GoldenDict Anki integration

[btrkeks]

I thought I'd drop some thoughts I had regarding further functionalities of Goldendict. Though I'm still not sure if GoldenDict is the right tool to replace all other dictionaries tbh

General

• Add right click on picture -> export to anki
• Export pronunciation to anki card
• Create script, which shows your anki cards containing the search term. Similiar to https://github.com/kanjieater/GoldenDict-Toolbelt
• Combine NHK accent dictionary and forvo dictionary to a 発音 dictionary, because everything is shown one under the other right now

Mecab Script

• Compile user dicts for mecab, to recognize bigger chunks
• Fix selection of arbitrary substrings, because only substrings inside links are selected right now
• Improve display of inflected words, e.g.: No Space between 動詞 and 助動詞 and combine to one link which points to the 動詞. With this the secondary clipboard would contain a perfectly formatted target sentence
• No link for 記号

[tatsumoto-ren]

Add right click on picture -> export to anki

The function that shows context menu in goldendict has to be modified. The "send to anki" action is added in that exact function.

https://github.com/xiaoyifang/goldendict/blob/staged/articleview.cc#L1711

But the task becomes complicated if you also want to convert the picture to WebP. I'd probably be satisfied with just copy-pasting to anki since it will trigger conversion to WebP automatically.

Export pronunciation to anki card

This will probably be redundant once audio dictionaries are added to AJT Japanese; but good as a backup feature in case there's an audio AJT Japanese doesn't have.

Create script, which shows your anki cards containing the search term.

Achievable with AnkiConnect and a call to https://github.com/FooSoft/anki-connect#findcards

[tatsumoto-ren]

One more thing is to introduce %GDSEARCH% in addition to %GDWORD% in order to give gd-mecab access to the whole search string without having to rely on xclip at all.

[btrkeks]

Do you think it's feasible to allow different anki settings for different groups? This way you could have word cards and sentence cards and include the gd-pics script only in the word card group.

[KonstantinDjairo]

Could we have this ?
With gd-mecab and deinflection, it would be a very cool combination

[tatsumoto-ren]

Could we have this ?

you can already have hunspell ja deinflection if you install the package and enable it as a dictionary in goldendict

https://aur.archlinux.org/packages/hunspell-ja-git

[tatsumoto-ren]

Do you think it's feasible to allow different anki settings for different groups?

once more settings are added to goldendict.

[btrkeks]

Another idea would be to add a script for Massif example sentences

[tatsumoto-ren]

a script for Massif example sentences

Done. It is in the repo now.

[tatsumoto-ren]

Some features to implement in goldendict

• Ability to export word reading along with the word.
• Ability to see if the word has already been added into Anki.
• Ability to open the card in Anki if it exists already.

[tatsumoto-ren]

since GD understands <a href="bword:..."> links, maybe it's possible to add other prefixes, e.g. <a href="ankisearch:..."> to open the anki browser and search for the word.

[tatsumoto-ren]

should work now.

[tatsumoto-ren]

The ankisearch: feature is needed so that we can see if a certain word has been already added to the Anki collection. If it has been added, we can open the corresponding card and learn or modify it.

See xiaoyifang/goldendict-ng#412 and https://github.com/Ajatt-Tools/gd-tools#gd-ankisearch

It is a "feature" but what's the point if no one will use it?

Thousands of people who are used to Yomichan will find this feature familiar.

[btrkeks]

I also think it would be better to use mecab's deinflection (%f[6]), because looking up います or します is pretty much useless right now. EDIT: nvm you noticed yourself

[btrkeks]

Though this deletes the whole sentence if we simply change the link

[tatsumoto-ren]

Yes this can happen

[btrkeks]

Maybe for our purposes it would make more sense to use a deinflection dictionary, which handles whole "words" like https://github.com/MrCorn0-0/hunspell_ja_JP

[btrkeks]

Also it would be very handy if one could use send to main window + pop up, eg. Clipboard -> main window, alt + selection -> pop up. Maybe ask the xiaoyifang guys...

[tatsumoto-ren]

https://github.com/MrCorn0-0/hunspell_ja_JP

how is this different from https://github.com/epistularum/hunspell-ja-deinflection?

If there are differences, I think we need to find a way to merge them into one dictionary.

[btrkeks]

It looks like it can handle verbs with their 助動詞, like "戻った", whereas epistularum's requires the word to be looked up as "戻っ". Using the other dictionary would fix boldness of the target word in a sentence card, but I don't now about it's quality. This would however also require gd-mecab to be rewritten, such that 戻った isn't split.

[btrkeks]

The problem with throwing away the 助動詞 is, that you get a lot of shit entries if you look up thinks like いきました, いた, いました, きた, ...

