XMELLT

1
XMELLT

• Cross-lingual Multi-word Expression Lexicons
• for Language Technology
• Multilingual Information Access and Management
• International Research Co-operation

Nancy Ide Department of Computer Science Vassar
College
2
Participants

• Department of Computer Science, Vassar College
• International Computer Science Institute,
  University of California, Berkeley
• Department of Computer Science, New York
  University
• Computing Research Laboratory, New Mexico State
  University

3
Framework

• Planning project
• one-year time frame
• Originally submitted as a joint NSF-EU project
  with additional European partners
• Istituto di Linguistica Computazionale, CNR, Pisa
  
• Institut für Maschinelle Sprachverarbeitung,
  Stuttgart
• LexiQuest, Paris

4
Overall goal

• define a core international infrastructure to
  support the creation of a multi-lingual
  multi-word expression lexicon incorporating both
  morpho-syntactic and semantic information

5
Specific aims

• determine the type and dimensions of information
  to serve the needs of critical NLP applications
• specify an overall architecture for a joint
  software and lingware development project

6
Aims...

• Explore the possibilities for recognizing and
  acquiring multi-word lexical units from corpora
  by means of partial parsing, statistics, etc.
• Outline a collaborative project to acquire and
  represent multi-word lexical entries for multiple
  languages

7
Motivation

• Multi-word constructions are extremely frequent
  in language
• 30of the lexical stock
• Existing resources do not adequately treat
  multi-word expressions

8
Limitations

• constructed for particular system or application
• incorporate tailored information (e.g., primarily
  syntax with little semantics)
• not reusable
• most devoted to a single language and/or approach

9
Limitations...

• not flexible, expandable to multiple languages
• MT systems' lexicons are typically little more
  than "translation memories"
• No interface among single-word entries,
  multi-word entries, syntax, and semantics

10
XMELLT Approach

• Broad view of multi-word expressions
• idioms, compounds, collocations, co-occurrence
  patterns
• focus on linking of individual language lexicons
• individual words and multi-word expressions
• different types of multi-word expressions
• e.g., English noun-noun vs Romance noun-PP

11
Considerations

• internal variation
• sub-categorization properties
• idiosyncratic constraints on inflection
• meaning (non-)compositionality

12
Encoding Model

• Compatible and integrated with existing and de
  facto standards
• e.g., EAGLES, PAROLE/SIMPLE, NOMLEX

13
Activities

• Assessment of existing lexical resources for
  multi-word expressions
• Delivery of survey

14
Activities...

• Creation of a small set of sample entries
• add lexical information on support verb
  constructions to 50 nouns drawn from NOMLEX for
  English, Italian, German, and French
• create lexical entries for 50 N-N English
  constructs from the PAROLE/SIMPLE lexicons and
  corresponding constructs in Italian, German, and
  French

15
Activities...

• Develop preliminary specifications for
  structuring and encoding multi-lingual,
  multi-word expression lexicons
• required linguistic information
• harmonized data architecture and encoding format

16
Activities...

• Exploration of techniques for automatic
  acquisition
• Months 1-6 Survey of acquisition techniques,
  typology of MWE
• Months 7-12 Design of architecture for MWE
  acquisition

17
Project information

• Start date June (?)
• Web site
• Contact

http//www.cs.vassar.edu/ide/XMELLT.html
Nancy Ide (PI) Department of Computer
Science Vassar College ide_at_cs.vassar.edu