A signed contract arrives as a scan, the deadline is close, and one faint handwritten change may alter who owes what. Before anyone translates a clause, the team has to make sure the source can be read, identify which version is authoritative, and distinguish uncertain wording from an OCR mistake.
Preparing a scanned contract for translation is the work of making the source readable, searchable, and clear about its status without changing what the contract says. OCR, or optical character recognition, converts the letters in an image into machine-readable text. OCR helps with searching and drafting, but it doesn’t certify that the recognized text is correct.
The safest workflow keeps the page image as the authority. A searchable text layer is useful, but a translator or reviewer must compare important text with the visible page, especially where a missed character could affect a party, payment, date, obligation, or exception.
What a translation team needs to prepare¶
A translation team needs more than a PDF that opens. The file should show every page clearly, represent the intended contract version, and include enough context to interpret marks, tables, attachments, and defined terms.
A scanned PDF may contain only images of pages, an OCR text layer, or both. An image-only PDF looks like a page on screen, but its words aren’t necessarily available for selection, search, or direct text processing. A searchable PDF usually combines the image with a hidden text layer that records what OCR software thinks the page says.
The distinction matters because the visible image and the text layer can disagree. A reader may see “15” on the page while the OCR layer says “I5,” or a line break may cause a clause to appear incomplete in extracted text. A translation team that receives only an OCR export may not know which characters came from the page and which came from recognition.
Start by checking the file, not by assuming its format tells the whole story. Open the PDF and try selecting a sentence. Copy it into a plain-text editor and compare it with the visible page. Search for a distinctive word, then check whether the result lands in the expected place. A selectable-text result is useful, but it doesn’t prove the text layer is accurate.
The University of Illinois OCR guide says, “The recommended resolution for best scanning results for OCR accuracy is 300 dots per inch (dpi).”
That recommendation is a useful starting point for creating a scan, not a promise of error-free recognition. Contrast, focus, page alignment, font, and the condition of the original still affect whether OCR can read a particular line.
A practical handoff should include the source scan and a short note about the job. The note can identify the target language, the version to translate, any known amendments, preferred terminology, formatting expectations, and questions that need the drafter’s attention. Keep that information separate from the contract itself so nobody mistakes an instruction for a contractual term.
The team should also distinguish between a source defect and a translation question. “The word after shall is unreadable” describes a source defect. “Does this clause require delivery at the buyer’s warehouse?” describes a question about meaning. The first needs inspection of the image or a better copy; the second may need clarification from the drafter.
For a broader file handoff checklist, see File Prep for Translation: A Practical Guide for Teams. Scanned contracts need the same basic discipline as other files, with extra attention to the accuracy of the text layer and the legal significance of small marks.
Improve the scan before running OCR¶
Scan quality affects how much work a person has to do after recognition. The goal isn’t to make a page look polished; the goal is to preserve every mark that could affect interpretation while giving OCR a clear image to read.
For clean, high-contrast printed text scanned in bitonal mode, NARA’s archival guidance gives a range of 300-600 ppi and recommends 600 ppi for that preservation use case. NARA specifies grayscale at 300-400 ppi for textual documents with poor legibility, fading, staining, annotations, or low contrast, and recommends 400 ppi. For content where color matters to interpretation, NARA’s 2002 guidance specifies 24-bit RGB color at 300-400 ppi and recommends 400 ppi (NARA guidance).
Those values come from archival imaging guidance, not a universal minimum for private contract translation. The University of Illinois OCR guide recommends 300 dpi for best OCR scanning results (OCR Best Practices). Use the guidance to choose a sensible scan setting, then inspect the actual pages. A higher resolution can’t restore a word that has been cut off, blurred, or hidden by a fold.
| What the page looks like | Practical preparation | Why the detail matters |
|---|---|---|
| Clean, dark printed text on a light background | Capture the full page in focus, with clear contrast and no clipped edges | OCR needs the shape of each character to remain distinct |
| Faded, stained, or low-contrast print | Preserve the visible page in grayscale or color rather than forcing a harsh black-and-white image | Faint strokes and annotations can disappear when tonal detail is discarded |
| Older or discolored paper | Consider RGB capture when the paper’s color or discoloration helps distinguish text from the background | The University of Illinois guide advises RGB scanning for older or discolored documents |
| Pages that lean or curve | Deskew pages and correct cropping where possible | Slanted baselines can confuse character and line recognition |
| Handwriting, initials, or pen changes | Keep a clear image and flag each mark for human reading | OCR cannot be relied on to recognize handwritten text accurately |
The University of Illinois guide warns that brightness set too high or too low can reduce recognition accuracy, and that low-contrast documents can produce poor OCR. A common mistake is to increase contrast until the page looks clean on screen, only to erase faint pencil, a light stamp, or the lower stroke of a handwritten number.
