To celebrate Independence Day and Bastille Day, we here in the KantanMT blogging workshop thought that we would use this opportunity to pay homage to the early contributions made by both American and French pioneers to the development of Machine Translation. In this first post, we are going to focus on America and one of the most important developments in the history of Machine Translation: The Georgetown-IBM Experiment. Background to the Experiment… Funnily enough, it all began with the Frenchman Léon Dostert who was Director of Georgetown University’s Institute of Languages and Linguistics. Dostert had previously worked as an interpreter for Eisenhower and liaison officer with Charles De Gaulle. Dostert also developed the translation system for the Nuremburg Trials. After attending the first ever conference on Machine Translation in 1952, an inspired Dostert decided to check out the feasibility of this new technology in a practical experiment. Dostert contacted the founder of IBM Thomas J. Watson, who agreed to support Dostert’s work. They established a team of both IT and linguistic specialists and the experiment was ready to begin. The Experiment… 12 machines, collectively known as the IBM type 701 electronic data processor, would translate 250 lexical items with six rules. The source language was Russian and the target was English. Why? Well, Russia was the biggest military threat to the US at the time, a machine that could translate Russian content to English would help the US to keep tabs on the Soviets. Watson said “I see this as an instrument that will be helpful in working out the problems (of world peace), we must do everything possible to get the people of the world to understand each other as quickly as possible”. Most of the sentences that were translated related to organic chemistry to show different uses of nouns and verbs. W John Hutchins, in his report The Georgetown Experiment-Demonstrated in January 1954, gives some examples:
- They prepare TNT
- They prepare TNT out of coal
- TNT is prepared out of coal
Associate Professor at the Institute Paul Garvin said that one of the major shortcomings of the experiment was that it was so limited – remember the experiment only consisted of 250 lexical items and six rules. But he defended its relevance, saying that the engine did have to make selection and arrangement decisions while translating the content.