AI context-based translation: amazing! But quickly out-of-context in practice?

[cspotcode]

Sorry for the thread title; let me explain.

The new context-based definitions are awesome! I love seeing a terse, relevant translation at the very top of the list. So many translations in Japanese include a lot of formatting noise or even the word written out as hiragana! (puts spoilers on reverse flash cards) So I’m really happy about this new feature!

However, the context-based translation is not shown any time I already have a saved LingQ. I wonder if it should be shown in those cases, too.

Here’s why:

1. I see a new word with multiple meanings, choose the context-based translation for meaning A in this context, create a LingQ
2. see the same word in different context, see my saved LingQ which is for the wrong context. What if I could also see the context-based translation, without clicking the “expand” button? What if it was always shown by default?

I think this would be the best way for LingQ to quickly show me what a word means, which is the whole point: to understand a word in context, as it is encountered in native materials.

Here is a real-world example of a LingQ that I created before this AI feature, which includes both definitions of a word:

We can see that the AI helpfully clarifies which definition is used in this context, yet I had to click the expand button to see it.

Consider an alternate history where I had created this LingQ on this lesson, choosing the AI suggestion “body.” Then, the next time I encounter it in a context where it means “health”, I would have an inaccurate LingQ. That would be fine if the AI suggestion was also shown directly beneath it, as a helpful cue to me. Then I can adjust my LingQ if I want.

I’m curious to hear if other users disagree with me or feel like I’m using the tools incorrectly, perhaps I should be creating my LingQs differently?

[cspotcode]

Here’s a picture of roughly what I imagine it might look like. I see my saved LingQ, but I also see the contextual meaning without clicking any extra buttons.

[gmeyer]

This is a very interesting example for me.

As I understand it, both “body” and “health” are somewhat loose translations of “体”.

“Body” in English has broader meaning than just the physical aspects of a living being. We have “bodies” of water and “bodies” of work and more. For “体”, at least when standing alone as a word in Japanese, we’re really just talking about the bodies of living beings.

“Health” in English generally has default positive connotation where “bad” can have to be explicitly added. For “体” in Japanese, we’re talking about the state of the bodies of living beings. [If I were to pick two common words in English, I might pick “bodily status”. (I find it interesting that “body” in the noun form is rather generic whereas “bodily” in the adverbial form is rather tied to that of living beings.)

Anyhow, I’d pick maybe “body/bodily status (health)” as a simple LingQ definition. Here, my ordering is intentional.

Now, if I dig into my ten-dollar English vocabulary, I can find such as “corporeality”! As I understand this specific Japanese word, that’s a really good one-word definition. However… I don’t think I’d ever use it in LingQ and I doubt anyone else would either.

What’s more important though for AI assisting in definitions for Japanese is kana! Many, if not most, LingQ students of Japanese put the kana reading of the word in as part of the definition as learning a word and learning its reading(s) are closely related yet somewhat separate endeavors.

In this example, what I personally would love to see LingQ offer is “[からだ] body/bodily status (health)”.

I love this example, because I’m going to have to see “体” in a good number of contexts that it doesn’t just mean “body” nor does it mean any kind of “body” as used in English.

To answer your question, YES, you are using it correctly. With Japanese, I’d just suggest you get the kana in the definition too.

But, the bigger question is whether or not LingQ is using the AI correctly. While the answer here IMO is a yes too, I’d love to see LingQ help me get to these kinds of understandings quicker.

Just now I asked ChatGPT 4o two questions:

What’s a simple definition of “体” in English using three or fewer words carrying as much nuance as possible?

Would you repeat that and put its kana in square brackets prior to the English definition?

It offered me:

[からだ] Body, physique.

While “physique” is too tied to musculature and “体” includes the bones, inner organs and pretty much everything else, it’s not a bad translation. “Physique” doesn’t carry the additional default positive connotations of “health” while yet emphasizing the state or status of the body.

Anyhow, this is a great example. While there is a lot of depth that could be grappled with here, just getting the kana in to the AI-generated translations would be AWESOME!

I love what cspotcode suggested, I’d just like to see LingQ suggest something more like:

[からだ] Body, physique

[obahtiyar]

I don’t know if I’m writing in the right place, but I would like to add a few things.

Maybe there could be more than one artificial intelligence translation. For example, one is the current context, the other is the other meanings of the word…

A suggestion about artificial intelligence translation.

If a word has more than one meaning, it doesn’t translate the other meanings. It only writes one meaning.

As far as I understand, LingQ sends the entire sentence to CahatGPT while translating the word. However, that word may have more than one meaning, and it would be great if we wrote these meanings with commas.

For example:
When LingQ translates it, it writes as follows.
assimilate: to adapt

When I want it, it writes as follows.
assimilate - to assimilate, to absorb, to digest

It would be nice if you do something like this.
Also I am eagerly waiting for ChatGPT translation to come to iPad and iPhone.

Thanks LingQ

[cspotcode]

I worry about missing the point of the AI-generated contextual meaning.

The AI is meant to generate only the meaning in this one context. There is no point in the AI trying to include all definitions, to duplicate the role of a dictionary. We already have perfectly good dictionaries with comprehensive definitions.

The issue is that I want to see the contextual meaning alongside a more comprehensive definition. But once I’ve saved a LingQ, the former is hidden by default.

[roosterburton]

On the IoS app GPT hints are currently being trialed along with the contextual definitions.

Your result will look something like this when the feature is ready.

On Rooster Reader it is displayed like this

[obahtiyar]

@cspotcode

I worry about missing the point of the AI-generated contextual meaning.

The AI is meant to generate only the meaning in this one context. There is no point in the AI trying to include all definitions, to duplicate the role of a dictionary. We already have perfectly good dictionaries with comprehensive definitions.

The issue is that I want to see the contextual meaning alongside a more comprehensive definition. But once I’ve saved a LingQ, the former is hidden by default.

Maybe there are proper translations of dictionaries for your native language, but it is a problem for my native language.

Translations from English to Turkish are sometimes given very briefly, in some places they are given unnecessarily extensively, you lose the meaning. (Sometimes I have seen examples where I did not understand the meaning of the dictionary, the dictionary did not fit the context, it did not settle until Artificial Intelligence came along. Now I just give the word, it expresses its different meanings completely and I learn faster.)

As a second example, there is no dictionary from Bulgarian to Turkish. Google translate first translates from Bulgarian to English, then from English to Turkish and gives the meaning of the word. This sometimes causes a very irrelevant translation. (I know that the translation of the word is not correct since I studied Bulgarian as a child, but I could not get its full meaning. I can say that Artificial Intelligence provided a great solution in this regard.)

It may seem contrary to your opinion, but I am also very happy to use it as a dictionary.

[cspotcode]

That never occurred to me, but now I understand:

It would be ideal to see both the context-specific meaning, and a more comprehensive definition, both generated by AI.

The latter does not require any context, so those definitions can be generated once for all users and cached by LingQ to save resources.