Kagi Translate Feedback

Giorgio

Giorgio Mahalo!

1) There is something weird going on with romanization. If I translate from Japanese and try the proofread target there are two different romanizations for the same text. As far as I understand, in the example provided, the left side is extremely formal, and I don't believe majority of people would use that pronunciation. The romanization on the right side is the one that is the most commonly used.

Why are these two different? It would be nice if it used the most common form on the left side as well. Changing the settings to formal/informal/natural/literal or adding context does not affect it at all.

These are extremely useful when you are learning the language, so you can figure out what kanjis are constructed of, and how you should pronounce it, but when it uses something that is not used in every day conversation, it's not very helpful. It would help if the settings for formality would affect these as well (on the input side, as I am trying to decipher the text that I paste in). If it's set to standard, it should use everyday conversation.

Here is the example Japanese sentence: "私は日本人です。" (watashi wa nihonjin desu)

2) When I change the text and languages, the URL should reflect that. Right now it stays the same as when I loaded the page (went via !t bang). So if I reload, the translation is lost and it's not possible to share the URL to it.

When trying to translate a telegraph article, the page mixes portuguese (the output language) with the original article's paragraphs:

I am using just the "Translated" option. If I use the "Both" option, it does the same thing, but in a more expected/organized way. And in "Original", it does the same as "Translated".

Link of example: https://translate.kagi.com/Portuguese/pt_br/https://telegra.ph/Lawfare-Alexandre-de-Moraes-strategy-to-persecute-opponents-02-19

Working well for the moment !

One feature that will make it even better would be the addition of alternatives translations.

Simply awesome. Thank you, @Giorgio!

1) I tried translating "Mars" (the planet) from English to Japanese. It offered incorrect romaji. The romaji displayed is in Chinese, not Japanese. Romaji should be "kasei". This is also visible from the hiragana that is offered in alternative version (かせい is kasei).

In fact, the alternative translation is exactly the same, only with added hiragana. Hiragana is helpful to decode the kanji, but it's not the alternative.

2) Why is it not possible to have a dark theme on Kagi Translate? At the very least there should be an option in the menu to the top right, and the best would be that it follows the same theme as defined for kagi.com search itself (I have that one set to dark theme). I see that some people are posting dark themed screenshots, so what am I missing, is it only reading the browser setting?

Two ideas:
DeepL has a popover dictionary for individual words that shows some useful info about a word, especially for those learning languages and using Kagi to help with that.

Second and somewhat similar: If possible it would also be useful to highlight a word on the source text, and have the corresponding translated word/phrase also highlighted. For example:
Here I have фотография highlighted:

The corresponding translated word photo should then be highlighted:

Not sure if this is possible, but if it is it would be appreciated!

frin I see that some people are posting dark themed screenshots, so what am I missing, is it only reading the browser setting?

It detects your browser's theme - if your browser is dark, the page switches to dark too. Still, I think a manual theme switch would be useful.

I see that "Slovenian" is on the list twice. The first one has language code "ls" which is wrong. The correct one is "sl". The "ls" is an invalid ISO language code, it does not exist.

Feature suggestion:
keep a dictionary in localstorage that remembers certain translations and prioritizes them, deepl does this quite nicely.

Giorgio Sweet but seems a little buggy, at least on the sample I tried.

В город приходит вечер should just correspond to Evening comes to the city

Again here, Я зажигаю свечи should just correspond to I light the candles

The second line doesn't activate Text Alignment at all.

There's a few bugs with editing too, once Word Insights is finished loading I'm incapable of editing the text (incl. deleting it) and the page must be reloaded. On longer text (that doesn't trigger Word Insights) sometimes clicking and dragging on the source text seems to treat it as an image/object? It doesn't highlight and instead drags sections of the text.

Finally very minor nitpick but I feel like the Text Alignment highlight color doesn't quite have enough contrast against the grey background, it's kinda hard to see depending on the lighting conditions.

Is there a token parameter? With regular kagi search I'm able to create an OpenSearch engine that works in private browsing (incognito) mode, by including my API token in the URL search params. It's great for one-off search engines in firefox. I also have an OpenSearch engine for Google Translate.

But when I try to do that for Kagi Translate, it's a little buggy in private browsing mode. First, red text appears saying "An error occurred during translation," and a Cloudflare widget appears temporarily that says "Verifying." And it doesn't translate my query until I hit "translate" button.

If, before using Kagi Translate in that private browsing window, I first use regular Kagi Search with my engine (which passes it my API token), everything works fine, and it knows who I am, and remembers my preferences. But it won't let me pass the token directly to Kagi Translate, so basically I have to use regular Kagi Search in that window before I can use Kagi Translate.

Not the end of the world, but it would be good to implement a token parameter. Maybe it already exists and it just has a different name. Nonetheless, are the search parameters documented anywhere? I would like to make sure my OpenSearch engine is exhaustive.

This would allow us to open Kagi Translate with a URL like https://translate.kagi.com/?token={api_token}&text=%s and have it log us in automatically, like Kagi Search does. Then it works just as well in private browsing/incognito mode.

Thank you. The translate widget is awesome btw. Much richer than Google Translate, in this respect. Word insights and the overall ML use are impressive. I love the extra options like translation style, context, etc. I hope these all get search parameters, and be documented so we can use them in automations, like one-off search engines. Rather than having to manually click each time (if for example we want to get formal translations by default).

Hi there,

The french pronunciation of « sourcil » is wrong, it sound like « surcil », that's funny but wrong.
https://translate.kagi.com/?from=auto&to=zh_cn&text=sourcil
See for instance https://www.google.com/search?channel=entpr&q=translate+sourcil for a correct pronunciation.