Page alignment matters for a different reason. OCR software uses the expected direction and shape of lines to decide where characters and words begin. A skewed line can cause the system to group letters incorrectly or mistake a character for part of the next one. The University of Illinois guide puts the mechanism plainly:
“The straightness of the initial scan can affect OCR quality. Skewed pages can lead to inaccurate recognition.”
A deskewed page can improve line detection, but preprocessing must not crop away page numbers, marginal notes, signatures, or text near the binding. Check the full page boundary after cropping. If the original is curved or folded, a flatter, sharper rescan is safer than aggressive image cleanup.
Color also has a practical role beyond appearance. A red handwritten insertion, blue signature, or colored stamp may be distinguishable from printed text only in the original image. When color contributes to understanding the page, retain it and explain why it matters in the handoff note. Don’t assume a monochrome copy will preserve every meaningful distinction.
The current eCFR digitization rules specify 300 ppi at the source document’s size for covered federal agencies digitizing modern textual paper records. The rule concerns permanent federal-record digitization, not a universal standard for privately held contracts (36 CFR Part 1236, Subpart E). That difference matters when someone cites a regulatory number as if it applied to every business scan.
Run OCR, then check the text against the page¶
OCR is a recognition step, not a proofreader. A machine-readable layer can help a translator search for defined terms and locate repeated phrases, but the source page remains the reference whenever recognition is uncertain.
A useful working sequence is to retain the original scan, make a searchable copy, and compare the OCR output with the image. Keep the two versions connected so a reviewer can return from a suspicious word to the exact page and location. If the scanning process produces a separate text file, preserve that file as a working aid rather than replacing the original PDF.
The eCFR digitization subpart applies to federal agencies digitizing permanent paper records. It requires covered agencies to use quality-management techniques and validate digitized records against its standards, while explicitly not requiring OCR during digitization; agencies may use OCR to meet business needs (36 CFR Part 1236, Subpart E). The rule is a useful reminder that image capture and OCR are separate choices, but it doesn’t impose a private-contract workflow.
NARA’s archival guidance says OCR text versions should accompany scanned-image transfers when available, as additional files that enable full-text searching linked to the appropriate scanned image (NARA guidance). That model is practical for translation teams too: keep the scan available, and use the OCR layer to find and work with text without treating it as a replacement for the page.
Review OCR output by risk, not just by page order. A quick pass should focus on text that could change rights or obligations if one character is wrong:
- Party names, company names, addresses, and defined terms.
- Dates, amounts, percentages, account details, and payment deadlines.
- Clause numbers, cross-references, paragraph letters, and exhibit labels.
- Negations and conditions, including “not,” “unless,” and “only if.”
- Tables, signature blocks, stamps, seals, handwritten insertions, and marginal notes.
- Text near page edges, folds, binding shadows, or low-contrast areas.
The list isn’t a substitute for reading the full contract. It gives the reviewer a way to spot-check high-impact parts systematically, then compare the remaining OCR text with the scan during translation and final review.
| Review target | Common recognition problem | What to do |
|---|---|---|
| Names and defined terms | A letter is mistaken for another letter, or a name changes spelling between pages | Compare each occurrence with the scan and record the approved spelling |
| Dates and amounts | A digit, punctuation mark, or decimal separator is misread | Verify against the visible page and check repeated references |
| Negations and conditions | A short word is omitted, merged, or misread | Read the entire sentence in context and compare the scan |
| Tables and clause references | Reading order becomes confused or a reference attaches to the wrong line | Check row and column relationships, numbering, and nearby text |
| Handwriting and stamps | OCR returns guessed text or omits the mark | Escalate for human reading and mark the text as uncertain if needed |
The University of Illinois guide says that if the goal is 100% text accuracy, OCR output must be checked and corrected after recognition because the system can’t perform that verification itself (OCR Best Practices). The point isn’t that every team must promise perfect recognition; the point is that OCR alone can’t confirm its own work.
