Translation files often carry more than the text itself: a game script may have a column saying who is speaking, a product sheet a column with a character limit per string. Wordbee stores these as segment custom fields, and your AI post-editing and quality assurance workflows can read them. Turn on one option, and every segment travels to the AI together with its custom field values, so your instructions can use them: "keep the translation within the character limit in field 2".
Enabling It
Open the configuration of the workflow and turn on Include segment custom fields. The option is available in post-editing and quality assurance workflows, in both Standard and Codyt projects:
-
Post-editing, or quality assurance run as a supplier service: the AI Agent tab of the supplier pricelist service.
-
Quality assurance run from a QA profile: the AI Analysis rule of the QA profile.
If you use the same quality assurance workflow in both places, turn the option on in both: the two configurations are independent.
The option is off by default. With it off, the AI receives exactly what it received before. With it on, every segment is sent with whatever custom field values it holds, and your Custom Instructions tell the AI what to do with them. Machine translation workflows are not covered by this option.
Custom field values reach the AI from any source that populates them: Excel columns, XML or JSON attributes, the API, or manual entry in the Editor. Turn the option on only where your instructions use the fields: the values travel with every batch, so each run takes slightly more processing time and AI usage.
Referencing a Field in Your Instructions
The AI receives each field under its number. Follow these steps to find a field's number:
-
Go to Settings > General Settings > Custom fields and click Configure.
-
Open the Segments tab and click Edit next to the field.
-
Read the Custom field ID at the bottom of the window, for example CustomStr2. The AI receives the number at the end: CustomStr2 arrives as field 2.
Write instructions that name the field by that number and say what to do with its value:
-
"Field 2 contains the maximum character length. Keep the translation within it."
-
"Field 12 names the speaker. Match the register to that character."
-
"If field 5 contains |Legal|, use formal register and do not rephrase."
Segment-level and language-level fields
A segment custom field is attached to the segment, to each language, or to both. The AI receives the levels separately:
-
A segment-level field has one value for the whole segment, for example the speaker.
-
A language-level field holds a value per language, for example a character limit that differs between languages. The AI receives the source-language value and the target-language value separately, so say which one you mean: "Keep the translation within the target-language value of field 15."
A field attached to both can appear at both levels.
What the AI Receives per Field Type
Every value arrives exactly as stored, as text. Write your instructions with the stored form in mind:
|
Field type |
What the AI receives |
|---|---|
|
Text |
The text as entered |
|
Pick list (single choice, custom text allowed, or autocomplete) |
The option text, or the text the user typed |
|
Pick list, multiple choice |
The options wrapped in pipes, for example |
|
Check box |
The text "1" when ticked; nothing when not ticked |
|
Hyperlink |
The full URL, always including the protocol (for example |
|
Date / date-time |
The date as stored |
|
Image |
The image's internal identifier, not the image |
Two consequences worth knowing: a check box is never "true" or "false", so write "if field 7 is present" rather than "if field 7 is true"; and a multiple-choice value keeps its pipes, so quote them when matching ("contains |Legal|").
An empty field is absent. An instruction referencing a field that is empty or does not exist does no harm: the run completes normally.
Values are sent as data, separate from your instructions. This reduces the chance that text inside a value is read as an instruction, but does not rule it out. Keep your rules in Custom Instructions, not in field values.
Who Sees What
The AI receives the custom fields that the user the workflow runs as is allowed to see:
-
Post-editing, and quality assurance run as a supplier service: the user the job is assigned to.
-
Quality assurance run from a QA profile: the user who started the QA analysis.
Visibility follows the field's Enable for internal users and Enable for external users settings, and access rights. A field hidden from external users is not sent when the job is assigned to an external linguist. A user without the right to view segment custom fields sends no custom field values at all. In both cases the run still completes.
Nothing hidden is ever sent. The only effect is less context: the same job with the same instructions can give different, more generic results when it is assigned to a user with restricted field visibility.
Learn More
-
Custom Fields: What segment custom fields are and where they appear.
-
Creating and updating Custom Fields: Creating segment custom fields and setting who can see them.
-
Configuring AI Workflow Parameters: All other workflow settings, including custom instructions.
-
AI Post-Editing and AI Quality Assurance: The workflows that can use custom fields.