AI and translation

[Humanic]

I have been thinking lately about if it will ever be possible to have Tibetan texts translated accurately by AI.

Are the days of the lotsawa numbered?

[Zhen Li]

It will be able to translate accurately, but there is greater value in a human translator—this is why they still exist for languages that have long been translated by machines. Humans can more accurately tell what words sound right to the ear. So, a human translated text will always end up being nicer to read. One thing that also sets computers and humans apart is verse. Those who know anything about poetry will be aware that AI writes very bad poetry, it can't count syllables, detect metre, or rhyme well—these are the reserve of humans, so far.

As for professional translation for high-resource languages, like French to English, translators are still employed. Translation memory tools that they use do provide machine translation of whatever text they are working on, but that is only used as a quick reference or look-up tool. In a way, a machine translation can act as a kind of quick jog to the memory about unfamiliar terms, or help with syntax for a translator, but is never a substitute for human composition.

[Kim O'Hara]

There's also the issue of background knowledge of the subject. AI knows the words but, in a sense, knows nothing about their meaning. When it's a simple matter of translating 'noir' to black' it will be okay but it doesn't know what either 'noir' or 'black' look like or feel like, or what their connotations are, so it will miss the occasions where a different word would have been better.
And that's a very mundane instance from two closely related languages. Extrapolate that to translating highly technical material between two unrelated languages and two unrelated cultures, and we really do need people - ideally, subject experts with an excellent grasp of both languages.


Kim