Multilingual Natural Language Processing

Uploaded by

Praveen Kumar

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

102 views2 pages

Multilingual Natural Language Processing

Uploaded by

Praveen Kumar

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 2

US-Korea Joint Workshop on Digital Libraries

San Diego, August 10-11, 2000

Multilingual Natural Language Processing

Ulf Hermjakob
USC Information Sciences Institute
http://www.isi.edu/natural-language
[email protected]

Multilingual digital libraries face particular challenges, including character sets and their
encoding, machine translation, and cross language information retrieval. To meet the chal-
lenges and complexity of machine translation, research in computational linguistics has
moved more and more towards empirical approaches, following a trend set by speech recogni-
tion, where empirical approaches are clearly dominant, as well as in parsing, which is increas-
ingly based on statistical (Collins, 1997 Charniak, 2000) and machine learning (Hermjakob,
2000) approaches. Building on initial research by (Brown, 1990 Berger, 1994), two examples
of current statistical machine translation research are (Knight, 1999) and (Ney, 2000).
While these approaches are often relatively language-independent in principle, they typi-
cally share the need for language specic training data for applications in specic languages.
Empirical parsing for example relies on treebanks, and machine translation on large par-
allel corpora (and often even larger monolingual corpora for the target language). Other
critical resources include monolingual lexicons and semantic ontologies, as well as extensive
dictionaries for multilingual applications.
This shows that Digital Libraries and Natural Language Processing can mutually benet
each other. Digital Libraries can clearly prot from machine translation, while various sub-
tasks of Natural Language Processing can greatly benet from resources such as bilingual
lexicons and parallel corpora, for which Digital Libraries are a natural repository. The size of
parallel corpora needs to be quite large, hundreds of thousands or even millions of sentences.
It is also desirable that these resources mirror application domains.
At the Natural Language Processing research group at the USC Information Sciences
Institute, we have pursued various applications of human language, including question an-
swering from the web (WEBCLOPEDIA project), machine translation (GAZELLE, QuTE,
EGYPT projects), automated text summarization (SUMMARIST), information retrieval
from web and text collections (MuST/C*ST*RD), and large ontologies of semantic (mean-
ing) symbols (SENSUS). Most of the research had a multilingual focus, with various projects
covering English, Spanish, Japanese, Mandarin Chinese, Korean, Arabic, Bahasa Indonesian
and French. ISI's current Webclopedia project, part of DARPA's TIDES program, uses our
empirical NLP approaches to answer natural language questions with short (<= 50 byte) an-
swers from web documents, at rst in English, and then with cross-language answer retrieval
at a later project stage.
As an example of adapting machine-learning based natural language techniques that were
initially developed for English, we ran a three-month three-person research project in 1999,
in which we annotated 1187 sentences from the Korean newspaper Chosun and used the
resulting treebank to train a Korean parser, achieving word level parse accuracy of 89.8%
recall and 91.0% precision. For initial pre-processing, the parser uses the KMA segmenter
and morphological analyzer and KTAG tagger provided by Prof. Rim of Korea University.
||||||||||||||||||
Dr. Ulf Hermjakob is a senior research scientist at the Information Sciences Institute of the
University of Southern California. He has worked on machine-learning based deterministic
parsing for English, Japanese and Korean, on machine translation, and is currently also on
ISI's Webclopedia project on question answering. Dr. Hermjakob obtained his Master's in
Computer Science from the University of Karlsruhe in Germany and his Ph.D. (1997) from
the University of Texas at Austin with a thesis on \Learning Parse and Translation Decisions
From Examples With Rich Context".

References
A. Berger, P. F. Brown, S. A. Della Pietra, V. J. Della Pietra, J. R. Gillett, J. D. Laerty,
R. L. Mercer, H. Printz, L. Ures 1994. The Candide system for machine translation. In
Proceedings of ARPA Workshop on Human Language Technologies.
P. Brown, J. Cocke, S. Della Pietra, V. Della Pietra, F. Jelinek, J. Laerty, R. Mercer,
P. Roossi 1990. A Statistical Approach to Machine Translation. In Computational
Linguistics 12 (2), pages 79-85
E. Charniak 2000. A Maximum-Entropy-Inspired Parser In Proc. of the Conference of the
North American Chapter of the Association for Computational Linguistics (NAACL),
pages 132-139.
M. J. Collins 1997. Three Generative, Lexicalised Models for Statistical Parsing. In 35th
Proceedings of the ACL, pages 16{23.
U. Hermjakob 2000. Rapid Parser Development: A Machine Learning Approach for Ko-
rean. In Proc. of the Conference of the North American Chapter of the Association for
Computational Linguistics (NAACL), pages 118-123.
K. Knight 1999. Decoding Complexity in Word-Replacement Translation Models. In Jour-
nal of Computational Linguistics, 25(4).
C.-Y. Lin 1998. Assembly of Topic Extraction Modules in SUMMARIST. In Working Notes
of the AAAI 1998 Spring Symposium on Intelligent Text Summarization.
H. Ney, F.-J. Och, C. Tillmann, S. Vogel Oct. 2000. Statistical Translation of Spoken
Dialogue in the Verbmobil System. In Proc. of Workshop Multilingual Speech Commu-
nication.