Re: anaphor means what? (was: oops! correction)

[cbmvax!uunet!math.ucla.edu!jimc]

> Date:  Mon, 29 Apr 91 11:43:05 EDT
> To:  lojban-list@snark.thyrsus.com
> From:  cbmvax!snark.thyrsus.com!cowan@uunet.UU.NET (John Cowan)
> Subject:  Re: anaphor means what? (was: oops! correction)
   
> kartr. jim. writes:
> > Do I interpret you right that the official Lojban doctrine is [already]
> > that anaphora copy words, not referents?
> 
> Only in the presence of "ra'o".

Let's see now...  (Pardon sloppy lujvo)

(1)	moi djim  
	(Jim speaking: -- Note, antecedent of "mi" is here)
(2)	.i mi zbacinpilka lemi zdani
	(I'm painting my house [assemble a skin of paint])
(3)	(Parser churns invisibly here)
(4)	.i moi djan
	(John speaking -- new antecedent of "mi")
(5)	.i dua				
	(supposed to be previous sentence anaphor, antecedent = (2))
(6)	.ije ra'o due
	(I'm sure that's the wrong anaphor, but again antecedent is supposed
	 to be (2) but under the influence of ra'o)

Sorry if I've got the wrong cmavo; I left my list home again.  Anyway, what
pops out from (5) and (6)?  Let's symbolize the referent of "la djim" by
[la djim].  In step 3, the referent from (1) would be plugged into sentence 
(2), which would become

(2r)	[la djim] zbacinpilka le [la djim] zdani

Then (5) would pick up a copy of (2r) so the English equivalent would be
"yep, that's what you're doing".  On the other hand,  (6) grabs (2) with
the original anaphora and resets them with local antecedents:

(6r)	[la djan] zbacinpilka le [la djan] zdani
	So am I.

Now the same exercise with copying words.  In step 3 the WORDS of the
antecedent (1) would be inserted into (2):

(2w)	la djim zbacinpilka le la djim ku zdani

and (5) would copy that.  Semantically, the result is virtually 
indistinguishable from the result of referent substitution (and similarly
under the influence of ra'o).  

I think people get confused when they hear "copy words" because it means to
them "copy the original anaphora then rebind them".  Clearly to make copying
work the copy source has to already have all anaphora replaced by antecedent
words, carrying with them sufficient context (tenses etc.) in word form so 
the copy will have the same referents as the original.  

Why bother?  Being oriented to actually producing results from a parser,
I see the process of converting words to referents as requiring practically
an act of God.  Thus I do not want to mess with referents until the last
possible moment, and similarly, I want the referent resolver to have to 
deal with the simplest possible input, namely selbri, and particularly not
anaphora.  If I copy words I triple or quadruple the number of sumti and
selbri I have to resolve (you've seen some -gua!spi examples with copied
tense info), but it's all repetitive and I'd rather quadruple the run time
and get the program written, than fail to finish a version that was 
supposed to be more efficient.  

Stated another way, if you do philosophical theory about anaphora 
separately from theory about referents, both jobs are easier.  True,
you probably can't do this successfully with English because it's too
tangled, but you can set up a language artifact to ease your work.

Also, I think that once you get over the "copy original anaphora" error,
students can unambiguously understand what they're supposed to do with
anaphora under the copying interpretation, whereas the idea of referents
is new to many people.  Thus the copying interpretation should be easier
to teach.

		-- jimc