[tatsumoto-ren]

https://github.com/Ajatt-Tools/hunspell-ja

Mashed them together.

[btrkeks]

I just noticed that the Juman dictionary for mecab doesn't split inflected words already. Can be obtained from the official site: https://taku910.github.io/mecab/

[Aquafina-water-bottle]

Have y'all tried hitting up kanjieater by chance, to maybe get some ideas, potentially integrate some of his work, get a brain dump, etc? He gave a detailed summary of how these current tools differ from his GoldenDict-Toolbelt and seems to have a lot of experience with using/developing for goldendict in general:

At a glance the ajatt-tools look like they copied (and in some cases improved) a few of the things I made over 4 years ago. Yes, [GoldenDict-Toolbelt is] all still up to date & I still use it myself, though yomichan w/ texthooker-ui is far more useful for normal mining.

Comparison:
ajatt-tools vs kanjieater tools

• Mecab vs jisho parsing - mine uses Jisho, which is online only, but responds in less than a second,
• Bing images vs google - this is probably the biggest advantage over plain yomichan
• Anki connect direct integration vs anki add-on

Missing features available still with mine:
The new ajatt tools still don't have a google "did you mean" correction, image tags (the things above the images in the screenshot of freq searched together or intended terms), search anki (useful for searching a saved search) & freq list feature.

There are two main advantages, and why I still use goldendict over yomichan at times:

1. Is for supplementing card info. If I have a cards I want to add defs or google images to, it makes it very easy to do (with my add-on at least, haven't tried out others)
2. Being able to search multiple online sources at once via global hotkey and get it all back in a ~second. I think being able to search subs2srs decks and books that you've loaded into anki and using that as a dict essentially is undervalued as a way to find a bunch of example sentences from things you're familiar with.

I really hope someone makes a yomichan google image dictionary at some point, that'd be at least half the battle. Not sure how feasible that is in terms of image size/dict size.

Anyways, glad goldendict is getting some more love, it's been around for forever and is still pretty handy after a bit of setup.

[tatsumoto-ren]

Hi @kanjieater! Great work on GoldenDict-Toolbelt! At Ajatt-Tools we usually talk on Matrix which is a popular messaging protocol that can also be seen as an alternative to Discord.

My current idea is to write PRs for GoldenDict itself to add more bridges between external programs and GoldenDict in order to allow external programs to be both more versatile and more simple and short.
Now GoldenDict can create Anki cards which lifts the responsibility from external apps, for example.

Your toolbelt looks impressive. I think initially we wanted to create something like that. The difference is that gd-tools is a set of smaller standalone scripts.

node.js is a pretty good choice for scraping web data, working with json & outputting html, which is pretty much all my scripts did.

Or it can be done with curl which is what we've been doing. I also want to experiment with libcurl at some point. For now we have a bunch of shell scripts just as a proof of concept.

Glad to see GoldenDict getting some love and better integrations.

For a long time I ignored GoldenDict because it couldn't load EPWINGs (at least on GNU/Linux). And at that time there were no MDX dictionaries that I could easily find. Now that @epistularum made a collection of MDX dictionaries and uploaded it to nyaa, it's easier for new users to get into GoldenDict.

[KonstantinDjairo]

Have y'all tried hitting up kanjieater by chance, to maybe get some ideas, potentially integrate some of his work, get a brain dump, etc? He gave a detailed summary of how these current tools differ from his GoldenDict-Toolbelt and seems to have a lot of experience with using/developing for goldendict in general:

At a glance the ajatt-tools look like they copied (and in some cases improved) a few of the things I made over 4 years ago. Yes, [GoldenDict-Toolbelt is] all still up to date & I still use it myself, though yomichan w/ texthooker-ui is far more useful for normal mining.
Comparison:
ajatt-tools vs kanjieater tools

• Mecab vs jisho parsing - mine uses Jisho, which is online only, but responds in less than a second,
• Bing images vs google - this is probably the biggest advantage over plain yomichan
• Anki connect direct integration vs anki add-on

Missing features available still with mine:
The new ajatt tools still don't have a google "did you mean" correction, image tags (the things above the images in the screenshot of freq searched together or intended terms), search anki (useful for searching a saved search) & freq list feature.
There are two main advantages, and why I still use goldendict over yomichan at times:

1. Is for supplementing card info. If I have a cards I want to add defs or google images to, it makes it very easy to do (with my add-on at least, haven't tried out others)
2. Being able to search multiple online sources at once via global hotkey and get it all back in a ~second. I think being able to search subs2srs decks and books that you've loaded into anki and using that as a dict essentially is undervalued as a way to find a bunch of example sentences from things you're familiar with.

