Sequentialized Language Models

We then turn to construction of a sequentialized grammatical model of linguistic objects in text compression. We develop the Prediction by Grammatical Match technique, a new compression framework employing a static context-free grammar and an adaptive finite-context statistical model. These compressors are adaptive, general compressors that operate in linear time and bounded space. We show these compressors can deliver substantial reductions in both bits-per-character rates and space usage, and suffer almost no penalty when the grammar does not apply. The new technique rests on three primary technical innovations: an algorithm for designing an optimal, strictly bottom-up parseable metalanguage for a compression scheme comprising multiple grammars; a principled approach to ambiguity and agrammatical text; and an incremental analysis selection algorithm. The metalanguage construction emphasizes lexical left-corner analysis descriptions, with each symbol in a description representing a maximal bundle of bottom-up and top-down information by naming the production introducing the next lexical left-corner item. These three innovations combine into a very powerful compression system that solves an important, long standing problem: efficient and effective use of context-free grammars in general data compression.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• Lake, John Michael

Contributors dc:contributor

• DeJong, Gerald F.

• Computer Science

Language dc:language

eng

Identifier

(MiAaPQ)AAI9990052

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/81987