Frederick Jelinek, a pioneer in creating the technology that allows computers to interpret human speech and translate languages, died Tuesday of a heart attack in his faculty office at the Johns Hopkins University's Homewood campus. The electrical engineering professor and Roland Park resident was 77.
In more than 40 years at IBM Research and Hopkins, Mr. Jelinek led the way in developing the statistical theory behind modern voice-recognition systems that do everything from starting up phone service to translating intelligence intercepts from Arabic to English. Essentially, he helped take an infant science that merely transcribed human speech to a sophisticated one that could interpret meaning and anticipate what the speaker would say next.
"He envisioned applying the mathematics of probability to the problem of processing speech and language," said Sanjeev Khudanpur, associate professor of electrical engineering at Hopkins. "This revolutionized the field. Fifty years ago no one thought that was possible. Today, it's the dominant paradigm."
Born to a Jewish father and Christian mother in Kladno in what is now the Czech Republic, Mr. Jelinek was barred from formal schooling by Nazi edict after he finished the second grade. Through much of World War II, he was educated in what were essentially underground academies.
William Jelinek, the professor's son, said his father's family moved to Prague after being ousted from their Kladno home by Nazi occupiers. Frederick Jelinek's father would eventually die of disease in the concentration camp at Terezin shortly after the Allied liberation.
In a 2001 speech he delivered when he received an honorary doctorate from Charles University in Prague, Frederick Jelinek said that after the war he resumed his schooling but struggled academically in the Czech equivalent of high school. "During my entire studies I never received a single A in any subject," he said.
In 1949, with her son blocked from taking the graduation exam by communist authorities, Mr. Jelinek's mother brought her family to the United States, where he completed high school in New York.
Mr. Jelinek would recall that engineering was not his original interest. "I somehow slid into my scientific profession," he told the audience in Prague.
He said his original dream was to become a lawyer and fight on behalf of the unjustly accused, but he concluded that with his heavy accent he would not be effective in court. He went into engineering, he said, when he learned it would take him only four years to earn a degree, compared with seven to become a lawyer.
Mr. Jelinek earned his bachelor's degree at the Massachusetts Institute of Technology in 1956. He stayed on at MIT to earn a master's degree in 1958 and a doctorate in 1962.
His son said Mr. Jelinek traveled in 1957 to a professional conference in what was then Czechoslovakia, where he met and fell in love with Milena Tobolova, a filmmaker and dissident against the Communist government.
For years after that, Ms. Tobolova was barred from leaving the country, her son said. But during a visit by Soviet leader Nikita Khrushchev to the United States, Mr. Jelinek's academic adviser, Jerome Wiesner, who was also a science aide to then Sen. John F. Kennedy, approached the Russian and asked him to intervene with the Czech authorities. Soon after Kennedy was elected president, Ms. Tobolova was allowed to emigrate.
"As an inaugural gift to Kennedy, the Czechs released nine dissidents and one of them was my mother," William Jelinek said. Marriage followed in February 1961.
While at MIT, Frederick Jelinek became interested in applying information theory to linguistics. After receiving his doctorate, he accepted an offer to teach and to study information theory at Cornell University.
In 1972, Mr. Jelinek accepted a summer position at IBM Research, which was just beginning to work on speech recognition. Eventually, he said, he was forced to decide between Cornell and his expanding role in IBM Research. He chose IBM, where he worked for 21 years and headed a team that sought to apply the power of supercomputers to the challenges of transcribing and translating the spoken word.
Mr. Khudanpur said that previous efforts at voice recognition and translation focused on codifying rules and applying them — an approach that was frustrated by the complexity and subtlety of language. He said Mr. Jelinek's approach was to assemble a huge database of text and to allow the computer to calculate the probabilities of words appearing in relation to other words — deriving meaning from context rather than rules.
The strategy was widely questioned at the time, but when the Defense Advanced Research Projects Agency sponsored a competition in the field in 1980, Mr. Khudanpur said, Mr. Jelinek's approach prevailed.
"By the '90s, everybody was on board," Mr. Khudanpur said.