ISMTCL

International Symposium on Data and Sense

Mining, Machine Translation and Controlled

Languages, and their application to

emergencies and safety critical domains

July 1-3, 2009

Centre Tesnière

University of Franche-Comté

Besançon, France

Presses universitaires de Franche-Comté, 2009

[ISBN 978-2-84867-261-8]

 

Selected abstracts

 

pp.13-20:

Multiple Uses of Machine Translation and Computerised Translation Tools

John Hutchins
web: http://www.hutchinsweb.me.uk

Abstract

For many years MT systems and tools were used principally for the production of good-quality translations: either MT in combination with controlled (restricted) input and/or with human post-editing; or computer-based translation tools by translators. Since 1990 the situation has changed. Corporate use of MT with human assistance has continued to expand (particularly in the area of localisation) and the use of translation aids has increased (particularly with the coming of translation memories). But the main change has been the ever expanding use of unrevised MT output, such as online translation services (Babel Fish, Google, etc.), applications in information extraction, document retrieval, intelligence analysis, electronic mail, and much more.

---

pp.38-42:

French to Arabic Machine Translation

Isomorphic Syntax, Use of Terminal Sequences

Mohand Beddar

Centre Lucien Tesnière, Université de Franche-Comté, France
mohand.beddar@edu.univ-fcomte.fr

Abstract

Languages are different and each of them requires special processing. A machine translation system should take into account syntactic and semantic particularities of each language. In fact, syntactic and semantic models should be defined in both the source and target languages and a link should be established between both the syntactic and semantic models. In security protocols, French and Arabic syntax are quite similar. Structures can be formalized to be isomorphic structures and terminal sequences link these structures to the semantics of the language, therefore limiting semantic ambiguities.

---

pp.43-48:

Remarks about Linguistic Analysis, Normalization and Translation of
Spanish "What to Do in Case of Fire" Texts

Xavier Blanco

Universitat Autònoma de Barcelona
Xavier.Blanco@ uab.cat

Abstract

We intend to discuss, in a understandable way for no trained linguists, the key points that can allow us to see the main invariable meaning of a cluster of texts through the enormous multiplicity of possible paraphrases. As means of example, we show how to formalize the core messages of a collection of Spanish “What to Do in Case of Fire” texts. We keep our analysis to the simple phrase level (i.e. no textual constraints as, for instance, cohesion or thematic progression, will be discussed). At this level, we discuss mainly the following topics: semantic labelling of facts; semantic labelling of entities; paradigmatic lexical functions; syntagmatic lexical functions; grammatical meanings. We give examples in different European languages concerning the translation of our “What to Do in Case of Fire” texts. We show that linguistic formalization can be regarded as sort of an interlingua particularly well suited for translation of alert texts.

pp.49-55:

Controlled Languages and Machine Translation

Krzysztof Bogacki

Warsaw University

kbogacki@gmail.com

Abstract

We will examine controlled languages in the context of machine translation. First we will present principles governing the conversion of standard natural language texts into controlled Polish. We will compare 6 converted texts with their standard sources and comment on them. Then we will present the results of an experiment which has turned out to be disappointing in this respect: the number of mistranslations and various sorts of mistakes was sufficiently big to make us query the reasons of such a result.

---

pp.69-73:

Achieving a Better Machine Translation from French to English
via a Controlled Language

Tessa Cornally
Centre Tesnière, Université de Franche-Comté, France

tessa.carnally@gmail.com

Abstract

This article discusses how the use of a controlled language can improve Machine Translation results. More specifically it is concerned with the sublanguage of oenology and examines the structures specific to this domain.

---

pp.82-89:

English/Veneto Resource Poor Machine Translation with STILVEN

Rodolfo Delmonte, Antonella Bristot, Sara Tonelli, Emanuele Pianta

Università Ca' Foscari - Department of Language Sciences

E-mail:delmont@unive.it

Abstract

The paper reports ongoing work for the implementation of a system for automatic translation from English-to-Veneto and viceversa. The system does not have parallel texts to work on because of the almost inexistence of such manual translations. The project is called STILVEN and is financed by the Regional Authorities of Veneto Region in Italy. After the first year of activities, we managed to produce a prototype which handles Venetian questions that have a structure very close to English. We will present problems related to Veneto, basic ideas, their implementation and results obtained.

