~s -° 
Computerized Linguistics ,A~commentary 
by 
Martin Minow, KVAL Group. Stockholm 
In reading through the various preprints, I was struck by the fact that 
many were concerned with specific computer implementations of one kind 
or another. Unfortunately, there seems to be little discussion of the program- 
ing from the programmer's viewpoint. This is under standable in a congress 
of computational linguistics but, I feel, somewhat regrettable. Especially 
nce the reports of what has already been done surely will stimulate other 
rticipants to use computers. I think that an informal session on program- 
ming in computational linguistics would be both interesting and valuable, 
allowing an exchange of ideas and experiences which would be much more 
difficult in the even more informal atmosphere of the dining room. 
In order to initiate such a session, I am enclosing the following questions, 
hoping that they will stimulate a certain informal correspondence before the 
congress begins. If enough interest is shown in this idea to warrant a session, 
I intend to edit the replies received (to the extent they show similarities), 
making summaries available to participants before the conference. 
In answering the questions, it is of course not necessary to confine your- 
self to the particular project you are discussing at the conference if another 
project will yield a more interesting reply. This is not, in other words, a 
final examination. 
~ at type of project? 
IR 
Modeling of linguistic system 
Data- pr oc e s sing 
Something else or more than one area 
One-time run or continuous production 
~mt language (machine, etc) was used? (Please note that when I say language 
I really mean language ÷ haxdware ÷ system.) 
Was batch processing, time sharing, or conversation used? 
Why this choice? Were other languages available? Suitable? 
How did the choice of language influence the definition of the problem? the pro- 
g r axnming ? 
About how much time (thinking, physical, machine) was required for - and how 
did the choice of language influence - the following: 
Statement of ~oblem (flo'." ~J~rting, definition of algorithm, pure linguistic 
considerations) 
-2- 
Defining data structures (lists, how character strings are stored) 
"System programming" (character manipulation subroutines, for example; 
even definition of input-output formats, external storage, etc.) 
Programming the algorithm 
Initial and final debugging (i. e. correcting the program versus correcting 
the algorithm) @ 
Documentation 
How large is the program in source form? What percent commentary? 
If you used time sharing, did it help? 
How did the software "bureaucracy" (for example, Job Control Language on th~ 
IBIVt 360) affect the programming for better or worse? 
Can the program be easily implemented on a different machine or system? 
(Was this one of the objectives?) If this was tried, how painful was the experience 
of transferring the program? 
If you were to start all over again, would you use the same language? 
Have you been teaching programming to linguistics students? If so, what has been 
their experience? Yours? Should linguists learn programming? Why? 
KVAL-is beginning to plan a fairly large computer system for translation 
from Swedish to "condensed" Braille. This requires a large dictionary, fairly 
sophisticated character manipulation routines, and the ability to perform purely 
linguistic operations on the input text (to recognize syllables, morphemes, etc). 
Additionally, the input-output problems are not at all trivial. We also want to 
produce a systemwhich can be moved easily - both to different computers within 
Sweden, and even, if possible, to other languages, The program should also b O 
self-documenting and permit easy modification in a modular fashion as new words, 
new syntactic transformations, etc prove useful. It would also be nice if the 
"linguistic work" could be written by linguists with only a little programming 
training, rather than the other way around. We wouldn't even mind if it was 
cheap and fast. g 
Suggestions please: What language, machine, type, etc? How should the dictionary 
be stored (assume any random-accesss device you are familiar wit'h)? Discuss'the 
goals, ranking them in t~,~ light of your own experience. 
- 3- 
There is much to criticize in the choice and form of the above questions. 
The naivety is obvious - to a certain extent it is due to my own inexperience in 
this field (KVAL" s problem has surely been faced - if not solved - by every IR 
project, for example). To a greater extent, however, the naivety is present to 
prevent the questions from containing their own answers. By all means, criticize 
the questions. If you do not feel they are pertinent to your own experience, or to 
the problem itself, ask your own, and answer them. 
Please inform me if you would be interested in taking part in a panel dis- 
cussion during the conference. 
