mechanisms that are shared across species and are also involved in non-linguistic
tasks.
It should be noted, fi nally, that Merge cannot be said to be speeded up by
grammar, as the multiple types of oscillations required to embed content that
Merge should consist of are distinct because of their different temporal frequen-
cies. What grammar could be doing instead is to make Merge waste less time
merging irrelevant information for the completion of a cognitive task. If there
is a system of chunks or frequent patterns of activation in place that is acquired
through experience, this should make cognition effi cient without speeding it
up. This phenomenon should be of the same nature as symbol training, which
seems to enhance the representational capacities of animals (e.g. Thompson
et al. 2001, Srihasam et al. 2012): it would make them focus on specifi c aspects
of processing that are relevant, while suppressing others. Also related to this,
the external guiding of attention by placing visual clues in front of prelinguistic
children seems to facilitate their chunking abilities, giving them higher-order
chunking capacities that were once believed to be an exclusive domain of a fully
developed grammar (Rosenberg and Feigenson 2013). Grammar, like the WM
capacities that support it, is a selective serializer.
Conclusion
Throughout this paper, the multicomponent model of Baddeley and Hitch (1974),
whic h was designed as an independent system to capture the phenomenon of
domain-specifi c short-term memory, has been reduced to an emergent property
that models the general way in which neuronal networks interact, giving rise to
a fruitful exchange of knowledge among multiple disciplines that were initially
mute regarding this topic, such as developmental psychology, animal cognition,
psycholinguistics and neurocognitive science. This exercise is still pending in many
bastions of linguistic theory, where language is taken as a distinctive component
or organ whose inevitable particularities seem to grant its separation from other
areas of cognition, creating a bad tendency to label the discovery of similarities
as little else than metaphors and to banish interdisciplinary dialogue.
The implementation of language should leave behind the traditional Broca-
Wernicke model to include the interaction of multiple brain regions and networks.
The study of language processing can be informed by the study of domain-
general cognition, and the structure of language knowledge and the stages of
language acquisition should be inextricably linked to the natural constraints
imposed on processing mechanisms. Likewise, the phenomena of processing,
short-term memory and long-term memory cannot be separated if our objective
is to design a realistic cognitive model of how the brain works.
Acknowledgements
Research funded by the FI-DGR 2015 Program from the Generalitat de
Catalunya.
122 Gonzalo Castillo