Motif

See what your FieldWorks grammar does. Try a change safely. Decide for yourself.

Motif is a tech demo, not a finished product. Keep a FieldWorks backup and try it on a copy of your project.

Motif is a desktop window and a command line for linguists who build grammars in FieldWorks. A grammar that works for a few dozen words gets harder to check and improve as it grows. Motif shows you what the parser does with your own texts, lets you try a change without touching your project, and hands the result to you to decide.

What it helps with

When you build a FieldWorks grammar, three questions come back again and again.

• Which of my words parse, and which don't? Motif parses the texts you choose and shows the results word by word, with Text Coverage, how many of your approved analyses the grammar still builds, how long each word takes, and what the grammar check found.
• Why did this word fail? Try a Word shows the parser's reasoning for one word, step by step, including where it gave up.
• If I change the grammar, will it help or hurt? A Proposal is a set of changes. A Trial applies it to a throwaway copy of your project and parses again, so the new numbers sit beside today's. Your FieldWorks project is not touched.

Motif reads your saved FieldWorks project and writes nothing to it until you open Review changes, look over each change, and choose Apply to FieldWorks project.

How it works

PanGloss measures, Motif shows, an AI assistant drafts, and a person decides.

1. Measure. PanGloss, the parser, parses the words you select. Motif keeps the results of each run.
2. Show. The Overview, Texts, Timing and Warnings pages put those results where you can read them.
3. Draft. Motif can hand the whole picture to an AI assistant, as a folder of files to drag into a chat or through a connected assistant. The assistant can propose changes. It cannot apply them.
4. Decide. You review each change, Keep or Undo it, and apply the ones you want.

See it

Overview: how much of your text the grammar parses, how many approved analyses it still builds, how fast, and what to look at first.

Try a Word: type a word and see how the parser read it.

Warnings: problems the grammar check found, to fix in FieldWorks.

Review changes: every change has Keep and Undo, and nothing reaches FieldWorks until you apply.

AI Handoff: the files and starter prompt for a chat model such as Claude, ChatGPT or Gemini.

These images come from the walkthrough tests, which run the window against a tiny test project, so the numbers are small.

Who it is for

Linguists and language teams who keep their grammar in FieldWorks (FLEx) and want a clearer view of how it behaves on their own texts. You do not need to write code.

What it is not yet

• It is a tech demo. The window and the commands will change, and before 1.0 a Motif file from an older version may be refused rather than upgraded.
• It does not replace FieldWorks. You still edit the grammar there. Motif applies only the changes you have reviewed.
• Installers are for invited testers. There is no automatic update, and the Windows beta may be unsigned.

Try it

Download the Windows x64 Setup.exe from the Releases page, choosing the beta named in your invitation, then follow the install guide. The Windows release includes the matching PanGloss parser. The same release offers a Linux x64 AppImage and a macOS Apple silicon ZIP as previews.

Then open your project, press Refresh, and follow What Motif does. The first setup for a project starts with an analysis-attempt limit of 200,000 per word; a project's saved limit stays in effect.

Build from source

Install the .NET 10 SDK, then from the repository root:

./build.ps1    # comment and design-token checks, then compile
./test.ps1     # the above, then Unit and Integration tests; add -All for every level

Use these rather than dotnet build or dotnet test: they run the checks a bare build skips. AGENTS.md is the guide for contributors. It covers the test levels, the build layout, staging SIL ICU on Linux and macOS (bash tools/stage-sil-icu.sh), building against a local libpalaso, and the comment and design-token rules. CONTEXT.md is the glossary, and its terms are binding. All Motif projects target net10.0.

After building, run the command or the window from bin/<Configuration>:

./bin/Debug/motif.exe help                 # Windows; on Linux or macOS ./bin/Debug/motif help
./bin/Debug/SIL.Motif.App.exe              # Windows; on Linux or macOS ./bin/Debug/SIL.Motif.App

Motif parses with PanGloss, pinned by version and SHA-256 in pangloss-release.json. build.ps1 stages the pinned parser beside the build. To use a parser of your own, set MOTIF_PANGLOSS_EXE to its path; that override is authoritative. Evaluation sets live in the separate motif-test-grammars repository: fetch the locked revision with pwsh ./evals/Get-TestGrammars.ps1, or tests that need them skip with a reason.

To stage a release payload, install the pinned Velopack CLI (dotnet tool install --global vpk --version 1.2.158) and run ./tools/package-release.ps1. The package workflow builds the Windows Setup, Linux AppImage and macOS ZIP; see the install guide for what each preview does and does not promise.

Documentation

The motif command, the window and the documentation site share one set of authored Help in src/SIL.Motif.Help/Content/. The site in site/ is built from it: export the catalog with motif help --all --json > bin/help-export.json, then in site/ run npm ci, npm test and npm run build with Node 22.12 or later, with MOTIF_HELP_EXPORT pointing at the export. ./tools/Build-Media.ps1 rebuilds the generated media. In CI, the CI workflow's publish_website option builds and validates the site artifact; it does not deploy it.

Read next:

• Current architecture: project references, command sharing and cache ownership.
• CLI and command API: where command usage and Guides are kept.
• Building a grammar: a suggested order: the order linguists usually work in, with the reasons; advice, not a rule.
• Semantic change contract and Proposal lifecycle.
• Assessment model and parser handoff.
• Motif in Brief, a short illustrated pitch.
• Older plans, kept as history: Plan A, product architecture plan, architecture proposal.

Licence

Motif and its synthetic language sets are MIT-licensed. Future real-language data may carry share-alike or non-commercial terms recorded in each set's LICENSES file; see the external evaluation-data licence notice.