Working with a scanned document requires that document to be turned into text first. This step comes before translation and carries its own risk of error: errors that are not in the source arise exactly here. That is why text extraction is planned as a separate work item and is reviewed separately.
Definition of the step
In files that cannot be edited, the text is turned into type by hand or with a tool. This is a separate job that comes before translation. Its duration is added to the plan.
A new source of error
Retyping is a step in which errors not present in the source may arise. For this reason the output is compared with the source. The check is not skipped.
Automatic recognition output
Character recognition output is not used directly, because it replaces characters it cannot recognise with similar ones. The output is compared with the source. The error is silent.
Checking numeric fields
Recognition errors in figures cannot be spotted from the context and are the riskiest. Numbers are verified separately against the image. The list method is used.
Handwritten sections
Handwritten areas cannot be recognised automatically and are read by eye. An illegible section is not guessed at. The situation is recorded.
Rebuilding table structures
When tables are retyped, the cell layout is preserved. Rows that get mixed up silently produce wrong information. The structure is checked.
Column layout
On pages with multiple columns the reading order may be disrupted. The columns are handled one by one. The order is verified.
Carrying over the formatting
Emphasis such as bold and underlining may carry meaning and is preserved. A text without emphasis loses information. The decision is made at the outset.
Heading hierarchy
The heading levels are rebuilt and show the structure of the document. A flattened text becomes unreadable. The structure is preserved.
Marking page boundaries
Which text is on which page is tracked, and this makes comparison possible. If the page information is lost, matching becomes difficult. The marking is done.
Showing illegible areas
Illegible sections are marked with a uniform expression. They are not filled in by guesswork. A list is kept separately.
Descriptions of seals and signatures
Non-text elements are recorded by describing them. These descriptions are carried into the translation. Leaving them out makes the document look incomplete.
Text inside images
The writing in diagrams and charts is put on a separate list. This text does not show up in automatic extraction. The list is drawn up by hand.
Second reading
The extracted text is compared with the source by a second person. A one-person check is not enough at this step. The method is applied.
Effect on the turnaround
Text extraction lengthens the total turnaround noticeably. The client is told about this time openly. The plan is kept realistic.
The source file alternative
If the client has an editable file, this step disappears completely. That is why the request is made persistently. The gain is explained.
Approval of the extracted text
In some jobs the extracted text is approved by the client. The approval secures the later stages. The decision is made according to the job.
The next step
Once the text is ready, the visual content is gathered in a separate list. The two jobs can run in parallel. The order is kept.