elicitation versus discourse as data source

[Spike Gildea]

I agree with John that one would produce a very strange, skewed grammar if
one based it solely on utterances elicited via requesting translations of
sentences in a contact language.  In fact, I go a bit further in that I
suspect a "context-free" sentence grammar will be wierd no matter WHAT
language  sentences are elicited in.  Recorded text data (ideally with the
linguist out of the picture and the microphones as unobtrusive as possible)
is clearly a methodological priority in order to get a sample of how
speakers really speak to each other.  But I have been using two methods of
elicitation which I have found to be more-or-less reliable in producing
sentences that all speakers seem to be happy with (as opposed to those
context-free sentences at the margins of the grammar, the things that
"nobody would really say", which leave a substantial proportion of the
speakers in disagreement, or just scratching their heads and saying they
don't know... y'know, the kind that separate speakers into dialect groups
with no other real correlates (cf. Labov's 1985 presidential address to the
LSA).

The first method I learned from Scott DeLancey and Colette Craig (although
I assume it goes way back), which is to collect a text and then mine it for
all the information possible by eliciting all sorts of variations on each
sentence: "what if *I* had done this", "what if it had been *YESTERDAY*",
what if you *couldn't see* it happen", etc.

The other is to record a session in which you get one speaker who is fluent
in your contact language, and one or more speakers who DO NOT UNDERSTAND
your contact language, and then conduct "interviews" designed with both
content and linguistic structure in mind -- here, the CONTENT is critical
as a means to distract speakers away from a focus on the formal structures.
 e.g., I ask a series of personal information questions, kinship relations
to other people I know, background in whichever languages, future questions
(like what hopes the speaker might have for the future of the language
community), counter-factual (like what the speaker would do if ...<fill in
the blank>), etc.  In a second phase, I ask some more language-oriented
questions, focused usually on issues I'm working with at the moment (e.g.
"what's the difference between X and Z?", where X and Z are minimally
different sentences, involving only some change in word order or the
presence versus absence of some elusive particle).  The recording of such a
session yields quite rich data, not only in (a) a range of examples of
grammatical sentence types, and (b) some direct answers to the linguistics
questions themselves, but also in (c) what appears to be relatively
unmonitored conversational data, as people argue over what REALLY is the
difference between two sentences, etc.

Then this recorded text itself becomes an object of sentence-by-sentence
study after it is transcribed and translated, and each sentence in it is
available for permutation, leading to more questions which can be asked in
the next recording session, ...

So while I think you always need a coherent context to ground it in, I
think there IS a place for elicitation in field work.

Spike