Translation Project Brainstorming Meeting 2 Agenda and Notes

Meeting Information

Wednesday, June 3 11 AM-12 PM EST

Join our Cloud HD Video Meeting Smithsonian link

Participants

• @Corey Schmidt

• @Laura Schroffel

• @Brian Zelip

Helpful Links

• https://archivesspace.atlassian.net/jira/software/projects/TD/boards/57

Notes

• How should we feed the LLM?

  • If local instance TechDocs repo - point the LLM to the markdown file.

  • If doing a web upload, upload the specific markdown file to LLM - https://raw.githubusercontent.com/archivesspace/tech-docs/refs/heads/main/src/content/docs/architecture/database.md

    • IF people try to copy and paste the content from the TechDocs website itself, risks having extra stuff included which may result in a wrong translation.

    • What about code blocks?

      • We will set up uniform prompts to address code blocks and internal links, etc.

    • Brian to draft some prompts for us to test run.

• Need to adjust the TechDocs

  • Between docs and content - new directory for specific languages.

  • PRs per page - one PR has all 5 translations per page.

  • Example filepath: tech-docs/src/content/docs/<LANG_CODE>/administration/backup.md

  • Every language folder will be a child of the docs folder

    • WE DO NOT want english in it’s own folder - that may mess with links

    • English is the default.

• Testing

  • @Laura Schroffel to test https://github.com/archivesspace/tech-docs/blob/main/src/content/docs/administration/docker.md

  • @Corey Schmidt to test https://github.com/archivesspace/tech-docs/blob/main/src/content/docs/about/authoring.md

  • To test: Dutch, French, German, Spanish, Japanese, Ukrainian

• What should the prompt look like?

  • Brian to draft some and have us do some test runs.

• Process for assigning project assignment

  • Corey to make a spreadsheet of sections of pages for individuals to sign up for.

  • Keep PRs per section - minimize PRs - one PR per section.

  • TODO: convert https://archivesspace.atlassian.net/browse/TD-21 to a planning ticket

  • TODO @Corey Schmidt: create an Epic Jira ticket - link subissues to tasks related.

• How are we going to get people to review these translations?

  • This is the crux - we will need community organizing and outreach. Not sure how to do this yet.

• Summary

  • Decided to initiate planning for this project, set to complete planning before August meeting for TechDocs

  • Brian to get AI prompts and TechDocs set up to be ready to ingest new language files. Check-in June 22 meeting - have ready before is preferable for Corey and Laura to start testing.

  • Advised to directly attach markdown file to LLM, rather than retrieving using URL to markdown file in GitHub. Brian encountering difficulties with LLM confusing copyright status.

  • Flow of project:

    • Planning phase - get TechDocs structure ready for ingest; get AI prompts ready; Corey and Laura to test individual pages for translations; check outputted files for markdown validity and issues translating headers, code blocks, style (italics, bold, underline, strikethrough, lists, etc. referencing https://github.com/archivesspace/tech-docs/blob/main/src/content/docs/about/authoring.md#markdown); iterate AI prompts with Brian to get better results; create step-by-step guide on translating.

    • Initial translation phase: have TechDocs team members pick sections to translate; review translations through PR.

    • Secondary translation phase: identify community members to review translations; make corrections based on community feedback

Action Items