---

pp.100-103:

Syntactic Problems in French-Russian Machine Translation of Periodicals

Ekaterina Ershova

Centre Tesnière, University of Franche-Comté, Besançon, France
ekaterinarshv@gmail.com

Abstract

French-Russian machine translation is not a new-found field of study. Nevertheless this is a domain that is very rich and fruitful in problems which still need to be resolved, particularly syntactic problems. Moreover, limited to a certain subject field, i.e. periodicals, MT “of high quality” is certain to be possible.

---

pp.104-113:

Translating Composite Sentences in Azerbaijani-English MT System

Rauf Fatullayev               Sevinc Mammadova             Abulfat Fatullayev

National E-Governance       National E-Governance        Institute of Cybernetics

Network Initiative Project    Network Initiative Project          Baku, Azerbaijan

Baku, Azerbaijan                 Baku, Azerbaijan                    fabo@box.az

fatullavev@gmail.com            sevinc@dilmanc.az

Abstract

This article is dedicated to the automation of the translation process of the composite sentences in the Azerbaijani language in an Azerbaijani-English MT system. First the Azerbaijani composite sentence is divided into simple sentences and English translation of the sentence is synthesized by using the translations of the simple sentences.

---

pp.138-142:

Some Problems in a French-Chinese Machine Translation System

Gan Jin

Centre Tesnière, Faculté des Lettres
Universiée de Franche-Comté

Besançon, France
jingan1982@hotmail.com

Abstract

The Chinese language is very different from European languages with respect to morphology, lexica, syntax and semantics. This complexity causes many problems in machine translation systems. ‘Rang' is a Chinese character morphologically simple, but its usage is very complicated. For a long time, linguists have been interested in its complexity. With all the efforts, the linguists have not come to agreement to this day as to the correct grammatical category. Some linguists consider it as a preposition, others as a verb. In this article, we try to explain the real usage ofRang' sentence in a French-Chinese machine translation system for a specific domain where safety is extremely important and we show our methods of disambiguating verbs with respect to not only Chinese grammar but also French grammar because the latter is our source language in the translation system. Our objective is to create a reliable machine translation system.

---

pp.158-163:

French-Vietnamese Noun phrase Translation

Le Thi Sinh

Centre Tesnière, Université de Franche-Comté, France
Stephanie_cuhb@yahoo.com

Abstract

Vietnamese is a noun classifier language, while this is not the case for French. This is why we encounter some problems when translating French noun phrases (NPs) into Vietnamese. This article will suggest a simple algorithm for French-Vietnamese NP translation after building a Vietnamese classifier-noun combination system which makes up a considerable knot to be solved in the algorithm. All these are realized basing on the results from the condensed comparative analyses of NPs of the two languages under consideration.

---

pp.172-178:

Cognitive Models of Yesterday and Today in Machine Translation and their
Implication for Controlled Languages

Henri Madec

Tesnière research Center in linguistics

University of French County, France

henri.madec@univ-fcomte.fr

Abstract

In this article we present a contradiction between the 50 years NLP epistemology based on behaviourism and the today's one that requires taking into account advances in brain imaging. The science of translation is certainly an idea developed after the Soviet experience of Pavlov on a stimulus-response model. Today we know better how the human brain works. It would be necessary to make a new science of translation, based on other principles and a new epistemology. This should be a necessity. And perhaps the probabilistic and statistical models in fashion now with Google lead us in this direction. But it is doubtful that the two last approaches go in the same direction.  Also would it be better to defend the Soviet science of translation which doesn't prove quite outdated in the current state of the art and allow the controlled languages.... even if it does not contribute to build a stable and consistent pattern of TAL.

---

pp.185-189

Abduction Alerts in Greek and Spanish

Eleni Papadopoulou                                            Marcel Puig Portella

Universitat Autònoma de Barcelona                        Universitat Autònoma de Barcelona

Eleni.Papadopoulou@uab.cat                                    Marcel.Puig@uab.cat

Abstract

In this paper, we intend to present a redaction model of abduction alert messages for both Greek and Spanish languages, in the frame of the MESSAGE project. The main object of the present work is to describe the lexicographic groundwork needed in order to construct models applicable to NLP (Natural Language Processing) software, which would be able to generate and translate automatically texts in Greek and Spanish and to be further used in the controlled languages' field.

