Léon Emile Dostert

Born: May 14, 1904, Longwy, France (Meurthe-et-Moselle) Died: September 1, 1971 (aged 67)

Overview

Léon Dostert was a French-born American linguist, military officer, and interpreter who invented the simultaneous-interpretation booth system used at the Nuremberg trials – still the standard at the United Nations, European Parliament, and European Commission. He was personal interpreter to General Dwight D. Eisenhower throughout World War II, served the United Nations from 1946 to 1947, founded Georgetown University’s Institute of Languages and Linguistics in 1949, and in 1954 staged the Georgetown-IBM experiment: the first public demonstration of machine translation. His prediction of imminent fully automatic translation became the founding myth of the MT field’s first funding cycle, and its failure helped trigger the first AI winter for language processing.

Note on birth date: The Prabook World Biographical Encyclopedia lists March 14, 1904; the date May 14, 1904 appears in Wikipedia and is the more widely cited source.

Early Life

Dostert was born in Longwy, a small steel-industry town on the Belgian border, at the time of heavy German industrial influence in Lorraine. When German forces occupied Longwy in August 1914 at the opening of World War I, the ten-year-old Dostert was forced to attend German schools for four years. He proved so capable in German that on finishing elementary school he was taken from cargo-loading work and made secretary and translator for a German occupation officer.

When American troops arrived in 1918, Dostert quickly acquired English and became an informal mascot for the regiment. An American soldier, Henri St. Pierre (Occidental College class of 1921), was so impressed that he arranged for Dostert to emigrate to California in the spring of 1921.

Dostert had effectively been orphaned by the war; the emigration was his way out.

Education

Year Degree / Institution Notes
1925–1926 Occidental College, Los Angeles Entered 1925; transferred after one year
1928 B.S. in Foreign Service, Georgetown University  
1930 B.Phil., Georgetown University  
1931 M.A., Georgetown University  

He is listed as a member of the Occidental College class of 1928 (sometimes written ‘28) because he graduated from Georgetown that year, though he had transferred from Occidental.

Military Service (1942–1946)

At the outbreak of World War II Dostert joined the U.S. Army and rose from major to colonel, 1942–1946. He was deployed to North Africa as a liaison officer to French Commander-in-Chief Henri Giraud and as staff officer and interpreter for General Dwight D. Eisenhower. He served Eisenhower from North Africa through the liberation of France. When Charles de Gaulle decorated Eisenhower at the Arc de Triomphe in Paris on June 1945, it was Dostert who translated Eisenhower’s speech in real time.

He also served with the Office of Strategic Services (OSS), the wartime predecessor of the CIA, which gave him a personal connection to CIA director Allen Dulles that would later matter greatly to his Georgetown MT program.

Nuremberg: Inventing Simultaneous Interpretation (1945–1946)

Dostert’s most enduring technical invention came at the Nuremberg war crimes trials of 1945–1946. The trial faced an unprecedented problem: proceedings had to be conducted in four languages simultaneously – English, French, German, and Russian – with all participants able to follow in real time. The previous international standard was consecutive interpretation, in which a speaker would pause after each sentence for the interpreter to render it. That would have tripled the length of an already long trial.

Dostert designed, staffed, and operated a system of sound-insulated interpretation booths – four booths, one per language – in which interpreters listened to the speaker through headsets and rendered the speech continuously into their target language. Delegates heard the appropriate channel through their own headsets. Interpreters were rotated every 30 minutes to maintain accuracy under the cognitive load.

The system worked. The Nuremberg trials ran from November 1945 to October 1946. Dostert’s simultaneous-interpretation architecture was subsequently adopted by the United Nations (where he served as director of the simultaneous interpretation division, 1946–1947), the Council of Europe, the European Commission, and the European Parliament. It remains the world standard.

Georgetown University (1949–1963)

In 1949, Dostert was invited back to Georgetown to found the Institute of Languages and Linguistics, in collaboration with Father Edmund A. Walsh, founder of the School of Foreign Service. He accepted and became the Institute’s first director.

In June 1952, Dostert attended a conference at MIT where Warren Weaver’s 1949 memorandum on machine translation – proposing that translation was in principle a mathematical problem solvable by statistical and information-theoretic methods – was being discussed in the context of early computers. Dostert became convinced that machine translation was achievable in the near term and began planning a demonstration.

