In English and many other languages, the acceptability of long-distance
wh-extraction out of a finite declarative complement clause appears to vary
based on the matrix verb: “bridge verbs” like "think" and "say" lead to
relatively high acceptability, verbs like "shout" less so. This variation
in relative acceptability, which we call “bridge effects,” cannot be easily
derived syntactically (Erteschik-Shir 1973; among many others) and has been
variously attributed to non-syntactic factors like information structure or
frequency. This study presents an experimental evaluation of a nearly
exhaustive list of 640 finite-clause embedding verbs in English in order to
critically assess three leading theories based on information structure,
frequency, and semantic similarity. Our results show that these
non-syntactic factors are unlikely the main driver of bridge effects,
indicating a need for theories with better empirical coverage.