|
Abstract:
Some current syntax-based approaches to parsing assume a
serial processing model for local ambiguities. As pointed out by
Mitchell (1994), one can argue that the relevant findings for
such models are also compatible with a ranked parallel model, in
which all structural possibilities are computed in the case of a
local ambiguity, but with a clear ranking. The studies we report
constitute a serious challenge for serial processing models. In a
reading time experiment, we investigated the processing of the
following constructions.
Die Sekretärin wusste nicht, ... the secretary didn't
know
(a) welcher Mitarbeiter wahrscheinlich unglücklicherweise
den Dekan eingeladen hatte which-nom assistent probably
unfortunately the-acc dean invited has
(b) welchen Mitarbeiter wahrscheinlich unglücklicherweise
der Dekan eingeladen hatte which-acc assistent probably
unfortunately the-nom dean invited has
(c) welche Mitarbeiterin wahrscheinlich
unglücklicherweise den Dekan eingeladen hatte which-amb
assistent probably unfortunately the-acc dean invited has
(d) welche Mitarbeiterin wahrscheinlich
unglücklicherweise der Dekan eingeladen hatte which-amb
assistent probably unfortunately the-nom dean invited has
The sentences c/d reflect the normal subject-object ambiguity,
which is resolved by the case marking of the second NP. As one
would expect, the reading time for the second NP is higher if
this element is marked as subject of the clause. This observation
reveals the subject preference for the initial wh-phrase. In
addition, the structures a/b show the unambiguous counterparts of
c/d: in the masculine singular paradigm, case marking suffices to
distinguish subject from object reading. A comparison of reading
times confirmed that those are higher for object-initial
structures than for subject ones. Such differences can be
analysed as a reflection of greater cognitive load for
object-initial structures. However, in spite of the fact that
there is a subject preference visible in c/d, we found that the
reading times for the first two phrases in c/d are in the same
time window as the dispreferred unambiguous accusative
alternative. This result is interesting since the time pattern in
a naming task shows that the ambiguous wh-phrase and the
preferred unambiguous nominatively marked ones are in the same
time window for word recognition. We take this as evidence
suggesting that in the ambiguous case both structural
alternatives are constructed, independent of a later visible
reanalysis effect. In this context we will present further
experimental results that suggest some new issues for the
discussion of the processing models.
Mitchell, D. (1994). Sentence Parsing. In: M.A.Gernsbacher,
Handbook of Psycholinguistics.
San Diego: Academic Press.
|
