Text translation systems:
Speech translation systems:
Examples: D-B MAT and VERBMOBIL
D-B MAT is a machine aided translation (MAT) system which is developed for human translation and, in a first phase, between German and Bulgarian languages. The innovative concept of D-B MAT is its capability to answer questions about the domain in which the text is situated.
Rationale: Translators claim that more than 40% of their time is spent for understanding the source text. Every action to achieve domain knowledge (looking up encyclopedias or domain overviews interrupts (and most often destroys) the "flow of translation", which only guarantees coherence and consistency of the translation.
D-B MAT3 has access to a formal (non-natural-language) representation of conceptual and referential domain knowledge (written in Conceptual Graphs) by the following technique:
- The translator can select one or more words in the text and
- choose corresponding questions from a menu ("What is ?", "Characteristics ?", "Who ?" etc.),
- the system follows (for technical terms only) the links from the lexical entry to the knowledge base entry, and
- evaluates the corresponding node and its environment along inference rules (inheritance of attributes, e.g.).
- The result set is passed to a chosen natural language generator, which produces natural language explanations according to the chosen degree of detail.
Verbmobil is a long-term project of the Federal Ministry of Education, Science, Research and Technology (BMBF). Its aim is to promote Germany´s international position in language technology and its economical application by national and international cooperation and concentration. National partners are about 7 industrial and 22 university institutes (among them Siemens, Daimler-Benz, IBM, Philips, the German Research Center for Artificial Intelligence and the universities of Hamburg, Karlsruhe, Munich, Saarbrücken ...).
The long-sighted aim is the development of a mobile translation system for the translation of connected spontaneous speech in face-to-face situations. The system is designed to translate speaker independent spoken input from German and Japanese to spoken English output.
The already existing laboratory version - the Verbmobil Demonstrator - has a lexicon capacity of almost 1200 words. It translates spontaneous utterances in the domain of appointment scheduling from German to English.
The system includes not only classical components like word recognition,
syntax, semantics, transfer and generation but also innovative
elements like the processing of prosodic information and discourse
acts. The Verbmobil system in its current demonstrator version
responds in an average of 7 times real time.
3 see v. Hahn, W. and Angelova, G.: Providing Factual Information in MAT ...
123
