Language/OS - Multiplatform Resource Library

home *** CD-ROM | disk | FTP | other *** search

/ Language/OS - Multiplatform Resource Library / LANGUAGE OS.iso / icon / contrib / pcw.lha / chap1.let < prev next >

Wrap

Text File | 1991-11-24 | 17.5 KB | 374 lines

.KF:chap1.toc .KW:59 .XT:2 .XB:0 .X:10 .L:59 .M:1 L---+----1----+----2----+----3----+----4----+----5T---+---R6----+----7----+----8 .H: .H: .H: .F: .F:...$$$... .M:1 CHAPTER 1 INTRODUCTION .K:1. INTRODUCTION .M:2 The purpose of this thesis is: (1) to examine a variety of grammatical theories, and to discuss the applicability of these theories to machine translation, (2) to discuss the structure and approach of a machine translation program designed to translate a Biblical passage into Spanish, and (3) to critique the results of the machine translation of this Biblical passage. .H: .H: $$$ .H: .F: .F: The bulk of this introductory chapter will be devoted to recounting the history of machine translation from 1946 to the present. The second chapter will discuss the grammatical analysis of some selected clauses taken from the Greek text of Matthew 26. Five different analyses will be done, and in each analysis a different theoretical approach will be used. The focus of each analysis will be the relationships of words to each other within clauses. For each of the theoretical approaches there will be a discussion of the suitability of that approach for use in machine translation. 2 The third chapter will focus on how sentences relate to each other to form discourses. The discussion will be centered around a Semantic Structure Analysis of the first half of Matthew 26. This chapter will be of particular importance because the machine translation program developed in connection with this thesis relies heavily on this type of analysis. The fourth chapter will discuss the structure and approach of the machine translation program itself. The complete source code for the program is listed in Appendix G. This chapter will also critique the translation of the Biblical passage which was produced by the program. A Brief History Of Machine Translation .K: A Brief History Of Machine Translation The material for this synopsis of the history of machine translation comes primarily from three sources: Hutchins (1982 & 1986:19-39), and Nagao (1986:17-48). Studying these sources it can be seen that the dawn of machine translation can be traced to the end of the Second World War when digital computers became generally available. In March of 1947 Warren Weaver, a member of the Rockefeller Foundation, suggested that the same techniques used to break codes during the war could be used to 3 translate languages as well. Andrew D. Booth was consulted on this matter, and he concluded that the computer could be used as an automatic dictionary provided it had enough memory. In 1949 Weaver circulated a memo entitled Translation which suggested the possibility of translating natural languages by computer. This memo was the catalyst which caused machine translation research to begin in earnest. Research on machine translation began at the University of Washington in Seattle in 1949, at the University of California in Los Angeles in 1950, at Massachusetts Institute of Technology in 1951, at Georgetown University in 1952, and Harvard University in 1953. In 1951 Yehoshua Bar-Hillel produced a survey describing the state of machine translation as it stood at the end of that year. The paper raised a number of issues: (1) the feasibility of doing completely automatic machine translation, (2) the purpose of syntactic analysis, (3) the use of statistical information, (4) the possibility of a single grammatical model applicable to all languages, (5) limited vocabularies, (6) post-editing, (7) the logical basis of language, (8) a universal interlingua, and (9) transfer grammars. Bar-Hillel argued that due to the difficulties of resolving semantic ambiguities, completely automated highly accurate translations could not be 4 produced by machines at that time. Bar-Hillel went on to propose a semi-automatic translation process in which human analysts would be involved in pre-editing and post-editing the text. In June of 1952 the Rockefeller Foundation sponsored the first conference on machine translation at the Massachusetts Institute of Technology. At this conference Stuart C. Dodd of the Washington Public Opinion Laboratory presented his ideas for regularizing the morphology and syntax of English. Erwin Reifler of the University of Washington proposed that writers be encouraged to write with machine translation in mind. Leon Dostert of the Institute of Languages and Linguistics at Georgetown University introduced the concept of using a pivot language when translating into a multitude of languages. Later, Erwin Reifler suggested that English would best serve as this pivot language. Finally, Yehoshua Bar-Hillel presented a completely new type of syntax which lent itself to arithmetic and logical manipulations. It could be said that this conference produced the framework for machine translation research for years to come. After the 1952 conference a machine translation project was set up at Georgetown University under Leon Dostert in collaboration with IBM. The goal was to produce 5 a small scale working model of a Russian-to-English translating system. The program was finished by the end of 1953. It required no pre-editing, and it went beyond word- for-word translation in that it did syntactic analysis of its Russian input. However, it was limited in that it had a vocabulary of only 250 words, just six