Small marks can carry meaning
Leave vertical space around lines and verify marked characters before translation smooths over them.
Protect marks above and below each letter
Character evidence
A crop with space above and below each line so small marks remain visible.
Identity relationship
A diacritic changes a character and can change a word, name, or grammar.
Pre-translation check
Compare marked characters before autocorrection or translation smooths them away.
Protect the space around letters
Leave a little space above and below every line so marks are not clipped. This is especially important for tight crops, underlined text, superscripts, and fonts with tall capitals. Make sure shadows and page texture do not merge with small marks.
Focus and compression should preserve dots, cedillas, rings, strokes, and combining marks. Inspect names and place names character by character because autocorrection may produce a familiar but different form.
Do not normalize away identity
A version without marks may be common in a URL or legacy database, but it is not always interchangeable with the person's or place's preferred spelling. Keep the source form and use an official Latin form when one is published.
For search, it can be useful to retain both marked and unmarked variants. For identity, academic citation, or legal use, follow the source and governing document rather than choosing the easiest keyboard form.
Names may need their sound preserved instead of their meaning changed, so compare transliteration with translation before editing them.
Review before translation
Compare extracted text with the image before the translation model sees it. One missing mark can change grammatical form or distinguish separate words, and the translator may make the corrupted input sound natural.
When the mark remains unclear, preserve the ambiguity or find a second source. Do not infer a person's name solely from frequency or a likely spelling.
Preserve accents and marks above, below, or through letters when extracting and translating names, places, and multilingual text. Continue with the languages and scripts guide collection for adjacent workflows.