Barbara Hemforth & Lars Konieczny
University of Freiburg
barbara@cognition.iig.uni-freiburg.de
We will present a series of on-line (eyetracking) and off-line (magnitude estimation) experiments on pronoun resolution and relative clause attachment in German sentences like (1), (2). For pronoun resolution a robust preference for antecedents in topic position was established, i.e., it was easier (took less time in total reading and was more acceptable) for a pronoun to be bound to the first NP (the default discourse topic) in the matrix clause irrespective of its functional role (subject or object).
| (1) | a. | Der Dirigent faszinierte die Pianistin, als er/sie sich vorstellte. |
| lit. | The conductor (masc, subj) fascinated the pianist (fem, obj) when he/she introduced her/himself. | |
| b. | Den Dirigenten faszinierte die Pianistin, als er/sie sich vorstellte. | |
| lit. | The conductor (masc, obj) fascinated the pianist (fem, subj) when he/she introduced her/himself. |
Hemforth, Konieczny, and Scheepers (2000a,b) argued that the preference pattern across languages that can be found for relative clauses can be explained by a combination of anaphoric and syntactic processes. The basic pattern in German is a slight to modest N1-preference for two-site attachment ambiguities like those in (2). If the topic preference also applies to RC-attachment, it would make sense to expect a stronger N1-preference in cases like (2a), where the head of the NP is the default topic, than in (2b) where it is not. However, we found a reliably stronger N1-preference for NPs in object position. In topic-position, we found a very modest numerical N1-preference or even an N2-preference in some of the experiments.
| (2) | a. | Der Chef des Kochs, der sehr krank war, wurde von einer Nachbarin gesehen. |
| lit. | The boss of the cook who was very sick was seen by a neighbour. | |
| b. | Eine Nachbarin hat den Chef des Kochs, der sehr krank war, gesehen. | |
| lit. | A neighbour has the boss of the cook who was very sick seen. |
This difference in preference patterns can be explained by the informational function of pronouns and relative clauses. Pronouns serve the cohesion of the text without adding any information and therefore tend to search for the discourse topic as the most active antecedent. Relative clauses, however, are more strongly informationally loaded and are therefore preferentially attributed to focused elements of the sentence (see Amor, 1999). Preliminary data from two experiments on prosodic marking in RC-attachment in production and comprehension corroborate these assumptions (see also Shafer et al., 1996, or Wijnen, 1998).