Translation Issues 3-98 (November 1998) by Anne Lehrndorfer by Stefanie Schachtl Germany

Preface by the Editor

The following paper is a machine-translated text from German into English. And at the same time it explains the technology applied. The German source-text was produced by Anne Lehrndorfer from an interview with Stefanie Schachtl. The final editing of the German text, the preparations for the machine-translation and the translation-process were performed at the Siemens-Nixdorf laboratories.
For the following text I have accepted only two paragraphs of the pre-processed text in German language and I have positioned them left of the English result of the translation-process. This result is the exact output of the machine-translation, without any additional editing. As I learnt while communicating with the authors, this quality of machine translation could only be reached by using a machine-controlled pre-processing of the German source-text, which is a component of this system and described in the paper. This process demands a certain amount of preparatory work which is - so to say - shifted away from the translation-process into the pre-translation-phase. The investment of time into this process has yet to be quantified. Nevertheless, the authors agreed that it is not at all negligible - as one will understand. The authors also announced that further investigation will be done throughout the winter, after which the results will be made available.
Here follows the English version as it was output from the machine translation - without any post-editing.
Editor.

Till now, controlled German only existed as a scientific and pedagogical concept for technical documentation. The linguistic or computational linguistic aspects of language control were treated in dissertations and doctoral theses. Rules for syntax and choice of words were trained in industrial tutorials for translatable and intelligible writing. For winter 1998, Siemens researchers now announce the demo version of the TopTrans machine translation system. In the context of TopTrans, the machine-aided Controlled Siemens Documentary German (CSDG) is also presented. This project combines the experiences in controlled German from scientific and pedagogical work in the long term. The following summary shall give a content insight into the history of development and functioning of CSDG/TopTrans. Moreover, this summary shall procure a formal insight into the present stage of development of CSDG/TopTrans. For this purpose, a pre-edited source text in CSDG (text A) is put opposite its English translation. (text B) The translation was generated with a preliminary version of TopTrans automatically and has not been post-edited. For the demo version in winter, the Siemens researchers want to concentrate on the improvement of the stylistic quality of the target language. (word order)

Controlled language and machine translation

The concept of controlled language is a mental offspring of machine translation. Also CSDG arose out of the problems the traditional translation software faces today. However, the developers of CSDG/TopTrans emphasise that TopTrans is essentially different from machine translation systems of today's generation. TopTrans translates a text only if it understands it. But, it translates the text in such a way that you need not post-edit it any more. Therefore, the focus of CSDG doesn't lie on the generation of texts that are simple (e.g. Caterpillar Fundamental English) and intelligible for the reader in the first place. The most important aim of CSDG is the increase of effectiveness of the machine translation components. However, intelligibility for the reader results as a positive side effect.
Functioning of CSDG/TopTrans

The text prepared for the translation

The result of the translation

Die Software arbeitet in zwei konzeptuellen Modulen. Zunächst wird interaktiv der deutsche Quelltext auf Einhaltung analysierbarer, syntaktischer und lexikalischer Strukturen überprüft (KSDD). Das System mahnt jede Abweichung vom zugelassenen Vokabular und jede nicht syntaktisch vorhergesehene Struktur an. Mehrdeutige Strukturen werden ebenfalls kenntlich gemacht. ... Die Verbesserung der Textstellen, die als Fehler erkannt wurden, erfordert Übung und linguistische Grundkenntnisse. Für die Auflösung der Textstellen, die als Mehrdeutigkeiten erkannt wurden, muss der Autor interaktiv Fragen beantworten. Diese Fragen generiert das System automatisch.

The software works in two conceptual modules. At first, the German source text is checked (CSDG) for adherence to analysable, syntactic and lexical structures interactively. The system indicates every deviation of the licensed vocabulary and every syntactically not anticipated structure. Ambiguous structures are also marked. ... The correction of the text passages that were recognised as errors requires practice and linguistic basic knowledge. For the resolution of the text passages that were recognised as ambiguities, the author must answer questions interactively. The system generates these questions automatically.

So, only the structures the software engineers have defined as controlled German before are translated. Only if the structures are unambiguous and correct in this sense, they are passed on as semantic representation (meta-language) to TopTrans. Then, TopTrans carries out the transfer into the semantic representation of the target language and the grammatical and lexical generation. Also means of the layout (e.g. fonts or labels of document formatting) are transferred into the target language. However, the TopTrans editor only supports HTML labels at present.

Advantages

The text prepared for the translation

The result of the translation

Die Vorteile dieses Systems entsprechen den bekannten Vorteilen von kontrollierten Sprachen. An die Stelle des Aufwands für die Nachbereitung von MÜ-Texten in den jeweiligen Zielsprachen tritt der Aufwand für das Pre-Editing in der Quellsprache. Wenn man in nur eine Zielsprache übersetzt, ermöglicht TopTrans eine Produktivitäts-steigerung um den Faktor 2. Dieser Faktor erhöht sich um 1 für jede weitere Zielsprache. TopTrans fungiert zudem als automatische Stil- und Rechtschreibprüfung und arbeitet somit autoren-spezifischen Stilabweichun-gen entgegen. Darüberhinaus prüft TopTrans lexikalische Konsistenz bezüglich der zugeschalteten Fachterminologien.

The advantages of this system conform to the well known advantages of controlled languages. The effort for pre-editing in the source language replaces the effort for post-editing of MT texts in the respective target languages. If you translate into only one target language, TopTrans permits an increase of productivity by the factor 2. This factor increases by 1 for every further target language. TopTrans functions as automatic correction of style and orthography moreover and therefore works contrary to idiosyncratic stylistic deviations. On top of it, TopTrans checks lexical consistency with respect to the connected technical terminologies.

Future

TopTrans transfers syntactic structures into a meta-language at first and only generates the target language from there. Therefore, the development of every new target language is confined to the respective language specific generation component. Thus, Chinese has been announced as further target language already for the end of the year 2000. In further steps, the languages of the EC shall be added.

Marketing

The different technical operational areas of CSDG/TopTrans require different terminologies and restrictions. Therefore, the software shall be adapted to the individual customer. Especially the training for CSDG shall be adjusted to the needs of the user. Because of a narrow contact with the customers, the software engineers of CSDG/TopTrans expect to achieve a further increase of the productivity and quality of software and languages.
 

"transline Deutschland - Übersetzungsdienst für technische Übersetzung"