OTTO { Ontology Translation System

Abstract

To enable knowledge access across languages, ontologies that are often represented only in English, need to be translated into di erent languages. For this reason, we present OTTO, an OnTology TranslatiOn System, which enhances ontologies with multilingual information. Rather a di erent task than the classic document translation, ontology label translation faces highly speci c vocabulary and lack contextual information. Therefore, OTTO takes advantage of the semantic information of the ontology to improve the translation of labels.

Introduction

The task of ontology translation involves generating an appropriate translation for the lexical layer, i.e. labels stored in the ontology. Most of the previous related

1 https://translate.google.com/

2 Translation performed on 25.06.2015

work focused on accessing existing multilingual lexical resources, like EuroWordNet or IATE [ 2, 3 ]. Their work focused on the identi cation of the lexical overlap between the ontology and the multilingual resources, which guarantees a high precision but a low recall. Consequently, external translation services like BabelFish, SDL FreeTranslation tool or Google Translate were used to overcome this issue [ 4, 5 ]. Additionally, [ 4 ] and [ 10 ] performed ontology label disambiguation, where the ontology structure was used to annotate the labels with their semantic senses. Di erently to the aforementioned approaches, which rely on external knowledge or services, we focus on how to gain adequate translations with a domain-aware SMT system, which is supported by the ontology hierarchy. 3

System Implementation

Based on the lexical and semantic overlap with the ontology labels, the OnTology TranslatiOn System { OTTO3 identi es, from a large set of parallel corpora, the most relevant source sentences containing the labels to be translated. The goal is to translate the ontology labels within the textual context of the targeted domain, rather than in isolation. For instance, with this selection approach, we aim to retain relevant sentences, where the English word vessel or injection belongs to the medical domain, but not to the technical domain. Statistical Machine Translation For the translation approach, OTTO engages the Moses toolkit [ 9 ]. To have a broader domain coverage of the generic parallel dataset necessary for training the SMT system, we merged the JRC-Acquis 3.0 [ 13 ], Europarl v7 [ 8 ] and OpenSubtitles2013 [ 14 ], thus obtaining a training corpus of 8.5M parallel sentences for English-German, 18.9M for English-Italian and 33.6M for the English-Spanish translation directions. To train OTTO for the (under-resourced) English-Irish translation direction, we collected around 723K parallel sentences from various parallel corpora, like DGT (DG Translation at the European Commission), EUbookshop or KDE4, from the OPUS webpage.4 Relevant Sentence Selection In order to improve the translation of ontology labels, we select from the concatenated corpus only those source sentences, which are most relevant to the labels to be translated. The rst criterion for relevance is the n-gram overlap between a label and a source sentence coming from the generic corpus. Due to the speci city of the ontology labels, just an n-gram overlap approach is not su cient to select all the useful sentences. For this reason, we follow the idea of extending the semantic information of the labels using Word2Vec5 for computing distributed representations of words [ 1 ]. The technique is based on a neural network that analyses the textual data provided as input, in our experiment ontology labels and source sentences, and outputs a list of semantically related words [ 12 ]. Each input string is vectorized and compared to other vectorized sets of words in a multi-dimensional vector space, which was trained with Word2Vec on the Wikipedia articles.6

To further improve the disambiguation of short labels, the related words of the label are concatenated with the related words of its direct parent in the ontology hierarchy. Given a label and a source sentence from the generic corpus, 3 http://server1.nlp.insight-centre.org/otto/ 5 https://code.google.com/p/word2vec/

4 http://opus.ling l.uu.se/

6 Wikipedia dump id enwiki-20141106

related words and their weights are extracted from both of them, and used as entries of the vectors to calculate the cosine similarity. Finally, the most similar source sentence and the label should share the largest number of related words.

4 OTTO Demo

OTTO takes as input an ontology represented in OWL or RDF and extracts the labels stored in it. To improve the translations of the labels stored in the ontology, the most relevant sentences, which contain the labels, are obtained from the concatenated generic corpus. Once the labels are identi ed in the context of the relevant sentences, OTTO engages Moses and translates the labels within the context into German, Italian, Spanish and Irish. After the translation process is done, the translated labels are identi ed in the relevant target sentences.

Since the translation of the extracted labels may take several minutes (or even hours), the OTTO user can provide an optional e-mail,7 which is used to inform the user about the completion of the translation process, as well as the web address where the provided data will be stored. Without this information, the address of the stored data is given trough the OTTO interface (Figure 1).

In the last step, the translated labels are represented in di erent formats, for example in a HTML table and CSV le to allow a better visualisation. Furthermore, the multilingual information is injected into the original monolingual ontology and represented as a multilingual ontology as well as in lemon8 [ 11 ], a model for linking linguistic information with ontologies.

