Recent experimental results (e.g., Sprouse & Messick, 2015) confirm the
observation that D-linking ameliorates Weak Islands (e.g., Whether,
Complex-NP Islands) but has no such effect on Strong Islands (e.g.,
Subject, Adjunct Islands). We present a theory of this difference: the
lexical items that that block dependencies in Weak Islands (e.g.,
“whether”) are structurally analogous to lexical items that do not block
dependencies (e.g., “that”) and thus can be coerced into supporting the
dependencies; the corresponding lexical items in Strong Islands have no
such good analogies. We present data from a computational analysis of
English grammar in support of this theory.