When the Soviets successfully placed
"Sputnik," the first satellite, into earth
orbit in 1957, it took America by surprise. Behind the scenes, the
news was more distressing because technical details of the satellite
had actually appeared months before its launch in a Soviet hobby magazine.
However, it had gone unnoticed because American intelligence did not
have the means to quickly translate Russian to English.
While not nearly as high profile or
sexy as the race to the moon in the 60s, the government
also set out to achieve automatic translation of Russian into English,
dumping millions into academic and industrial MT research. Agencies,
such as the CIA, relished the thought of having immediate access to
thousands of Soviet papers and publications, with an eye on the huge
advantage it would give them in counter-intelligence.
Unfortunately, almost a decade and
20 million dollars later, results were not meeting
these overly optimistic expectations. In 1966, the Automated Language
Processing Advisory Committee (ALPAC) issued a highly critical report,
citing the lack of significant progress, which effectively halted
government spending on MT.
Star Trek - Universal TranslatorIn part, one can attribute the
apparent failure of MT research to the unrealistic
expectations set in this field's early days. The public's first general
introduction to the concept of MT came from the classic 1960s TV series
Star Trek, where the crew of the starship Enterprise used a device
called the "Universal Translator" to communicate with alien
races across the galaxy.
With little more than a few snippets
of dialogue from a newly encountered race of sentient
beings, the Universal Translator deduced the meaning of their languages
entire lexicon and flawlessly, in near real time, translated speech.
In retrospect, not only was this unrealistic for the times, but a
downright impossible goal.
Fully automatic, high quality
text-to-text Machine Translation across vastly different
knowledge domains is challenging. However, throw in a scarcity of
training data and speech-enabled front and back ends, and the ideal
symbolized by the Universal Translator becomes unachievable even with
today's best technology. Due to overly optimistic expectations and
a subsequent collapse of government funding, research into MT survived
in only a few institutions that could afford going it alone, such
as IBM and strangely enough -- the Mormon Church.
The Mormon Connection
In the late 1970s, the Church of
Latter Day Saints undertook a massive MT project in
hopes of making it relatively
|| easy to translate their
religious literature into different languages. A key figure in that
effort was Steve Richardson, first an undergraduate and then graduate
student at Brigham Young University in Provo, Utah, who used his computer
science and linguistic education to further the Mormon MT effort.
Upon completion of his bachelor's degree in 1977,
Richardson worked full-time for the Mormon MT project until he completed
his master's degree in 1980. At that point, after five years, they
canceled the undertaking. Although not successful at producing cost-effective
MT because of the high cost of computing power on the
IBM mainframes, the project inspired a number of MT start-ups in the
Utah area, the descendents of which continue in operation today. With
a growing family to support, Richardson took a job as an associate
programmer with an IBM product group in Endicott, New York.
The IBM Connection
In 1983, Richardson contacted a group at IBM's
T. J. Watson Research Center, dedicated to pushing
the limits of natural language processing. Richardson met George Heidorn
and Karen Jensen on an incredibly snowy day in mid-February. "I
remember my first meeting with Karen and George clearly, on February
11, 1983, because it was snowing so hard that the Watson Research
Center had to close," says Richardson.
Heidorn was the manager of the Natural Language
Processing group at IBM Watson and Jensen, a leading
authority in English grammar, his close colleague. Heidorn, Jensen,
and Richardson formed a powerful trio of talent that weathered many
technical and corporate storms to eventually build their shared dream
- one of the largest and most successful Natural Language Processing
(NLP) Projects in the world.
In the 1980s, government and industry funding was
again flowing for MT research and development. The
launch of the Japanese Fifth Generation Project, aimed at building
an intelligent computer within 10 years, was the equivalent of another
"Sputnik-scare,"spurring the U.S. government to again open
its purse strings for MT research. This funding launched large-scale
efforts, such as CYC (short for encyclopedia), to create software
capable of understanding natural language via common sense reasoning.
Realizing the need to demonstrate the practical
application of their NLP Project, the IBM trio transferred
to Big Blue's development side, with hopes of including a grammar
checker in a software suite tentatively dubbed "OfficeVision,"
intended to compete directly with Microsoft's highly successful Office.