00100	I would like to question the continueing validity of the program for
00200	the ARPA Speech Understanding Research activity as outlined in the report
00300	prepared something over two years ago.
00400	
00410	This does not mean that the original report was in error, quite the
00455	contrary, it simply means that much has happened in two years.
00500	The progress that the different ARPA groups have made, the increased
00600	level of activity in the general field of Language Understanding,
00700	the entry of several profit making organizations into the field
00800	of Speech Understanding, Xerox, IBM and BTL to name but three,- all
00900	of these factors have greatly changed the situation. As a matter of
01000	fact one can make the point that the ARPA project has
01100	succeeded beyond the committee's reasonable expectations and
01200	that the work has demonstrated that speech understanding for a very
01300	limited range of discourse and for a narrowly restricted range of
01400	speakers can be achieved in the estimated time span.
01450	
01500	With this demonstration as a fait accompli one can then raise a
01600	question as to the proper course to follow from now on.
01650	
01700	It is my belief that ARPA should effectively abandon the study of
01800	Speech Understanding as applied to the problems associated with
01900	rigidly limited fields of discourse and undertake the much harder
02000	and longer range study of Speech Understanding of general English,
02100	perhaps not completely unrestricted English but certainly a rich
02200	enough subset so that one could address the system much as one would
02300	speak to an eight grade student. The work should also have as its
02400	objective that of responding to a reasonable group of untrained
02500	speakers, either with out a training session or at most a session
02600	limited to the requested repeating of one or two sentences.
02700	
02800	By opening up the objectives in this way, several potential
02900	problems would be averted. In the first place it should then be
03000	possible to assign work to the different projects or to as many of
03100	them as can be supported on the basis of their specific
03200	competences with a minimum of unnecessary duplication. The fact
03300	that the immediacy of the commercial application had been postponed
03400	should do much to delay the introduction of organizational rivalry.
03500	The various groups have cooperated to an amazingly degree in the
03600	past but this could not be expected to continue for much longer
03700	as soon as it became obviouus that a commercial product was in the
03800	offing.
03900	
04000	Problems as to the extendability of the work would be answered
04100	by forcing the consideration of the extended problem at once.
04200	Greater reliance would have to be placed both on tha development
04300	of better acoustic recognition and at the same time on tieing
04400	the work closer to the general problem of language understanding.
04500	
04600	This would solve the problem of how ARPA could support several viable 
04700	projects without requiring them each individually to be above the critical
04800	size required to man an entire system. There would still be problems
04900	as to critical size but they would be greatly reduced. Greater
05000	reliance would probably have to be placed on the ARPA net, but this
05100	might have benificial side effects.
05150	
05200	Finally, the expansion of the objectives would make a place for the
05300	smaller projects which have been able to help out in the past but
05400	whose role would largely disappear as an operational system of the
05500	very limited sort now anticipated neared completion.