Translation Memory

4 min read

Translation Memory allows Translator V2 to reuse previously approved translations exactly. It is intended for wording that must remain consistent, such as product names, legal clauses, standard disclosures, and recurring sentences.

Translation Memory is different from a glossary. A glossary provides terminology guidance to the language model. Translation Memory replaces matching source text with its approved target text, ensuring that the approved wording is reused verbatim.

How it works

Translator V2 loads the configured source and target language columns from the Translation Memory workbook.
It finds approved source text within each translation unit, such as a paragraph or table cell.
Matched source text is protected while the remaining content is translated.
After translation, the protected content is restored using the approved target wording.
If a translation unit is fully covered by Translation Memory, Translator V2 can return the approved translation without sending that unit to the language model.

Approved target text is restored after text post-processing and is therefore not modified by configured post-processors.

Translation Memory and glossary behavior

Capability	Translation Memory	Glossary
Primary purpose	Reuse approved source-target text exactly	Guide terminology used by the language model
Result	Approved target wording is restored verbatim	The language model decides how to apply the glossary entry
Matching	Exact matching rules; no fuzzy matching	Can identify close terminology matches
Language-model usage	Fully covered translation units can skip the language model	Glossary entries are included in the language-model prompt
Configuration file	Uses a dedicated Translation Memory workbook	Uses a separate glossary file

Both features can be active at the same time. Translation Memory protects approved wording first; the glossary is then applied to the remaining content.

Prerequisites

Create an XLSX workbook containing approved source and target translations.
Upload the workbook to the Knowledge Base.
Keep the Translation Memory workbook separate from the glossary file.
Record the Knowledge Base scope ID and the exact workbook filename for the configuration.

Workbook structure

Translator V2 reads the first worksheet. The first row must contain language column names, and each subsequent row contains approved translations of the same text.

English	German	French
The agreement is effective immediately.	Die Vereinbarung tritt sofort in Kraft.	Le contrat entre en vigueur immédiatement.
Past performance is not indicative of future results.	Die Wertentwicklung in der Vergangenheit ist kein Indikator für zukünftige Ergebnisse.	Les performances passées ne préjugent pas des résultats futurs.

Language columns

Use recognizable language names such as English, German, French, Simplified Chinese, Traditional Chinese, Traditional Chinese (Taiwan), or Traditional Chinese (Hong Kong).
When the source language is known, Translator V2 uses the matching source column and the selected target column.
If the known source language has no dedicated column, a Default column can provide fallback source text.
If the selected target language has no dedicated column, a Default column can provide fallback target text.
When the source language cannot be detected, all named language columns other than the selected target are considered possible source columns. The Default column is not used as an unknown-language source.
Empty cells and non-text values are ignored.

Matching behavior

Text that does not contain Chinese, Japanese, or Korean characters is matched case-insensitively and must respect phrase boundaries. For example, rig does not match inside righteous.
Text containing Chinese, Japanese, or Korean characters uses exact, case-sensitive substring matching.
Fuzzy matching is not applied.
If source entries overlap, the longer matching source text takes precedence. Equal-length entries retain their workbook order.
Entries without usable letters or numbers are ignored.
Approved target values are restored exactly as written in the workbook.

Configuration

Default

Translation Memory is disabled by default:

json

{
  "translationMemoryConfig": null
}

Enabled example

json

{
  "translationMemoryConfig": {
    "scope_id": "scope_111111111111111111111111",
    "filename": "translation-memory.xlsx",
    "active": true
  }
}

Parameters

Parameter	Description	Default
`scope_id`	Knowledge Base scope containing the Translation Memory workbook	`null`
`filename`	Exact filename of the XLSX workbook	`null`
`active`	Whether Translation Memory is enabled	`false`

PowerPoint formatting

PowerPoint text can contain multiple runs with different formatting. When Translation Memory matches text within a paragraph, the approved wording is preserved, but run-level formatting and hyperlinks within that paragraph may be simplified. Test representative presentations before enabling a large Translation Memory.

Fallback behavior

If Translation Memory is inactive, missing, empty, unreadable, or does not contain compatible language columns, Translator V2 continues the translation without Translation Memory substitutions.

Recommended practices

Use Translation Memory only for translations that have been reviewed and approved.
Keep source text specific enough to avoid unintended matches.
Use longer entries for complete approved phrases and shorter entries only when they are valid in every context.
Use consistent language-column names across workbooks.
Review duplicate source entries; the last applicable value may take precedence.
Test the configuration with representative text, Word, Excel, and PowerPoint content before enabling it broadly.