Cover -- Title Page -- Copyright -- Contents -- Acknowledgments -- Introduction -- PART 1: Applicative and Scientific Context -- Chapter 1: Leveraging Comparable Corpora for Computer-assisted Translation -- 1.1. Introduction -- 1.2. From the beginnings of machine translation to comparable corpora processing -- 1.2.1. The dawn of machine translation -- 1.2.2. The development of computer-assisted translation -- 1.2.3. Drawbacks of parallel corpora and advantages of comparable corpora -- 1.2.4. Difficulties of technical translation -- 1.2.5. Industrial context -- 1.3. Term alignment from comparable corpora: a state-of-the-art -- 1.3.1. Distributional approach principle -- 1.3.2. Term alignment evaluation -- 1.3.2.1. Precision at rank N or TopN -- 1.3.2.2. MRR -- 1.3.2.3. MAP -- 1.3.3. Improvement and variants of the distributional approach -- 1.3.3.1. Favoring distributional symmetry -- 1.3.3.2. Using syntactic contexts -- 1.3.3.3. Relying on trusted elements -- 1.3.3.4. Improving semantic information representation -- 1.3.3.5. Using second-order semantic affinities -- 1.3.3.6. Improving the bilingual resource with semantic classes -- 1.3.3.7. Translating polylexical units -- 1.3.4. Influence of data and parameters on alignment quality -- 1.3.4.1. Data -- 1.3.4.2. Parameters -- 1.3.5. Limits of the distributional approach -- 1.4. CAT software prototype for comparable corpora processing -- 1.4.1. Implementation of a term alignment method -- 1.4.1.1. Implementation and data -- 1.4.1.2. Extraction of the terms to be aligned -- 1.4.1.3. Collecting context vectors -- 1.4.1.3.1. Monolexical term context vectors -- 1.4.1.4. Polylexical term context vectors -- 1.4.1.5. Translation of the source context vectors -- 1.4.1.6. Term alignment -- 1.4.2. Terminological records extraction -- 1.4.3. Lexicon consultation interface -- 1.5. Summary.

Chapter 2: User-Centered Evaluation of Lexicons Extracted from Comparable Corpora -- 2.1. Introduction -- 2.2. Translation quality evaluation methodologies -- 2.2.1. Machine translation evaluation -- 2.2.1.1. Automatic evaluation measures -- 2.2.1.2. Human MT evaluation -- 2.2.2. Human translation evaluation -- 2.2.2.1. Quantitative models -- 2.2.2.2. Non-quantitative models -- 2.2.3. Discussion -- 2.3. Design and experimentation of a user-centered evaluation -- 2.3.1. Methodological aspects -- 2.3.1.1. Evaluation criteria and purpose -- 2.3.1.2. Subject matter expertise -- 2.3.1.3. Basis for comparison -- 2.3.2. Experimentation protocol -- 2.3.2.1. Data -- 2.3.2.1.1. Comparable corpora and extracted lexica -- 2.3.2.1.2. Texts to be translated -- 2.3.2.1.3. Resources used in the translation situation -- 2.3.2.1.4. Translators and judges -- 2.3.2.2. Evaluation progress -- 2.3.2.2.1. Translation phase -- 2.3.2.2.2. Translation quality evaluation phase -- 2.3.3. Results -- 2.3.3.1. Lexicons usability -- 2.3.3.1.1. Translation speed -- 2.3.3.1.2. Use of resources -- 2.3.3.1.3. Translators' impressions on the lexicons extracted from comparable corpora -- 2.3.3.2. Quality of the generated translations -- 2.3.3.2.1. Inter-annotator agreement -- 2.3.3.2.2. Judgment task -- 2.3.3.2.3. Ranking task -- 2.3.3.3. Lexicon coverage -- 2.3.3.4. Reproducing the protocol on a wider scale -- 2.4. Discussion -- Chapter 3: Automatic Generation of Term Translations -- 3.1. Introduction -- 3.2. Compositional approaches -- 3.2.1. Compositional translation principle -- 3.2.2. Polylexical units compositional translation -- 3.2.2.1. Lexical variation and multiple decomposition -- 3.2.2.2. Morphological relations -- 3.2.2.3. Extracting translations from mixed documents -- 3.2.2.4. Bag of equivalents -- 3.2.2.5. Hybridization with distributional method.

3.2.3. Monolexical units compositional translation -- 3.2.3.1. Translation of words formed by means of prefixation -- 3.2.3.2. Translation of words formed by means of neoclassical compounding -- 3.2.3.3. Translation of word compounds -- 3.2.4. Candidate translation filtering -- 3.2.4.1. Looking for an attestation -- 3.2.4.2. Filtering based on context similarities -- 3.2.4.3. Supervised learning filtering -- 3.3. Data-driven approaches -- 3.3.1. Analogy-based translation -- 3.3.2. Rewriting rules learning -- 3.3.3. Dealing with morphological variation -- 3.4. Evaluation of term translator generation methods -- 3.5. Research perspectives -- PART 2: Contributions to Compositional Translation -- Chapter 4: Morph-Compositional Translation: Methodological Framework -- 4.1. Introduction -- 4.2. Morpho-compositional translation method -- 4.2.1. Scientific positioning -- 4.2.2. Definitions and terminology -- 4.2.2.1. Polylexical units -- 4.2.2.2. Monolexical unit -- 4.2.2.3. Word -- 4.2.2.4. Complex word -- 4.2.2.5. Morpheme -- 4.2.2.6. Simple word or free morpheme -- 4.2.2.7. Bound morpheme -- 4.2.2.8. Prefix -- 4.2.2.9. Combining forms -- 4.2.2.10. Suffix -- 4.2.2.11. Notations -- 4.2.3. Underlying assumptions -- 4.2.3.1. Compositional meaning -- 4.2.3.2. Compositional translation -- 4.2.3.3. Fertility -- 4.2.3.4. Distortion -- 4.2.3.5. Lexical divergence -- 4.2.3.6. Morphological variation -- 4.2.4. Advantages of the proposed approach for processing comparable corpora -- 4.3. Issues addressed and contributions -- 4.3.1. Generating fertile translations -- 4.3.1.1. Semantic fertility -- 4.3.1.2. Surface fertility -- 4.3.2. Dealing with diverse morphological structures -- 4.3.2.1. Word compounding composition -- 4.3.2.2. Neoclassical compounding -- 4.3.2.3. Prefixation -- 4.3.2.4. Suffixation -- 4.3.3. Candidate translations ranking.

