First Machine Translation Demo: Computers Speak Russian

A Russian sentence about chemistry appeared on a screen in English, and the room full of reporters at IBM''s New York headquarters understood they were watching the future arrive. On January 7, 1954, Leon Dostert of Georgetown University and IBM researcher Peter Sheridan demonstrated the first public machine translation system, converting sixty Russian sentences into English using the IBM 701 computer, the most powerful machine IBM had built. The Georgetown-IBM experiment was deliberately limited in scope. The system used a vocabulary of only 250 words and six grammar rules, translating sentences exclusively about organic chemistry and general science. The translations were crude but comprehensible. "Mi pyeryedayem mislyi posryedstvom ryechyi" became "We transmit thoughts by means of speech." The demonstration was designed to show that machine translation was possible in principle, not that it was ready for deployment. The press coverage was breathlessly optimistic. Headlines predicted that machine translation would be perfected within three to five years, that computers would soon render human translators obsolete, and that the language barrier between nations would dissolve. The Georgetown researchers privately knew better. Even their carefully selected demonstration sentences required extensive pre-processing, and the system could not handle ambiguity, idiom, or context. The optimism triggered a massive influx of government funding, particularly from the CIA and military intelligence, which were drowning in untranslated Soviet scientific literature. But progress stalled. Natural language proved far more complex than anyone anticipated. In 1966, the Automatic Language Processing Advisory Committee published a devastating report concluding that machine translation was slower, more expensive, and less accurate than human translation, and recommended cutting federal funding. The field entered what became known as an "AI winter" that lasted nearly two decades. Practical machine translation did not arrive until the 2000s with statistical methods and neural networks. Google Translate launched in 2006. The technology that Dostert demonstrated in 1954 took fifty years to become useful and seventy to become good.