question on Preferred Arg. Str.
tgivon at uoregon.edu
Thu Oct 14 20:56:10 UTC 2004
I think there are (at least) three words of caution one would want to add before
christening these (strong) distributional tendencies "preferred argument
structure (in discourse)". First methodological: There's nothing magical about
trhe absolute values in Dubois' text counts. Different oral text will yield
differen t absolute numbers. Dubois' distributions tends to correspond, in my
experience, to long oral narratives, particularly traditional two-actants
stories. Different texts types will show considerable variation, tho the
relatives tendencies seem to hold, in the main.
Second, theoretical: "preferred argument structure" was a theoretical notion
about the lexical semantics of verbs (or verb senses) and how it maps onto the
syntax of "simple" (main-declarative-affirmative-ACTIVE) clauses. But the mapping
onto "simple" clauses (using Keenan's term from 1976) already disguises some
discourse-distributional facts about the most common clause type in natural, oral
discouse (by a whopping 90%, on the average). The so-called Dubois distributional
observations, on the other hand, have to do with a rather different aspect of
oral communication: The tendency to chunk information in oral language under
very small "intonational clauses", on the average 2-3 words per "clause" of 1-2
seconds duration. You can see some discussion of this temporal dynamics, albeit
from a different perspective, in my 2002 book "Bio-Linguistics", ch. 5. (Both
Walley Chafe and myself have written on this previously). Typically, this
"chunking" strategy relies on extensive anaphora (zero or PRO) of both areguments
AND verbs. And the cognitive foundations of this apparent temporal-chunking
restriction seems very different from the traditional "preferred argument
structure". So perhaps using the same term doesn't gain us all that much.
Third, also theoretical: Calling these regularity "principles", be they one or
more (here I agree with Martin) seems to simply defer the need to EXPLAIN them in
the way at least functionalists are honor-bound to explain: By some reference to
communicative, cognitive or--God forbid--even neuro-biological properties of
human information preocessing (including adaptively-driven evolution, if
necessary...). Calling these distribution al facts "principles" is, at best, a
heuristic convenience preliminary to searching for more satisfying explanations.
Martin Haspelmath wrote:
> Dear Mira and Funknetters,
> >(a) Ditransitives. It's not necessarily the case that if A is nonlexical
> >there won't be 2 other core lexical arguments.
> Yes, but this is not part of the original claims, I believe. Moreover,
> since the Recipient has an equally strong tendency to be nonlexical,
> again it is unclear whether an additional constraint is needed.
> >(b) If there's an additional quantity constraint (additional to "avoid
> >lexical As"), then when you get lexical A and lexical O you have two
> >violations. When you only get lexical A (with a nonlexical O) you only have
> >one violation. One can check whether the cases with two violations are
> >rarer, for example. In Sakapultek (Du Bois 2003 in Du Bois et al eds) there
> >are indeed no 2 new arguments, but there are some new As (6%).
> Yes, but is there a *statistically significant* number of new As? The
> figures are as follows in Du Bois's Sakapultek data:
> new A given A total
> new O 0 47 47
> given O 6 134 140
> total 6 181 187
> This distribution is not significant. Although there are fewer newA-newO
> combinations than newA-givenO combinations, this is expected, because
> given O is more common than new O overall. The same was found by Östen
> Dahl in his much larger spoken Swedish corpus.
> >2. The identification between "new" and "lexical". I beg to differ. The two
> >are NOT interchangeable. Some Given entities may be lexical (and marginally,
> >the opposite is also true).
> So the question is: Do the lexical given NPs show an effect of the
> constraints that specifically refer to lexical (as opposed to new)
> status? Has anyone examined this?
> >3. Getting rid of "Avoid new As": It's true that agents are human and
> >topical in many cases, but not in the percentages that PAS findings show. Of
> >course the correlations you point to are relevant. This must have motivated
> >the choice of As for Given entities and O for potentially new entities in
> >the first place. But then, there's Ss too! Their profile is similar to that
> >of As in terms of humanness and topicality. If so, why do they split off
> >from As (in allowing new entities)?
> It seems that in many languages/many data sets Ss are intermediate
> between As and Os. I don't see that theire behavior provides good
> evidence for the Given A Constraint ("Avoid new As").
More information about the Funknet