Japanese solutions display standalone hiragana as katakana

[stochaztic]

Solutions in Japanese are displaying all "standalone" hiragana as katakana, as shown in the first image below. The expected hiragana answers are not in the solution list at all. The only hiragana shown are kanji replacements. This highly reduces the usefulness of the Japanese solution list; while using katakana instead of hiragana everywhere is technically permissible in Duolingo, it is not at all how the language is used, and makes reading or filtering the solution list extremely hard.

I am unsure when this started happening, as it was not the case when I used this extension regularly many months ago. As noted in the second image, the data in compact_translations in the https://excess.duolingo.com/batch at question submission time does not show katakana. I was not sucessful diving deeper to try to find the cause of the issue in my cursory attempt.

[blmage]

Thanks for the heads up!

The solutions are currently extracted from the "grader" graph, as I originally wanted to have maximum completeness, but one of the goals when I started working on v5 a while ago (before life and other projects caught up with me) was actually to switch to using the "compact translations" instead.

If I remember well, most of the code related to this was ready to test, so I should be able to reuse it in v4, even if only to parse Japanese solutions. Would you be willing to test a pre-release of the extension, and let me know if everything looks good, and/or if anything could be improved (especially regarding the Japanese course)?

[stochaztic]

I can do so next week, yes.

…

On Thu, May 11, 2023 at 08:42 blmage ***@***.***> wrote: Thanks for the heads up! The solutions are currently extracted from the "grader" graph, as I originally wanted to have maximum completeness, but one of the goals when I started working on *v5* a while ago (before life and other projects caught up with me) was actually to switch to using the "compact translations" instead. If I remember well, most of the code related to this was ready to test, so I should be able to reuse it in *v4*, even if only to parse Japanese solutions. Would you be willing to test a pre-release of the extension, and let me know if everything looks good, and/or if anything could be improved (especially regarding the Japanese course)? — Reply to this email directly, view it on GitHub <#197 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACBHL64L5D2LOMXYYJQKI5TXFTNCHANCNFSM6AAAAAAXWG6HH4> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[blmage]

Sorry for the delay! it took a little longer than I expected, especially because I wanted to give a try to the syllabary filter (the extension now uses Wanakana.js to automatically convert between syllabaries, using the compact translations as the basis).

Version 4.13.0-beta.1 can be downloaded here and loaded as-is in your browser, if you want to give it a try.

Thanks!

[blmage]

The changes mentioned are now part of version 4.14.0.

If and when you have some time, your feedback would be greatly appreciated!

[stochaztic]

Sorry I wasn't able to get to this in the beta. I'm still experiencing this issue on the Chrome Web Store release of 4.14.0.

[blmage]

No worries, thank you very much for taking the time to give it a try!

The compactTranslations are actually not always available, which is probably why the extension fell back to using the graph, but until now I thought this was only the case for fill-in-the-blank challenges - maybe this is more common on the Japanese course than the others.

I'll try to implement the syllabary filter for solutions parsed from graphs as well, then, hopefully it won't be too costly in terms of performance. I guess I'll have to fix and rethink it first, though, as it doesn't seem to work quite as expected when applied to the compactTranslations, in addition to not being very useful (or at least quite confusing) in its current state.

[blmage]

Just wanted to give you a quick update since I just tested the new filter on a list of 72k solutions, and using it to exclude katakana tokens removed 99.5% of them. Also, adding the flag to all solutions seems to be a sweet spot for memoization, as it took only a fraction of the total time (if the profiler is to be trusted, at least).

All that's left then is to integrate it back into the UI!