grammatical rules, and worked on only select sentences. The program was first demonstrated on January 7, 1954, and the demonstration received a good deal of publicity. More demonstrations were given throughout the year; one of which was to D. Y. Panov of the USSR Academy of Sciences. After Panov returned to the USSR, research was begun on machine translation in the Soviet Union. Research into machine translation began at Cambridge, England in 1954 under Margaret Masterman. Also in this year the Massachusetts Institute of Technology published the first issue of the journal Mechanical Translation edited by Victor Yngve. In 1955 news began to come out of the Soviet Union about several machine translation projects which were underway in Moscow and Leningrad. In the same year the Cambridge Language Research Unit held its first meeting, a machine translation institute was set up under the auspices of the Centre Nationale de la Recherche Scientifique at Grenoble 6 University in France, and the first machine translation project of Italy was started by Silvio Ceccato in Milan. Machine translation research also began in Japan in 1955 at Kyushu University centered around Professors Toshihiko Kurihara and Tsuneo Tamachi. At this time Professor Tamachi received a Science Research Grant from the Ministry of Education to build a translation computer called the KT-1. It was intended to translate between Japanese, English, and German. It could be said that 1955 marked the beginning of a period of enthusiastic worldwide machine translation research. About this time the United States Government began to provide grants for research on machine translation to universities and research institutes throughout the country. By 1957, the year that Russia launched its Sputnik satellite, an impressive amount of research had been completed in the United States. Sputnik was a considerable shock to the American people, and because of this it was considered essential that Russian-to-English translation be automated as quickly as possible. In the meantime the Soviets were aggressively pursuing their own machine translation programs, especially in the Soviet Academy of Sciences, so that by 1960 more researchers in 7 the Soviet Union were engaged in machine translation research than in the United States. A second machine translation project was begun in Japan in 1957 at the Electrotechnical Laboratory of the Industrial Institute of the Ministry of International Trade and Industry. This project was placed under the supervision of Hiroshi Wada, and encompassed optical character recognition as well as machine translation. The system, informally known as 'Yamato', was completed in 1959. Its capabilities were somewhat rudimentary, but it was the first English-to-Japanese machine translation program in the world. Another significant event which occurred in 1957 was the publication by Noam Chomsky of Syntactic Structures, the first of his books on Transformational Grammar. This development was significant for two reasons. First, Chomsky's grammar was one of the first which lent itself to manipulation by means of computers since transformations were governed by rules which could be implemented as computer programs. Second, Chomsky's grammar posited a 'deep structure' which represented the semantic component of language. Until this time researchers in the field of machine translation had studiously avoided the issue of 8 semantics, preferring to limit their investigations to surface structure. In 1960 the Special Committee on Mechanical Translation (part of the Committee on Science and Astronautics) submitted to the US House of Representatives a report entitled Research on Mechanical Translation. It gave a synopsis of the history of machine translation up to that point as well as an analysis of the state of machine translation in 1960. Among its many conclusions and recommendations was that translation is heavily dependent on human understanding--a semantic issue. The Japanese were the first to confront the semantic issue head on. This was necessitated by the fact that the syntax of Japanese is radically different from that of the European languages. While researchers working with European languages could produce acceptable machine translations between those languages while confining their analyses to surface structure, successful translations between the European languages and Japanese could not be produced in this manner. In 1964 Makoto Nagao proposed a "method for investigating the usefulness of 'meaning tables' or 'semantic networks', which have been widely employed ever since." (Nagao 1989:24) Nagao showed that 9 through the use of semantic networks the relationships between the nouns in a sentence could be established. .H: .H: $$$ .H: Another significant event of 1960 was the publication of Yehoshua Bar-Hillel's paper titled A Demonstration of the Nonfeasibility of fully Automatic High Quality Translation which stated emphatically that high quality, fully automatic machine translation was impos sible. Then in April of 1964 the US Academy of Sciences formed the Automatic Language Processing Advisory Committee (ALPAC). The committee was commissioned to report on the present state of affairs in machine translation, and to give its prognosis for the future. On August 20, 1965 the committee published Language and Machines: Computers in Translation and Linguistics, better known as the ALPAC report. The committee had concluded that machine trans lation was more costly than human translation, and produced inferior results. They saw no hope for improvement in the