Glossary imposition in translations
In very simple terms, glossary imposition is the act of searching for certain terms in a text to translate or replace them with preferred terms. This can be for example to make sure a term is consistently translated or to ensure, as applicable, that preferred terms are always used, or conversely that undesired terms are never used.
Glossary imposition can happen before the content is translated (pre-translation), after (post-translation), or both.
For the sake of simplification, we will refer to language as source and target and to the glossary as search and replace.
Pre-translation imposition
- The process starts with the creation of the glossary itself, where Language IO and the customer compile search and replace pairs that typically correspond to the source language and target language respectively.
- When translation is requested, but before the content is sent out for translation, the relevant terms or phrases are flagged in the source content according to this glossary.
- As the content is sent for translation, glossary terms that are identified in the source content are substituted, and terms that are flagged as “do not mutate” or ”apply no translate” mean that the MT engine does not modify the replaced term. Otherwise, the replaced term is included in the translation as normal content and may be modified through translation.
Note: “Do Not Translate” vs. “Do Not Mutate”.
“Do Not Translate” is an instruction to not translate the detected word at all. This is marked in the source content, typically with [[ and ]] tags.
“Do not Mutate” means that the detected word must be taken as is for translation. This means that other forms of the word (such as singular or plural, conjugation or other grammatical variations) cannot be used for the translation. This setting is defined in the glossary entry itself.
Post-translation imposition
Post-translation imposition works on the same basis but the process is simpler.
- Language IO creates the search and replace pairs in the customer’s glossary, in the target language only. This must be set up accordingly prior to the translation requests, so that terms are correctly flagged to apply the glossary after translation (this is the post-translation imposition).
- The user requests a translation.
- When the server processes the translation and the translated content comes back, the system performs a secondary search in the translated content to find any post-glossary terms (the terms with the setting mentioned in step 1) and replaces them with the approved term, as applicable.
- The final, edited translation is delivered to the user.
Example of imposition
Suppose that you do not want conversations to use trademarks in speech. You set up your glossary rules so that, if the system detects a trademark, for example in French “kleenex” being used to mean “tissue”, it replaces it with the generic word for “tissue”:
- Input: J'ai essuyé l’excès de pâte thermique avec un kleenex mais mon processeur n’est toujours pas détecté.
- Imposition: J'ai essuyé l’excès de pâte thermique avec un [mouchoir] mais mon processeur n’est toujours pas détecté.
- Output: I wiped off the excess thermal paste with a tissue but my CPU is still not detected.