Radek Šimík (Potsdam): Formal expression of information structure in Slavic - experiments and modeling (joint work with Marta Wierzba)

We present a number of acceptability rating experiments that shed new
light on the formal expression of information structure in three West
Slavic languages: Czech, Slovak, and Polish. On the formal side we look
at constituent order (scrambling) and accent placement (stress shift),
on the pragmatic side we look at givenness and focus. Our core findings
are:

1. Givenness requires deaccenting in West Slavic (just like it does in English or German).
2. Non-rightmost sentence stress is penalized in Czech and Slovak but less so in Polish.
3. Given objects need not scramble (to a preverbal position) if they are not placed rightmost.
4. Scrambling of given objects is penalized in Polish but not in Czech and Slovak.
5. Scrambling of focused objects is prohibited in Czech and Slovak but only dispreferred in Polish.

We interpret the results as supporting the hypothesis that information
structure relates to prosody (givenness-deaccenting, focus-accenting)
and only indirectly to syntax, via the tendency for rightmost accent
placement (findings 1, 2, 3). This conclusion (standard for English) is
rather controversial for Slavic languages, where (at least some)
information structural categories are standardly taken to be encoded in
syntax (Junghanns & Zybatow, Biskup, Kučová, Mykhaylyk, a.o.).
Another consequence that can be drawn from our results is that a
Reinhart/Fox-style economy constraint (movement only licensed if there
is an interface effect) might not be universal after all: we do find
evidence for such a constraint in Polish but not in Czech and Slovak
(finding 4).

The experimental results show a complex and gradient pattern suggesting
that a set of interacting violable constraints is at play; some
constraints cause a comparable decrease in acceptability across
languages when violated (e.g., DESTRESS-GIVEN), whereas for others, the
effect-size seems to be language-specific. We propose to model the
results using the framework of Linear Optimality Theory (Keller 2000) by
assigning language-specific weights to constraints that are operative at
the prosody-syntax interface and the prosody-information structure
interface. The predictions of the resulting model are fairly precise and
can be tested by further experiments.