00001							March 26, 1973
00002	
00003	
00010	Dear Raj:
00020	
00030		Having just returned from six months down under, I am  in  no
00040	position  to  give  a  firm committment as to whether or not we could
00050	meet the dead-lines set in your letter to Cordell of Mar. 9th for the
00060	participants of your June 11-15 workshop, but I would like to try.
00070	
00080		What I would propose to do would  be  to  use  our  signature
00090	table  approach  to  effect  segmentation  only. This is not the most
00100	direct procedure, I realize, but I would like  to  see  how  well  it
00110	stacks up with other methods.
00120	
00130		There is one hitch in that we require  data  for  a  training
00140	sequence,   and  preferable  these  data  should  be  from  the  same
00150	population as the data to be used for the test. Would it be  possible
00160	for  us to get, say, 30 sample utterances frome the same speakers who
00170	produce the test utterances? We would use these for training and then
00180	we  would  immoblize  the  learning during the running of the actuual
00190	test.
00200	
00210		Your  letter did not define the type of segmentation that you
00220	have in mind. Do you want all linguistically significant  changes  to
00230	be  noted,  such for example as the onset of prenasalization during a
00240	vowel, etc. or are you interested in grosser effects only?
00250	
00260		We  are  already  looking  into  the  problems  that we might
00270	encounter but we do not plan to do much serious  work  until  I  hear
00280	from you or from Lee.
00290	
00300		A lot has happened while I was away and I feel that I am very
00310	much  out of touch with current developments. Perhaps I should make a
00320	trip around to some of the more active places. When would you suggest
00330	that  I  visit  your place? I will have to come east on some personal
00340	business soon in any case.
00350	
00360					Sincerely,
00370	
00380	
00390	
00400	
00410										
00420						Arthur L. Samuel