One model of the way in which words are stored in the mind is that of Taft and Foster (1975) which was
refined by Taft (1979, 1981). In this model roots (including bound ones) and prefixes are stored separately
in the mind: complex words are not stored in a pre-assembled state. Word recognition crucially involves
MORPHOLOGICAL PARSING. According to Taft and Foster, the dictionary in the mind engages in PREFIX
STRIPPING, a process where a word is parsed and all the prefixes are identified. This is followed by
looking up the root in the dictionary. Stanners
et al. (1979) took the logical step of extending the same
model to inflectional suffixes. This is the model that has been loosely assumed and implicitly used
throughout this book.
An alternative model, and one that has won the approval of most psycholinguists, is that of Butterworth
(1983). It is called the Full Listing Hypothesis (FLH). Advocates of FLH assume that familiar words are
entered in the lexicon already fully assembled, but if confronted with unfamiliar words, one might resort to
parsing. Routinely, speech recognition need not entail any morphological parsing. In other words, normally
there is no affix stripping.
There are two versions of FLH. Version 1 assumes full listing, with information of morphologically
complex entries spelt out. A complex word like
recovering is listed, complete with its morphological
analysis. So it would appear in the lexicon as (re(cover)ing). Version 2 of FLH assumes
that every word has
a separate entry in the dictionary. But the entries of related complex words are linked as SATELLITES forming
part of a constellation whose nucleus is a simple un-affixed bound root or word (cf. Bradley (1978, 1980),
Bybee (1987)). For example, the dictionary would indicate that there is the word
cover which functions as
the nucleus of the constellation containing the satellites listed in [11.7]
[11.7]
cover
covers
covering
coverings
covered
uncover
uncovers
uncovered
recovered
recovering
coverable
recoverable
recovery
uncovering
recoverability
As already mentioned, parsing is not entirely ruled out in this model. Butterworth (1983) suggests that it
is resorted to,
in extremis, when a hearer needs to cope with a novel or unfamiliar word such as
Lebanonisation
or
deprofessionalisation.
A theme that has run through this book is that listing of totally unpredictable lexical items is essential. But
many lexical items are compositional and hence not unpredictable. Such items need not
be listed in the
lexicon. FLH takes the opposite view and assumes that the morphological parser is used sparingly: listing of
preassembled words is the norm. I am inclined to disagree. In the remainder of this section, following
Hankamer (1989), I will argue that FLH is not a plausible model of how to produce or recognise words.
Hankamer examines the morphological complexity of Turkish, a typical agglutinating language. He shows
that the number of word-forms corresponding to a single lexeme is so large that even assuming that a
speaker has a very modest word-hoard, it is simply impossible that they would have the storage space to
store the billions of word-forms they know. This is how Hankamer puts it:
It seems that agglutinative morphology is even more productive than has been thought. Given a
lexicon containing 20,000 noun roots and 10,000 verb roots, which does not seem unreasonable for an
educated speaker of Turkish, the FLH would require over 200 billion entries. Furthermore, most of
the entries would necessarily be complex, and thus would take up significant storage space in the human
brain.
162 ENGLISH WORDS
(Hankamer 1989:403)
Hankamer shows that this is what the word-cluster containing some of the satellites of the nucleus
ev
‘house’ would look like:
[11.8]
ev
‘house’
evler
‘houses’
evlerimiz
‘our houses’
evlerimizde
‘in our houses’
evlerimizdeki
‘the one in our houses’
evlerimizdekiler
‘the ones in our houses’
evlerimizdekilerin
‘of the ones in our houses’
evlerimizdekilerinki
‘the one of (belonging to) the ones in our houses’
We have here already eight forms, some of them quite complex, without going through all the
grammatical cases and without listing the various persons (
my, your, her etc.) Since house
is pretty much an
everyday word presumably it would have to reside permanently in the mental lexicon, all pre-assembled.
That would require a lot of storage space. Now, if you look at the data in
Chapter 3
from Latin, a typical
inflecting language (cf. [3.17]), and those from Eskimo, a typical incorporating language (cf. [3.20]) you
will see that the idea of listing pre-assembled words would be no less implausible in those languages.
The storage implications make the FLH untenable. The human brain is estimated by Sagan (1985) to have
a storage capacity of 12,500,000,000,000 (12.5 billion) bytes. This estimate is based on the number of
neurons in the brain. If the storage of a fairly simple word takes, say 10 bytes (and that of complex words
like
evlerimizdekilerinki requires considerably more), it would be possible to store a maximum of 125
billion word-forms. And that would only be possible if the brain was exclusively dedicated to storing
words. Since at a conservative estimate educated speakers of Turkish would need to store 200 billion word-
forms, the FLH model fails to account for how these speakers manage to list the words they know. But, of
course, a bigger problem is the fact that even to cope with the lower figure of 125 billion words, the FLH
would need to make the absurd assumption that our Turkish speaker uses the brain exclusively as a word-
store.
While conceding that the FLH cannot work for Turkish, one might say that as English is not an
agglutinating language like Turkish (we established in Chapter 3 that English is essentially an isolating
language), there is no reason to be pessimistic about the chances of the FLH accounting for the storage of
English words in the mind. The problem is that English does have a sizeable number of words, many of
them not rare by any means, which contain several affixes. When we listed the satellites of the nucleus
cover
in [11.7], we stopped at 14 word-forms. We could have gone on. In this respect
cover is not
particularly unusual. Clearly, even in English there is quite a big number of agglutinative words. If we take
the vocabulary of an educated speaker, any attempt to apply the FLH would mean listing millions of word-
forms, many of them quite complex, in the mental lexicon. Again the pressure this would exert on storage
space would be intolerable.
If the human mind is equipped with a morphological parsing device to cope with extensive agglutinating
languages like Turkish, there is no reason to assume that it cannot be turned on whenever the need arises to
THE MENTAL LEXICON 163
deal with morphologically complex words in a language like English. Given the existence of words like
those in [11.7] it is reasonable to assume that morphological parsing is needed fairly often in English.
Do'stlaringiz bilan baham: