Hacker News Reader

And as the parent commenter points out, not only do they have different information encoded, but they may require different information to be expressed so that it's not permissible to simply omit information that the source language omitted.

https://en.wikipedia.org/wiki/T%E2%80%93V_distinction

https://en.wikipedia.org/wiki/Pro-drop_language

https://en.wikipedia.org/wiki/Evidentiality

View on HN

But there absolutely must be a common denominator. A Japanese sentence might encode the metric "relationship to subject" which an English sentence will encode as "NULL".

The issue is if you're translating from a language that doesn't explicitly encode that information into a language that requires it; you need some way of filling in the blanks. This is all well and great when that information can be inferred from the explicit context provided; the real problem is that with translation there are a non-trivial number of cases that require the translator to consider the pragmatic context of what's being translated to figure out a correct/good translation (e.g. who's speaking/writing, who are they addressing, what is the overall societal context of content, what might be the purpose of the content, etc.). For a lot of these problems it's entirely reasonable to get a computer to fill in the blanks for 90%+ of cases, but the last few percent of cases require AI that is equivalent to that of a human.

I'm not positive that I understand what you mean by "common denominator", but if you mean "it's possible to somehow express or explain what one language has encoded using another language", I would agree with that.

Reply on news.ycombinator.com