Kagi Translate: Grammar Insights Feature

AntonMarchukov

I am currently learning Hungarian and it is helpful for me to use Kagi Assistant with custom prompt to break down sentences grammatically, list the infinitives, present forms etc.

Here is the example of how this works for me there.

The prompt I use: "I am learning Hungarian and want you to translate the Hungarian sentence and explain its grammar. The sentence is:"

And the result it gives for "Szeretem a jó bort.":

Translation:
"I love good wine."

Grammar explanation:

Szeretem - 1st person singular present definite conjugation of "szeret" (to love). The definite conjugation (-em ending) is used because the direct object ("wine") is specific (with the definite article "a").

a jó bort - Accusative case noun phrase:
a = definite article
jó = adjective "good" (nominative form, unchanged)
bort = accusative form of "bor" (wine). The "-t" suffix marks the direct object.

Hungarian word order: Subject (implied "én") + Verb + Object.

I thought that it could be a useful feature integrated into Kagi translate depending on circumstances as it seems that a lot of this might already be there at least for the target language with insights feature and the proofread option.

I have two options in mind:

Create special virtual target language "Explain Grammar" or maybe use existing "Proofread" mode and than copy the input text with highlights which are when clicked give grammatical insights (visually same as current insights work).
Use already existing source language text area and add a toggle "Grammar insights" there which when enabled highlight specific words with ability to click and get grammar explanations (again same as current insights work).

I personally would prefer the second option.

Giorgio

Hey, thanks for the idea. We're always open to adding more features to translate - especially for language learners.
We'd prefer option 2, but we need to find a way to implement it with good UX that doesn't clutter the interface.

There's only so much we can display simultaneously, so it would be really helpful to know all the data/questions that you ask LLMs for learning languages - this would allow us to prioritize showing the most common/useful data. Everyone is welcome to contribute their thoughts! 🙂

AntonMarchukov

Hey, I was thinking about it a bit more as this UI clutter is indeed a problem. Also probably not all features that AI provide are easily integrated into the UI without clutter.

I think the grammar comments generally fit into two groups (based in the example above):

Break down of individual words/group of words depending in the language. It would be useful for me if individual elements of the original text (not a linguist, but I guess it could be either words of group of words depending on the language) gets underlined and when clicked a pop up with explanation appears. The explanation will need to list at least the word present form infinitive, the gender (if any) and the case used, maybe some information about how this case is formed from the infinitive.

For the original example this is how AI separates "Szeretem a jó bort." into relevant parts: "Szeretem - 1st person singular present definite conjugation of "szeret" (to love). The definite conjugation (-em ending) is used because the direct object ("wine") is specific (with the definite article "a").", "a = definite article", "ó = adjective "good" (nominative form, unchanged)", "bort = accusative form of "bor" (wine). The "-t" suffix marks the direct object.". All those can go into individual popups for specific parts.

General information about the sentence. E.g. the translation, word order used, is it a question or not, etc. This is sentence level related, so harder to fit. One of the idea is maybe to have some pictogram (a bulb maybe) appear after a sentence with grammar explanation about the sentence when clicked.

In the example this would be "Translation: I love good wine, word order: Subject (implied "én") + Verb + Object.".

If all this is only enabled when some button "explain grammar" is toggled below the clutter is minimized as nothing is enabled till enabled. And once enabled it only provides highlights and additional information is available only when specific highlight marker is clicked for specific element (which also possibly reduces server/LLM load).