SYSTRAN in daily use
Around 1970, by its developer's account, Peter Toma's SYSTRAN, developed with US Air Force sponsorship, became an operational Russian-to-English translation system, and it was still one in March 1976, when Toma presented it at a seminar of the Foreign Broadcast Information Service. By his account the system, on an IBM 360 or 370, processed 300,000 words per hour of CPU time.
Why it matters
A few years after the ALPAC report, which saw no practical future for machine translation, a fully automatic system was translating government text by the millions of words, much of it sent out unedited. This is an editorial judgement resting on the vendor's own presentation.
The source is the talk of Latsec Inc.'s president Peter Toma, that is, the vendor's own account; no report by a customer (the Foreign Technology Division or the European Commission) was found. By the talk, FTD's system monitor Robert Wallace said at a bidders' conference in September 1975 that nearly half of the 15 million words translated had been distributed without post-editing; NASA used raw output for working papers of the Apollo-Soyuz project. Pairs in 1976: Russian to English, English to Russian, English to French, and weaker German to English and Chinese to English. Air Force sponsorship dates from July 1967 and Latsec was formed in 1968. The English-French system was 'recently installed' for the Commission of the European Communities; the talk gives no month and no contract date. The record does not claim pages per year, a dictionary size or the dates of pairs added in 1978-1979: a document of 1976 does not have them.