted phrase highlighting?
3 Likes
I think you mentioned above, “We will be working on refining those meanings so they restrict the meaning to that of the word so that should be corrected.” I think this is part of the issue I was trying to communicate to you. I’m glad a solution is on its way. In the meantime, perhaps the AI translations for Popular Meanings could appear at the bottom of the list instead of the top?
Or, perhaps, even roll back the way the Popular Meanings worked to the way they worked before? I haven’t had problems with the old translations method for Popular Meanings, especially when selecting multiple words. The new AI translations for Popular Meanings are providing solutions to people looking for a context specific translation, which is slightly different from people who just want to look up the popular meaning of a word or phrase.
I like @gmeyer 's idea of displaying these context specific translations clearly so we don’t mix them up with dictionary definitions or human/user-curated definitions. Perhaps we can display these context specific AI translations under a section called “Context Specific Meanings” instead of “Popular Meanings”?
4 Likes
I use a Boox reader to cut down on eye strain. The new green theme helps distinguish the blue and yellow colours of words. Otherwise, I can’t see most yellows. Variations that would work on the Boox would be helpful. Maybe underline styles that aren’t very colour dependent. But this seems to help. Thanks.
A next sentenece+translation + audio button and shortcut is still needed to make sense for comprehensible input.
Make today the default please for stats. Who cares about 7 days?
Midel aligned stats are deply untrustworthy. Please right alight numbers.
If you can use AI, you can translate into Irish. Please add a slot and release it with AI translations. We don’t need mini-stories or a grammar guide.
1 Like
Hmm… This is an interesting idea. That is, having the AI Contextual Definitions ranked by occurence, based on how many times the Contextual Definition has appeared for the individual user. As in, my AI Contextual Definitions would rank based on the content I read, which may differ from yours. I imagine this wouldn’t be too challenging, as you just need to compare the current AI Contextual Definition with the previous ones (which they are probably already doing) and then add +1 to the counter, if it’s the same. Then sort for order of appearance. @mark
Well said.
What about the refresh button rotates which source is used? As in, first the translation is from ChatGPT, if you don’t like that translation, the refresh button translates with Google Translate, and if you don’t like that translation, the refresh button translates DeepL. So the rotation would be: 1) ChatGPT, 2) Google Translate, 3) DeepL. Or this is what is already being done?
3 Likes
Beyond the technology of translation itself, the prompts given to the AI play a crucial role. For instance, when I ask ChatGPT to translate a few sentences or paragraphs, I often guide it with instructions like “literally” and “at a B2 level” or “with dynamic equivalence” and “at a B1 level of vocabulary.”
Consider the concepts discussed here: Getting to Grips with Translation Theory.
With the Bible being the most translated book, translation theory is arguably most advanced in its context, though it is also arguably the most troubled due to the complexities introduced by interpretive aspects and doctrinal biases. I bring this up because my manual interaction with ChatGPT in translation varies significantly.
I often find it effective to translate a passage in different styles to observe the choices it makes. This method helps with translation of “comprehensible input,” and I apply a similar approach when I write to create “comprehensible output.”
For instance, in addition to comprehensible input and LingQ, I’ve been taking more traditional online classes with Lingoda. For each one-hour class, I dedicate at least an hour to preparation. I read everything in advance, complete all exercises, and often write two or three pages of content. Typically, I draft my prep work in Word and then use ChatGPT to confirm the grammar and improve the fluidity. At this stage, I explore prompts for different levels of vocabulary, grammar, and idiomatic structures. I sometimes experiment with different registers of output. This practice has helped me understand how much output is shaped by the contours of my mother tongue. Beyond these limits, I want to discover more native ways of presenting ideas in the target language. I frequently ask ChatGPT to explain why it has phrased things a certain way.
I’ll offer a revealing anecdote. As I advance in French, I often ask ChatGPT for French idioms similar to those I know in English. The other day, I referred to some of Lingoda’s content on a topic au ras des pâquerettes to describe the shallowness of their treatment of a complex topic. These examples illustrate the limits of “AI word translation” because word-for-word translation has profound limitations. For instance, word-for-word, au ras des pâquerettes is “at the level of the daisies.” However, if I ask ChatGPT for the dynamic equivalence of that phrase, it tells me it means “shallow” or “superficial.” I’ve now learned several things: the word ras (level with, near, flush), the word pâquerettes (daisies), and the idiom itself. Further, I clarified by asking ChatGPT, “What’s the difference between pâquerettes and marguerites?” I learned they are two different species: pâquerettes are the small daisies that pop up in lawns, often as weeds, whereas marguerites are the larger ones typically cultivated by gardeners. This nuance helps me internalize the idiom’s pejorative nature.
