Skip to content Skip to navigation

Flexible Repair of Escher Sentences

Roumyana Pancheva
University of Southern California
October 16, 2015 - 3:30pm
Margaret Jacks Hall, Greenberg Room (460-126)

(This talk reports on joint work with Ellen O’Connor)

Illusory Escher sentences like (1) sound remarkably acceptable in spite of their grammatical anomaly. Lacking a suitable bare plural than-clause subject, (1) leaves no possible host for a degree variable, and it is unclear what, if anything, the sentence means.


(1) a. More people have been to Berlin than I have. (Montalbetti 1984)

     b. * [more than wh2 d2-many I have been to Berlin]1 [d1-many people have been to Berlin]


Some authors have proposed that Escher sentences are not interpreted at the point at which they are judged acceptable (Townsend & Bever 2001). However, prior results by Wellwood, Pancheva, Hacquard & Phillips (2009, 2015) suggest that these sentences are interpreted, since their acceptability depends on the possibility of deriving an event comparison interpretation. When illusion sentences contain a repeatable predicate – one that a single individual can participate in multiple times, such as (2) – a plurality of events can be generated within the than-clause, serving as a source for the degree variable and leading to an illusion of grammaticality; sentences with non-repeatable predicates, such as (3), allow no such shift and are rated as significantly less acceptable.


(2) More people attended class than I did. → People attended class more than I did.

(3) More people graduated law school than I did. → # People graduated law school more than I did.


Online measures also reveal slower reading times for illusion sentences even when they receive high ratings (O’Connor, Pancheva & Kaiser 2012), indicating that a repair operation is triggered by detection of the anomaly. However, the breadth of the repair mechanism remains unknown: is event comparison the only possible interpretation of Escher sentences?


I will discuss a series of acceptability rating studies that probe whether Escher sentences show differential sensitivity to semantic versus morphological plurality of direct object noun phrases, as in (4) vs. (5), and to plurality of the than-clause subject, as in (6) vs. (7), in sentences with stative predicates that do not support cardinality measures, precluding repairs to event comparison (O’Connor and Pancheva 2015).


(4) More cats have striped tails...→ #Cats have more striped tails...        

(5) More cats have mouse toys...→ Cats have more mouse toys...                      


(6) More foreign-born diplomats are familiar with the trade agreement than the American citizen is.

(7) More foreign-born diplomats are familiar with the trade agreement than the American population is.


Our results support prior claims that Escher sentences are analyzed deeply and are semantically repaired. Additionally, we find evidence for a flexible repair mechanism, allowing for a comparison of events orindividuals, so long as a suitable plural predicate is available. The results illustrate that research on Escher illusions can be informed by semantics theories of measurement, and in turn, such research can provide experimental evidence for a link between plurality and cardinality, formulated on formal grounds. 

Event Type: 
Linguistics Colloquium