Possibility to make translator software better?

The more languages a translation is apart from the original text, the more meaning is lost; it's almost a universal law. Translations directly from source to target language are always valued more, and are usually considerably more accurate, than translations by proxy.

If you're curious about translation software, the main problem at its current stage is word sense disambiguation. The problem is more apparent when translating from, say, Japanese to English, because idioms and subtle meanings of words can be very different in these two languages. However, the problem exists in all languages.

It's an ‘’AI-complete’’ problem; that is, effective artificial intelligence is necessary to completely solve the issue. AI is still a ways off, so currently there are two other approaches to translation software.

First, however, consider these two sentences — ‘’the bass part of the song is very moving’’ and ‘’I went fishing for some sea bass.’’ The two meanings of bass will have different translations, in most languages. How to differentiate between them to produce a proper translation?

A) Deep approach

Deep approach presumes access to a comprehensive body of world knowledge. Knowledge such as ‘’you can go fishing for a type of fish, but not for low frequency sounds’’ and ‘’songs have low-frequency sounds as parts, but types of fish do not’’ is used to determine in which sense the word is used.

This approach is not very successful in practice, mainly because we do not have access to such a body of knowledge, except in very limited domains.

B) Shallow apprach

Shallow approach does not try to understand the text. It just considers the surrounding words, using information like ‘’if ‘’bass’’ has words ‘’sea’’ or ‘’fishing’’ nearby, then it probably is used in the fish sense; if ‘’bass’’ has the words ‘’music’’ or ‘’song’’ nearby, then it is probably used in the music sense’’.

This approach, while theoretically not as powerful as deep approaches, gives superior results in practice; it's the approach that's used in most translation software. However, as I said earlier, the translations produced this way can be used just as an aid. Actual automated translations require AI.

Originally Posted by eriol33

Most of translator softwares dont have much trouble when it translates romance language like french, german, spanish, portugese, and italy.

Hah, mind sharing that software?

Jam it back in, in the dark.

Originally Posted by eriol33

has there been any progress to implement this on translator so far?

No, because no such AI exists. Basic AIs that are used in areas like computer games and robotics ('weak' AIs) are not enough. Proper translation is such a complex task that 'strong' AI is needed — that is, it needs to be able to reason logically, solve unexpected problems and have an immense data-base that is constantly updated; some degree of self-awareness is required.

There's nowhere I can't reach.

I'm not trying to imply anything, I'm just saying that AI hasn't been implemented in translation software because no sufficiently advanced AI exists. I'm sure it will be developed some day, but who knows when.

This thing is sticky, and I don't like it. I don't appreciate it.