near future. The report was widely criticized as biased, inaccurate, and misleading, but the damage was done. Soon after the report's publication funding for research in machine translation all but dried up. Researchers in the field became objects of social stigma. Research virtually stopped. As a consequence, much of what had been learned in the previous ten years was simply lost. 10 The period following the ALPAC report through the late 1970's could be called the Dark Ages of machine translation. However, not all machine translation efforts came to a halt. At the time of the ALPAC report there were two working Russian-to-English machine translation systems in existence. One was IBM's Mark II system, and the other was the Georgetown University system. The Georgetown system was moved in 1965 to the Rome Air Defense Center in Rome, New York where it has been in continuous use for more than two decades. Another surviving project was the TAUM/METEO English-to-French weather forecast translator. This system was developed at the University of Montreal, and is still in use today. Peter Toma, one of the pioneers of the Georgetown program, established a private company which developed its own machine translation system which imitated the Georgetown model. Called SYSTRAN, it was completed in 1969, and installed at Wright-Patterson Air Force Base in Dayton, Ohio. Originally another Russian-to-English system, it was later expanded to include Arabic, Dutch, French, German, Italian, Japanese, Portuguese, and Spanish. Today SYSTRAN is used at approximately twenty organizations in the US, Europe, and Japan (Homer 1987:387-388). 11 In 1970 Brigham Young University established its Translation Sciences Institute with Eldon G. Lytle as its leader. The purpose of the Translation Sciences Institute was to develop an interactive machine translation system called ITS (Interactive Translation System). ITS was to work with English, Chinese, French, German, Portuguese, and Spanish. This system had two distinctives: First, ITS was developed using a totally new grammatical model called junction grammar. Second, ITS was intended only for trans lations of a single source text into multiple target languages. Because its accuracy was obtained by heavy interaction with translators during the analysis of the source text, the process was not efficient for translations into a single language. ITS was successfully completed, but was abandoned in 1979 because it was thought to be too elaborate. Interactive analysis of the source text was time consuming, and translators resented having to answer the system's trivial questions (Hutchins 1986:299-301). Another system which was developed during the so- called Dark Ages of machine translation was named LOGOS. This system, developed during the 1970s, was used to translate from English-to-Vietnamese during the Vietnam War. Little research was carried out in Europe during this period except in Grenoble, France at the Automatic Trans lation Research Institute (CETA) in association with the 12 National Center for Scientific Research (CNRS). Work continues there on Russian-to-French and French-to-Russian to this day. More recently, the institute has also been developing a French-to-Malayan translator. The Language Research Center at the University of Texas at Austin began developing its German-to-English translation system, METAL, in 1970. Despite a two year funding gap in 1977 and 1978, the project has continued up to the present with additional work going on in Dutch, French, and Spanish (Bennett 1987:382). All told, there were thirteen new major machine translation projects started during the difficult period between 1967 and 1975 (Hutchins: 1986:336-337). More recently, in 1976 the European Economic Community embarked on a machine translation project of unprecedented proportions. Originally intended to provide for translation among the four official EEC languages (Dutch, French, German, and Italian), it was later expanded to include five more (Danish, English, Greek, Portuguese, and Spanish). This means that there are nine EEC languages which must be translated to and from eight others, giving a total of seventy-two language pairs to be translated. The researchers thought at first that the translation needs of the EEC might be met by the SYSTRAN system, but they 13 abandoned this approach in 1978 to develop their own machine translation system called EUROTRA (van Eynde and ten Hacken 1989:5). The project is currently in the prototype stage (van Eynde and ten Hacken 1989:38). The Japanese government began to officially promote research in machine translation in 1982. Between 1984 and 1985 more than ten Japanese companies put machine trans lation systems on the market (Nagao 1989:43-45). One particularly interesting system is the mu machine trans lation project which began in 1982 and was completed in 1986. This Japanese-to-English system is noteworthy because of its heavy reliance on semantic and pragmatic information during analysis of the source language text (Nagao 1987:262-277). Researchers in Japan have also begun work on an interpreting telephone system (Nagao 1989:3). Current working MT systems in the United States include the Automatic Language Processing System (ALPS) which was derived from Brigham Young University's ITS system, and a keen competitor the Weidner system of Weidner Communications Corporation. Both are machine-aided trans lation systems. The ALPS system works with English, French, German, Italian, and Spanish. The Weidner system includes all of these plus Arabic, Japanese, and Portuguese.