Three constraints on Relative Clause Attachment ambiguities

Barbara Hemforth and Lars Konieczny
University of Freiburg

Relative clause attachment preferences in N1-N2-RC ambiguities such as (1-3) have been claimed to be based upon case matching between the relative pronoun and the potential host (Sauerland and Gibson, 1998). This view was challenged by Hemforth, Konieczny, Seelig, and Walter (2000), who demonstrated that nominative and accusative RCs show a N1 preference regardless of their host?s case. However, they left open the question of whether or not RCs bearing lexical case (genitive and dative) show a preference for matching hosts. We conducted two completion studies, using sentences like (1) to (3), where the relative pronoun varied between nominative case (1), genitive (possessives) (2), and dative case (3). In Experiment I N1s were dative, as in (1) to (3), in Experiment II, N1s were nominative (N1 in Subject/topic position, not presented here). N2s were always genitive marked.

(1)

Der Manager hatte der Band [dat] der Musikerin [gen], die [nom] häufig Fehler machte, einen großen Auftritt besorgt. Die .................... machte häufig Fehler.
--[lit..] The manager had the band [dat] of the musician [gen] who [nom] made many mistakes ?

(2)

Der Manager hatte der Band [dat] der Musikerin [gen], deren [poss] Fehler sich häuften, einen großen Auftritt besorgt. Die Fehler der .................... häuften sich.
--[lit.]The manager had the band [dat] of the musician [gen] whose [gen] mistakes became more and more frequent ?

(3)

Der Manager hatte der Band [dat] der Musikerin [gen], der [dat] viele Fehler unterliefen, einen großen Auftritt besorgt. Der .................... unterliefen viele Fehler.
--[lit.] The manager had the band [dat] of the musician [gen] to whom [dat] many mistakes happened ?

The results can be taken to partially support the case matching hypothesis. In N1[dat]-N2[gen] sentences (N1 in indirect object position), nominative RCs (non-matching) showed the well-established N1 over N2 preference. Dative RCs, matching with N1, showed an even stronger preference for N1 attachments, while no reliable preference could be found for possessive RCs (matching with N2). For nominative N1s (in topic position) however, dative RCs (non-matching) showed a preference for N2 attachment, whereas nominative RCs (matching with N1) appeared neutral, and possessives (matching with N2) yielded an even stronger N2 preference. Case matching, however, cannot explain the generally higher preference for N1-attachments in non-topic positions which has been established across several experiments now. Taken together with results from other published and more recent studies, the data fit best with an approach integrating the following three constraints:

(1)

A general tendency to attach RCs high, possibly due to attachment-binding,

(2)

a prosodic constraint by which RCs in topic-position tend towards N2-attachment (Frazier et al., forthcoming), which are both finally modulated by

(3)

case matching, possibly restricted to lexical case or cases low on the case hierarchy.

In two further experiments, currently running in our lab, we investigate in how far the results can also be established in an on-line reading paradigm as well as with magnitude estimations.