Natural Language Modelling and Parsing with Typed Unification Grammars

Joachim Niehren

provides
x-ozlib://niehren/typed-unif-grammar/parser.ozf
x-ozlib://niehren/typed-unif-grammar/convert.ozf
x-ozlib://niehren/typed-unif-grammar/test-parser.oz
x-ozlib://niehren/typed-unif-grammar/test-converter.oz
x-ozlib://niehren/typed-unif-grammar/syntax.ozf
x-ozlib://niehren/typed-unif-grammar/types.ozf
x-ozlib://niehren/typed-unif-grammar/grammar.ozf
x-ozlib://niehren/typed-unif-grammar/lexicon.ozf
x-ozlib://niehren/typed-unif-grammar/sentences.ozf
x-ozlib://niehren/typed-unif-grammar/map.ozf
x-ozlib://niehren/typed-unif-grammar/type-check.ozf
requires
mogul:/niehren/base
mogul:/niehren/assignment
mogul:/niehren/error
mogul:/niehren/context-free-grammar
mogul:/niehren/unif-grammar

Purpose

Typed unification grammars extend context free grammars so that they become useful for natural language modelling. This package provides parsers for typed unifications grammars and converter to context free grammars (for which the types are essential).

The package support atomic types, record types, disjunctive types, and recursive types. A record type defines the features that syntactical entity may record. This permits to omit features in the rules, which are then automatically inherited before parsing.

The parser is obtained by reduction to parsing with unification grammars as described in the Oz-course, which was offered repeatedly since 1998 by Joachim Niehren, Denys Duchier and Claire Gardent at the departments of computational linguistics and computer sience of the University at Saarbrücken.

Converting a typed unification grammar into a context free grammar means to compute all ground instances of the grammar rules and lexicion entries. This set of possible ground instances is defined with respect to the types. Conversion to a context free grammar terminates if and only if there are finitely many ground instance for all rules and lexicon entries.

The actual package is in practial use in a spoken language control and information dialog system which was develloped for the elevator at the department of computational linguistics in Saarbrücken. It serves for creating a context free grammar as required as input of speech recognition systems with bounded language coverage.

Installation

Download the file niehren-typed-unif-grammar-1.0.pkg, invoke ozmake and execute in a shell ozmake --install --package=niehren-typed-unif-grammar-1.0 ozmake --extract --package=niehren-typed-unif-grammar-1.0 The second command will provide the necessary file in the current directory.
You can test it out from the Mozart Emacs user environment by feeding test.oz.


Joachim Niehren