2 ms·
The whole article is essentially about this not being the case, such that it's surprising when it is. To summarize: Not all translations can be to words whose r
by derefr 7y ago
The whole article is essentially about this not being the case, such that it's surprising when it is. To summarize: Not all translations can be to words whose referents are one-to-one mappings with referents in the source language, because not all cultures embodied by ethnolinguistic enclaves have the same "knowledge-base" of referents to attach words to.
There are languages (e.g. historical Japanese) with no single word meaning "blue" or "green", only a word meaning "blue-green." In my made-up notation, that'd be:
en["blue"].referent ≭ en["blue"].translate(ja).referent
...and that's, to be clear, because:
en["blue"] ≭ en["blue"].translate(ja).translate["en"]
Instead, you'd be able to make these assertions:
en["blue"].referent ⊂ ja["青"].referent
en["green"].referent ⊂ ja["青"].referent
en["blue"] ⊂ ψ(ja["青"].translate(en))
en["green"] ⊂ ψ(ja["青"].translate(en))
(i.e., the Japanese ao ('青') has a referent containing both the referents for the English 'blue' and 'green'; and translating 青 back into English gets you a word that's kind of "in superposition" between being English 'blue' and 'green'—a probability density function containing both possibilities—such that a translation AI—or a human—would likely be using context to decide what the correct translation of the word is.