[Author Login]
Title:Robust Interpretive Parsing in Metrical Stress Theory
Authors:Bruce Tesar
Abstract:Robust Interpretive Parsing in Metrical Stress Theory

Bruce Tesar

Rutgers University

Most computational work to date within Optimality Theory (Prince &

Smolensky 1993) has focused on generation, the mapping from an

underlying form to its full structural description (Eisner (1997),

Ellison (1995), Frank & Satta (to appear), Tesar (1995)). While that

is a natural starting place, it is not the only function of interest

related to an OT grammar. Language comprehension involves the

interpretation of overt forms, forms consisting of the auditory

information directly available to a listener. The function is not the

inverse of generation, but the mapping from an overt form to its full

structural description. The process of computing this latter function

is here called interpretive parsing.

This paper presents an efficient algorithm for the interpretive

parsing of forms using an optimality theoretic system of metrical

stress grammars. The system has 8 freely rankable constraints, and

can account for a wide variety of metrical phenomena. The full

structural descriptions include foot structure and the assignment of

stress levels to the syllables, while the overt forms include the

syllables and their stress levels, but not the foot structure. Overt

forms are inherently ambiguous; the same pattern of stress levels is

consistent with multiple foot structures. Interpretive parsing must

recover the correct foot structure from the overt form, using the

constraint ranking of the grammar. The algorithm is based upon the

dynamic programming approach of Tesar, but includes non-trivial

extensions to efficiently deal with the non-localities of metrical


Interpretive parsing poses an interesting issue that does not arise in

generation: the issue of dealing with overt forms that are ungramma-

tical according to the learner's current grammar. By definition, an

optimality theoretic grammar assigns a description to every possible

input. But there can be overt forms which do not correspond to any

optimal structural description of a grammar. The approach presented in

this paper assigns the best possible interpretation to an overt form,

whether it is grammatical or not. Thus, the procedure is called robust

interpretive parsing.

Robust interpretive parsing has great significance for language

learning. A learner in the process of learning their native language

does not yet have the correct grammar, and thus needs to be able to

interpret the form even when it is ungrammatical with respect to their

current grammar. Robust interpretive parsing allows a child to use what

they have already learned to estimate a best interpretation of overt

forms, which they can then use to perform further learning. Significant

learning results, reported elsewhere, have already been obtained using

this approach to learning.

This paper will appear in the Proceedings of the WCCFL XVII.
Type:Paper/tech report
Article:Part 1
Part 2
Part 3
Part 4
Part 5
Part 6
Part 7
Part 8
Part 9
Part 10
Part 11
Part 12
Part 13