Should a ZWNJ at the end of a word be ignored when looking a title up? (مهندس‌ علی رضایی gives given مهندس‌) #1335

A ZWNJ (U+200C) joins nothing at a word's edge, but it stays in the token, so the word no longer matches the vocabulary. One Expediency Council decree types a ZWNJ after every word: جناب‌ حجةالاسلام‌ والمسلمین‌ آقای‌. parse("مهندس‌ علی رضایی") gives given مهندس‌, where مهندس علی رضایی reads the title.

decisions/T1 (separators, not joiners) leaves an edge joiner to the caller, at the tokenizer (#851). Stripping one only when looking a word up, and leaving the token text alone, is a narrower question than T1 asked.

Questions

• Should the lookup (_normalize, or only the title lookup) drop a ZWNJ at a word's edge?
• If it goes in _normalize, the match's _fold has to change alike. Read Persian titles and Persian spellings of the Arabic ones (#1151) #1331's tatweel fix first changed only _normalize and made محمـــد and محمد read as a listed spelling, found by review.
• What about a ZWNJ inside a word that the vocabulary spells without one (آیتالله against آیت‌الله)?

What to measure: how often a source types an edge ZWNJ outside that one decree. A Persian-script source of running names would show it better than the member pages.

Follows #1151.

🤖 Generated with Claude Code