words, how is a selective gating mechanism such as Merge or the focus of
attention able to select the most appropriate content to let in and sustain? The
start of an answer was already given from a domain-general point of view by
appealing to a self-sustaining mechanism of cognitive control, but the domain of
language allows us to distinguish different dimensions (pragmatic, semantic.. .)
that should influence processing, grammar being perhaps one of the most
important.
The non-locality of grammar, and grammar as a whole, can be interpreted
as a set of chunks or learned content that fulfill the specific role of guiding and
constraining the merging operation during processing so that it becomes an
efficient mechanism for extracting meaning out of the world. Regarding the
non-locality of many syntactic objects, it can be understood as a trigger of
specific refreshing operations across time that operate automatically (without
the need for cognitive control). While the processing of an utterance is taking
place, the labeling of some structures will generate expectations about how to
proceed (e.g. a subject will want to find its verb) which will not depend on
explicit rules or be encoded in any specific format, but emerge from the act of
labeling itself, from processing an item. Crucially, these expectations would not
need to carry on across cognitive cycles, so further processing resources will be
made available as normal, and the operation of Forget will still operate. Fol-
lowing the model put forth by Jonides et al. (2008), what would happen is
that the recently processed content will automatically refresh when, within a
new cognitive cycle, a new item is processed that partially matches it. A way to
represent this can be found in Figure 7.5.
The idea that non-locality is not necessarily a controlled process is in line
with experiments by McElree and colleagues (McElree et al. 2003, Öztekin and
McElree 2007, Va n Dyke and McElree 2011) o n memory retrieval during
language comprehension. First, these experiments examined the search mecha-
nism employed in language comprehension to detect and understand non-local
relationships, reaching the conclusion that, since no time differences are observed
in retrieval despite the amount of material that is placed in between the two
elements, it is probably a direct-access mechanism in which information becomes
readily available for interpretation without the need to go (Merge) step by step
Figure 7.5 A representation of the online establishment of dependencies as the
parallel reactivation of already processed material. The processing of each
word leaves a trace in short-term memory that is refreshed (represented
here by the color getting darker) by the ensuing processing of specific
items that have a grammatical relationship with it. The most active trace
at each step is the one the syntactic relationship corresponds to.
120 Gonzalo Castillo