Question about potential GSoC 2026 project ideas for Inflection

[BHK4321]

Hello maintainers,

I’ve been contributing to the Inflection repository recently and really enjoying the codebase and problem space.

I wanted to ask whether there is any possibility that Inflection-related work might be proposed as a project under Unicode for GSoC 2026, or if there are areas in this repo that you feel could evolve into a student project.

I understand that plans may not be finalized yet, just wanted to discuss some of them.

Thank you for your time!

[BHK4321]

@grhoten @nciric If we can discuss possible student projects, please put your thoughts, I would like to discuss.

[grhoten]

Hi,

We're still thinking of the topics. Possible projects include the following. We will probably only be able to do one or two of them based on mentor availability.

• Provide a tool to generate a report based on CLDR data in Wikidata
  • At minimum, the units should be covered.
  • There are likely other localizations in CLDR that should be checked for coverage.
• Improve the way to customize the data
  • Improve how SemanticConcept can be used to create app specific or domain specific vocabulary end to end.
  • Provide a way to permanently add/remove/modify words with their inflection patterns with the dictionary-parser.
  • Finish fixing the dictionary-parser unit tests.
• Provide a better way to combine lexemes that differ by gender in dictionary-parser.
  • The current filter properties files can do very focused combinations, but a more scalable solution should be provided, especially for a language like German, or Italian.
  • This may require analysis of more data from Wikidata in some way.
• Improve Message Format integration (sorted by priority and ease of implementation)
  • Support text transformations in icu::message2
  • Support lists in icu::message2
  • Support quantify in icu::message2
  • Support SemanticConcept in icu::message2 for custom vocabulary
  • Support PronounConcept in icu::message2
  • Support the print and speak lines from SpeakableString in icu::message2 (probably unlikely to do due to incompatible design, but I'm open to hearing options)
• Port to Windows
  • This primarily involves modifying the CMake files.
  • The implementation for ResourceLocator::getDirectoryFromEnvironment probably has to be rethought.
  • A new GitHub hook is required to test this configuration.
• Add a new language
  • The prioritized list provides a sorted list of candidates.
  • If there isn't enough data for a language, there should be a way to more quickly synthesize, review and contribute the data to Wikidata. This would be a part of the task.
• Implement better sound property derivation for words
  • This is a continuation of work with Derive sound properties from Wikidata for Korean and Turkish #113.
  • Better parsing of phonetic information in Wikidata is likely required.

[BHK4321]

Provide a better way to combine lexemes that differ by gender in dictionary-parser.
The current filter properties files can do very focused combinations, but a more scalable solution should be provided, especially for a language like German, or Italian.
This may require analysis of more data from Wikidata in some way.

Hi, I think I have worked on this issue before and for a more scalable solution we should first try brute forcing through the wikidata and searching for pairs which defer only by gender using a script to check how many such pairs are there and then do the further processing as usual in ParseWikidata without using the properties files. I think I can work on this. If this involves more analysis or optimization than this, then I would like to know more options.

[grhoten]

I'm not sure that brute force is necessary. Take Organisatorin as an example. It combines another lemma with an -in feminine suffix in the combines lexemes section. Perhaps data can be derived from that kind of structure in German and Italian.

Also the Unicode organization wasn't picked this year for GSoC. So it looks like these tasks would be independent from GSoC this year.