Professional Documents
Culture Documents
Modeling Phonological Competence Coursework Notes 7: Dangers and Deficiencies in The Model
Modeling Phonological Competence Coursework Notes 7: Dangers and Deficiencies in The Model
So far so good. But the system we have created is extremely (overly) powerful.
You can easily be more extreme. In an account of Spanish, James Foley proposes that the
underlying representation of 'es' (BE 3rd sing present) is /s+e+twe/ and that of 'amó' (LOVE
3rd sing perfect) is /am+a++m+twe/.
4. Constraints on abstractness
This problem of the 'abstractness' of underlying forms has not surfaced before now, partly
because we have tacitly assumed that the underlying form of a lexical element is the same as
one of its surface alternants. In other words, we have been imposing (intuitively) some form
of constraint on the way in which related deep and surface structures may diverge. You may
think, therefore, that a simple answer to the abstractness problem is to make some such
constraint explicit. But it is worth stopping for a moment to think over the consequences.
Effectively, this means that the base form is the form occuring in morphologically simple
word forms – uninflected, underived bases. (On the grounds that such environments are the
most 'neutral'.) But how about the effects of word-final neutralisation in Catalan?
(ii) The basic form is always a member of the same grammatical category
This would allow us to choose feminine singular forms as basic in Catalan. But then how
about Russian which shows not only neutralisation of stop contrasts (p:b) word finally but
ALSO neutralisation of the contrast between /a/ and /o/ in unstressed syllables?
CUT picena; HOE notoxo; LICK netale (none of which appear on the surface)
cf Russian HEAD galavá nom sg; gólavu acc sg; galóf gen pl – hence basic form golov.
(iv) Each of the constituent SEGMENTS of the basic form appears in some alternant
We are now getting pretty relaxed. But how about Turkish?
Turkish
NOM kap kep tat at yt sat merak inek renk
GEN kab kepi tad at ydy sat merak inei reni
The alternation inek ~ ine and the absence of VgV suggests an underlying form /ineg/!
(v) Each of the constituent segments of the basic form appears contrastively in at least some
forms in the language
This is now quite weak. The aim is to exclude pseudo-phonological segments being used to
mark superficially similar forms which alternate differently from one another. COUGH
plural is /kofs/ but LEAF plural is /livz/. Why? An answer might be that COUGH has the
underlying representation /kof/ but leaf has the underlying representation /li/ But obviously
/f/ and // never contrast in any environment in English. To rectify this on the surface we
would need a process of absolute neutralization - which might well be outlawed on principle.
But even then there are at least plausible cases which we might not want to exclude. Does
French have an underlying /h/ to account for /ariko/ behaving like a consonant-initial word?
les souris les animaux les haricots les habitants
/le suri/ /lez animo/ /le ariko/ /lez abitã/
5. How well founded are patterns of alternation?
In the light of the above examples it is clearly difficult to impose restrictions tighter than the
last without invalidating many commonly accepted analyses. However, there is perhaps
another tactic we could use to ensure an appropriate distance betweeen surface and
underlying forms - access native speaker intuitions directly (experimentally). We might, for
example, reject base forms whose construction depends on the acceptance of alternations
which speakers do not recognize as such?
Are English speakers aware, for example, of a phonological relationship between 'sign' and
'significance' or 'night' and 'nocturnal' or 'moon' and 'month'? Do French speakers connect
'oeuf' and 'ovipare'? Do Spanish speakers identify the roots of 'leche' and 'lactal'? All of these
pairings might well be used to motivate underlying forms – but should they be?
In fact, nothing in the structure of our model stops us from expressing processes which are
completely unnatural. For example, we could just as easily have used [+lab] as the change in
the rule above and come up with a rule allowing the counterfactual gaan#Æ gaam#
The problem is that the while our model operates with phonetic features, it deals with them
as if they had no phonetic meaning, no real connection with physical events. They are treated
simply as abstract symbols which can be freely manipulated. Switch every feature value in a
set of phonological rules - the results might be crazy but the system would still 'work'.
But before accepting that proposition without question, look at the following Somali forms
deliberately excluded from the dataset of Coursework Notes 2. Do we want to insist that all
phonological rules must be natural? What about the processes which convert underlying
holta to hooa , especially the chane lt Æ ?
Singular Sing Def Plural
downpour hool hooa holo
hole bohol bohoa boholo
month bil bia bilo
mule baal baaa balo
place meel meea meelo