4.4. Evaluation methodology -- 4.4.1. A priori reference -- 4.4.2. A posteriori reference -- 4.5. Conclusion -- Chapter 5: Experimental Data -- 5.1. Introduction -- 5.2. Comparable corpora -- 5.3. Source terms -- 5.4. Reference data for translation generation evaluation -- 5.4.1. A priori reference -- 5.4.2. A posteriori reference -- 5.4.2.1. Exact -- 5.4.2.2. Acceptable -- 5.4.2.3. Close -- 5.4.2.4. False -- 5.5. Translation ranking training and evaluation data -- 5.6. Linguistic resources -- 5.6.1. General language bilingual dictionary -- 5.6.2. Thesaurus -- 5.6.3. Bound morphemes translation table -- 5.6.4. Lexicon for word decomposition -- 5.6.5. Morphological families -- 5.6.6. Dictionary of cognates -- 5.7. Summary -- Chapter 6: Formalization and Evaluation of Candidate Translation Generation -- 6.1. Introduction -- 6.2. Translation generation algorithm -- 6.2.1. Decomposition -- 6.2.1.1. Morphological decomposition (SPLIT) -- 6.2.1.2. Morpheme concatenation (CONCATENATE) -- 6.2.2. Translation -- 6.2.3. Recomposition -- 6.2.3.1. Permutation of translated elements (PERMUTATE) -- 6.2.3.2. Concatenation into words (CONCATENATE) -- 6.2.3.3. Filtering (FILTER) -- 6.2.4. Selection -- 6.2.4.1. Definition of a candidate translation -- 6.3. Morphological splitting evaluation -- 6.4. Translation generation evaluation -- 6.4.1. Reference data and evaluation measures -- 6.4.1.1. A posteriori reference -- 6.4.1.1.1. Coverage (C) -- 6.4.1.1.2. Precision (P) -- 6.4.1.1.3. Usability (U) -- 6.4.1.1.4. Results obtained -- 6.4.1.2. A priori reference -- 6.4.1.2.1. Precision (P) -- 6.4.1.2.2. Recall (R) -- 6.4.1.2.3. F1-measure (F1) -- 6.4.1.2.4. Results obtained -- 6.4.2. Model genericity influence -- 6.4.2.1. A posteriori evaluation -- 6.4.2.2. A priori evaluation -- 6.4.2.3. Synthesis -- 6.4.3. Linguistic resources influence -- 6.4.3.1. A posteriori evaluation.

6.4.3.2. A priori evaluation -- 6.4.3.3. Synthesis -- 6.4.4. Fallback strategy influence -- 6.4.4.1. A posteriori evaluation -- 6.4.4.2. A priori evaluation -- 6.4.4.3. Synthesis -- 6.4.5. Fertile translations influence -- 6.4.5.1. A posteriori evaluation -- 6.4.5.2. A priori evaluation -- 6.4.5.3. Synthesis -- 6.4.6. Popular science corpus influence -- 6.4.6.1. Fertility and discourse type -- 6.4.7. Qualitative analysis -- 6.4.7.1. Silence analysis -- 6.4.7.2. Noise analysis -- 6.5. Discussion -- 6.5.1. Findings -- 6.5.1.1. A priori versus a posteriori reference -- 6.5.1.1.1. Comparing language pairs -- 6.5.1.1.2. Fertile translations -- 6.5.2. Research perspectives -- 6.5.2.1. Improving linguistic resources -- 6.5.2.2. Fertile translations -- Chapter 7: Formalization and Evaluation of Candidate Translation Ranking -- 7.1. Introduction -- 7.2. Ranking criteria -- 7.2.1. Context similarity -- 7.2.2. Candidate translation frequency -- 7.2.3. Parts-of-speech translation probability -- 7.2.4. Components translation mode -- 7.3. Criteria combination -- 7.3.1. Value standardization -- 7.3.2. Linear combination -- 7.3.3. Learning-to-rank model -- 7.4. Evaluation -- 7.4.1. Reference data and evaluation measures -- 7.4.2. Bases of comparison -- 7.4.3. Results -- 7.4.3.1. Ranking English ---> French translations -- 7.4.3.2. Ranking English ---> German translations -- 7.5. Discussion -- 7.5.1. Findings -- 7.5.1.1. Usefulness of criteria combination -- 7.5.1.2. Inconclusive results of the methods based on training -- 7.5.2. Research perspectives -- 7.5.2.1. Language and domain independent models -- 7.5.2.2. Better suited evaluation methods -- 7.5.2.3. Identification of correct translations -- 7.5.2.4. Combining different methods -- 7.5.2.5. Looking for new ranking features -- Conclusion and Perspectives -- PART 3: Appendices -- Appendix 1: Measures.

Appendix 2: Data.