Sluicing is affected by focus, but how?

Katy Carlson$^1$, Michael Walsh Dickey$^1$, Lyn Frazier$^2$, & Charles Clifton, Jr.$^2$
katyc@northwestern.edu, dickey@ling.nwu.edu, lyn@linguist.umass.edu, cec@psych.umass.edu
1. Northwestern University; 2. Universtity of Massachusetts at Amherst

In an auditory interpretation study, Frazier and Clifton (1998; F&C) tested sluices with two indefinite possible antecedents:

(1)

Some tourist suspected that the hotelkeeper was hiding someone. Guess who?

With the embedded object (someone) accented, object responses (who was hiding) reached 69%, while accenting the matrix subject (tourist) produced 46% object responses. Accent position significantly affected interpretation, presumably by helping focus the accented element, but object responses remained even with subject accent. This object bias was traced to focus more indirectly, through perceivers' general tendency (in English) to assign focus clause-finally. However, these results could also result from a preference specifically for object antecedents, or for antecedents not separated from the sluice by additional clause boundaries. Additionally, this work explored focus only in indefinite sluices, though other types exist (Romero 1997, Merchant 2001).

An auditory questionnaire addressed alternative explanations using sluices with two VP-internal indefinites:

(2)

Gary received some pamphlet from some organization but I don't know which.

With an accented object (pamphlet), object responses reached 40%, while accenting the source/goal PP (organization) lowered object responses to 28% (p's$<$.02). F&C 's focus-based theory predicted the general preference for the clause-final argument here, and this suggests that their earlier results should not be traced to an object-specific or distance-based preference. Instead, with accent position also modulating preferences, these results support both the direct and indirect focus-based effects they proposed.

Two further experiments extended this work to the processing of contrast sluices (with 'else' indicating contrast with a definite antecedent, instead of an indefinite) as in (3):

(3)

Alice insulted Bill, but I don't know who else.

In a written questionnaire, sentences like (3) received 86% object responses, showing that the preference for a clause-final antecedent holds for contrast sluices. In an auditory questionnaire, object responses fell to 42% with the subject (Alice) accented, vs. 88% with the object (Bill) accented (p's$<$.001). With the verb or both arguments accented, object responses were at 76% (each differed from the subject/object accent conditions, p's$<$.05). These results show that exclusive accent on only one argument is needed to raise focus assignments to that argument, not just the presence of an accent. Finally, despite strong accent effects, an overall bias towards the lower antecedent remains, showing that the position of default focus is important in all sluices tested. The implications of these results for ellipsis processing and the relation between accent and focus will be discussed.