MULTIWORD UNITS IN MACHINE TRANSLATION AND TRANSLATION TECHNOLOGY
Editorial page
Title page
LCC data
Table of contents
About the editors
Multiword units in machine translation and translation technology
1. Introduction
2. Multiword units in natural language processing
2.1 Historical notes
2.2 POS tagging and parsing
2.3 Word sense disambiguation
2.4 Information extraction and information retrieval
2.5 Other applications
3. Multiword unit processing in machine translation
3.1 Historical notes
3.2 Multiword unit processing in RBMT
3.3 Multiword unit processing in EBMT
3.4 Multiword unit processing in SMT
4. Multiword units in translation technology
References
Analysing linguistic information about word combinations for a Spanish-Basque rule-based machine tra
1. Introduction
2. Definitions, challenges and treatment of MWUs in MT
3. Linguistic analysis of Basque and Spanish noun + verb combinations
3.1 Noun + verb combinations in bilingual dictionaries
3.1.1 Basque and Spanish noun + verb combinations in the dictionary
3.1.2 Translations of noun + verb combinations in the dictionary
3.1.3 Equivalences of noun + verb constructions in translations
3.2 Contrasting information with parallel corpora
3.3 Classification of the Spanish MWUs
3.3.1 Syntactic flexibility
3.3.2 Semantic compositionality
4. Evaluation of MWU detection and translation adequacy
4.1 Evaluation of MWU detection
4.2 Evaluation of MWU translation quality in an RBMT system
5. Conclusions and future work
Acknowledgements
References
How do students cope with machine translation output of multiword units? An exploratory study
1. Introduction
2. Experimental set-up
3. Analysis
4. Conclusion
5. Future work
References
Aligning verb + noun collocations to improve a French-Romanian FSMT system
1. Context and motivation
2. Handling MWEs for MT
3. Collocation definition
4. Translation problems
5. The Architecture of the FSMT system and verb + noun collocation integration
6. Preprocessing Verb + Noun collocations
7. The MWE dictionary
8. The collocation alignment algorithm
9. Experiments
9.1 MWEs and the lexical alignment system
9.2 MWEs and FSMT system
9.3 MWE identification before aligning
10. Conclusions and future work
References
Multiword expressions in multilingual information extraction
1. Introduction
2. Application context
3. MWEs in multilingual information processing
3.1 MWE extraction
3.2 MWE lexical representation
3.2.1 Design
3.2.2 The lexicon
3.2.2.1 General annotations. Such annotations refer to the definition of a lexicon entry, and to information needed for processing.
3.2.2.2 MWE extensions. Multiword expressions share the above-mentioned annotation with the other lexicon entries. The entry definition features are as follows: the value of the lemma feature is the lemma of the multiword (e.g. take into account); the va
3.3 MWE analysis and identification
3.3.1 Design
3.3.1.1 MWE processing after analysis. Early attempts in MT systems like LMT (McCord, 1989) or METAL (Thurmair, 1990) only analysed single words, and used the transfer component for MWE treatment: The transfer lexicon contained entries with tests on the
3.3.1.2 MWE processing before analysis. An alternative approach is to do MWE processing before the analysis. A special component looks over the input string and marks MWE candidates. Analysis then starts with such MWEs as single nodes in the input.
3.3.2 MWE treatment in analysis
3.3.2.1 Preprocessing. Incoming texts are deformatted, and then split into sentences. Next, instead of tokenisation, an allomorph-based segmentation component is used; as already mentioned, tokenisation would address the fact that there is much more varia
3.3.2.2 Chart initialisation. The process of chart initialisation looks up each allomorph in the lexicon. The lexicon is compiled such that all allomorphs of an entry point to the entry in a hash table. In the case of MWEs, an allomorph instantiates all
3.3.2.3 Analysis design. The analysis component is designed as an X-bar scheme, extended such that the MWE readings are integrated in the earliest possible steps. Both single words and multiwords can take modifiers and specifiers to form XP constituents;
3.3.2.4 Multiword analysis rules. Multiwords are processed on the X1 level. But before the multiword rules fire, basic morphology rules are applied (on the X0 level) which attach inflections to the lemmata, as in No → No N-Flex and Vb → Prefix Vb V-Flex.
