=Paper=
{{Paper
|id=Vol-1167/CLEF2001wn-adhoc-CarpuatEt2001
|storemode=property
|title=CLEF 2001 Bilingual Task: Simple Dictionary-Based Query Translation
|pdfUrl=https://ceur-ws.org/Vol-1167/CLEF2001wn-adhoc-CarpuatEt2001.pdf
|volume=Vol-1167
|dblpUrl=https://dblp.org/rec/conf/clef/CarpuatF01
}}
==CLEF 2001 Bilingual Task: Simple Dictionary-Based Query Translation==
https://ceur-ws.org/Vol-1167/CLEF2001wn-adhoc-CarpuatEt2001.pdf
CLEF 2001 Bilingual Task
Simple Dictionary-Based Query Translation
Marine Carpuat and Pascale Fung
HKUST
Human Language Technology Center
Department of Electrical and Electronic Engineering
University of Science and Technology
Clear Water Bay, Hong Kong
feemarine,pascaleg@ee.ust.hk
Abstract
In this paper, we describe our approach to the French English Bilingual
Task in CLEF 2001. A simple dictionary-based method is used to translate
the French query into a bag of weighted English words, the English query,
which is submitted to the SMART retrieval engine. Despite the simplicity
of the method, the results happen to be reasonable.
1 Introduction
For our rst participation to CLEF, our goal was to evaluate what level of per-
formance can be expected from a simple CLIR system, which does not make
use of sophisticated resources. We decide to work on a fully automatic method,
based on freely available resources. We participated to the French-English bilin-
gual task and present the results obtained using a simple dictionary-based query
translation method.
2 Method
2.1 Dictionary-Based Keyword Translation
The easiest and most common way to cross the language boundary for CLIR
is to translate the original query in the target language and to input this new
query to a monolingual IR Sytem. In this work, French queries are simply
translated into English by dictionary look-up. Our "bilingual dictionary" is in
fact a bilingual word list generated from a free online Bilingual French-English
dictionary[3]. The lists consists of around 200,000 entries, but many di erent
forms are listed for each French word ( singular/plural, male/female, tenses ).
The number of terms actually represented is therefore much lower. Multiple
translations are all included according to the online dictionary.
This naive construction of the bilingual lexicon allows us to perform very little
pre-processing on the French queries. We only focus on keyword selection. In
order to translate meaningful terms, we try to remove as many stopwords or
function words as possible. We chose a basic stoplist [5] and augmented it.
1
CLEF2000 French topics are used to identify function words that are used to
express the information need but do not carry information themselves. These
words typically occur in a large number of queries with a low frequency in
each query, whereas content-bearing words occur in fewer queries with a high
frequency in each query. Potential function words are therefore easily selected
with tf.idf information.
The English query is a simple bag of words, taking into account all the possible
translation of the keywords given by the dictionary. If no translation candidate
can be found in the dictionary, the French words are directly copied and pasted
in the English query. For closely related languages such as French and English,
keeping the original word in the source language is helpful, since the missing
translations happen to be proper nouns, English words included in French topics
or cognates.
2.2 Text Retrieval Engine
The Text Retrieval Engine used in this work is the well-known IR sytem Smart[2][6].
It is well adapted to index and query a large corpus such as the CLEF Los An-
geles Times corpus. Smart implements the vector space model, in which queries
and documents are represented by vectors containing tf.idf weights. English
queries consist of all the translation candidates found for the French topics
keywords, weighted by the frequency of the original French keyword and the
translation probability when multiple candidayes are found. We assume these
translation probabilities are uniform.
3 Results
Our experiment was conducted using the title and description elds of the top-
ics only. We obtain average results, which are reasonable given the simplicity
of the method used. The majority of unique translation candidates given by
our lexicon happen to be correct. When multiple translations are proposed,
co-occurence with the translation of other keywords performs a simple disam-
biguation. Not surprisingly however, when the number of translation candidates
grows larger and when the terms are more ambiguous, the query results drop
below the average.
Recall Precision
0.00 0.5634
0.10 0.4996
0.20 0.4388
0.30 0.3630
0.40 0.3040
0.50 0.3234
0.60 0.2010
0.70 0.1743
0.80 0.1554
0.90 0.1302
1.00 0.0932
Table 1: Interpolated Recall-Precision Averages
2
Improvements could be achieved by adding a statistical disambiguation method[1]
and making use of a more sophisticated translation weighting scheme.
4 Conclusion
In summary, our simple dictionary-based keyword translation method performs
reasonably well with a simple lexicon. Of course, query expansion and disam-
biguation methods are much needed to improve these results. But this shows
that even a simple lexicon based on the most frequent translations is a reason-
able basis for Bilingual Information Retrieval.
References
[1] Ballesteros, L. and Croft, B. (1997). Phrasal Translation and
Query Expansion Techniques for Cross-Language Information
Retrieval. In: Proceedings of the Association for Comput-
ing Machinery Special Interest Group on Information Re-
trieval (ACM/SIGIR) 20th International Conference on Re-
search and Development in Information Retrieval; 1997 July
25-31; Philadelphia, PA. New York, NY: ACM, 1997. 84-91.
[2] Buckley, C. 1985 Implementation of the smart information
retrieval system, Technical Report 85-686, Computer Science
Department, Cornell University, Ithaca, New York, May 1985.
[3] English-French bilingual dictionary. Available: http://sun-
recomgen.univ-rennes1.fr/FR-Eng.html
[4] Grefenstette, G. (1998). The Problem of cross-language in-
formation retrieval. In G.Grefenstette (Ed.) Cross Language
Information retrieval. pp 1-9. Kluwer Academic Publishers.
[5] Veronis, J. Un Antidictionnaire. Available:
http://www.up.univ-mrs.fr/~veronis
[6] Smart. Available: ftp://ftp.cs.cornell.edu/pub/smart.
3