How to retain formulas when translating a scanned paper?

Key points of this article

First determine whether the PDF is a scanned copy or a text version: drag and select a section of text. If not, it will be a scanned copy. Be aware of this before translating.
Scans can be uploaded directly without OCR first; when you need to copy text, use the desktop client to generate a selectable text layer.
When there is LaTeX source code provided by the author, the source code translation will be used first, and the original writing of the formula will not be lost due to scanning.
Select bilingual comparison output, so that the original formulas and the translations can be side by side, making it easier to check whether symbols, superscripts and subscripts have been changed item by item.
Word will not directly convert formula images into LaTeX code. Formula conversion must rely on source code or a specialized editor.

When reading scanned English papers, the most troublesome thing is often the formulas in them. The scans are essentially pictures, and the formulas do not have a text layer that can be copied directly. Common problems after translation are that the formulas become garbled, the superscripts and subscripts are misplaced, or the entire paragraph is scattered. What many people care about is not "whether it can be translated", but "whether the formula is still there after translation and whether it is still correct."

This article is for those who are reading literature, doing coursework, or preparing to write a thesis. After reading, you will know how to determine the file type, how to choose the output method, and how to check the formula before delivery. At the end of the article, a frequently searched question will also be explained: How to recognize formulas as LaTeX code in Word.

How to retain the formulas in the scanned paper? One sentence conclusion

Whether the formulas of the scanned paper can be retained mainly depends on whether you have the original LaTeX source code. When the source code is available, use the source code translation directly, and the formulas are saved in the form of text commands, and the results are stable; only when scanning PDF, select bilingual comparison after uploading, so that the original formulas and the translations appear side by side, which not only retains the layout, but also facilitates your verification.

your situation	Recommended practices	reason
There is a LaTeX source code compressed package disclosed by the author	Translate directly from source code	Formulas are saved in source code and do not rely on image recognition.
There is only a scanned version of PDF and you need to read it carefully.	Upload the scanned document and select bilingual comparison	The original formula and the translation are side by side for easy checking.
I just want to understand the general idea without comparing them one by one.	Upload the scanned copy and only output the translation	When reading, focus on the content and skip the comparison steps.
After translation, you still need to modify the formula in Word.	Translate first, export to Word and check the formulas one by one.	Whether you can continue to edit the formula after it is exported depends on the actual file.
Text needs to be copied for citation or retrieval	First use desktop client OCR to generate text layer	There is no selectable text in the scanned document. It can only be copied after OCR.

Before you start: Confirm these points first

Whether the document is a scanned copy or a text version. Drag and select a text in the PDF reader. If you can select and copy it, it means it is a text version; if you cannot select it or copy it with garbled characters, it is probably a scanned copy. The judgment result determines whether you should do OCR first.
Whether there is LaTeX source code. The authors of many papers will disclose source code packages on preprint platforms or personal homepages. The formulas in the source code are text commands, and the retention effect is more reliable than the scanned copy. It is worth looking for it first.
If you understand it, you still need to change it. If you are just reading, bilingual comparison is enough; if you want to modify the formulas or write your own documents later, you need to reserve time for exporting and checking.
Formula density and layout complexity. Papers with dense formulas, spread-sheet tables, and double-column formatting are prone to misalignment after translation. It is recommended to use one or two pages for trial translation first, and then process the full text after confirming the effect.
The translation requires manual verification. Machine translation may use inaccurate words, and OCR may also misrecognize similar characters. Contents involving definitions, symbol meanings, and conclusions must be confirmed against the original text.

Steps to translate a scanned paper in OmniDoc and retain formulas

Confirm the file type. Follow the method in the previous section to drag and select a piece of text to determine whether the PDF is a scan.
Prioritize checking the source code. If the author provides a LaTeX source code package, you can use the source code for translation first and skip the subsequent scanned copy processing.
Generate text layer on demand. If you need to copy text or perform a search, open the download.omnidoc.top download desktop client, perform OCR on the scanned document in the PDF editor, and obtain a PDF with optional Chinese characters.
Upload files. Upload PDF at omnidoc.top. Scanned documents can be uploaded directly for translation without having to do OCR first.
Select the output method. Select bilingual comparison when you need to compare formulas; select only output translation when reading only translations.
Submit and download. Wait for the translation to finish downloading the result file.
Spot check formula-intensive pages. Open the results, first turn to the pages with the most formulas, and check them one by one against the original text.
Export on demand. After confirmation, if you need to continue editing, you can export it to Word or Excel; whether the formula is still editable in Word should be based on the exported file and open it one by one to check.

Check each item after completion

After you get the result, do not quote or forward it directly, press the table below to go over it.

Check item	What do you think?	What if there's a problem?
Whether the formula is complete or not	Compare the original text to see if the fractions, summation signs, integral signs and superscripts are complete.	Go back to the original PDF corresponding page to check; if the deletion is serious, use the source code translation instead.
Is the symbol rewritten?	Compare variable names, such as x and χ, uppercase and lowercase, superscript levels	The original text shall prevail and be marked next to the translation
Is the layout misplaced?	See if the original text corresponds to the translation paragraph by paragraph in the bilingual control, and whether the picture and table follow the text.	Retranslate the misplaced pages separately and, if necessary, only view the translation output.
Spread formulas and tables	Check if the spread is broken and the header is missing	Complete the original text, or intercept the corresponding page and re-translate it.
Is the terminology uniform?	Is the translation of the same concept consistent in the full text?	After unified replacement, go back to the definition and check it again.

# frequently asked questions (FAQ)

1. How does word identify latex formula?

Word itself does not recognize the formula picture directly into LaTeX code. It is safer to find the LaTeX source code exposed by the author first; if you have only Word files, you can handle the conversion between Word and LaTeX in the UniDoc page editor (app.unidoc.top) and check the formulas one by one.

2. What does it mean to recognize latex by formula? Can you just identify the formula?

Formula recognition usually refers to converting the formula in the picture into editable code such as LaTeX. The process introduced in this paper is mainly translation, not the output of formula code alone. If you only need the formula itself, it is recommended that the source code prevail, and the formula in the scan should be manually confirmed one by one.

3. If the scanned piece is not OCR, what will happen if it is translated directly?

You can upload directly without having to do OCR first. The retention effect of the formula and layout is subject to the downloaded result. Only when you need to copy text and do keyword search, you need to do OCR through the desktop client to generate a selectable text layer.

4. Can the translated formula be directly used for contribution?

Direct use is not recommended. Both machine translation and OCR can make mistakes, formulas and symbols need to be confirmed one by one against the original text, and the cited data and conclusions should also be based on the original paper.

5. What should you look at when choosing free tools to process papers?

Focus on file size restrictions, whether there are watermarks in the export results, privacy policies, and whether typesetting and formulas are retained. If the paper contains undisclosed data, confirm how the platform handles user files before uploading.

# Summary

The key to whether the formula of the scanned paper can be retained is to judge whether the document is scanned or not, whether there is LaTeX source code, and then choose bilingual control output. Check the formula and layout item by item according to the checklist before delivery, and the risk can be controlled within an acceptable range.

Open OmniDoc to edit, convert or translate your LaTeX document