Menu
Home
Log in / Register
 
Home arrow Language & Literature arrow Word and paradigm morphology
Source

Learning without overgeneralization

The correspondence between features and markers is relatively transparent in Table 8.3. Yet, as discussed in Chapter 6.3.2, the contrasts that discriminate the

Table 8.3 Agreement properties and markers

SuBj

lsg

ipi

Obi

2sg

2pi

3p

isg

g—

g—t

v—

g-

ipl

g—t

g—t

v—t

g-/-t

2sg

m—

gv—

2pl

m—t

gv—t

—t

-t

3sg

m—s

gv—s

g—s

g—t

—s

-s

3pl

m—en

gv—en

g—en

g—en

—en

-en

m-

gv-

g-

g-/-t

forms of a transitive verb cannot be disassembled straightforwardly into a set of context-independent form-meaning relations, whether expressed as morphemes or as realization rules. Recall that the sub-paradigm of forms with ip object features are effectively discriminated, with the prefixal opposition between m- and gv- distinguishing lsg from lpl objects, and the suffixal opposition between 0, -t, -s and -en distinguishing subjects with 2sg, 2pl, 3sg and 3pl features. In particular, the contrast between 2sg and 2pl object features are unambiguously marked in the pair damxatav^dagvxatav by the opposition between m- and gv-. There is no motivation within this sub-paradigm for marking the second of these forms by -t, and hence no need for a constraint to ‘block’ this generalization. On the contrary, marking dagvxatav redundantly by -t would have the clearly harmful effect of eliminating the contrast with dagvxatavt, which is associated with 2pl subject features.

It is precisely with the goal of preventing the addition of a redundant -t that Anderson (1992:132) proposes extending a disjunctive ordering condition to apply across rule blocks. As discussed in Chapter 6.3.2, this extension allows lpl gv- to preempt a ‘general plural’ marker -t. The need for such an extension derives from the treatment of -t as a single marker introduced by a single realization rule Rt that spells out the intersection of the features associated with -t in all of the environments in which it occurs. This is of course an extreme analysis but the problem it raises, capturing context-dependent contrasts by context-independent rules, is endemic to a realizational approach.

Consider next the marking of ip subject properties in Table 8.3. In forms with (formally unmarked) 3p objects, ip subjects are marked by the prefix v-. In a realizational account, this pattern is expressed by a realization rule Rv that spells out ip subject features (however represented) by the marker v-. Just as the general plural rule Rt introduces -t whenever a feature bundle contains a plural feature, Rv introduces v- whenever a bundle contains ip subject features. This rule will have the desired effect in transitive forms with 3p objects, as well as in ‘absolute’ intransitive paradigms with a single subject argument (Harris 1981: i37f.). However, in transitive forms with 2p objects, Rv does not apply, since the prefixal slot is occupied by the 2p object marker g-. As discussed in Chapter 6.3.2, realizational approaches explore a variety of strategies to regulate the ‘competition’ between the rules introducing v- andg-.

From a learning-based perspective, invoking a notion of ‘rule competition’ appears to misconstrue the problem. There is no case in Tables 8.2 and 8.3 in which -t marks object plurality in the context of ip obiect features, and no case in which v- marks ip subject features in the context of 2P obiect features. The rules Rt and Rv ‘overapply’ in these contexts because they are overly general. Instead of characterizing the sets of cells in which -t and v- actually occur, the rules express context-independent associations between agreement markers and natural classes of cells (i.e., the class with plural features or with ip subject features). Hence the statement of the rules overgeneralizes the distribution of the markers that they are meant to describe.

The goal of‘resolving’ conflicts involving overgeneralized rules seems even more fundamentally misconceived. Most formal accounts of Georgian verb morphology attempt to provide some kind of principled account for the fact that the 2p object marker g- prevails over the ip subject marker v- in forms with 2p object and ip subject features. There may, again, be an illuminating historical explanation for this pattern but for a speaker of Modern Georgian, there is no indeterminacy in the resulting synchronic system. All forms with 2p object features are marked by g-, irrespective of the features of the subject. There is no ‘competition’ for a learner to resolve because v- never occurs with 2p objects. The same is true of the other cases of‘slot competition’. All forms with 3pl subjects are marked by-en, even when this prevents the expression of a contrast between 2sg and 2pl objects. In present paradigms, forms with 3sg subjects and 2pl objects are marked by -t, leaving the 3sg features unmarked.[1]

  • [1] Though syncretism is unavoidable, since a form with -s instead of-t would neutralize the contrastbetween 2sg and 2pl objects, as in the forms with 3pl subjects.
 
Source
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
 
Subjects
Accounting
Business & Finance
Communication
Computer Science
Economics
Education
Engineering
Environment
Geography
Health
History
Language & Literature
Law
Management
Marketing
Mathematics
Political science
Philosophy
Psychology
Religion
Sociology
Travel