Living in a global economy has definite advantages. One can access and disseminate information from anywhere and to anyone. Being part of this community generates issues as well, as companies consider other cultures and languages when selling products or communicating worldwide.
Overcoming the language barrier is a critical challenge. Especially when a website or document needs to be translated from an Asian to European language; and vice versa. These types of translations are particularly complex, because both the language and the character sets are completely different.
To help with the translation effort, Japan Science and Technology Corporation has created an internet-based, English-Japanese machine translation system using their Transfer-Driven Machine Translation engine.
This system originated in the early 1980s as a collaboration between Japanese government and industries, in an effort to create an application which allowed users to run their own translation system on a personal computer. It has evolved over time to incorporate new features and capabilities.
Most of these original commercial MT systems used the Syntactic Transfer Method. Syntactic Transfer is a method which first analyzes sentences of the original language based on a built-in knowledge of vocabulary and grammar, then generates a dependent structure reflecting the analysis and finally translates it into the stream of words and sentences of the target language. This method produces fairly good output because of the continuous research of vocabulary and grammar over the past 15 years. However, the Syntactic Transfer method doesn't work as well if the sentences it needs to translate are long, or the grammar isn't perfect. Further, the translations can sometimes be stilted and awkward.
To counter this problem, Advanced Telecommunications Research Institute (ATR) developed a Transfer-Driven machine translation engine (TDMT) which reduces semantic ambiguity through a combination of Syntactic Transfer and Example-Based methodology. This engine stores a large number of sample sentences from the document that it is translating, and compares them against user input. This method enables TDMT to provide more accurate and precise translations, with less ambiguity. Further, the sample is automatically incorporated once the fundamental engine is made, therefore drastically reducing the maintenance load.
Japan Science and Technology Corporation (JST) recently enhanced ATR's engine by deploying it over the Internet as a server-based application. Commercial J-E MT users needed to access English-language science and technology papers published on the internet, and the new engine makes these resources available -- for free.
Ohta Hiroko, Chief planner of this project says that Lisp was used in all elements of the E-J MT System -- from original language analysis, to conversion of the display format from the original to target language, to final generation of the target language
Further, the machine translation is a difficult problem in which the non-monotonicity and context dependence requires advanced intelligence. Lisp was the natural choice in the development of MT, given its stability and robustness when compared to other languages, and it is still considered a powerful weapon when maintenance and updates of the system are needed.
Yoshida Yoshio, Chief Architect of the system says, ``Lisp is [the language] best suited for the intelligent applications. Only Lisp can express irregular assembly of thesaurus and examples. In our J-E MT system, all procedures and all data (such as dictionaries) are expressed in Lisp.''
Although the TDMT engine runs on most Common Lisp systems such as Allegro CL, Harlequin CL and GNU CL, Allegro CL was chosen for this latest version due to its speed and advanced functionality. Yoshida compared sentence translation rates of Allegro CL and other CL systems. The Allegro CL performed 3 to 10 times faster in every case. Allegro CL's powerful socket library also made it the most suitable option for writing the server portion of the application.
|Copyright © 2019 Franz Inc., All Rights Reserved | Privacy Statement|