I share this frame of reference and this specific anecdote to illustrate that the power AI offers goes beyond contextual translation at the word level. Additionally, I find multiple translations “a good thing” for helpful for internalizing nuance. Moreover, I find AI as powerful, if not more so, in aiding my learning through what I consider “comprehensible output.”
In my opinion, LingQ’s human community-based Writing Exchange needs to be completely scrapped and rethought. With AI, IMO, there are now much better ways to acquire language usage skills through the struggle of creating “comprehensible output.” Over the last year, I’ve written hundreds of pages of French, most of which I’ve reviewed back-and-forth with ChatGPT to explore and grapple with various ways to convey my desired meaning. Wanting to, and being able to, better express oneself has incredibly positive feedback for the affective filter.
Anyhow, I look forward to seeing what LingQ does with all of this proceeding ahead.
2 Likes
I completely agree, but I think there are other priorities first.
Same here:
Previously, the translations of phrases / shorter sentences (in English, French, Spanish, etc.) were quite good.
Now L2 phrases / sentences are narrowed down to one (or a few) word(s) as hints, which are often not correct L1 translations.
Example:
The phrase " a master of a certain kind of paranoid fiction" (from the Wikipedia article, “Neuromancer”) is translated into German as: “Meister paranoider Fiktion” = English: “master of paranoid fiction”.
That’s ok as a hint when reading, but rather useless when trying to learn / acquire phrases.
So, the end result is:
Almost every short single sentence / phrase has to be translated manually by using Deepl & Co and then pasted into LingQ’s “saved meaning” box.
@mark
Sorry, that’s not an improvement, but just terrible! 
And the underlying idea that language consists of “single” words (with multiple meanings) somehow woven together is even worse (who on Earth learns languages like that - in 2024!?)
Unfortunately, LingQ is probably no longer a good solution for Reading And / While Listening approaches - maybe we need something else (also: AI first) here…
1 Like
Good point, but we’ve known for many (!) years that languages don’t simply consist of “single / individual” words, but are composed of “conventionalized word groups” (idioms are just a more extreme example of this phenomenon). See, esp., all the research on “collocations”.
“Anyhow, I look forward to seeing what LingQ does with all of this proceeding ahead.”
I have given up on LingQ and I cancelled my subscription a few minutes ago because I’m unhappy with almost everything now (the design decisions, the SW engineering, the poor QA, the integration of genAI. etc- even their SLA philosophy in general).
In my opinion, with the rise of genAI, it’s probably better to look for “new” (AI-centric) SLA approaches and also take a look at what the competition is doing…
2 Likes
Let us know when you find something really valuable.
One problem I see is that there are a lot of clones popping up. There are a lot of copycat softwares that promise to learn languages with AI, but they could be out of business in a year. It’s hard to invest time in learning a new, random platform with all this uncertainty.
I think the LingQ team is listening and trying to find the right solutions. But it’s true that I personally use AI more and more these days. I basically use it every day when I use LingQ.
2 Likes
I’ve noticed that when i click on a word, all my definitions are bunched together on one line so they are difficult to read. E.g as below -
Can you tell me how to get back to the old system where each definition has it’s own line.
I know I can put the dictionary pop-up to the side of the page, but I don’t want to do this since it de-centres the transcript. I can click on the line to open up all the definitions but if I’m doing this every time it’s a step back in efficiency.
Thanks.
1 Like
Awesome job on adding the translation refresh! But is it only in Sentence View? It would be nice to have it for the Page View if there are multiple errors in translation that need a refresh.
1 Like
Since we are on the topic of translation, would it be possible at some point to get a literal, word-for-word translation? Some languages have different word orders than English and have a literal translation would help to learn them. Cheers!
1 Like
I asked Mark that same question two days ago and it’s not a bug, but a new way of doing things.
Apparently they will add a feature to deactivate it.
1 Like