“OCR systems cannot recognize handwritten documents with any degree of accuracy,” the University of Illinois guide cautions.
A handwritten amendment shouldn’t be treated as dependable OCR text, even if a system produces plausible words. Ask for a clearer image, a typed copy, or confirmation from someone authorized to explain the mark. The translation team should preserve the uncertainty rather than silently turning a guess into contract language.
The same caution applies to small type, unusual typewriter fonts, and low contrast. The University of Illinois guide identifies those conditions as factors that can reduce OCR accuracy. A clean-looking text export can still contain errors in precisely the places where the scan is hardest to read.
For a comparison of ways to turn a scan into an editable file, see Scanned PDF to Editable Word Before Translation: Options Compared. Whichever route a team chooses, the comparison against the source image is what catches recognition errors.
Set the contract version and terminology before translation¶
Accurate text is only one part of accurate contract translation. A translator also needs to know which source version controls, how recurring terms should be handled, and which source ambiguities need a decision rather than an editorial guess.
Start by naming the file clearly and recording its status. A filename such as Supply_Agreement_signed_2026-10-18.pdf is more informative than scan_final2.pdf, but the name alone doesn’t prove that the document is the approved version. The handoff note should identify who confirmed the version and whether amendments, schedules, exhibits, or side letters belong with it.
When several copies differ, don’t merge them into a single working file without permission. A clean copy may omit handwritten changes; a signed copy may have missing pages; a later draft may contain revisions that weren’t executed. Keep each version distinct and ask the document owner which one to translate. Record the answer where the translator and reviewer can see it.
A terminology note should define how the team will handle recurring terms, company names, abbreviations, product names, and references to other contract documents. The aim is consistency, not a glossary so large that nobody can maintain it. Include a preferred translation only when the client or responsible legal contact has approved it. Otherwise, list the source term and flag the choice for review.
Formatting instructions belong in the brief too. A team may need the translated file to preserve clause numbering, tables, signature spaces, defined-term capitalization, or the position of stamps and notes. Those expectations affect how the output should be checked. A translation that preserves the words but scrambles a table can obscure which amount belongs to which service.
The American Translators Association’s document-preparation page is included in the research dossier, but its content wasn’t available for verification for this article. Rather than attribute specific advice to that page, use the practical handoff checklist below to make decisions visible before work begins.
| Handoff item | What to record | What it prevents |
|---|---|---|
| Source identity | Filename, date received, and who confirmed its status | Translation of an outdated or unapproved copy |
| Related documents | Schedules, exhibits, amendments, and referenced attachments | Missing context or incomplete deliverables |
| Terminology | Approved names, recurring terms, abbreviations, and preferred treatment | Inconsistent wording across clauses |
| Layout requirements | Numbering, tables, headings, signatures, and annotations to preserve | A usable translation being mistaken for a visually complete one |
| Open questions | Unclear words, ambiguous clauses, and the person responsible for answering | Silent guesses presented as settled meaning |
Legal translation isn’t simply word substitution. The translator has to understand what a clause is meant to do and convey that purpose in the target language without inventing a clearer source than the parties wrote.
Paula Arturo’s ATA Law Division article describes the first task as identifying what the parties intend to achieve with each clause. The article also advises translators to flag vague language and seek clarification from the primary drafter (A Methodological Approach to Contract Translation).
“First, when the source text is vague, it’s always a good idea to flag that vagueness and ask the primary drafter for clarification.”
A translator can explain that a phrase is ambiguous, but shouldn’t decide which commercial outcome the parties intended. If the drafter can’t clarify the text, the team can agree on a translation that preserves the ambiguity and document the unresolved question for the reviewer.
Compare preparation and translation workflows¶
A scan can be handled by an agency, an independent translator, or a document-translation tool. The right route depends on the file’s condition, the level of review required, the expected output, and whether anyone must certify the translation.
