What machine translation options are available in SpeakUp?
SpeakUp offers two machine translation options:
- Default Machine Translation
- OpenAI-Based Translation
The table below compares how they work, which languages they cover, and where your data is processed, so you can choose the option that best suits your organisation.
|
|
Default Machine Translation |
OpenAI-Based Translation |
|
Model type |
Purpose-built, narrow translation model (see explanation below) |
General-purpose large language model (see explanation below) |
|
Language coverage |
Narrower (Intake language list) |
Broader (Intake language list) |
|
Translation quality |
Standard |
More natural, context-aware |
|
Text data location |
EU (standard hosting footprint) |
EU (standard hosting footprint) |
|
Voice transcription location |
EU (standard hosting footprint) |
May process data outside EU/EEA (covered by completed Transfer Impact Assessment) |
Narrow (purpose-built) translation models are trained specifically to do one job: convert text from one language to another. They're smaller and more specialized, which typically means faster and cheaper to run, but they translate more literally and can struggle with idioms, ambiguity, or context that depends on the surrounding conversation.
Large language models (like the one behind the OpenAI route): are general-purpose models trained on a much broader range of text and tasks, translation being just one of them. They carry a wider understanding of language and context, they tend to produce translations that read more naturally and handle nuance, tone, and ambiguous phrasing better. The trade-off is that they're bigger models, so this route is the one where voice transcription can involve processing outside the EU/EEA.