Possibility to make translator software better?
 

I wanted to ask this question a long-long time ago, but I keep pending it because my question sounds so stupid.

Well, I actually want to ask the possibility to make translator software better. So far, most of translator softwares are based on english input. Most of translator softwares dont have much trouble when it translates romance language like french, german, spanish, portugese, and italy. But it tells different story if we want to translate english into non romance language such as chinese, japanese, and korean or vice verca. usually the translation is bizarre and confusing.

So I'm thinking the idea of step by step translator if it's might be possible. As we know, language has its own branch and probably by doing translation step by step to another language which is close to the targeted language before going to the targeted language itself, we might get more accurate result.

maybe to illustrate my idea:
English -> japanese (the usual translation software)
English -> language a -> language b -> language c -> japanese (step by step translation, the point is to translate english into another language which have traits of similarities to japanese before it's translated to japanese)

but I dunno, is this a good or bad idea? is it even possible? If you are expert in language or in this field, hope you dont mind to share your opinion about this.

The best approach is the direct one, because of the possible degradation of the original message that can happen in each translation operation.

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.

Hah, mind sharing that software?

Hope you aren't talking about Google's translator wannabe, heh.

oh I'm using software called Easy Translator Deluxe 4.0 any of you familiar with that software? It's quite helpful when translating from/to romance languages.

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

Linguistics is not an easy subject, plain and simple. Human languages are nearly impossible to represent in terms of nice logical series and sequences, and converting between them is incredibly difficult to do in computers. There have been hundreds of doctorate-level papers written on language translation via computer and many different efforts. The flaws in current implementations are not the fault of lack of R&D, I can assure you.

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.

Hope u know some lisp or something.

So Aardark are you implying that it's currently impossible/impractical to develop such AI or simply that no one has bothered to do it yet?

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.