Retraction Decisions: Parsing vs. Interpretation

Markus Bader
Fachbereich Sprachwissenschaft, Universität Konstanz

(Note: I haven't included glosses and translations into the word count)

Due to ambiguity, incremental language comprehension will often force language comprehenders to retract earlier decisions. This paper will present three experiments that have compared retracting syntactic decisions with retracting semantic decisions. These experiments have focused on the ambiguity of German "sie" which is fourway ambiguous, corresponding to English "she", "her", "they", or "them", depending on syntactic and semantic context.

The difference between retracting syntactic and retracting semantic decisions is illustrated in (1).

(1) a. Was Maria betrifft, so habe ich gehört,
What M. concerns so have I heard
dass sie die Kinder besucht haben.
that she/her the children visited have

"Concerning Maria, I have heard that the children visited her"

b. Maria hat gesagt, dass sie auf die Kinder gewartet haben.
M. has said that she/they for the children waited have

"Maria has said that they have waited for the children"

In (1a), "Maria" is the obligatory antecedent of "sie". "Sie" must therefore be a singular pronoun and the sentence has an
object-subject structure which is unpreferred in German, causing a garden-path-effect in sentences like (1a).

In sentence (1b), "sie" is unambiguously the subject. Due to the plural morphology on the clause-final verb, "sie" is a plural
pronoun. This contradicts the initial expection deriving from taking "sie" as being coreferent with "Maria". In sum, the clause-final auxiliary triggers a syntactic reanalysis in (1a) but a semantic reanalysis in (1b).

Two major findings of the to-be-presented experiments are as follows:

(A) Experimental evidence shows that it is much easier to arrive at the correct reading in sentences like (1a) than in sentences like (1b): Retraction of syntactic decisions is substantially easier than retraction of semantic decisions. This difference cannot be due to "sie" having no antecedent in (1b) because having no antecedent by itself does not harm language comprehenders, as shown by (B).

(B) Wenn "sie" has no antecedent, sentences in which "sie" unambiguously is a subject are easy to process whether a clause-final auxiliary is marked for singular (sie = she) or for plural (sie = they). This suggests that in the absence of an antecedent, "sie" remains unspecified for number.

Given that disambiguation was achieved by syntactic means in (1a) and (1b), these findings, together with additional findings concerning gender information in pronoun resolution (using masculine antecedents in sentences similar to (1)) and the combined effect of retracting syntactic and semantic decisions, argue for a modular conception of the HSPM: On receiving the disambiguation information, the HSPM first tries to resolve the conflict within the syntactic module; only if this fails, as in (1b), is the possiblity considered that the actual error lies outside of the parser in an earlier semantic decisions.