Overview
Machine translation converts content between languages. Neural systems translate whole sentences in context rather than word by word, which is why modern output reads naturally where older statistical systems read like a dictionary lookup.
How it works
An encoder builds a language-agnostic representation of the source, and a decoder emits the target language from it. Context windows spanning whole documents let the system resolve pronouns and keep terminology consistent across paragraphs.
Use cases
Website localisation
Publish the same catalogue and marketing pages across many locales.
E-commerceMultilingual support
Let agents answer in their own language and deliver in the customer's.
SupportSubtitling
Generate and translate captions for video libraries at volume.
MediaDocument translation
Convert contracts and manuals while preserving layout and terminology.
LegalBenefits
- Opens content to markets without a translation vendor per language.
- Keeps terminology consistent through a glossary.
- Turns support and documentation multilingual almost immediately.
- Round-trips quickly enough for live chat and subtitling.
Limitations
- Idiom, humour and cultural register still need a human.
- Low-resource language pairs lag badly behind major ones.
- Legal, medical and safety-critical copy needs certified review.
- Gendered and formal/informal forms are frequently guessed wrong.
What to look for when choosing a tool
- Glossary and termbase support
- Formality and tone controls
- Document-level context, not sentence-by-sentence
- Coverage and quality for your specific language pairs
- Data residency if you translate confidential documents
FAQ
How accurate is AI translation compared to a human?
For general content between major languages it is close enough for most business use. For marketing, legal or safety-critical text, a human reviewer who speaks the target language is still necessary.
Can I enforce my own terminology?
Yes — most business-grade tools accept a glossary or termbase that pins product names and industry terms to a fixed translation.
Does it handle formal and informal address?
Some tools expose a formality setting. Without one, the system guesses, which produces awkward results in languages that distinguish the two grammatically.