[lbo-talk] "Machine translation" (was: Protection of Chinese language urged)

Jon Johanning jjohanning at igc.org
Sun Jun 6 09:55:10 PDT 2004


On Saturday, June 5, 2004, at 04:34 PM, dano wrote:


> Machine translation - getting better, still expensive.
> about: <http://www.languageweaver.com/how_it_works.php>
> Chinese:
> <http://www.languageweaver.com/language-translation-for-chinese.php>
>
> I have not seen their Chinese translation (subject of the thread) but
> have seen them translate Arabic (Al Jazeera) on the fly with expert
> Arabic speakers who confirmed that the translation is high quality.
> The Chinese may be as good or it may not.

This web site, of course, like many other similar ones, is an advertisement for a commercial company, which puts the best possible face on the products it has for sale.

The quality of "machine translation" software is difficult to evaluate in a simple manner, since it depends on the purposes the user of the software has in mind, and what the user is willing to accept. The statistical method used by this company's software is one of the more adaptable kinds of MT systems, but no current MT system can replace human translators for all purposes, and it is very doubtful that MT systems ever will, though this is a hotly debated question in the translation and computational linguistics communities. Basically, in my opinion, as well as that of many other translators, MT will only be "high-quality" when robots which can essentially think and operate in the world like human beings (e.g., Commander Data on Star Trek) are made, which obviously will take a while.

Where communications of a political nature are concerned, especially, it is very unlikely that "machine-translated" texts will be considered satisfactory. The most satisfactory employment of MT currently is in fields where the vocabulary is limited, the source texts are frequently written on purpose in a very simplified and controlled manner (in syntax, etc.) to facilitate the MT processing, and the resulting texts in the target language are extensively edited to make them comprehensible. Thus, MT is normally used where very large volumes of very similar texts, which can be effectively analyzed statistically, must be churned out: technical manuals, for example. Note that Language Weaver states, concerning its Chinese module:

"Especially in cases of domain specific text where there is repetition in the text and a limited universe of phrasing, such as technical documentation handbooks, Language Weaver’s SMTS technology can save customers considerable money and time through automation of the translation process, as well as enable cost effective, real-time translations of previously untranslated text, like Web based customer and product support information."

These are very typical MT applications, and very different from political contexts, in which the terms used vary rather wildly in meaning, disagreements over their meanings are very frequent, and misunderstandings consequently common (see for example this list, which is conducted in only one language). The satisfactoriness of MT rises in proportion as the language becomes more stereotyped and rigid, which is not conducive to good political activity.

Jon Johanning // jjohanning at igc.org __________________________________ When I was a little boy, I had but a little wit, 'Tis a long time ago, and I have no more yet; Nor ever ever shall, until that I die, For the longer I live the more fool am I. -- Wit and Mirth, an Antidote against Melancholy (1684)



More information about the lbo-talk mailing list