Why syntactic ambiguity is costly after all: Reading time and ERP evidence

Stefan Frisch, Matthias Schlesewsky, Douglas Saddy, and Annegret Alpermann
frisch@rz.uni-potsdam.de, schlesel@rz.uni-potsdam.de, saddy@ling.uni-potsdam.de
alperm@rz.uni-potsdam.de
Institute of Linguistics, University of Potsdam, Potsdam, Germany

One of the core issues in psycholinguistic research is the question of how our parser deals with syntactic ambiguity. The processing of the ambiguity itself can tell us whether the presence of syntactic alternatives induces processing cost or not, which would allow us to distinguish between serial and parallel approaches to parsing (cf. Mitchell, 1994). On the other hand, the point of disambiguation is of interest since it is crucial for deciding whether our parser prefers one alternative over others or whether it considers all continuations to be equally likely (cf. Mitchell, 1994). German provides an excellent testing ground for both questions, seeing that in German, arguments can be either ambiguous with respect to their grammatical function or they can be unambiguously case marked. We addressed the question of how syntactic ambiguity is processed by using transitive German declarative structures in which either the first (1) or the second argument (2) was ambiguous with respect to its grammatical function and in which either the first (a) or the second argument (b) was the subject.

(1)

a.
Die Detektivin sah den Kommissar ...
--[the detective]FEM.AMB saw [the cop]MASC.ACC

b.
Die Detektivin sah der Kommissar ...
--[the detective]FEM.AMB saw [the cop]MASC.NOM

(2)

a.
Der Kommissar sah die Detektivin ...
--[the cop]MASC.NOM saw [the detective]FEM.AMB

b.
Den Kommissar sah die Detektivin ...
--[the cop]MASC.ACC saw [the detective]FEM.AMB

In two experiments using different on-line techniques (self-paced reading and event-related potentials/ERPs), we presented the above sentences with a phrase-by-phrase segmentation. Results for the sentence initial NP showed that syntactic ambiguity induces enhanced processing cost which is reflected in higher reading times as well as in a late positive ERP component (P600). On the second NP, the initially ambiguous sentences being disambiguated in the dispreferred way (object-before-subject, 1b) elicited higher reading times as well as a P600 ERP effect compared to all other conditions. In sum, both techniques revealed a higher cost of processing for a syntactic ambiguity as well as for a dispreferred disambiguation. Results speak in favour of a parsing architecture according to which the parser is sensitive to syntactic ambiguity but then makes a decision for one of the alternatives (local parallelism).



References

Mitchell, D.C. (1994). Sentence parsing. In M. A. Gernsbacher (ed.): Handbook of Psycholinguistics (pp. 375-409). San Diego: Academic Press.



AMLaP Conference, Saarbrücken, September 2001