An agency workflow may combine scanning, OCR, translation, and review under one project. A single contact can coordinate version questions and return a formatted deliverable. The trade-off is that the client should still ask who checks the OCR against the image, how unclear handwriting is handled, and whether changes to the source after work starts affect the process.
An independent translator may work well when the scope is defined and the contract’s terminology is familiar. A direct handoff can make questions easier to discuss with the translator, but a client may need to coordinate separate OCR, layout, and review support. Ask whether the translator wants an editable OCR file, the original scan, or both.
A document-translation tool can help when the source exists only as a scan and the team needs a draft in an editable layout. In ChatsControl, you upload a scanned PDF or a supported image file, the system reads the page, rebuilds a formatted Word copy, checks that copy against the scan, and translates it as a Word document. The product also grades scan quality and counts words the check still finds missing; its documentation warns that handwriting and very degraded scans are read unreliably. The bilingual review view can then help a reviewer compare source and translation. ChatsControl is a document-translation option, not a full CAT tool or a substitute for human legal review.
The workflow fits a typical situation in which an operations or legal team has a legible printed contract as a scan and needs a draft that can be reviewed in Word. A fully handwritten contract, a badly damaged page, or a scan subject to strict on-premise requirements calls for another route or a human-led process.
| Workflow | Useful when | Questions to settle before work |
|---|---|---|
| Translation agency | The project needs coordinated OCR, translation, layout, and review | Who checks the recognized text, how are questions escalated, and what file will be delivered? |
| Independent translator | The source is manageable and the client can provide a clear brief | Does the translator handle OCR and formatting, or should the client prepare those separately? |
| Online document tool | A digital draft and editable layout are useful for review | Can the scan be read reliably, how will uncertainty be exposed, and who checks legal meaning? |
A translation tool’s automatic checks can help identify differences in numbers, names, terminology, or omissions, but a check is not a legal opinion. A reviewer still has to decide whether the source itself is clear and whether the translation preserves the intended relationship between clauses, tables, and defined terms.
For software-specific workflows, How to Translate a Scanned PDF in Trados Studio 2024 and 2026 and How to Translate a Scanned PDF in memoQ 12.5: OCR, Import, Export explain other routes. A CAT tool (computer-assisted translation software) may be part of a professional workflow, but it doesn’t make a poor scan legible by itself.
A step-by-step handoff checklist¶
A short, repeatable handoff makes it easier to see what has been checked and what still needs a decision. The following sequence works whether the translation will be completed by an agency, a freelancer, or a document tool.
-
Collect the complete source set. Gather the contract, amendments, schedules, exhibits, and any attachment the contract references. Ask the document owner whether every page is present and which file represents the version to translate.
-
Preserve the original scan. Keep an untouched copy of the received file. Create any deskewed, cropped, or OCR-processed copy separately so the team can compare edits with the source image.
-
Check page readability. Review each page for clipped edges, blur, shadow, skew, low contrast, faint writing, and missing pages. If a mark is unclear, ask for a better scan or flag the location rather than guessing.
-
Create or inspect the text layer. Test whether the PDF has selectable text, then search and copy a few passages. If OCR is needed, keep the resulting text connected to the scanned page so reviewers can trace questionable words back to the image.
-
Review high-risk details. Compare names, dates, amounts, negations, clause references, handwritten changes, stamps, and table entries against the page. Record unresolved items with a page and location.
-
Write the translation brief. State the target language, approved source version, terminology preferences, layout expectations, and questions for the drafter. Mark any instruction that is not part of the contract itself.
-
Translate, review, and close open questions. Compare the translation with the source, verify the high-risk details again, check cross-references, and confirm that unresolved source ambiguity hasn’t been silently removed.
A checklist is only useful if the team records the result. A short note such as “page 7, handwritten amendment after clause 4.2, unclear final word” is more actionable than “some text is hard to read.” The note should identify the issue without inserting a guessed interpretation.
Translation should pause for a decision when an illegible word or unclear clause changes the possible meaning. A minor layout imperfection can sometimes be corrected during formatting; a missing “not” or uncertain amount shouldn’t be repaired by choosing the most plausible option.
