Information Technology Reference
In-Depth Information
duces nonword pronunciation deficits with even rela-
tively small amounts of damage.
Interestingly, as the level of damage increases, the
model makes increasingly more semantic errors, such
that the profile of performance at high levels of damage
(e.g., 90%) provides a good fit to deep dyslexia, which
is characterized by the presence of semantic and visual
errors, plus the inability to pronounce nonwords. The
semantic errors result from the learning-based division
of labor effect as described previously (section 10.3.2).
Furthermore, we see another aspect of deep dyslexia in
this data, namely a greater proportion of semantic er-
rors in the abstract words than in the concrete ones (es-
pecially when you add together semantic and visual +
semantic errors).
Finally, the last case of direct pathway damage with a
completely lesioned semantic pathway produces mostly
visual and “other” errors.
10.4
The Orthography to Phonology Mapping
The next model explores the important issues of regu-
larities and exceptions in the orthography to phonology
mapping. A much larger number of words is needed to
establish what counts as a regularity, especially given
the complexity of the nature of this mapping in English,
as described below. The mapping from written word
spelling orthography to phonology has been studied ex-
tensively for roughly 3,000 English monosyllabic words
in a series of influential models (Seidenberg & McClel-
land, 1989; Plaut et al., 1996). These models confront
two central issues: (1) the relationship between the pro-
cessing of regular versus exception words, specifically
whether a single system can process both; (2) the ability
to simulate the systematic performance of skilled hu-
man readers in pronouncing novel nonwords, which de-
pends on properly encoding the often subtle regularities
that govern how letter strings are typically pronounced.
To understand these issues, we must first understand
the nature of regularities and exceptions. A regular-
ity can be defined as a mapping (between a letter and a
phoneme) that is present in a relatively large number of
examples from the language. For example, consider the
pronunciation of the vowel i in the words mint , hint ,and
flint. In every case, it is pronounced the same way (as
a short vowel), making this the regular pronunciation.
The group of words that define a regularity like this is
called a neighborhood . In contrast, the word pint is
pronounced differently (with a long vowel), making it
an exception word. Note that the regular pronunciation
is context dependent because it depends on other letters
in the word, particularly for the vowels. For example,
the words mind , hind ,and find form another neighbor-
hood of regular pronunciation, but with the long form
of the vowel this time. Another neighborhood of long
vowels comes from the familiar “rule” regarding the ef-
fects of a final e ,asin mine , fine ,and dine.
Thus, as many non-native learners of English are
probably painfully aware, the regularities in English are
not simple (Rosson, 1985). A single system may be
well suited to perform this mapping, because many fac-
tors need to be considered simultaneously to determine
what the “regular” response is. Furthermore, the irreg-
Go to the PDP++Root window. To continue on to
the next simulation, close this project first by selecting
.projects/Remove/Project_0 . Or, if you wish to
stop now, quit by selecting Object/Quit .
10.3.6
Summary and Discussion
This model illustrates how words can be represented
in a distributed fashion across a set of different spe-
cialized areas (layers), and how damage to various of
these layers produces behavioral results similar to those
observed in different types of dyslexia. The general
framework for the interactions between orthographic,
semantic, and phonological representations instantiated
by this model will be elaborated in subsequent models.
One general conclusion that can be drawn from these
results is that it can be difficult to localize damage based
just on patterns of error. For example, certain lev-
els of both direct and semantic pathway lesions pro-
duce only visual errors —- only with larger direct path-
way lesions do semantic errors start to appear. Thus,
purely behavioral approaches to cognitive neuroscience
are often underconstrained, and require converging ev-
idence from additional methodologies, including com-
putational models like these (see also, Plaut & Shallice,
1993; Plaut, 1995).
Search WWH ::




Custom Search