4 ms·
I've often wondered if it would be not just possible, but user-friendly, to write in a way that guarantees machine translation is 100% accurate. Specifically,
by crazygringo 5y ago
I've often wondered if it would be not just possible, but user-friendly, to write in a way that guarantees machine translation is 100% accurate.
Specifically, imagine you write a few sentences in English. Then when you hit "publish", it would look for the potentially amibiguous parts of your sentences, and then give you "nearby synonyms" for you to select which one you meant. It wouldn't change the text in your source language, but it would remove the ambiguity so all the output translations would be correct, no guessing.
It would do things like ask which noun "it" referred to since some languages have male/female versions, ask for clarification of verb tenses as necessary, etc.
After all, the hardest part of machine translation is inferring meaning. If that part could be guaranteed to work accurately, it seems like machine translation could become a ton more useful.
- yorwba 5y agoAbstract Wikipedia is kind of trying to do this, but using structured data instead of English. https://meta.wikimedia.org/wiki/Abstract_Wikipedia https://meta.wikimedia.org/wiki/Abstract_Wikipedia
- a1369209993 5y ago> not just possible, but user-friendly, to write in a way that guarantees machine translation is 100% accurate. Not in (human-readable source) English, no, since it's too heavily[0] optimised for general-purpose colloquial communication, but it's probably possible to engineer some cross between text localization systems, procedural content generation for text games, and early-AI-style natural language processing data structures that could serve as a backing representation subject to linting/sanity-checking/etc. Edit: 0: To be clear, "heavily" is not necessarily the same as "well-".