---

pp.190-197:

Grammatical and Lexical Errors Analysis of English-Vietnamese Translation
Texts with the Google & EVTRAN Engines and Post-editing Tasks

Phan Thi Thanh Thao,

Wolverhampton University

T.Phan.Thi.Thanh@wlv.ac.uk; thaohuy7269@yahoo.com

Abstract

Evaluation of machine translation (MT) is a challenging task for computational linguists due to the variety of translation engines used to “decode” different pairs of languages. Although a large number of automatic evaluation measures have been proposed and studied over the last years, human judgements of MT quality still remain the best method for comparing and evaluating different MT systems. Moreover, to achieve the high quality of MT, manual post-editing tasks are taken into significant consideration. It is important to study the methods of post-editing the MT output fast and effectively, which requires the typical error analysis of different translation engines. This paper mentions the grammatical and lexical error analysis of English-Vietnamese translation texts extracted from BBC News from January to May 2009 with the Google and EVTRAN engines. Based on the error analysis, some manual post-editing tasks are suggested to improve the quality of translation engines to some extent.

---

pp.198-202:

Treatment of the Imperative Forms in the Machine Translation between

Catalan, Spanish and Greek

Marcel Puig Portella                                          Eleni Papadopoulou

Universitat Autònoma de Barcelona                     Universitat Autònoma de Barcelona

Marcel.Puig@uab.cat                                   Eleni.Papadopoulou@uab.cat

Abstract

This paper is a fruit of our research on controlled languages and their automatic/machine translation in the frame of Alert Messages and Protocols project and its purpose is twofold. Firstly, we outline the imperative and alter-imperative forms in Catalan, Greek and Spanish. Following to this description, we propose a systematic treatment of these forms in the framework of the automatic translation and the controlled languages.

---

pp.231-235:

Polish controlled language and its machine translation into French

Zuzanna Rudas

Centre Tesnière, Université de Franche-Comté, Besançon
z.rudas@gmail.com

Abstract

This paper presents some difficulties encountered during the analysis of Polish controlled protocols about the fire regulations in their aim for a machine translation into French. The applied method to solve the problems uses the systemic linguistics.

---

pp.249-255:

Post-editing Experiments with MT for a Controlled Language

Irina Temnikova and Constantin Orasan
Research Institute in Information and Language Processing

University of Wolverhampton, UK
E-mail: I.Temnikova2@wlv.ac.uk, C.0rasan@wlv.ac.uk

Abstract

This paper aims to establish whether a new controlled language (CL) for emergency-related texts could facilitate both human and machine translation. To achieve this, an experiment involving an MT engine and human translators and post-editors was conduced. In order to estimate whether the CL pre-editing has an impact on human and machine translation, the time to manually translate, the time to post-edit and the edit distance between the original and the simplified texts were measured. The results of the experiment confirm the hypothesis.

---

pp.256-259:

Research into the Practicality of Machine Translation of Administrative

Documents

Tsai Yi-Jung

Centre Tesnière

30, Rue de Mégevand,

25000 Besançon, France

charlotte0421@gmail.com

Abstract

This paper discusses Machine Translation (MT) in the domain of administrative documents. MT is not only an interesting field for research but also has an immense practical value to society. Due to the fact that natural languages contain many ambiguities and there is too much information for the corpus to be complete, there are still some problems to resolve and some obstacles to overcome. However, administrative documents issued by governments and other official institutions, which are written in a standard form, with an identical structure, limited vocabulary and rarely contain ambiguities are just an ideal sublanguage domain to treat in which fully automatic high quality translation is achievable.

---

pp.260-268:

Building a Linguistic Database for Chinese Interrogative Sentences

Xiaohong Wu

Centre Tesnière, Faculté des Lettres
Université de Franche-Comté, France

Faculté de Langues Etrangères

Minzu Université de Qinghai, Chine

wuxiaohongfr@yahoo.com.cn

Abstract

Analysis of the interrogative sentences plays an important role in systems that focus on work such as question-answering and/or human-machine communication. In this paper we present the work we have done for a multilingual MT system. Texts collected for the building of the parallel corpora are those from the domains where the accurate interpretation of the texts is extremely crucial. Therefore, exact and accurate translation is not only necessary but also obligatory. To reach high quality translations, we adopt the controlled language technique. Here we lay focus on building a linguistic database for the analysis and transfer of the French interrogative sentences into Chinese, which play an important role in some texts in our corpora. We will introduce how we classify and control the interrogative sentences in our work. We will describe the classifications and the linguistic information needed when processing the interrogative sentences according to their differed usages.