3.3.2.5 Analysis output. As the system does not intend to do full parsing, the output structure is a flat collection of XP constituents under an S node, created by recursive rules like S → S XP. The XP nodes can contain named entities, terms (built from s
3.4 MWE translation and generation
3.4.1 Transfer
3.4.2 Generation
4. Evaluation
4.1 MWE coverage: rule – lexicon compatibility
4.2 Lexicon coverage
5. Conclusion
Acknowledgements
References
A multilingual gold standard for translation spotting of German compounds and their corresponding multiword units in English, French, Italian and Spanish
1. Introduction
1.1 Related work
2. Resources
2.1 Selection and preprocessing of the gold standard material
3. Evaluation and discussion
3.1 Quality of universal part-of-speech (UPOS) tagging
3.2 Aligned UPOS tags across languages
3.3 Complexity of the compounds and aligned MWUs
3.4 Evaluation of the quality of the automatic GIZA++ word alignment
3.5 Optimisation of the directed word alignments through symmetrisation
3.6 Frequency effects
3.7 Effects of morphological complexity
3.8 Lexicalisation and variability
4. Conclusion
Acknowledgements
References
Appendix
Dutch compound splitting for bilingual terminology extraction
1. Introduction
2. Dutch compound splitter
2.1 Domain adaptation
2.2 Data Sets and Experiments
3. Impact on word alignment
3.1 Data sets and experiments
4. Impact on terminology extraction
4.1 Experiments
5. Conclusion
References
A flexible framework for collocation retrieval and translation from parallel and comparable corpora
1. Introduction
2. Phraseology
2.1 Typologies of collocations
2.2 Transfer rules
3. Related work
3.1 Collocation retrieval
3.2 Parallel corpora
3.3 Comparable corpora
4. System
4.1 Candidate selection module
4.2 Candidate filtering module
4.3 Dictionary look-up module
4.4 Parallel corpora module
4.5 Comparable corpora module
5. Evaluation
5.1 Experimental setup
5.2 Experimental results
5.3 Discussion and future work
Acknowledgements
References
On identification of bilingual lexical bundles for translation purposes
1. Introduction
2. Background and related work
3. Research material and methodology
4. Results
5. Discussion and conclusions
References
Appendix 1. Pivot LBs in EPILs
The quest for Croatian idioms as multiword units
1. Introduction
2. Theoretical background
2.1 Idioms as a type of MWU in Croatian language
2.2 Importance of idiom detection in translation
2.3 Previous work
3. Corpus of Croatian idioms
4. NooJ – NLP tool of our choice
5. Dictionaries and syntactic grammars
6. Classification of idioms
6.1 Idioms of Type 1
6.2 Idioms of Type 2
6.3 Idioms of Type 3
6.4 Idioms of Type 4
6.5 Idioms of Type 5
7. Results
8. Conclusion
Acknowledgements
References
Corpus analysis of Croatian constructions with the verb doći ‘to come’
1. Introduction
2. Lexicographic description of the verb doći in contemporary Croatian dictionaries
3. Corpus analysis of the verb doći
3.1 Collexeme analysis of lemma doći and constructions doći do and doći na
3.1.1 Corpus analysis of the construction doći do ‘to come to’
3.1.2 Corpus analysis of the construction doći na ‘to come onto’
4. Conclusion
References
Anaphora resolution, collocations and translation
1. Introduction
2. Collocations in Translation
3. Translating collocations with Its-2
3.1 Anaphora resolution
4. Results and evaluation
4.1 Two experiments
4.2 Evaluation of the precision of the AR procedure
5. Conclusion
References
Index