Typo generation

This standalone workflow generates and filters typo mappings without loading AutoCorrect2 data.

run_typo_generation() accepts source words, an ordered sequence of TypoGenerationTask objects, a TypoGenerationConfig, and an optional n_workers value.

Typo Generation Workflow

flowchart TD
    A["Source words"] --> D["Execute typo-generation<br>tasks"]
    B["Generation tasks"] --> D
    C["Configuration<br>and worker count"] --> D
    D --> E["Raw MULTYPO samples"]
    E --> F["Aggregate and classify<br>generated mappings"]
    F --> G["TypoGenerationResult"]

TypoGenerationResult.candidates maps each unambiguous noisy form to exactly one target word. clashes contains noisy forms that mapped to multiple target words and are therefore removed before any AutoCorrect2 processing.

This workflow has no dependency on AutoCorrect2 hotstring models or option policies.

Command-line usage

Run standalone typo generation with typo-generation:

uv run python -m hotstring typo-generation `
    --words-file .\words.txt `
    --single-attempts 100 `
    --multi-attempts 100 `
    --multi-min-length 5 `
    --workers 4 `
    --report .\reports\typo-generation.txt

The command accepts either repeated --word arguments or one --words-file, but not both. It does not load or modify AutoCorrect2.

See also: Command-line interface for all generation options.