GreekGram Reporting on the progress of the implementation of a Modern Greek Grammar Fragment using the XLE parser

1
GreekGramReporting on the progress of the
implementation of a Modern Greek Grammar Fragment
using the XLE parser

• Kakia Chatsiou
• achats_at_essex.ac.uk
• Dept of Language and Linguistics
• University of Essex
• Parts of this project have been funded with an
  ESRC Award PTA-2004-031-00112 , support which is
  gratefully acknowledged.
• Previous versions of this report have been
  presented at the LangUE 2007 International
  Postgraduate Conference and the Language and
  Computation Day 2007, University of Essex.

2
Outline

• Introduction Overview of the ParGram Project
• Project Objectives
• Participating Members
• The XLE Platform
• Grammar Architecture
• The GreekGram Project
• Overview
• Assumptions
• Coverage
• Demonstration
• Future Development Directions

3
Outline

• Introduction Overview of the ParGram Project
• Project Objectives
• Participating Members
• The XLE Platform
• Grammar Architecture
• The GreekGram Project
• Overview
• Assumptions
• Coverage
• Demonstration
• Future Development Directions

4
An Overview of the ParGram Project

• Project Objectives
• Broad coverage grammars
• Inclusion of important and frequently occurring
  constructions
• Linguistically motivated analyses
• Parallel and crosslinguistic development of
  grammars between the participating members
• All grammars are guided by a common set of
  linguistic principles and a commonly agreed upon
  set of grammatical analyses and features
• Identical treatment of core crosslinguistic
  phenomena
• Methods in grammar engineering
• Common test methods and evaluation strategies
• Balance between efficiency, performance,
  reliability and maintainability across grammars

5
An Overview of the ParGram Project
Essex Greek, Welsh Manchester Arabic Oxford Malaga
sy

• Participating Members

Bergen, Norway Georgian, Norwegian, Tigrinya
DCU, Ireland Chinese, English, French, German,
Japanese, Spanish
Fuji XEROX Japanese
Ho Chi Minch Vietnamese
PARC, CA Chinese, English, French
Debrecen Hungarian
IMS, Stuttgart German Konstanz Urdu
Sabanci, Istanbul Turkish
6
An Overview of the ParGram Project

• The Xerox Linguistics Environment (XLE) Platform
• Under current development at PARC (Palo Alto
  Research Center, USA)
• An implementation of the Lexical Functional
  Grammar (LFG) Formalism
• Implemented in C works in Unix, Linux, MacOS. MS
  Windows version under development.
• Integrates a morphological analyser employing
  Finite State Technology
• Can be used for both parsing and generation
• Includes tools for various grammar development
  activities (such as analysing performance,
  test-suites)
• The core technology used in the consumer search
  engine based on natural language processing which
  is currently under development by

7
An Overview of the ParGram Project

• a Silicon Valley Company, currently building a
  transformative consumer search engine based on
  natural language processing
• It is based on technologies that take advantage
  of the structure and nuances of natural language
• It offers an innovative approach to searching
• It breaks the confines of keyword search ?
  queries using both keywords, short phrases and
  natural language
• Makes search more natural and intuitive
• Aims at fundamentally changing how we search the
  web and at the same time delivering higher
  quality results
• currently searching Wikipedia with the help of
  FreeBase as the semantic knowledge database
• (source http//www.powerset.com )

8
Search Engine Demo
9
Demo sample queries

• Who is Indiana Jones?
• Which are the films that Spielberg directed?
• I am looking for Van Goghs paintings
• Paintings by Van Gogh
• Albums by westlige
• Books by Charles Dickens

10
The XLE parser interface
11
An Overview of the ParGram Project

• Basic Grammar Architecture

12
Outline

• Introduction Overview of the ParGram Project
• Project Objectives
• Participating Members
• The XLE Platform
• Grammar Architecture
• The GreekGram Project
• Overview
• Assumptions
• Coverage
• Demonstration
• Future Development Directions

13
The GreekGram Project

• Overview
• A preliminary effort to develop a large-scale LFG
  computational Grammar for Modern Greek
• Shares the objectives and principles of similar
  ParGram projects (parallel and crosslinguistic
  balancing of maintainability and achieving large
  coverage)
• Current main focus is on development of grammar
  rules and syntactic phenomena relevant to my
  thesis the lexicon is kept as minimal as
  possible

14
The GreekGram Project

• Assumptions
• Focus on syntax (but in such a way that the
  representations produced could serve as direct
  and useful input for incorporation into semantic
  interpretations)
• All four possible Modern Greek Word orders are
  treated as equally acceptable in terms of
  markedness and acceptability
• Modern Greek word Order is represented
  non-configurationally (Tzanidaki, 1996
  Alexopoulou, 1999)
• No treatment of morphology (using the
  Morphological analyzer or some other Finite state
  tool) there is a separate lexical entry for
  each form.

15
The GreekGram Project

• Coverage
• All 4 possible word orders (irrespective of
  markedness)
• Pro-drop character of the language
• Subcategorisation frames of intransitive,
  transitive and ditransitive verbs
• Number, case and gender agreement within the NP
  or PP subject-verb agreement
• Basic Relative Clause structure
• Coordination of relative clauses
• Some punctuation

16
The GreekGram Project

• Demonstration

17
The GreekGram Project

• Future Development Directions
• Account for a greater variety of verb
  subcategorisation frames
• Enrich the lexicon adding more lexical entries
• Account for the morphology within the lexicon,
  using the XFST (or other finite state tools)
• Expand the coverage of the grammar to other
  constructions such as interrogatives, imperatives
  and negation

18
Thank you! For more information and updates on
the progress of the project visit
http//privatewww.essex.ac.uk/achats/projects/gr
eekgram/index.html