Working memory constraints on syntactic ambiguity resolution as revealed by electrical brain responses.

Parsing strategies in temporarily ambiguous sentences were investigated in readers with different sentence memory capacities using event-related brain potentials (ERPs). Readers with a high memory span as well as readers with a low memory span were required to read subject and object relative sentences which were either ambiguous until the last word (late disambiguation) or were disambiguated by case marking either the clause initial pronoun (immediate disambiguation) or the noun phrase following it (early disambiguation). ERPs registered during sentence reading elicited the following effects: In the late disambiguation condition, high span readers, but not low span readers, displayed a more positive going wave at the disambiguating number marked auxiliary for the object relative sentences than for the subject relative sentences. This positivity is taken to reflect processes of revision that become necessary at the disambiguating element if the initial structure considered is a subject relative clause. When case marking was available in the clause initial at the relative pronoun, both high and low span readers showed a positivity at the disambiguating element for the object relative sentences, suggesting the immediate use of case marking information for revision. When case marking was available in the noun phrase following an ambiguous pronoun both groups showed no clear effect of revision at the disambiguating element, but only at the sentence final number marked auxiliary. This non-immediate use of the case marking information seems to be due to an inherent ambiguity in the German case marking system which interacts with the disambiguating element's position in the sentence. The combined data indicate that morphological information can be used immediately by high and low span readers to resolve syntactic ambiguity during sentence processing whenever the information given is clearly unambiguous. In addition they suggest that possible processing differences in ambiguity resolution between high and low span readers may only appear when the ambiguous regions are long.