Morphology-aware matching (inflection / lemmatization support)

[moritzfl]

Check for existing issues

• Completed

Describe the feature

Summary
Vale currently relies on regex and dictionary-based matching, which makes it difficult to handle inflected word forms (e.g., pluralization, conjugation, declension). This can lead to incomplete rule coverage or complex and hard-to-maintain patterns.

Problem
When defining rules, users often need to account for multiple inflected forms of the same word manually. For example, a rule targeting a base form like run will not match:

• runs
• ran
• running

To compensate, users must either:

• enumerate all variants explicitly, or
• write complex regex patterns (e.g., run(s|ning)?)

This approach is error-prone, and reduces readability and maintainability of rules.
This limitation becomes significantly more severe in languages with richer morphology, such as German.

This gives a short overview about what you can expect for most languages other than English: https://youtu.be/ettP9Ayrho8?is=g0hMRMZqVJE74K8p

Minimal example (in German)

Rule:

extends: substitution
message: Consider using '%s' instead of '%s'
level: warning
ignorecase: false
swap:
  gut: hervorragend

Text:

Das ist eine gute Lösung.

Current behavior:

• No match

Expected behavior:

• Match based on shared lemma (“gut” → “gute”)

For example, the adjective „gut“ can appear in many forms depending on case, gender, and number:

• gut
• gute
• guten
• gutem
• guter

Similarly, verbs like „gehen“ produce forms such as:

• gehe, gehst, geht
• ging, gegangen

Covering these via regex or explicit lists quickly becomes impractical. As a result:

• rule definitions become bloated
• important variants are easily missed
• false negatives increase significantly

This makes Vale harder to use effectively for non-English content and limits its usefulness in multilingual environments.

Discussion / possible approaches
I understand that Vale is intentionally lightweight and primarily regex/dictionary-based, and that performance and simplicity are key design goals.

With that in mind, a possible direction could be:

• We could basically do some sort of macro expansion with a syntax that triggers an expansion to all valid forms through a dictionary: "Word A" in a rule becomes "Word A - Variant 1|Word A -Variant 2| ... | Word A - Variant n". * All of that happens before the actual linting so it is only done once per linting run. Potentially, vale could even cache results for performance.

Benefits

• Simpler and more maintainable rules
• Better coverage with fewer false negatives
• Improved support for morphologically rich languages (German, Finnish, Slavic languages, etc.)
• Better usability in multilingual teams

Question
Would morphology-aware matching be considered within Vale’s scope? I could also try working on it if we agree on an approach that would be accepted as PR.