Where and When AI translation is a No No
The good news is that we can logically identify where and when AI translation is a No No.
WHERE – Languages not (yet?) fitted for AI translation:
- Less common languages or dialects, as there is not enough parallel data to train AI. For example some African, Oceanic and Southeast Asian languages.
- Highly regionalized languages. Ex: various flavors of Arabic.
- Agglutinative languages, which attach prefixes and suffixes to a root word. Ex: Greenlandic.
- Tonal and context-dependent languages. For example Chinese, Thai, and Japanese.
WHEN – Project not (yet?) fitted for AI translation:
- Technical jargon.
- Software localization as AI translations tend to be wordy and the space is limited.
- Branding as there is no emotion.
- Poorly written and/or cryptic original text, since humans can logically eliminate some possible interpretations, but machines cannot.
- Highly creative materials such as poetry.
- Content that refers to a cultural element that does not exist in the target country.
- Some idiomatic expressions.