I really hope someone makes a yomichan google image dictionary at some point, that'd be at least half the battle. Not sure how feasible that is in terms of image size/dict size.
Anyways, glad goldendict is getting some more love, it's been around for forever and is still pretty handy after a bit of setup.

first, it's not that good to rely on google to do the suggestions, you can already achieve this by using the built-in morpholy feature with that

also, jisho is a website that already has a morphological analyzer, but it's unnecessary to use it when you already have mecab, it's simple and tiny and we can costumize it to use the dicts that we want (flexibility)

[tatsumoto-ren]

Jisho.org most likely relies on Mecab internally. There should be no significant difference.

[KonstantinDjairo]

Jisho.org most likely relies on Mecab internally. There should be no significant difference.

it's far better to run mecab alone in your machine than using a script to communicate with jisho, imagine the difference in latency

[kanjieater]

Have y'all tried hitting up kanjieater by chance, to maybe get some ideas, potentially integrate some of his work, get a brain dump, etc? He gave a detailed summary of how these current tools differ from his GoldenDict-Toolbelt and seems to have a lot of experience with using/developing for goldendict in general:

At a glance the ajatt-tools look like they copied (and in some cases improved) a few of the things I made over 4 years ago. Yes, [GoldenDict-Toolbelt is] all still up to date & I still use it myself, though yomichan w/ texthooker-ui is far more useful for normal mining.
Comparison:
ajatt-tools vs kanjieater tools

• Mecab vs jisho parsing - mine uses Jisho, which is online only, but responds in less than a second,
• Bing images vs google - this is probably the biggest advantage over plain yomichan
• Anki connect direct integration vs anki add-on

Missing features available still with mine:
The new ajatt tools still don't have a google "did you mean" correction, image tags (the things above the images in the screenshot of freq searched together or intended terms), search anki (useful for searching a saved search) & freq list feature.
There are two main advantages, and why I still use goldendict over yomichan at times:

1. Is for supplementing card info. If I have a cards I want to add defs or google images to, it makes it very easy to do (with my add-on at least, haven't tried out others)
2. Being able to search multiple online sources at once via global hotkey and get it all back in a ~second. I think being able to search subs2srs decks and books that you've loaded into anki and using that as a dict essentially is undervalued as a way to find a bunch of example sentences from things you're familiar with.

I really hope someone makes a yomichan google image dictionary at some point, that'd be at least half the battle. Not sure how feasible that is in terms of image size/dict size.
Anyways, glad goldendict is getting some more love, it's been around for forever and is still pretty handy after a bit of setup.

first, it's not that good to rely on google to do the suggestions, you can already achieve this by using the built-in morpholy feature with that

also, jisho is a website that already has a morphological analyzer, but it's unnecessary to use it when you already have mecab, it's simple and tiny and we can costumize it to use the dicts that we want (flexibility)

I used the hunspell dict and found it to be significantly worse than Google's suggestions and also it gives outdated suggestions. I documented my experience with it here in 2021 goldendict/goldendict#1352 (comment)
Not to mention you won't get related searches, which is different than an autocorrect that Hunspell provides. But hey, it's offline, so that's cool.

I think I already touched on the latency being a non-issue since it responds consistently in under a second, in which time you can progressively load it underneath immediately loading offline sources. Again, my tool already exists, so you can try any of these for yourself - it's meant to be standalone, and not incorporated directly into GoldenDict. I've used it for 4 years with consistent results.

Jisho also has wikipedia entries and various other features. It can map english keywords to japanese and it's sentence deinflection was more robust than mecabs last time I investigated. Though I do think mecab sits at it's core - it's not that different, but it is a little bit better.

That being said, clearly offline has some big advantages, I just wouldn't be so quick to brush off online sources as it's one of the big benefits of Goldendict, being able to query multiple online & offline sources asynchronously at the click of a global hotkey.

The difference is that gd-tools is a set of smaller standalone scripts.

Most of my scripts are standalone as well. I did start to get lazy and combined the Anki & frequency list one though if memory serves me.

Now GoldenDict can create Anki cards which lifts the responsibility from external apps, for example.

I look forward to trying this out & your future work! Thanks for sharing 👍

[tatsumoto-ren]

xiaoyifang/goldendict-ng#412 merged

[KonstantinDjairo]

that would be nice

https://github.com/Doublevil/JmdictFurigana

[tatsumoto-ren]

how would this help us?

[KonstantinDjairo]

how would this help us?

yomichan has this feature of showing furigana on top, we could add an option to enable it

[tatsumoto-ren]

Looking at furigana is considered a bad practice, so it's better kept hidden until the user clicks on a word.

[btrkeks]

Though furigana could be nice, if it can tell you the context dependent reading of a word like 避(さ)ける vs 避(よ)ける or when counting 一日、４日、一分 . But I think most furigana generators can't do that

[tatsumoto-ren]

Yes, mecab will output one reading and you don't know which one.