[Author Login]
[Home]
ROA:267
Title:Lexical Stratification and Ranking Invariance in Constraint-based Grammars
Authors:Haruka Fukazawa, Mafuyu Kitahara, Mitsuhiko Ota
Comment:15 pages. "Other" file is a Mac WordPerfect 3.5 file
Length:15
Abstract: Lexical stratification and ranking invariance in

constraint-based grammars.



Haruka Fukazawa

University of Maryland at Collage Park



Mafuyu Kitahara

Indiana University



Mitsuhiko Ota

Georgetown University



Lexical stratification -- the partitioning of the lexicon into

distinct subsets with different phonological characteristics -- appears

to be at odds with the fundamental OT assumption that the grammar of

each language is represented by a single invariant ranking of

constraints. In fact, previous OT accounts of lexical stratification

have generally compromised ranking invariance by allowing, for instance,

faithfulness constraints to re-rank according to the sublexicon (Cf.

Ito^ and Mester 1995). To remedy this situation, Fukazawa (1997)

proposes a Correspondence Theoretic analysis of lexical stratification

which fully adheres to the concept of invariant ranking. In this model,

each sub-lexicon is posited to have its own unique Input-Output

Correspondence relation. Each IO-relation then bears a full set of

faithfulness constraints and all stratum-specific constraints are ranked

independently in the constraint hierarchy.



The main goal of the paper is to further demonstrate the empirical

advantages of this approach and to tackle some related conceptual

issues. Using Japanese as a domain of investigation, we will show that a

model of lexical stratification that employs multiple faithfulness

constraints not only fares better than the re-ranking approach from a

theory-internal point of view, but also covers a wider range of

empirical facts about the Japanese lexicon. The decisive evidence is

found in hybrid words which consist of elements from more than one

sublexicon. Our analysis shows that the well-formedness of such words

must be evaluated simultaneously against distinct faithfulness

constraints, each specific to the sublexicon to which the components of

hybrids belong. Without stratum-specific faithfulness constraints that

are separately rankable, attempts to obtain the same results run into

serious empirical problems.



We will further address two conceptual issues involved in the claim

that multiple projections of faithfulness constraints derive the effects

of lexical stratification. First, there are concerns that such a move

may lead to learnability problems and/or unrestricted projections of the

same type of constraints. We show how these problems can be circumvented

by employing simple learning procedures. The crux of the idea is that

the propagation of faithfulness constraints is touched off during the

course of acquisition by a specific type of data in the ambient

language. Since this process takes place only under pressure from the

data, it serves as an evaluation metric which prevents excessive

multiplication of faithfulness.



Second, our model opens up the possibility that stratum-specific

constraints can be ranked freely with respect to each other, allowing

some amount of flexibility in the relation between lexical strata.

Despite previous claims that lexical strata form a core-periphery

structure, we find evidence in the Japanese lexicon that this is not

necessarily the case -- an indication that there are no a priori

restrictions on the relative ranking of stratum-specific faithfulness

constraints across different FAITH types. This leads us to the conclusion

that the core-periphery organization is only a tendency that may have

resulted from the process of lexical assimilation, but not an inherent

property of the grammar.
Type:Paper/tech report
Area/Keywords:
Article:Version 1