Language - Computational Explorations in Cognitive Neuroscience - page 320

fails to account for the dissociations between syntactic
deficits in production and comprehension observed in
people with Broca's area damage (Berndt, 1998). Al-
though this data is damaging for theorists who posit a
strictly modular distinction between syntax and seman-
tics (e.g., Berndt & Caramazza, 1980), we see it as con-
sistent with the kind of variability that occurs through
general cortical learning mechanisms discussed previ-
ously. In a neural network framework, specialization is
typically a matter of degree.
The neural basis of semantic representations is a very
complicated and contentious issue, but one that neural
network models have made important contributions to,
as we will discuss in more detail in section 10.6. Part
of the complication is that semantic information is only
partially language-specific — for example, there are vi-
sual, auditory, and functional semantics that are most
likely associated with the cortical areas that process the
relevant kind of information (e.g., visual cortex for vi-
sual semantics). The result is that virtually every part
of the cortex can make a semantic contribution, and it is
therefore very difficult to provide a detailed account of
“the” neural basis of semantics (e.g., Farah & McClel-
land, 1991; Damasio, Grabowski, & Damasio, 1996).
Certainly, Wernicke's area is only a very small part of
the semantics story.
There are a number of other brain areas that typically
produce language impairments when damaged, but we
won't discuss them in detail here. A useful general-
ization that emerges from the locations of these areas
is that language seems to be localized in the areas sur-
rounding the superior part of the temporal cortex, in-
cluding the adjacent parietal, frontal, and occipital cor-
tex (figure 10.1). Further, language function appears to
be typically localized in the left hemisphere, often even
in left-handed individuals. Although right-hemisphere
damage often leads to measurable effects on language
abilities, these are not as catastrophic as left-hemisphere
damage. For more detailed overviews of the neural ba-
sis of language, see Alexander (1997), Saffran (1997),
and Shallice (1988).

nasal cavity

hard palate

alveolar
ridge

velum
(soft palate)

teeth

lips

uvula

tongue

epiglottis

glottis
(vocal cords)

Figure 10.2: Major features of the vocal tract, which is re-
sponsible for producing speech sounds.

10.2.2 Phonology

Phonology is all about the sounds of speech, both
in terms of speech production and the physiologi-
cal/acoustic characteristics of the human sound produc-
ing hardware, and speech comprehension of the result-
ing sound waves by the auditory system. Clearly, there
is a relationship between the two, because differences
in the way a phoneme is produced will typically yield
corresponding differences in its auditory characteristics.
We focus on the productive aspect of phonology, be-
cause our models focus on speech production in the
context of reading aloud.
The human speech production system is based on vi-
brating and modulating air expelled from the lungs up
through the vocal cords (also known as the glottis )and
out the mouth and nose. This pathway is called the
vocal tract (figure 10.2). If the vocal cords are open,
they do not vibrate when air passes through them. For
speech sounds made with open cords, the phoneme is
said to be unvoiced , whereas it is voiced if the cords are
closed and vibrating. Changing the positions of things
like the tongue and lips affects the acoustic properties
of the (vibrating) air waves as they come up from the
lungs. The different phonemes are defined largely by
the positions of these parts of the system. In our dis-
cussion we will gloss over many of the details in such
distinctions between phonemes; the critical characteri-
zation of the phonemic representations for our models