Add Ukrainian stemmer

[polaz]

Summary

Add a new stemming algorithm for the Ukrainian language.

Implementation

The stemmer follows the general design of the existing Russian stemmer, adapted for Ukrainian morphology:

• Regions: pV (after first vowel), R2 (standard Snowball definition) used to guard aggressive stripping
• Suffix removal order: perfective gerund → reflexive → verbal noun → (professional / diminutive) → adjective / verb / noun → diminutive stem → professional → derivational → tidy-up
• Compound suffix patterns: Single-step removal of derivational+inflectional combinations (e.g. -аційний, -ічний, -ійний, -уальний, -онний) protected by R2
• Ukrainian-specific features: apostrophe handling (U+0027, U+02BC), soft sign cleanup, superlative най- prefix, comparative -іш- suffix

Design decisions

• Single-char adjective endings removed: Unlike some approaches that include -а, -е, -і etc. in the adjective section, these are handled by the noun section to avoid stealing endings from verb/noun patterns (matches Russian stemmer approach).
• Professional before diminutive: Professional suffixes (-ар, -ник, -ельник, etc.) are tried before diminutive (-ик, -ок, etc.) to ensure будівельник → будів rather than diminutive -ик intercepting first.
• R2 guards compound suffixes: Long derivational suffixes like -ація, -ічний only fire in R2 to prevent over-stemming short words (e.g. акація → акаці not ак).
• Verbal noun with R2: The verbal noun suffixes (-ання, -іння, -ення) use R2 to protect short words (e.g. мигтіння, боління) from over-stemming.

Testing

• 216 hand-crafted test cases covering nouns, adjectives, verbs, diminutives, professionals, verbal nouns, compound derivational patterns, reflexives, gerunds, participles, and negative tests
• 57,868-word vocabulary from Ukrainian Wikipedia (frequency threshold 300) — all pass

References

• Vysotska, V. (2024). "Methods and means of NLP for Ukrainian morphological analysis" — used for comparison of compound derivational stems
• Based on structural patterns of the existing Russian and Serbian Snowball stemmers

PRs

Matching PRs will be submitted to snowball-data and snowball-website repositories with branch name matching this repo's PR branch.