Chapter 1: Preliminaries 1
1.1 Challenges in MT - Relevance to the European environment 1
1.2 A brief review of MT development history. 2
1.3 Advantages and disadvantages of main MT paradigms 3
1.4 The PRESEMT methodology in a nutshell 7
1.5 Closing note on implementation. 9
1.6 References 9
1.7 Glossary of Terms 12
Chapter 2: Implementation. 14
2.1 Introduction: Summary of the approach. 14
2.2 Linguistic resources: Data and existing linguistic tools 15
2.2.1 External processing tools 16
2.2.2 Lemma-based bilingual dictionary. 17
2.2.3 The parallel corpus 19
2.2.4 The TL monolingual corpus 22
2.3 Processing the parallel corpus 22
2.3.1 Phrase Aligner Module. 23
2.3.2 Phrasing Model Generation. 28
2.4 Creating a language model for the target language. 30
2.5 References 32
Chapter 3: Main translation process 34
3.1 Introduction. 34
3.2 Translation Phase one: Structure Selection. 35
3.2.1 The Dynamic Programming algorithm.. 37
3.2.2 Example of how Structure Selection works 39
3.3 Phase two: Translation Equivalent Selection. 40
3.3.1 Applying the language model to the task. 42
3.3.2 Example of how TES works 44
3.4 References 45
Chapter 4: Assessing PRESEMT. 47
4.1 Evaluation dataset 47
4.2 Objective evaluation metrics 48
4.3 System evaluation. 49
4.3.1 Evaluation objectives 49
4.3.2 Evaluation results 50
4.3.3 Expanding the comparison. 51
4.3.4 Experimenting with further data. 52
4.4 Comparing PRESEMT to other MT systems 53
4.5 Conclusions 56
4.6 References 57
Chapter 5: Expanding the system.. 58
5.1 Preparing the system for new language pairs 58
5.2 Examining language-pair-specific issues 60
5.2.1 Agreement within a nominal phrase. 60
5.2.2 Case mismatches 61
5.2.3 The null subject parameter 61
5.2.4 Word order 62
5.3 Notes on implementation. 62
5.4 Conclusions 63
5.5 References 63
Chapter 6: Extensions to the PRESEMT methodology. 64
6.1 Splitting SL sentences into phrases more accurately. 64
6.1.1 Design and implementation of TEM.. 65
6.1.2 Experimental evaluation. 68
6.1.3 Conclusions 70
6.2 Combining language models of different granularity. 71
6.2.1 Extracting the n-gram models 72
6.2.2 Experimental results 74
6.2.3 Discussion. 75
6.3 References 76
Chapter 7: Conclusions and future work. 78
7.1 Review of the effectiveness of the PRESEMT methodology. 78
7.2 Likely avenues for improvements in translation quality. 79
7.2.1 Automatic enrichment of dictionary. 79
7.2.2 Design and implementation of TEM.. 80
7.2.3 Grouping of tokens and PoS tags into related classes 81
7.2.4 Revision of the Structure Selection translation phase. 81
7.2.5 Improving the alignment of words/phrases 82
7.2.6 Augmenting the TL language model to cover supra-phrasal segments 83
7.2.7 A closing evaluation of translation accuracy. 84
7.3 References 85
About the Author: George Tambouratzis graduated from the Electrical Engeneering Department of the National Technical University of Athens (1989), and received his M.Sc. (1990) and Ph.D. (1993) degrees from Brunel University. Since 1996 he has been with the Institute for Language and Speech Processing (ILSP), working on machine learning, neural networks and evolutionary computation algorithms for computational linguistics. He is the Director of Research at ILSP and the Head of the Machine Translation Department. He co-ordinated several EU-funded projects.
Marina Vassiliou studied Linguistics and holds a Master's degree in Generative Syntax from the University of Athens. As a research associate at ILSP since 2000 she has worked on various, mainly European, research projects concerning specifications for syntactic analysis, machine translation, stylometry, controlled languages, multilingual thesauri and business ontologies as well as the development of a coreference resolution system for Greek language.
Sokratis Sofianopoulos graduated from the University of Ioannina in 2002 and holds a M.Sc. from Heriot-Watt University (2003) and a PhD from the National Technical University of Athens (2010). Since 2005 he is a research associate at ILSP. He has worked in several European R&D programs in the field of NLP and machine translation, such as METIS-II (FP6-IST-003768), PRESEMT (FP7-ICT-248307), QTLaunchPad (FP7-ICT-296347).
