CSLI Publications logo
new books
catalog
series
contact us
for authors
order
search
CSLI Publications
Facebook
 

Wasow, Thomas & Clausen, David: Weak Cross-over and Informativity

Wasow (1972) introduced the distinction between weak and strong cross-over, illustrated in (1) because examples like (1a) seem possible in appropriate contexts, whereas the indicated coreference in (1b) is totally impossible.
  
(1) a.  ??Whoi did you talk to the boy shei liked about?  
    b.  *I know whoi Charlie think hei hurt.
Following a suggestion of Culicover's, Wasow proposed that the gaps in long-distance dependencies contain traces coindexed with the fillers. Under this analysis, the low acceptability of an example like (1a) is related to the low acceptability of cataphora with indefinite antecedents:
(2) ??I talked to the boy shei liked about a girli.
Work by Hofmeister, Sag, Clausen and others has shown that at least some island phenomena can be explained in terms of factors that cause processing difficulty. They have shown that the graded nature and context sensitivity of many island phenomena are better explained in terms of demands on memory during processing than through appeals to universal grammatical constraints. This talk explored the possibility of explaining weak-crossover (WCO) in similar terms. It reported on three studies in which participants identified the referent of a pronoun and then rated sentences on a seven-point Likert scale. The test sentences were embedded in a context designed to favor the intended interpretation of the pronoun. Each study contained examples with short (uninformative) antecedents for the pronoun as well as examples with longer (informative) antecedents. The three studies investigated WCO in questions, WCO in relative clauses, and cataphora. Each study also contained controls without the phenomenon under study, as well as both grammatical and ungrammatical fillers. Sample stimuli are given in (3)-(5).
(3) WCO in Questions: 
The hiring committee was presented with only a subset of the
applicants. Some had been eliminated because their files were
incomplete, but the committee chair dropped one applicant because of
negative comments from colleagues who knew that applicant well.  THE
REST OF THE COMMITTEE WAS NOT TOLD WHO/WHICH JOB APPLICANT PEOPLE THAT
KNEW HIM WELL HAD CRITICIZED
(4) WCO in Relative Clauses
In the bottom of the fourth inning, a questionable call elicited jeers
from the visitors' dugout. One of the umpires evidently heard
something sufficiently offensive to stick his head into the dugout and
issue a warning.  THE RADIO ANNOUNCER COULDN'T SEE THE PERSON/VISITING
PLAYER WHO THE UMPIRE HE HAD OFFENDED HAD WARNED.
(5) Cataphora
The news media are already beginning to cover the race for the 2012
Republican presidential nomination.  At a recent convention of GOP
bigwigs, many potential candidates were soliciting endorsements.
EVERYONE HE SOLICITED HAD PURPORTEDLY AGREED TO ENDORSE ONE
PERSON/POTENTIAL CANDIDATE
Our expectations of these experiments included the following:
  • In all three studies, we expected the controls (non-WCO or anaphora) to be judged better than the test sentences (WCO or cataphora).
  • In all studies, we expected informative antecedents to be judged better than the uninformative antecedents.
  • We expected WCOs to be judged better than the ungrammatical fillers.
  • We expected the embedded question WCOs to be judged worse than the relative clause WCOs.
  • We expected the cataphora examples to be judged better than the WCOs.
Our studies collected 1111 ratings of embedded question WCOs, 1447 relative clause WCOs, and 1220 ratings of cataphora examples, from a total 361 participants. The results confirmed some of our expectations but not others. Specifically:
  • As expected the controls (non-WCO or anaphora) were judged better than the test sentences (WCO or cataphora).
  • As expected, informative antecedents were generally judged better than the uninformative antecedents, but not in embedded question WCOs or in RC controls. This difference is a mystery to us.
  • Contrary to our expectations, the embedded question WCOs were judged slightly better than the relative clause WCOs.
  • As expected, the cataphora examples were judged better than the WCOs.
  • Even informative WCOs received low acceptability scores, but were rated higher than ungrammatical fillers.
We tentatively concluded that processing factors play a role in the low acceptability of WCO. In particular:
  • The processing cost of filler-gap dependencies makes WCO hard.
  • The processing cost of establishing pronoun-antecedent pairings makes WCO hard.
  • WCO involves the same extra processing cost as cataphora.
  • Informative antecedents mitigate the processing costs, at least sometimes.
  • Whether the cumulative processing costs are sufficient to account for the low acceptability of WCO examples remains unclear.
The question of whether weak cross-over effects can be fully explained in terms of constraints on processing remains unresolved but deserves further study.

References

  • Clausen, David R. (2010) Processing Factors Influencing Acceptability in Extractions from Complex Subjects. Unpublished ms., Stanford University
  • Hofmeister, Philip (2011) Representational Complexity and Memory Retrieval in Language Comprehension. Language and Cognitive Processes, Volume 26.
  • Hofmeister, Philip, and Ivan A. Sag (2010) Cognitive Constraints and Island Effects. Language, Volume 86.
  • Wasow, Thomas (1972) Anaphoric Relations in English. MIT dissertation


Maintained by Stefan Müller

Created: November 15, 2011
Last modified: January 09, 2019

pubs @ csli.stanford.edu