PRESEMT Phrase Model Generator

by Michalis Troullinos, November 2013, 18 pages.

FIMU-RS-2013-3. Available as Postscript, PDF.

Abstract:

The PRESEMT (Pattern REcognition-based Statistically Enhanced MT) project is intended to lead to a flexible and adaptable MT system, based on a language-independent method, whose principles ensure easy portability to new language pairs. This report describes the Phrase Model Generator (PMG) of the PRESEMT system. PMG supports two distinct operations. The first operation processes the output of Phrase Aligner Module to train a phrasing model for the SL of the specified language pair. The second operation makes use of the phrasing model established to parse any SL text input and split it into phrases in preparation for the translation process.

Phrase Aligner

by Michalis Troullinos, November 2013, 44 pages.

FIMU-RS-2013-2. Available as Postscript, PDF.

Abstract:

The PRESEMT (Pattern REcognition-based Statistically Enhanced MT) project is intended to lead to a flexible and adaptable MT system, based on a language-independent method, whose principles ensure easy portability to new language pairs. This report describes the Phrase aligner module (PAM) of the PRESEMT system. PAM processes the bilingual corpora by performing text alignment at word and phrase level within a language pair. It operates in offline manner, processing the set of parallel sentences so as to determine how phrases are transformed from SL to TL.

Dynamic maintenance of an accepting run

by Florent Peres, Ivana Černá, August 2013, 46 pages.

FIMU-RS-2013-01. Available as Postscript, PDF.

Abstract:

Finding an infinite sequence of transitions (a run) going through some special states (called accepting) is of importance for fields like formal verification. Whereas finding such a sequence as been extensively studied, the problem of maintaining this sequence upon changes in the model or the specification has received less attention. In this work, we propose a solution to the maintenance of an accepting run when the transition system representing the product of the model and of the specification is changed, using the Tarjan`s algorithm as a base algorithm.