This conflation of Merge with the focus of attention would forcefully imply
that Merge is not a linguistic operation per se, but a fundamental process that
is shared across species, so further clauses would have to be added to the model
to explain what in Merge, or in grammar (for example, the use of labels), or
WM, may constitute the human-specificity that supports the acquisition of
language. For now, it should be noted that the building of trees, that is, the
production of structures that can be defined as hierarchical, may not be one of
those human-specific features. If integration of information is subject to the
goal of processing, it would be necessary to select an order in which the items
in the fringe of attention enter the focus of attention, completing subtasks in
order to fully encode the item and the context in which a response has to be
selected. Thus, an argument that follows this reasoning states that domains like
motor planning, which is phylogenetically prior to language, have, despite sig-
nificant differences (e.g. Moro 2014, cf. also B oeckx and Fujita 201 4), much
in common with the hierarchical structures that Merge builds in language. As
an example, Pulvermüller (2014) s hows how the action of walking to the bath-
room to brush one’s teeth can be described by a center-embedding operation
that not only is hierarchical in nature, but also exceeds the usual limit of three
embeddings characteristic of language use.
It should be noted that this domain-general capacity is not Merge per se, as
Merge as it has so far been described produces only unordered sets that would
then need to be connected to the interfaces to linearize and become speech.
What we have here is the notion of sequential hierarchy (Fitch and Martins
2014 ), in which the building of hierarchy, like integration in WM, is subject to
a sequence of time in which, crucially for language and many types of actions,
the order matters. Even if all processing is sequential, we should distinguish
between hierarchical information that is presented sequentially, such as the act
of recognizing a face, in which the order of the components does not alter the
product, and hierarchies that can become meaningful only if they follow a
specific order of operation, that is, if they are serialized in a fixed order. If Merge
is to be equated with processing, it cannot remain being a producer of unor-
dered sets unless we propose a complementary mechanism that serializes in a
similar way as the global broadcast serializes localized processors in the global
workspace framework. The problem with this option is that it would amount
to saying that Merge is only a parallel mechanism that is not capable of produc-
ing the type of hierarchies that characterize language, in which order matters.
A more parsimonious solution than adding a complementary serializer is to
implement the property of forming sequential hierarchies in Merge itself, but
crucially leave outside of syntax any judgments about the rightness of the
sequences that Merge produces, keeping in place the separation between syntax
and grammar. If Merge is to be identified with the focus of attention, it also
has to be deployed along a full cognitive cycle, which is a process that is parallel
and serial at different stages, so it would make sense that Merge is also defined
as a process that is parallel and serial at different stages. The difference would
be that, while the cognitive cycle captures all localized processors trying to reach
Language and working memory 117