Creating a corpus

[Angus B. Grieve-Smith]

        An elicited corpus can be very useful, but I wanted to mention a
topic that came up at the North American Symposium on Corpus Linguistics a
little more than a week ago: representativeness.  It's definitely valuable
to say "native signers of x language produced this utterance and didn't
act as though it were a speech error."  But if you do any frequency
counts, you need to make sure that they can be generalized to the entire
language (or at least an entire register, genre or dialect), and for this
you need to have a representative sample.  It might be possible to get
this through elicitation, but you'd first need to study the balance of
"text types" that a typical Deaf person is exposed to in a typical day,
week or month, and then recreate that balance.

                                        -Angus B. Grieve-Smith
                                        Linguistics Department
                                        University of New Mexico
                                        grvsmth at unm.edu
                                        grvsmth at panix.com