In this thesis I define and evaluate from a cognitive perspective some fundamental properties of a computational model of linguistic competence that aims to be descriptively/explanatorily adequate, flexible (effectively usable in performance tasks such as parsing and generation) and realistic (tractable in computational terms and reasonable with respect to human performance data).
With the purpose of doing that, in the first chapter of this dissertation I introduce the linguistic concepts that are used in most frameworks, seeking to highlight their cognitive nature (essentially comparing processing at other cognitive levels, such as vision); in particular, consideration will be given to defining what features are, how they combine and which relations shall be defined among them; additionally some essential concepts will be provided in order to understand cost functions (complexity metrics) of the proposed devices.
Generative frameworks will be systematically reviewed in the first part of chapter 2 (Principle and Parameters / Extended Standard Theory, the Minimalist Program and a 'unusual' processing perspective for this framework, that is Phillip's (1996) left-to-right processing model; finally the Cartographical Approach will be explored). The second part of this chapter will show some (partially successful) attempts to implement/formalize these linguistic frameworks (the Principle-based parsing approach, Stabler's minimalist grammar formalization, and Fong's minimalist parser).
The rest of the thesis will try to solve many standing problematic issues: in order to provide a precise context, in chapter 3 I will firstly formalize the idea of structural description then the (performance) tasks that must access the grammar (essentially parsing and generation, which are specific cases of comprehension and production respectively). Afterwards, the most problematic aspects of the language (ambiguities and long distance dependencies) will be formalized with respect to the proposed performance tasks. Finally the formalization of a minimalist grammar (inspired by Stabler's 1997 work and the Minimalist Program) enriched with considerations on the articulate geometry of the functional feature structures proposed within the Cartographic Approach will be provided.
From the theoretical perspective, the standard minimalist approach sketches
a model that does not fit in a clear way with specific performance
algorithms such as parsing or generation even though much emphasis is put
on 'interface properties'. The grammar formalized considers some important
constraints posed both by the generation and parsing problems, in the end
defining a model that is:
- usable both in parsing and in generation;
- cognitively motivated;
- tractable;
- as much as possible deterministic.
To achieve these results I focus on three properties that are not only cognitively plausible, but also formally and computationally advantageous:
- structure building operations can be embedded within the grammar if they apply either top-down (in generation) or from-left-to-right (in parsing);
- using a Linearization Principle (inspired by Kayne's 1994 LCA) and fixing the functional structure by mean of a universal hierarchy (Cartographic Approach) makes the algorithm mostly deterministic;
- formalizing phases (cf. Chomsky 1999) helps us to make the algorithm (in relation to dealing with ambiguities and long distance relations) tractable.
The last chapter, chapter 4, will provide a specification of thealgorithm/grammar used, showing which empirical coverage this model can reach. In particular cases of Argumental movement, Criterial movement, Locality effects, Strong Island conditions, cross-serial dependencies, covert movement and parameter setting will be explored.
|
E-mail this message to a friend
