Add eflomal word aligner: Bayesian IBM1→HMM→fertility with N parallel Gibbs chains

[johnml1135]

Summary

Wires EflomalAlignmentModel from thot into machine's alignment pipeline. eflomal is a Bayesian IBM1→HMM→fertility cascade trained by collapsed Gibbs sampling. Depends on Thot 3.5.0 (released to NuGet; implementation in sillsdev/thot#11).

Changes

ThotWordAlignmentModelType / ToolHelpers — Eflomal enum value + "eflomal" string alias.

ThotEflomalWordAlignmentModel — new class. Because eflomal is an HMM-based aligner, it implements IHmmWordAlignmentModel and shares the alignment-scoring logic with ThotHmmWordAlignmentModel via a new abstract base, ThotHmmWordAlignmentModelBase (extracted from the HMM model so the two are siblings rather than one subclassing the other). Its alignment scores come from swAlignModel_getEflomalAlignmentProbability. Added to the ThotWordAlignmentModel.Create factory.

ThotHmmWordAlignmentModelBase — new abstract base holding the shared ComputeAlignedWordPairScores; concrete ThotHmmWordAlignmentModel and ThotEflomalWordAlignmentModel supply only their model type and the underlying C API call.

ThotFastAlignWordAlignmentModel — now uses the FastAlign-specific swAlignModel_getFastAlignAlignmentProbability instead of borrowing the IBM2 entry point (the new thot binary dispatches by model type, so the IBM2 path would return 0 for a FastAlign model).

ThotWordAlignmentModelTrainer — Eflomal branch creates a single model that runs the full IBM1→HMM→fertility cascade internally. Key behavior:

• Iterations reuse the standard counts. Ibm1IterationCount → IBM1 stage, HmmIterationCount → HMM stage, Ibm3IterationCount → fertility stage. The separate Eflomal*IterationCount properties are gone.
• Auto schedule by default. If none of those counts are set, the model derives its corpus-scaled schedule automatically (thot default). The trainer then reads the resolved count via swAlignModel_getEflomalScheduledIterations after startTraining and drives exactly that many sweeps — required because the schedule (and the staged cascade) isn't known up front. Progress is reported as indeterminate until it resolves.
• Applies the eflomal hyperparameters (below) when set.

ThotWordAlignmentParameters — reworked eflomal hyperparameters to match the thot 3.5.0 C API:

• EflomalNumSamplers — parallel Gibbs chains; marginals summed across chains before argmax (eflomal's n_samplers). Thot default 3 (left untouched unless set).
• EflomalSeed — RNG seed for the samplers.
• EflomalDeterministic — trains chains serially so a fixed seed is reproducible.
• EflomalLexNorm — plain 1/N(e) vs Dirichlet-smoothed lexical denominator.
• EflomalLexAlpha / EflomalJumpAlpha / EflomalFertilityAlpha — Dirichlet priors.
• EflomalP0 — NULL-alignment mixing weight (renamed from the removed EflomalNullProb).
• EflomalJumpWindow — jump-distribution half-window.
• Removed EflomalNullAlpha (no longer in the API). Partial schedules fall back to thot's per-stage defaults (8 / 8 / 32).

Thot.cs — P/Invoke declarations for the eflomal API (setEflomalSeed/NumSamplers/Deterministic/Iterations/AutoIterations/LexNorm/P0/AlphaLex/AlphaJump/AlphaFertility/JumpWindow, getEflomalScheduledIterations, getEflomalAlignmentProbability) and swAlignModel_getFastAlignAlignmentProbability.

AlignmentModelCommandSpec — eflomal reuses the --ibm1-iters / --hmm-iters / --ibm3-iters flags for its schedule; new --eflomal-samplers, --eflomal-deterministic, --eflomal-seed, --eflomal-lex-norm, --eflomal-lex-alpha, --eflomal-jump-alpha, --eflomal-fert-alpha, --eflomal-p0, --eflomal-jump-window flags.

Test plan

• ThotEflomalWordAlignmentModelTests: CreateTrainer+Align, AlignBatch, translation probability, vocab counts, save/load round-trip, symmetrized model, explicit-schedule training, deterministic reproducibility, seed + lex-norm
• Full machine Thot test suite passes (83/83) against released Thot 3.5.0 from NuGet
• Solution builds warning-clean in Release (TreatWarningsAsErrors=true)
• dotnet csharpier check . clean

🤖 Generated with Claude Code

---

This change is

[Copilot]

Pull request overview

Integrates the new Thot-backed Eflomal Bayesian word alignment model into the SIL.Machine.Translation.Thot alignment pipeline, exposing training configuration (iterations, parallel Gibbs samplers) and wiring it through the CLI and model factory.

Changes:

• Adds Eflomal as a ThotWordAlignmentModelType and wires it into ThotWordAlignmentModel.Create.
• Introduces ThotEflomalWordAlignmentModel, plus trainer support for EflomalNumSamplers and iteration scheduling via ThotWordAlignmentParameters.
• Extends Thot interop (Thot.cs) and CLI plumbing (AlignmentModelCommandSpec, ToolHelpers) and adds Eflomal-focused tests.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/SIL.Machine.Translation.Thot.Tests/ThotEflomalWordAlignmentModelTests.cs	Adds coverage for creating/training/aligning with Eflomal, batch alignment, save/load, and symmetric alignment.
src/SIL.Machine.Translation.Thot/ThotWordAlignmentParameters.cs	Adds Eflomal iteration and sampler parameters with defaults.
src/SIL.Machine.Translation.Thot/ThotWordAlignmentModelType.cs	Adds Eflomal model type and string mapping.
src/SIL.Machine.Translation.Thot/ThotWordAlignmentModelTrainer.cs	Adds an Eflomal training branch, including sampler configuration.
src/SIL.Machine.Translation.Thot/ThotWordAlignmentModel.cs	Wires Eflomal into the alignment model factory method.
src/SIL.Machine.Translation.Thot/ThotEflomalWordAlignmentModel.cs	Implements the Eflomal-specific ComputeAlignedWordPairScores behavior.
src/SIL.Machine.Translation.Thot/Thot.cs	Adds Eflomal alignment-model enum value, P/Invoke for sampler count, and mapping from ThotWordAlignmentModelType.
src/SIL.Machine.Tool/ToolHelpers.cs	Adds CLI string constant for eflomal.
src/SIL.Machine.Tool/AlignmentModelCommandSpec.cs	Updates help text and parameter mapping to include Eflomal iterations and samplers.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 78.32168% with 31 lines in your changes missing coverage. Please review.
✅ Project coverage is 73.20%. Comparing base (9a21ddd) to head (c897aec).

Files with missing lines	Patch %	Lines
....Translation.Thot/ThotWordAlignmentModelTrainer.cs	66.66%	11 Missing and 8 partials ⚠️
....Translation.Thot/ThotEflomalWordAlignmentModel.cs	25.00%	9 Missing ⚠️
...ranslation.Thot/ThotFastAlignWordAlignmentModel.cs	0.00%	1 Missing ⚠️
...Machine.Translation.Thot/ThotWordAlignmentModel.cs	0.00%	1 Missing ⚠️
...ine.Translation.Thot/ThotWordAlignmentModelType.cs	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #433      +/-   ##
==========================================
- Coverage   73.20%   73.20%   -0.01%     
==========================================
  Files         440      442       +2     
  Lines       36931    37011      +80     
  Branches     5077     5089      +12     
==========================================
+ Hits        27037    27094      +57     
- Misses       8781     8797      +16     
- Partials     1113     1120       +7     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[johnml1135]

All Copilot review comments have been addressed — see replies on individual threads for details.