parsing

[Matthew S Dryer]

I second Susanna Cumming's surprise that this parsing discussion
is taking place on Funknet, as well as her other comments, but
want to add a few additional comments.  Quite apart from the major
issue of context, there are a number of other ways in which the
discussion seems a number of steps removed from what people do
when they parse sentences.

The notion that the output of a parse is a syntactic tree is odd.
The "real" output is some sort of meaning.  It may very well be
that parsing the sentence syntactically plays a major role in
allowing people to determine the meaning, but that doesn't mean
that an entire tree is produced in the process.  Many parsers for
computer programming languages parse computer programs
syntactically, in that they identify the syntactic structure, but
this is only because of the extent to which certain aspects of
meaning are associated with syntactic structure, and identifying
these aspects of syntactic structure are crucial in determining
the intended meaning.  But even when they do, they do not
construct syntactic trees for the program; they construct a
representation of the meaning.  Parsing is only of interest if the
output is some sort of meaning.

I thus take issue with the claim

>>All parsers should be held to the task of labelling parts of
>>speech, parts of the sentence, sentence type, and tense and
>>voice as well as being able to manipulate strings: change
>>actives to passsives and statements to questions and so on.

>>This after all is what parsing is.

This has little to do with what parsing is, if by parsing we are
referring to something that people do.  Speakers of any language
can parse sentences in their language without being able to label
parts of speech or other grammatical features.  Nor do they need
to be able to change actives into passives.  What they need to be
able to do is parse active and passive sentences and come up with
the same denotative meaning.  Being able to change actives to
passives is not necessary for this.

Nor is this just a terminological issue about what "parsing" is.
If part of the test of a syntactic theory is its ability to parse
sentences, then part of the test of a syntactic theory is how
successful it is to assigning meanings in context, either as part
of the theory itself, or in terms of its interaction with other
systems.  Thus the idea that parsing simply involves producing
trees is reminiscent of the sort of modular view of syntax that
most functionalists reject.

When someone tells me about a web site at which one can have
conversations with a computer program attached to a knowledge
database, even if the area of knowledge is very limited, and the
vocabulary very limited, and only simple syntactic structures
permitted, then I'll be interested.

Matthew Dryer