National Technical University of Athens (NTUA)
Μοντελοποίηση γλώσσας για συστήματα μηχανικής μετάφρασης με μονόγλωσσο σώμα κειμένων
Abstract
dc:descriptionThis PhD thesis aims at finding an effective way of modeling the language in order to be used in a Machine Translation system that utilises a monolingual target language (TL) corpus and also at optimising the parameters of the language model using evolutionary algorithms and a combination of automatic MT evaluation metrics that ensure an objective evaluation of the quality of the produced translation. In the first section of the thesis the language model, model for the translation system was presented. The model allows the mapping of a clause in the source language (SL) to clauses from the TL corpus through a complex clause comparison process on two levels the first is at word level and the second at phrasal level based on the phrases identified after the syntactic analysis of the clauses SL and TL clause comparisons are performed using an implementation of the Hungarian pattern matching algorithm combined with a series of numerical parameters of the translation model. This approach does not require any bilingual data but only relies on a TL corpus and a bilingual lexicon combining characteristics from various MT paradigms rule based and stochastic tools for SL and TL text processing and statistical information for the translation process. More specifically translation involves determining alignments between items in the two languages. Unlike many MT systems instead of sequences of η grams the system handles syntactically defined phrasal segments of varying length generated by syntactic parser. We begin by giving a detailed description of the METIS-II translation system as well as the representation and storage method of all information regarding the monolingual corpus followed by the description of the process of applying the language model to the translation system. The evaluation of the translation results is performed by applying well established metrics of automatic evaluation and comparing the results with a standard commercial machine translation system. In the second part of the thesis we present the optimisation method of the model parameters. The absence of a parallel corpus prohibits the use of the training techniques traditionally employed in state of the art Statistical Machine Translation systems. The proposed approach for fine tuning the system parameters towards the generation of high quality translations is based on Evolutionary Algorithms and more specifically on a Genetic Algorithm and on the Strength Pareto Evolutionary Algorithm 2 (SPEA2). Even though the nature of Evolutionary Algorithms makes them ideal for the task they have not been widely used for the optimisation of MT system parameters. In order to confirm the effectiveness of this approach a number of experiments were conducted involving the use and combination of modern automatic MT evaluation criteria as fitness functions to the evolutionary algorithms. In this approach the problem of MT parameter optimisation is treated as a multiobjective optimisation problem attempting to optimise system parameters by applying multiple MT evaluation metrics. Previous approaches in the simultaneous application of two MT evaluation metrics for the training and optimisation of MT systems combined them either by simply adding them up or by interpolating them and had only been tested in statistical systems. The methodology developed here combines two automatic evaluation metrics for the optimisation of the system parameters in two ways, a) as a weighted sum and alternatively, b) as two distinct criteria in a Pareto based multiobjective optimisation using the SPEA2 evolutionary algorithm. The optimisation of the parameters of a hybrid MT system combining multiple MT evaluation criteria using either a GA or a Pareto based evolutionary multiobjective algorithm such as SPEA2 is novel in the field of MT while to the best of our knowledge a Pareto based combination has never been applied to either the combination of translation quality criteria or to the task of optimising the parameters of an MT system. The presentation of the proposed optimisation method is followed by a series of comprehensive experiments designed to objectively evaluate the method.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2009
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Sofianopoulos, Sokratis
- Σοφιανόπουλος, Σωκράτης
Subjects
dc:subject × 14- Μηχανική μετάφραση
- Αναγνώριση προτύπων
- Γενετικοί αλγόριθμοι
- Βελτιστοποίηση παραμέτρων
- Πολυκριτηριακοί εξελικτικοί αλγόριθμοι
- Machine translation
- Pattern matching
- Genetic algorithms
- Parameter optimisation
- Multiobjective evolutionary algorithms
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/23002