The Beginnings at Microsoft Research
However, IBM OfficeVision effort buckled under the company's stifling
bureaucracy, and its collapse threatened to end Heidorn, Jensen, and
Richardson's NLP Project. In a bold move, the trio jumped ship and
moved the project to Microsoft's fledgling research lab in 1991, becoming
the first researchers to join Microsoft Research (MSR) and the first
to leave the prestigious halls of IBM Watson for the relatively small
software company, mainly known for its DOS program.
To add insult to injury, this move took place at time when IBM and
Microsoft were going through a very public "divorce" over
the OS/2 personal computer operating system. Microsoft had decided
to abandon the joint OS/2 development effort in favor a Windows-only
strategy. Microsoft's decision was widely viewed within IBM as nothing
less than base treachery, and losing three top researchers to Microsoft
during this period made matters all the more infuriating.
Hence, Heidorn, Jensen, and Richardson's transition to MSR was fraught
with nasty episodes, such as IBM locking them out of their offices.
The rights to the group's intellectual property were a real point
of contention between the two companies. Fortunately for the trio,
they had placed most of the work on the NLP Project in the public
domain in various scientific journals and publications. Nonetheless,
the trio had to coauthor a book to document this fact before IBM relinquished
its claims to the concepts.
Once at MSR, the trio immediately began recruiting
theoretical and computational linguists, starting with the team
|
|
members they had originally assembled
at IBM. Bill Dolan, Lucy Vanderwende, and Joseph Pentheroudakis, the
first three recruits, were deemed critical in moving the NLP Project
along - and they possessed sufficient computer skills to pass muster
with Microsoft's development managers. Dolan and Vanderwende had worked
with the technology at IBM, while Pentheroudakis had been at Executive
Communications Systems (ECS), a leading developer of natural language
software.
The NLPWin System
The NLP Group at MSR, growing steadily in number, strove to make
their NLP Project's conceptual modules a reality in the NLPWin system.
This systems conceptual framework consists of the components shown
in Figure 1, with each module in this series designed to successively
abstract the structure and meaning of the words and sentences within
the language.
The NLP Group's bottom up approach to natural
language processing contrasts sharply with the approach typically
taken by researchers in the Artificial Intelligence (AI) field. AI
scientists primarily focus on creating a machine that reasons similar
to a human and on reach this ability to thinking, assume it is a relatively
trivial task to generate a natural language dialogue with humans.
However, after forty plus years of AI research, this top down approach
has shown little success. The mapping between the abstract concepts
found in machine reasoning and the highly rich, complex nature of
natural language is much more difficult than first imagined. |
