Context dependent AI word translation, new stats page, new themes and more

Hi Everyone,

We have added a bunch of new features over the past month or so. Here are the major announcements:

New theme options in the Reader on all apps

This is something frequently requested so we decided to spend the time and provide a range of page theme options on all platforms along with a quicker way to access these new settings. You will now find these new theme and font options right in the reader by clicking the Aa control in the top right.

  • Edit: New theme colours are not yet ready on mobile but will be there soon.

Context based AI word translations

Very often words have multiple words and the generic definition found in our Popular Meanings may not always be accurate. We have now made it so that the first meaning in the list of Popular Meanings is generated by Ai and is based on the context you are looking at. This was just launched on web and will be available on both mobile apps within a week. This is very interesting functionality!

**


**

You have probably seen the new Statistics interface by now which includes the ability to dive deep on the theory behind the LingQ method. We also added two new metrics based on feedback from you, our users - Reading Speed shown in Words Per Minute, which allows you to see your improvement over time, and Study Time which just tracks how long you spend using LingQ.

Refresh Sentence Translation

There are times when the sentence translation shown in Sentence View is not correct. There is now a control that will refresh and pull a new translation for that sentence from whichever translation source you have specified in your settings - Google Translate, DeepL, Ai.

Let us know your thoughts on any of these enhancements on the forum post. Keep the feedback coming as it very much can influence our future development.

10 Likes

The context-dependent suggestions are a great idea. One annoyance they sometimes cause is that they displace the second “popular meaning” from being shown. This sometimes results in showing a “popular meaning” under an identical AI-suggested meaning. But seeing two slightly different definitions can often be helpful in choosing the correct translation and understanding the word.

Would it be feasible to make it so that the AI-suggestion doesn’t duplicate what’s already being shown as a popular meaning? Perhaps the AI can merge with and “endorse” the popular meaning in case its own result coincides with it?

1 Like

This might be one of the most useful updates to the platform yet, nice work on this one!

7 Likes

Thanks for the introduction of AI translations. There is a new bug since this update though. When a user has several separate entries for one word, they end up being merged together in a single box and the text is all jumbled up.

4 Likes

I also like this suggestion! For my language cantonese, I find a lot of the suggestions not only have the meaning but also how to pronounce the character. Endorsing the popular meaning could help solve this issue as well as shrinking the number of redundant dictionary descriptions in the future.

1 Like

We are aware of that issue and will see what can be done there. Ideally, we can make sure the meanings shown are different.

2 Likes

This isn’t a bug. It groups all definitions together in the minimized popup. This becomes useful when you add an Ai translation to an existing LingQ. Rather than having to edit, copy, paste the new Ai translation into your existing meaning, it will add it as a separate meaning but display it as one. However, we are going to add a setting to disable this so that you can see them separately the way they used to display.

3 Likes

We wouldn’t likely endorse any meanings with additional information like you are showing. Those meanings cause review activities to be redundant since the answer is provided in the meaning. I think endorsing is unlikely anyway since the Ai meaning is always context specific.

2 Likes

Please add this setting as soon as possible as the current display looks terribly messy. When you have several definitions, they all end up being crammed into one box and the size of the box is too small when you have several lines of words. Some definitions overlap with others.

2 Likes

This is a big improvement visually and functionally. Thank you.

2 Likes

Are there any plans to make this feature optional?

  1. Right now, it often translates a single word that is a part of a phrase as the whole phrase.

As a somewhat experienced language learner, I can get around this problem without much stress. However, as a somewhat experienced language tutor who uses LingQ with my beginner students, I have seen them getting confused by simpler things…

  1. If it’s always enabled, it kind of robs me of one of my favorite parts of reading on LingQ, which is figuring out which meaning fits the given context.
2 Likes

We will be working on refining those meanings so they restrict the meaning to that of the word so that should be corrected. As for a setting, we would rather not have one. More settings typically lead to more confusion and, in general, settings don’t get used by the majority of users. This context based meaning is identified by the AI icon as different so can be used as a meaning or just as an indicator as you look at and choose from the list of available meanings.
At any rate, let’s use it as is for a while before we make further refinements.

