The challenge to digital translation from the beginning has always been Fully Automatic, High-Quality Translation (FAHQT). This challenge was laid down to digital translation--or mechanical translation as it was called in those days, and then later, machine translation--by the Israeli logician Bar Hillel in the 1960's.  It was partly because of the failure of early systems, after ten years of trying, to attain anything approaching FAHQT that the field was all but abandoned for a time (in the late 60's, the 70's, and early 80's).   And when interest finally re-awakened--largely because of the changing economic scene--it was not FAHQT that developers were after so much as simply a useful translation tool for leveraging the work of human translators, to make them more productive. 

In the past decade, there have been scores of successful installations where digital translation has been used cost-effectively, at very substantial savings to its users.  Nevertheless, the fact remains that, thus far, translation by computer still accounts for a negligible fraction of the total translation effort.  This is so because traditional methods, aided perhaps by the addition of translation memory, still seem able to satisfy the requirements of most multinational corporations.  This situation may be changing, as resources for manual translation are becoming increasingly strained and overloaded, but even so, any large-scale shift to digital translation will likely come about only very gradually.

But now a new situation is arising.  Because of the Internet, a burgeoning need for reliable (good-quality), low-cost, near instantaneous translation has emerged, one that traditional methods can never hope to satisfy.  Digital translation thus has been handed an unprecedented opportunity to contribute, not to the productivity of translators, but to the informational needs of a wide spectrum of modern society.  A vast opportunity indeed, but one it can only hope to satisfy if the quality of the computer's output is up to the job.  

Before discussing the prospects, there is an important caveat.

*    *   *   *   *

The first law of digital transition is that the translation output can only be as good as the input text.  

This rule applies no matter how powerful the system.  It is not likely that a digital translation system can ever improve upon the quality of writing of the source document.  Yes, it can flag sentences that are apt to cause problems for the system, but, unlike a human translator, it do little in the way of improving  those sentences. 

When the input document is well written, the quality of digital translation often surprises people as being human-like.  Near human-quality translation by a high-end digital translation system can indeed occur when the following three conditions are present:

	The input document is discursive in nature.  That is, it's purpose is to convey information rather than to edify.  In brief, this refers to documents where the writing style is of secondary importance. (See Written-Language Spectrum)
	The input document is written in a simple, straightforward (non-idiosyncratic) style. Sentences are not too long or complex, highly idiomatic expressions are kept to a minimum.
	The knowledge base of the digital translation system has been updated to cover the terminology of the text.  In brief, there are no missing words or noun phrases (where the translation of the noun phrase is more than the sum of its parts).

It is a fairly straightforward task to realize these conditions in a corporate environment whose documentation is characterized by good writing for specific domains.  Technical and training manuals are among the most ideal candidates for good quality digital translation, as many corporate users have abundantly experienced.

It is rather another matter to try to apply digital translation to a wide spectrum of documentation where the type of writing may vary from very formal to very informal, the quality from good to poor, and the domains may range all over the board.  This is the ultimate challenge developers of digital translation system like the Logos System must meet.

If the challenge has not yet been fully met, what are the chances?

 

Let's Get Specific

Below are some examples of the kinds of things a high-end digital translation system like Logos is presently capable of.  Following this is a discussion of the kinds of things that still pose problems, along with some suggestions of how the Logos System plans to solve them.

 

Handling Long, Complex Sentences