Human language is a set of stored lexical items and
a system of generative algorithms for combining them into larger
representations. So far, research on the neurobiology of lexical
access, on the one hand, and composition, on the other, have proceeded
largely separately, with little work addressing the impact of
composition on the neural representations of individual words. Here we
used data from a simple picture naming paradigm to inquire about the
impact of phrasal contexts on the representations of individual words
within the phrase. We use machine learning algorithms to investigate
the extent to which the representation of “house” is the same when
“house” is produced as a single word as opposed to when “house” occurs
in a phrase such as “red house”. This avenue of investigation
aims to answer whether composition affects word representations
equally, or whether structural factors, such as being the syntactic
head of a phrase, matter.
.