The history of machine translation
The first real research in the science of dynamic machine translation began in the late 1950s, when intercepted Russian broadcasts were partially translated "on the fly." The goal was to get a general idea of the intercep's message to determine if further, more advanced translation was required. The basic goal of Machine Translation (MT) technology hasn't changed much since then. MT provides a quick, inexpensive, convenient text translation for anyone who does not want, or cannot get, a professional translation because of availability, time, or cost.
With the advent of the Internet and globalization, MT has taken on a prominent role. A major goal of Internet entrepreneurs is the creation of global communities, and the language barrier is an obstacle to creating these global communities. MT, on a small scale for now, has successfully created language transparency within these communities through multilingual chat, e-mail translation, and Web page translation.
MT today
MT technology has matured significantly over the last 50 years. While most of the leading MT systems still take a semantic/syntactic approach, increases in processing power and inexpensive disk storage have allowed MT developers to create more complex transfer systems that devote more time to source analysis and translation synthesis rules. Early systems used pattern matching, doing little in the way of word sense disambiguation and syntactic rearrangement. Even today, some MT systems are little more than word-for-word translators, yielding highly unpredictable and often unintelligible results.
Today's MT systems run at extremely high speeds, translating a Web page from one language to another in a short amount of time. The accuracy of the translation is largely dependent upon the quality of the source content. If the page is unambiguous and grammatically sound, the quality of translation can be high. Measuring MT quality and accuracy, however, is difficult because translation quality is highly dependent on the quality of the source. Accuracy is expressed in terms of errors per hundred words instead of a percentage. Typically, in a large document, it is anticipated that about 20 errors per 100 words will occur. Word arrangement errors, for example, may or may not have an effect on the intelligibility of the translation. On the other hand, depending on the circumstances, as little as one error in a sentence can completely alter its meaning. For example, assume the translation was "the land mines were left of [should be "in"] the road, left by the army that had occupied the city." While there is only one error in 17 words (a remarkable 94% accuracy rate), the error completely changes the meaning of the sentence. |
require 