5 Conclusion

This paper is aimed at showing OTTO, an OnTology TranslatiOn System for multilingual enrichment of semantically structured data, i.e. ontologies or taxonomies. The system is based on an approach to identify the most relevant source sentences from a large generic parallel corpus, giving the possibility to automatically translate highly speci c ontology labels in context without particular in-domain parallel data. The demonstrated approach reduces the ambiguity of expressions in the selected sentences, which consequently generates better

7 The provided e-mail is stored as a variable and is deleted after the process nishes.

8 http://lemon-model.net/

translations of ontology labels. As an ongoing work, we further focus on improving the extraction of the lexical knowledge stored in ontologies. Additionally, we plan to enable knowledge enrichment for existing multilingual ontologies.

Acknowledgement

This publication has emanated from research conducted with the nancial support of Science Foundation Ireland (SFI) under Grant Number SFI/12/RC/2289.

References

1. Arcan , M. , Turchi , M. , Buitelaar , P. : Knowledge portability with semantic expansion of ontology labels . In: Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing. ACL , Beijing, China ( 2015 )

2. Cimiano , P. , Montiel-Ponsoda , E. , Buitelaar , P. , Espinoza , M. , Gomez-Perez , A. : A note on ontology localization . Appl. Ontol . 5 ( 2 ), 127 {137 (Apr 2010 )

3. Declerck , T. , Perez , A.G. , Vela , O. , Gantner , Z. , Manzano , D. , D-Saarbru cken: Multilingual lexical semantic resources for ontology translation . In: In Proceedings of the 5th International Conference on Language Resources and Evaluation ( 2006 )

4. Espinoza , M. , Montiel-Ponsoda , E. , Gomez-Perez , A. : Ontology localization . In: Proceedings of the Fifth International Conference on Knowledge Capture. K-CAP '09 , ACM , New York, NY, USA ( 2009 )

5. Fu , B. , Brennan , R. , O 'Sullivan , D. : Cross-lingual ontology mapping - an investigation of the impact of machine translation . In: Gomez-Perez , A. , Yu , Y. , Ding , Y . (eds.) ASWC. Lecture Notes in Computer Science , vol. 5926 . Springer ( 2009 )

6. Gomez-Perez , A. , Vila-Suero , D. , Montiel-Ponsoda , E. , Gracia , J. , Aguado-de Cea , G.: Guidelines for multilingual linked data . In: Proceedings of the 3rd International Conference on Web Intelligence , Mining and Semantics. ACM ( 2013 )

7. Gracia , J. , Montiel-Ponsoda , E. , Cimiano , P. , Gomez-Perez , A. , Buitelaar , P. , McCrae , J. : Challenges for the multilingual web of data . Web Semantics: Science, Services and Agents on the World Wide Web 11 ( 2012 )

8. Koehn , P. : Europarl: A Parallel Corpus for Statistical Machine Translation . In: Conference Proceedings: the tenth Machine Translation Summit. AAMT ( 2005 )

9. Koehn , P. , Hoang , H. , Birch , A. , Callison-Burch , C. , Federico , M. , Bertoldi , N. , Cowan , B. , Shen , W. , Moran , C. , Zens , R. , Dyer , C. , Bojar , O. , Constantin , A. , Herbst , E.: Moses: Open source toolkit for statistical machine translation . In: Proceedings of the 45th Annual Meeting of the ACL on Interactive Poster and Demonstration Sessions. Stroudsburg , PA, USA ( 2007 )

10. McCrae , J. , Espinoza , M. , Montiel-Ponsoda , E. , Aguado-de Cea , G., Cimiano , P. : Combining statistical and semantic approaches to the translation of ontologies and taxonomies . In: Fifth workshop on Syntax, Structure and Semantics in Statistical Translation (SSST-5) ( 2011 )

11. McCrae , J. , Spohr , D. , Cimiano , P. : Linking lexical resources and ontologies on the semantic web with lemon . The Semantic Web: Research and Applications ( 2011 )

12. Mikolov , T. , Chen , K. , Corrado , G. , Dean , J.: E cient estimation of word representations in vector space . CoRR abs/1301 .3781 ( 2013 )

13. Steinberger , R. , Pouliquen , B. , Widiger , A. , Ignat , C. , Erjavec , T. , Tu s , D., Varga , D. : The JRC-Acquis: A multilingual aligned parallel corpus with 20+ languages . In: Proceedings of the 5th International Conference on Language Resources and Evaluation (LREC' 2006 ) ( 2006 )

14. Tiedemann , J.: Parallel data, tools and interfaces in opus . In: Chair), N.C.C. , Choukri , K. , Declerck , T. , Dogan , M.U. , Maegaard , B. , Mariani , J. , Odijk , J. , Piperidis , S. (eds.) Proceedings of the Eight International Conference on Language Resources and Evaluation . Istanbul, Turkey (may 2012 )