Even in the absence of interferences, the contents stored in short-term memory
decay with time. Verbal rehearsal has already been mentioned as way of prevent-
ing the loss of phonological content, but the central executive also has the
domain-general capacity of going back to the contents stored in any buffer and
re-allocate attention to them in order to maintain them for longer. This process
is known as refreshing (Joh nson and Hirst 1993).
The last component that completes the standard WM model is the episodic
buffer (Bad deley 2000), a domain-general storage system where the central
executive integrates the information from the perceptual buffers into coherent
episodes known as chunks (Mil ler 1956). Chunks are considered as complex
units of thought updated by content stored in long-term memory, so their
formation marks the dividing line between perception and cognition. One of
the reasons that motivated the introduction of the episodic buffer was the
domain-specificity and limited capacity of the perceptual buffers, which could
not explain how it was possible that humans could produce and understand
linguistic utterances. Even if the phonological buffer’s capacity is exceeded dur-
ing the presentation of the stimulus, and verbal rehearsal is blocked by a con-
current task, understanding a list of words as a meaningful sentence increases
short-term storage dramatically, from 5 unrelated words up to more than 12
(Pott er and Lombardi 1990). This is known as the sentence superiority effect,
and the way it was captured by the model was by suggesting that long-term
memory has a boosting effect on multimodal integrations, including the associa-
tions between sound and meaning that characterize language (Badd eley and
Wilson 2002). The episodic buffer can be considered to have an approximate
limit of 4 simultaneous chunks, a limit that has been termed “the magical
number 4” (Mill er 1956; Cowa n 2001, 2010 ).
Although at first it was assumed that the central executive uses attention to
produce chunks, more recently, studies like Badd eley et al. (2009) and Alle n
et al. (2006) have shown that chunk formation is not significantly affected by
the realization of concurrent tasks that are designed to tax attentional resources.
This has forced a redefinition of chunk formation as a mostly automatic process,
while central executive functions are left for flexible chunk manipulation (Bad-
d eley et al. 2009), a central characteristic of language that in linguistics and
philosophy of language is known as compositionality.
One of the main criticisms of the multicomponent model is that its central
executive is little else than a homunculus, as it has the same capacities as the
system that it is trying to explain (e.g. Verb ruggen et al. 2014). But as Badd eley
(2012) correctly argues, the introduction of homunculi can be useful if they
are not used as explanations, but as a way of constraining the set of things that
are in need of explanation.
From this introduction to the multicomponent model a basic vocabulary of
concepts has emerged: task switching, inhibition, attention, refreshing, chunk
formation, chunk manipulation, the processing/storage distinction, and the
short-term/long-term memory distinction. We can ask ourselves how these
concepts connect with our language capacities, but in order to find satisfactory
Language and working memory 105