Machine Translation

John Cowan cowan at
Wed Sep 9 19:47:24 CEST 2009

Peter Constable scripsit:

> A generic tag ("machxlat"?) doesn't seem like a terrible idea.

With "machxlat-systran" vs. "machxlat-google"?

I think this is a good use case for a BCP 47 extension; translation source
information is closely associated with BCP 47 language tags, such that it
would make sense to pack them into a single string, yet distinct enough
from the language/origin/variant model to not fit into it well.

So we could have t-original, t-human, t-machine and possibly others.
These would be added to the end of any valid language tag to carry
information about translation sources.

To be sure, the administrative barrier against adding a new extension
is *much* higher than that against adding a new variant, although the
required RFC will be much less complex than BCP 47.

(This should silence Doug's whinging on the subject of extensions.)

Principles.  You can't say A is         John Cowan <cowan at>
made of B or vice versa.  All mass
is interaction.  --Richard Feynman

More information about the Ietf-languages mailing list