Parsing

[Susanna Cumming]

Folks,

I'm somewhat surprised that this parsing discussion is taking place on
Funknet. Analysis of isolated "sentences" into labeled trees without
reference to contextual factors is something that human beings never do
(in fact we never even encounter such objects), so why should we be
interested as functional linguists in whether a computer can do it?

Paul Deane's point that the range of constructions encountered in everyday
language use is still much wider than has yet been accomodated in any
actual grammatical description, on or off-line, is well-taken. Moreover,
he takes his examples from written language; far greater yet is the range
encountered in everyday informal speech -- the only universal form of
language use. This is the real test if we are interested in language
processing as scientists rather than as software engineers.

While I for one am certainly in favor of in computer modelling of natural
language production and understanding as a tool for testing our hypotheses
about the way contextual factors and linguistic form interact, for me such
attempts are only interesting to the extent that they reflect actual human
behavior in natural communication situations.

This is not of course to denigrate the commercial potential of
sentence-level parsing, a matter the marketplace can decide.

Susanna Cumming