Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!mit-eddie!uw-beaver!cornell!rochester!pt.cs.cmu.edu!cadre!pitt!unix.cis.pittsburgh.edu!jmast
From: jmast@unix.cis.pittsburgh.edu (John M Allen)
Newsgroups: sci.lang,comp.cog-eng,sci.psychology
Subject: Re: Regional accents (was: Spelling and Perceptual Mode)
Keywords: GB Shaw, orthography
Message-ID: <18156@unix.cis.pittsburgh.edu>
Date: 23 May 89 03:19:36 GMT
References: <2763@puff.cs.wisc.edu> <60340@yale-celray.yale.UUCP> <11726@bcsaic.UUCP> <5@minya.UUCP> <3368@tank.uchicago.edu>
Reply-To: jmast@unix.cis.pittsburgh.edu (John M Allen)
Distribution: na
Organization: Univ. of Pittsburgh, Comp & Info Sys
Lines: 87
Xref: utzoo sci.lang:4599 comp.cog-eng:1143 sci.psychology:1968

In article <3368@tank.uchicago.edu> goer@sophist.UUCP (Richard Goerwitz)
writes:
>The assumption here seems to be, e.g.:
>
>     /bar/ [ba:]
>
>because the /r/ is realized as [r] in forms like "barring."
>
[...]
>                               How do we know that "bar" and "barring"
>represent divergent hypostases of a single underlying form (in the
>Platonic sense)?  Why not have speakers simply lexicalize the relation-
>ship?

   There are several reasons that a single common base form is
prefered over several forms related by rules, but the most important
one is memory size.  Humans have a large but finite amount of memory.
If you assume that humans store a single lexical form for regualar
words and all of the spoken forms are derived by a small number of
rules then you have a fairly small memory load.  On the other hand if
you store all possible forms that a word might have the memory load is
dramatically increased because each morpheme may have several
different morphophonemic shapes.

   Consider the following data from Russian verbs.  According to one
analysis, there are several rules that operate on verbs.
	1)  vocalization.  A nasal between two consonants will change
to "a".
		z^m + t' ==> z^at'
		z^m + u  ==> z^mu

	2)  sonorant deletion.  A sonorant, [n,m,j,v], at the end of a
root is deleted when the suffix begins with a consonant.
		znaj + t' ==> znat'
		znaj + u  ==> znaju

	3)  palatalization.  Some of the verbal endings palatalize the
preceding consonant.
		z^m + u  ==> z^mu
		z^m + om ==> z^m'om


   Now comparing the two hypotheses.

	      Base+Rules		    Lexicalization
			Cumulative			Cumulative
	Forms	  	  Total		Forms		  Total
---------------------------------------------------------------------
Rules			    3				    0

z^m	z^m		    4		z^a,z^m,z^m'	    3
  `press'
znaj	znaj		    5		zna,znaj	    5
  `know'
nac^n	nac^n		    6		nac^a,nac^n,nac^n'  8
  `learn'
z^iv	z^iv		    7		z^i,z^iv,z^iv'	   11
  `live'

As you can see with only four words, the lexical approach takes more
space than derivations from a base.  Furthermore, every Russian verb
has as least two different morphophonemic shape under this analysis,
so the difference will continue to increase as you add more verbs to
the lexicon.

   Another reason that the single form is preferred is that many
alternations are automatic.  If you give a person a new word in some
form and ask for the same word in another form, they will apply all
the appropriate rules.  This is very difficult to explain using a
lexical approach because the person could not have heard the form
before and thus could not have stored it.  The rule based approach has
no problem explaining this.

   Finally, lexicalized items tend to resist change.  If every word
had all of its forms lexicalized, then we would expect many more
exceptions to the morphophonemic rules than there are.  The rule based
approach simply changes the rule in the appropriate way.

> -Richard L. Goerwitz

\		 |  |			John Allen
 \		\ ||			allen@mercutio.lcl.cmu.edu
/ \					formerly allen@mercurio.lcl.cmu.edu
					jmast@unix.cis.pittsburgh.edu

   "The bond that links your true family is not one of blood, but of
respect and joy in each other's life.  Rarely do members of one family
grow up under the same roof."  -Richard Bach, _Illusions_