3 Likes

Context based AI word translations
For Polish, I find that the popular meanings curated by users are much better. And the old, default Google Translate popular meanings for Polish are also much better than the new AI popular meanings. The AI translations are so specific to that sentence, that the LingQ won’t make sense when I see the same words again. I think I’d much prefer that LingQ return to the old popular meanings for Polish, or at least put the AI popular meanings in a separate section.

For example, the AI translation here is overly context specific:

3 Likes

Thanks for the update, @mark. We very much appreciate these posts you make. I’m very much interested in trying out the AI Contextual Translations, when I comes to Android. The two new metrics are also great adds.

Well done @mark and team!

2 Likes

I am finding that when I am LingQing a whole phrase, I am getting the meaning of only a word or two of the phrase. In order to get the meaning of the whole phrase I have to use the pop-up Google Translate window. Doing this is quite time consuming.

3 Likes

For Catalan also the popularly curated translations or Google Translate translations seem to be better still, so I agree!

1 Like

@mark
Not sure I understand this correctly because when I click nothing happens. I suppose this is because I use the same source that you use by default?
What are you using by default? So I change with a different one in order to have two options when I refresh.

The AI translations are very good. I also like when they are the same as other popular meanings because they give me an option to double verify similar translations. If both the popular meaning and the AI tell the same, it means there is more probability it’s right.
So far, the translations are very good but I usually often enter my own definitions. But when in a hurry, I can rely on AI to speed up the day.
It’s a very good tool.

I love the subtly by which you’re grafting in AI here!

Adding to the discussion, when learning a word, it’s important to grasp:

  1. It’s meaning in its context
  2. It’s general, primary, or as you put it "generic meaning
  3. But yet maybe too, any other additional meanings

A dictionary offers points 2 and 3 above, but not number 1.

Yet further beyond the traditional typography of a dictionary, you have several graphical treatments to work with in UI design.

Positioning, which you mentioned, is important; you’re putting the context-specific definition in first spot.

However, in addition to positioning, you also have other elements to work with…

Font weight: For instance, you could put the general/primary/generic meaning in bold.

There are other aspects to font emphasis, such as underlining and more.

There’s color too, not just of the font but of the background of the “cartouche” around the definition. For instance, you could instead of putting the contextual definition first, you could, much as in the method of a traditional dictionary, position the primary definition first, and then rank by usage yet go further aqnd bold the contextual definition and maybe even have a color border around the “cartouche” / “button” of that definition. At a glance, the primary definition could be grasped, as well as the nature of the contextual usage, as well as being reminded of any other definitions or usages of the word.

I’m not immediately suggesting what you should do, but going down a path of subtle cues is going to be the fastest way to incept the right ideas into the mind.

I’m also not suggesting all of this is immediately doable in mid-2024, but with the speed of change, it may be possible to roadmap things such as this already.

The historic dictionary copyright issues are a problem of the past. Rather than user contributed content being the only available means to bypass dictionary copyright issues, generative AI may be the means to not only surpass copyright issues but supersede the functionality of what a dictionary has historically provided for centuries.

Personally, I think dictionaries may go the way of sextants and serve only in the future role of a vintage collectable for the office.

Advanced AI behind the scenes with subtle cues at the surface may offer LingQ huge potential to differentiate in language learning. As a content-driven platform, you have the pole position.

3 Likes

It’s working fine for me, I think it pulls another translation from the same source you have specified. Maybe it’s clearer with more complex sentences, the ones where the translation may be more doubtful.

2 Likes

@D.lfzM I don’t know, I have tried to click the wheel on different sentences and there was no reaction. Maybe it would be useful to have a sort of visual feedback to be sure it was working and didn’t find anything, or not.