Presentation at Tsujii Lab, University of Tokyo

Tomaž Erjavec

Dept. of Intelligent Systems
Jozef Stefan Institute
Ljubljana
Slovenia

Tuesday, 22 January 2002

Overview:

• Slovenia
• IJS E8 and NLP
• My work and history
• Current research interests
• Work in Tokyo

Slovenia

• Facts and location
• Map of the country
• Slovenia in pictures

IJS AI and NLP

• Jozef Stefan Institute
• Dept. of Intelligent Systems
• about Natural Language Processing at our department:

  Corpus Linguistics

  Tomaz Erjavec

  Learning Language in Logic

  Saso Dzeroski, Tomaz Erjavec

  Text and Web Mining

  Dunja Mladenic, Marko Grobelnik

  Speech Synthesis

  Tomaz Sef, Ales Dobnikar

My work and history

• 1984 BSc, Computer Science, University of Ljubljana
• 1990 MSc, Computer Science, University of Ljubljana
• 1992 MSc, Cognitive Science, University of Edinburgh
• 1997 PhD, Computer Science, University of Ljubljana
  "Unification, Inheritance and Paradigms in the Morphology of Natural Languages"
  Supervisors: Ivan Bratko (Ljubljana), Ewan Klein (Edinburgh)

ILD UK SALT project (1994-1996)

The Integrated Language Database
(RA, Centre for Cognitive Science, 1994)

MULTEXT-EAST Copernicus Joint Project COP 106 (1996-1997):

Multilingual Texts and Corpora for Eastern and Central European Languages

ELAN MLIS EU Project (1998-1999):

European Language Activity Network
(local page)

TELRI Copernicus Concerted Action (1999-2001, 1995-1997):

Trans-European Language Resources Infrastructure II
(local page)

CONCEDE Copernicus Joint Project (1998-2000):

Consortium for Central European Dictionary Encoding
(local page)

Ministry of Information Society Project (2001)

Localisation of Open Source Spell Checkers ispell and aspell
(collaborator)

MZT L2-0461-0106 (1998-2001)

Development of Digital Publishing with Distance Learning Support
(project leader)

MZT T2-0409 (1998-2000)

Speech Copora and Tools for the Slovenian Language

FIDA (1996-1999)

Reference corpus of the Slovene Language
(TEI/SGML consulting)

GNUsl (1995--)

A GNU effort for the Slovene Language
(server maintenance, resource contribution)

• "Introduction to SGML"
  3-day course at EuroLan Summer School on Corpus Linguists, 1997.
• "Standards for Language Encoding"
  1-week course at ESSLLI'99.
• "Annotation of language resources" (with Anne-Marie Mineur)
  1-week introductory course at ESSLLI 2002, 5-16 August, Trento, Italy.

• President of SDJT, the Slovenian Language Technologies Society
• Advisory board member of EACL, the European Chapter of the Association for Computational Linguistics
• Council member of the TEI, the Text Encoding Initiative Consortium
• Editorial board member of IJCL, the International Journal for Corpus Linguistics (Bejamins publishers)
• Commissioning editor for CHum, the journal ``Computers in the Humanities'' (Kluwer Academic Publishers)

Research interests

• Morphology, relation to syntax and phonology
• Head-driven Phrase Structure Grammar
• Inheritance hierarchies, typed featrue structures, lexical rules
• Finite state automata and trasducers for morphology and phonology

• Corpus linguistics and "language technologies" for Slovene language:
  • Collection and annotation of Slovene and multilingual corpora
  • Development of other language resources (lexica, tagsets)
  • Ensuring Web and "open source" availability of these resources
  • Support for localisation of open source software
• Tools for annotation:
  • most work on part-of-speech tagging for Slovene
  • also application of segmenters, tokenisers, aligners and bi-lingual lexicon extractors
• Markup languages
  • Corpus Encoding Standard use and development
  • Design of EAGLES and MULTEXT compliant part-of-speech tagsets
  • Use of TEI in a number of projects / resources
  • Familiar with XML, XSLT, RDF, XML Schemas... (but still a lot to learn here!)
• Learning Language in Logic:
  • technique: inductive logic programming
  • domains: tagging, morphological analysis, lemmatisation
  • joint work with Saso Dzeroski and University of York Machine Learning Group

Work in Tokyo

TEI encoding

A preliminary SGML prolog:

<!DOCTYPE TEI.2  SYSTEM "tei2.dtd"  [
  <!ENTITY % TEI.prose "INCLUDE">
  <!ENTITY % TEI.dictionaries "INCLUDE">
  <!ENTITY % TEI.terminology "INCLUDE">
  <!ENTITY % TEI.general "INCLUDE">
  <!ENTITY % TEI.linking "INCLUDE">
  <!ENTITY % TEI.analysis "INCLUDE">
  <!ENTITY % TEI.fs "INCLUDE">
  <!ENTITY % TEI.corpus "INCLUDE">
]>

Multiple Hierarchies

Transformations

Machine learning

Also use of other information soruces connected to MEDLINE, i.e. MeSH and UMLS.