The Georgetown-IBM Experiment: January 7, 1954

Working with IBM’s Cuthbert Hurd (director of IBM’s Applied Science Division) and IBM mathematician Peter Sheridan, and with Georgetown linguist Paul Garvin, Dostert staged the first public demonstration of machine translation on January 7, 1954, at IBM’s World Headquarters at 590 Madison Avenue, New York, in the same ground-floor space where the IBM 701 was installed.

The setup: an operator with no knowledge of Russian sat at a keypunch. She was handed Russian sentences transliterated into the Latin alphabet and punched them onto IBM cards. The cards went into the IBM 701. Moments later, the high-speed line printer output English translations at two and a half lines per second.

The system used a vocabulary of 250 words and six grammar rules – a deliberately constrained and carefully curated demonstration vocabulary, not a general-purpose system. The six rules were mechanical operations on integer-coded dictionary entries: word-order reversal, selection between two English equivalents based on context, insertion or deletion of words without direct equivalents. There was no morphological analysis, no sentence parser, no semantic representation.

More than 60 sentences were translated in total across seven subject domains: politics, law, mathematics, chemistry, metallurgy, communications, and military affairs. The demonstration was covered on the front page of the New York Times the next day (January 8, 1954): “Russian is turned into English by a fast electronic translator.” The story ran in newspapers worldwide.

The IBM press release of January 8 quoted Dostert’s prediction:

“Five, perhaps three years hence, interlingual meaning conversion by electronic process in important functional areas of several languages may well be an accomplished fact.”

That sentence – “five, perhaps three years” – followed Dostert for the rest of his career, and has been cited as an archetypal example of premature AI optimism in every history of natural language processing since.

CIA Funding

The Georgetown MT program became, quietly, the largest machine translation research effort in the world. Historian Michael Gordin (Princeton) established from declassified records that the CIA funneled money into the program through the National Science Foundation. Of $411,000 in NSF awards to Georgetown between 1956 and 1958, $305,000 originated with the CIA. After 1958, the CIA paid Georgetown directly – approximately $1,314,869 in subsequent years (over $9.7 million in 2014 dollars). Dostert’s OSS/CIA connections – particularly his friendship with CIA director Allen Dulles – facilitated this arrangement. The CIA wanted rapid translation of Soviet scientific literature; after Sputnik in 1957, the urgency was acute.

Departure and the ALPAC Report

The CIA cut its direct Georgetown funding in 1963, and Dostert left Georgetown. He returned to his undergraduate institution, Occidental College, where he spent his last years.

In November 1966, the Automatic Language Processing Advisory Committee (ALPAC), convened by the National Research Council and chaired by Bell Labs executive John R. Pierce, published its devastating report: “Language and Machines: Computers in Translation and Linguistics.” It concluded that machine translation had not achieved useful results after more than a decade and $20 million of funding, singled out Georgetown for requiring heavy human post-editing, and recommended redirection of federal funds. U.S. government MT funding collapsed for approximately twenty years.

Dostert died in 1971, five years after the ALPAC report, before the eventual revival of MT in statistical form in the 1990s, and long before neural MT in the 2010s validated, in a roundabout way, the claim he had made too early in 1954.

Legacy

Two legacies, on entirely different timescales:

  1. The simultaneous-interpretation system designed for Nuremberg in 1945–1946 is still in daily use worldwide, unchanged in its essential architecture.
  2. The Georgetown-IBM experiment of January 7, 1954 is the public founding event of the machine translation field. Every system since – statistical, neural, large-language-model – traces its public debut to the 701 printout that ran at 590 Madison Avenue that afternoon.

Connections to Others

  • Cuthbert Hurd – Hurd co-organized the IBM side of the Georgetown-IBM experiment and was Dostert’s principal IBM contact
  • Paul Garvin – Georgetown linguist who worked with Dostert on the linguistic design of the MT demonstration
  • Warren Weaver – Weaver’s 1949 memorandum proposing machine translation as a mathematical problem was the intellectual impetus Dostert acted on
  • Allen Dulles – CIA director; Dostert’s personal connection from OSS days; conduit for CIA funding of the Georgetown MT program
  • Dwight D. Eisenhower – Dostert served as Eisenhower’s personal interpreter throughout the European theater of WWII
  • Thomas Watson Sr. – Present at the January 7, 1954 demonstration at 590 Madison Avenue, one of Watson Sr.’s last major IBM public appearances before his death in June 1956

Sources