🚀 go-pugleaf

           Department of Computer Science, University of Toronto
              (GB = Galbraith Building, 35 St. George Street)

       -------------------------------------------------------------

                                    AI
                  GB221, at 11:00 a.m., 10 December 1992

                              Peter F. Brown
                   IBM Thomas J. Watson Research Center

             " A Statistical Approach to Machine Translation "

 In 1949, Warren Weaver proposed that statistical techniques from the
 emerging field of information theory might make it possible to use
 modern digital computers to translate text from one natural language
 to another automatically.  Although Weaver's scheme foundered on the
 rocky reality of the limited computer resources of the day, we feel
 that the increase in computer power over the last forty years makes
 reasonable a new look at the applicability of statistical techniques
 to translation.

 We have developed a rudimentary model of the translation process and
 have used it to construct a glossary of English words and French
 phrases.  We assume that each English word is either deleted during
 translation, or used to produce one or more French words.  We further
 assume that if French words are produced for an English word, then
 each is produced independently of the rest and depends only on the
 English word itself.  Finally, all of the French words are shuffled
 around in the sentence in a way that depends on the position of the
 English words in the English sentence.

 We have also implemented a procedure for searching for the most
 probable English translation of a French sentence and have used this
 procedure to attain preliminary translation results.