Language selection changes which characters and word patterns an OCR engine expects. Mixed scripts also change direction, segmentation, and review. This collection treats language as a document property, not a dropdown afterthought.
Use this collection when a page contains multiple languages, scripts, directions, vertical text, names, or diacritics. Model selection changes expected characters and word patterns, while document layout determines how those character runs should be ordered and recombined.
You should finish with a page language map, installed-model check, test of combined versus separate passes, and a review list for names, punctuation, digits, direction changes, and code points that ordinary spell-check may incorrectly normalize.
Test this cluster with Bilingual OCR, then begin the detailed review with How to Choose an OCR Language Model.
To separate a language-model mistake from a capture problem, follow the OCR error diagnosis process.
A document language setting affects expected characters and word patterns, but one page can contain names, numbers, quotations, formulas, or a second script that needs different handling.
| Document signal | What it means | Useful next action |
|---|---|---|
| One dominant language | Most prose uses one installed model. | Test the smallest text, names, punctuation, and borrowed words. |
| Several scripts | A combined pass may favor one script and damage another. | Compare a full-page pass with script-specific regions. |
| Direction changes | Numbers and Latin fragments may reorder inside right-to-left text. | Inspect copied logical order, not only visual appearance. |
| Vertical layout | Column progression and glyph orientation differ from horizontal prose. | Map column order and separate horizontal inserts when needed. |
Keep coordinates when recombining separate passes. Without page regions, correct tokens from different scripts can be joined into an output that no longer represents the source sequence.
Inventory scripts and directions, test the hardest regions, use separate passes when one model damages another, and review names without forcing them into dictionary words.
Use its installed model and test names, numbers, and borrowed words.
Compare a combined pass with script-specific crops.
Verify both character content and the sequence in which regions are exported.
Compare code points and visible marks instead of relying on spell-check.
Choose model coverage and region handling for mixed scripts, right-to-left sequence, vertical text, diacritics, and identity-bearing words.
Choose the model for the language actually printed on the page, confirm it is installed, and test names, numbers, punctuation, and borrowed words.
Use one combined pass only when it preserves both languages. Otherwise recognize script-specific regions separately and recombine them with page coordinates.
Verify both the recognized characters and their logical sequence. Mixed right-to-left text, numbers, punctuation, and Latin fragments can look correct visually while copying incorrectly.
Set the expected orientation and line progression before recognition, then verify vertical columns and horizontal inserts as separate layout decisions.
Review identity-bearing text character by character. Dictionary plausibility is weak evidence for a name, code, citation, or unfamiliar word.
Compare a combined recognition pass with focused script regions, then keep names, punctuation, and page order tied to the original image.
Open Bilingual OCR