Teams that quote scan-heavy work should keep the scan and the recognized text available while scoping the job. How to Quote a Scanned Document When Word Count Finds Nothing explains why a word-count tool may not find words in an image-only PDF and what information helps make the scope clear.
Common pitfalls and how to prevent them¶
Most avoidable problems come from treating a convenient file as a reliable source. A searchable PDF, a clean OCR export, or a polished Word recreation can all look finished while hiding a mistake.
Treating searchable text as verified text. A PDF can contain selectable text that came from OCR, and selection only proves that a text layer exists. Compare key terms with the image and review the full text according to the project’s risk and review requirements.
Over-cleaning a faint page. Excessive brightness or contrast changes can erase pencil marks and pale print. The University of Illinois guide notes that brightness set too high or too low can reduce recognition accuracy, and that low contrast can cause poor OCR (OCR Best Practices). Keep an untouched original and check any processed copy against it.
Cropping the page too tightly. Page numbers, handwritten notes, seals, and words near the binding can carry meaning. Inspect the complete page boundary after cropping, especially where a scan was made from a bound or folded contract.
Trusting plausible handwriting recognition. A confident-looking OCR result may still be wrong. The University of Illinois guide says OCR systems can’t recognize handwritten documents with dependable accuracy. Ask for a typed copy or human reading where the mark matters, and keep uncertainty visible until someone authorized can resolve it.
Mixing contract versions. A draft, a signed version, and a version with later amendments aren’t interchangeable. Keep them separate, identify the version approved for translation, and make the decision traceable in the handoff note.
Repairing ambiguous language without approval. A translator may see a phrase that seems incomplete or inconsistent with another clause. The apparent mistake could reflect the parties’ wording, a missing amendment, or a source-file defect. Ask the drafter instead of silently changing the legal effect.
Losing table relationships. Extracted text may list labels and values in an order that doesn’t match the page. Check rows, columns, footnotes, and totals visually, then verify the translated file preserves the relationship between each heading and its value.
Paula Arturo’s ATA Law Division article advises the translator to flag vague language and ask the primary drafter for clarification (A Methodological Approach to Contract Translation).
That advice matters most when the source is both hard to read and legally ambiguous. The translator can identify the problem and explain its possible effect, but the drafter or other authorized contact must resolve the underlying intent.
A useful quality-control record separates three outcomes: confirmed text, uncertain source text, and confirmed source wording that remains ambiguous. That distinction helps a reviewer decide whether to correct OCR, request a better image, or preserve the ambiguity in translation.
FAQ¶
What resolution should a scanned contract be for accurate translation?¶
The University of Illinois OCR guide recommends 300 dpi for best OCR scanning results (OCR Best Practices). NARA’s archival guidance gives different ranges for clean text, poor-legibility text, and content where color matters, so resolution should match the page condition rather than serve as a guarantee.
How can I tell whether a contract PDF has selectable text or only scanned images?¶
Try selecting a sentence and copying it into a plain-text editor. If the words don’t copy and the page behaves like a picture, the PDF may be image-only; if words do copy, compare them with the page because selectable text can still contain OCR errors.
Should I run OCR on a contract before sending it for translation?¶
OCR can make a scan searchable and provide text for translation workflows, but recognition doesn’t verify itself. Send the original scan with any OCR output, and check important text against the page; NARA also describes OCR text as an additional file linked to scanned images in its archival guidance (NARA guidance).
Which parts of a scanned contract need checking after OCR?¶
Check names, dates, amounts, clause references, negations, defined terms, tables, stamps, signatures, and handwritten additions. Compare those details with the visible page, and flag any text that remains unclear rather than guessing.
How should a translation team handle unclear clauses and contract versions?¶
Identify the approved source version, keep amendments and attachments with the relevant contract, and record terminology and formatting instructions in a separate brief. Ask the drafter to clarify vague language; the ATA Law Division article describes clarification as preferable to silently translating an uncertain meaning (A Methodological Approach to Contract Translation).
Can an online tool translate a scanned contract without losing its layout?¶
Some document-translation tools read scanned pages and rebuild a formatted file, but results depend on scan quality and still need review. ChatsControl can produce a formatted Word translation from scanned PDFs and supported image files, while its product documentation identifies handwriting and very degraded scans as limitations.