MACHINE TRANSLATION: SESSION INTRODUCTION 
Gary R. Martins 
Intelligent Software, Inc. 
15450 Cohasset Street 
Van Nuys, CA 91406 
Having agreed, with gratitude and delight, to 
serve as chairman of the machine-translation 
session of this Conference, it became my duty to 
comment for the record on the set of four papers 
to be presented to us. Thus have events laid on 
me an opportunity to review, in a sense, the state 
of a field which was my professional home In 
earlier and headier days. The task has given me 
both enjoyment and knowledge, but few surprises. 
More than two decades ago, scholars of the 
stature of the late Y. Bar-Hillel were stating in 
detail their perceptions of the futility of 
automated language translation by the methods then 
under development. Some years (and many "progress 
reports') later, others of us name to see that 
general translation, without the benefit of a 
robust and dominant semantics and praEmatlca, was 
not thinkable. The very slow and scattered 
development of these latter, in both principle and 
substance, is the most fundamental impediment to 
the computer manipulation of language; in those 
tiny domains where the work has been done, the 
results are Impresslve--across a variety of 
methods. 
Why isn't more of this work being done, with 
more focus? Ferhaps both the potential market and 
the (mostly military) research sponsors are awed 
by the magnitude of the challenge, and skeptical 
of the payoffs. One might envision a world in 
which, once the skepticism has been answered, such 
answers would stimulate rather than intimidate 
research. 
Meanwhile, we should be busy with other goals 
in mind, of which aids-to-translatlon is an 
obvious example. Two of the papers before us 
describe approaches to improving the human 
translation process. A third valiantly wades into 
the struggle with computational semantics. The 
fourth paper offers unsurprislng results form an 
odd experiment. I should note that all I had 
available to me were the abstracts; my cemments 
below reflect that limitation. 
The Slocum paper is more a "statement of 
work" for a paper, rather than a substantive 
draft. We expect a report on the state of the 
Siemens-Texas translation project, and a defense 
of its approach in engineering terms. The good 
news in this paper is the commercial sponsorship 
of the work; the success of such a venture mtght 
stimulate stronger investment in natural-language 
projects and studies. 
The Melby paper offers an architectural 
sketch of an open-ended translator's system which 
could include an MT system as one component. This 
modest paper deserves reading for the author's 
perspective on the professional translator. 
The paper by Nishlda and Doshlta describes 
the use of Montague-style functional notatlon to 
create a sentence-level semantic Interlingua. The 
obvious benefits in syntactic simplification are 
illustrated. An extended example, in Japanese, 
may be crucial for appreciating the scope and 
delicacy of the authors" work to date. 
The paper by Somers discusses a project to do 
machine translation on a home computer, which 
turns out to be difficult. 
