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

[David Elworthy <cbmvax!uunet!computer-lab.cambridge.ac.uk!David.Elworthy>]

(This is a message about anaphora, sparked off from some comments in recent
mailing. I've not been following this thread, so please forgive me if I'm
repeating things that have already been said, or "picking up the ball and
running in the wrong direction").


I don't know much about Lojban, and less about -gua!spi, but in a recent
messages by James Carter and John Cowan there was mention of interpreting
anaphors by copying the words of the antecedent. This is an approach which was
suggested by linguists some time ago, and which has been firmly rejected. What
is wrong with it? Consider this (famous) sentence:

(1) Every farmer who owns a donkey beats it.

The simple word copying approach gets us:

(2) Every farmer who owns a donkey beats a donkey.

which certainly does not mean the same thing: there is no requirement for the
donkey beaten by any particular farmer to be own by that same farmer. An
alternative is to take the words of the antecedent and somehow manipulate them,
as in:

(3) Every farmer who owns a donkey beat a donkey which he owns.

which is better, but still leaves a problem. It isn't clear whether (1)
commits each farmer to beating all the donkeys he owns, but in an example like

(4) Every man who loves a woman likes her.

(assuming "love" implies "like"), there does seem to be such a commitment. The
paraphrase:

(5) Every man who loves a woman likes a woman who he loves.

does not seem to have the same reading.


The copying problem can also be seen in another classic sentence, known as the
Bach-Peters paradox:

(6) The boy who deserves it will get the prize he wants.

On the simple copying approach, we have "it" = "the prize he wants", and "he"
= "The boy who deserves it". It is fairly clear that this leads to an infinite
regress if you try to expand it out.


The conclusion from this is that the interpretation of anaphors is *not* the
business of the parser, or of the syntactic component in any sense. Indeed, in
English, there are very few syntactic constraints on anaphora: number, gender,
and a restriction on reflexivity in sentences like:

(7) John talks to him. (wrong if him=John)
(8) John talks to himself. Jane talks to *himself. (him if we want Jane
talking to John).

Consequently, you have to do anaphor resolution as part of the semantic
component, and even there you can't just do it by copying parts of the
"logical form" or whatever around: it must come in at the level of
interpretation.


If you want to follow up semantic ways of doing anaphora, I can give you some
references. There are theories which are reasonably good for singular NP
anaphora, and a few which make brave attempts at plural NPs, VPs and temporal
anaphors. (But if you want to know how to *really* do anaphora, you'll have to
wait until I write my PhD thesis!.)


-- david elworthy, Natural Language Group, Cambridge Computer Laboratory