Bilingual Information Retrieval
                  with DesIRe
       and Internet Translation Services
                                                         ∗
                                        Norbert Gövert


                       University of Dortmund, Germany




1        Introduction
DesIRe is the Dortmund extensible structured Information Retrieval engine 1 . Its extensi-
bility is based on the implementation of physical data independence; it's query interface
consists of datatypes with respective search predicates. This concept enabled us to add
bilingual search predicates for the datatypes Text::English and Text::German (for En-
glish and German text, respectively). Our implementation uses free Internet resources
for translating topics from English to German and vice versa.


2 Search predicates for bilingual retrieval
Having a system which is extensible w. r. t. datatypes and their respective search predi-
cates we decided to extend the Text::English and Text::German datatypes by search
predicats for bilingual text retrieval. These predicates needed to peform the translation
of topics and queries from German to English in case of datatype Text::English and
vice versa in case of datatype Text::German.
    For translation of queries we adopted two rather naive, but fully automatic ap-
proaches. In both approaches we used free internet resources:
      • Approach 1 uses the Babelsh translation service2 of Altavista. This service allows
         to translate passages in a source language to a given target language. Besides the
         translation from German to English and vice versa, Babelsh is capable of various
         other languages.
 ∗
     <goevert@ls6.cs.uni-dortmund.de>
 1 http://ls6-www.cs.uni-dortmund.de/ir/projects/DesIRe/
 2 http://babelfish.altavista.com/
   • Approach 2 uses an ordinary online dictionary for word-by-word translations.
     We chose the Leo Dictionary service3 for this purpose. Leo provides for a En-
     glish / German dictionary with about 223 900 entries. Translations can be done
     in both directions. Since also composed words and phrases are included in the
     dictionary, we exploited this by not translating the original topics word-by-word
     but by interpreting each two neighbouring terms as phrases. Adopting a real naive
     approach we even didn't take measures in order to tackle the word disambiguation
     problem.

              query       result

                      1
                                             Query
                                     source/target language
                  bilingual                                             Internet translation
                                               2              wrapper         service
              search predicate
                                        tranlated query

                      3




                   index




                                   Figure 1: bilingual search predicates

    Figure 1 shows the general scheme of our search predicates for bilingual text re-
trieval. The user gives the query in a source language, which is translated by means of
a translation wrapper. The task of the wrapper is to give a uniform interface to free
translation resources on the internet: It accepts the query as given by the user plus
source and target language and then handles the translation through the service it was
implemented for.




 3 http://dict.leo.org/