Why one structural alternative is not always enough: Evidence for local parallelism during the processing of German

Matthias Schlesewsky¹, Ina Bornkessel², and Stefan Frisch¹
schlesel@rz.uni-potsdam.de, bornke@cns.mpg.de, schlesel@rz.uni-potsdam.de
1 Department of Linguistics, University of Potsdam
2 Max-Planck-Institute of Cognitive Neuroscience, Leipzig

We present a new perspective on the serial vs. parallel debate in sentence processing by means of two studies using event-related brain potentials (ERPs) to track the on-line processing of argument order variations in German. The first experiment varied not only the word order of the clause but also incorporated either a full NP (1) or a pronoun (2) as the type of dislocated argument. The latter variation was included in view of the observation that pronouns behave like subjects with respect to unmarked word order in definite German main clauses, thereby offering the opportunity to contrast two unmarked but distinct structural continuations.

(1)

Dann hat dem Jäger der Lehrer den Roman gegeben.
then has theDAT hunter theNOM teacher theACC novel given
'Then the teacher gave the novel to the hunter.'

(2)

Dann hat ihm der Lehrer den Roman gegeben.
then has himDAT hunter theNOM teacher theACC novel given
'Then the teacher gave him the novel.'

The ERPs revealed several interesting observations: (a) in sentences with non-pronominal arguments (1), the determiner of a non-canonical noun phrase elicited a negativity between 300-450 ms post onset; (b) in sentences with pronominal arguments (2), no differences were observed for non-canonical word orders. Seeing that this pattern of results is predicted neither by the relative frequencies of the constructions examined (according to the German IDS corpus) nor by information structural differences, we interpret the contrast between non-pronominal NPs and pronouns as showing that the human parser may locally compute structural alternatives (predictions) in a parallel manner.

This hypothesis was tested in a second ERP study in which the word order of embedded German dass ('that')-clauses was varied (e.g. 3).

(3)

dass dem Jäger der Richter folgte
that theDAT hunter theNOM judge
'Then the teacher gave him the novel.'

In constructions such as (3), initial dative-marked arguments may be interpreted as the sole argument of a passivised verb and need therefore not have been dislocated, while initial accusatives must have undergone dislocation to a position preceding the subject. Accordingly, only initial accusative-marked arguments gave rise to a negativity in the ERP, while initial datives did not differ from initial nominatives.

In summary, our results call for a parsing architecture allowing for the availability of several structural alternatives during syntactic prediction in a structurally ambiguous region.