Patrizia Paggio

1
A Modular and Scalable Environment for the
Semantic WEB

• Patrizia Paggio
• Center for Sprogteknologi

2
Goals

• To develop an innovative methodological
  environment and software that will enable content
  providers to build a knowledge grid in which
• the content of WEB pages can be managed in a
  modular and scalable way, and
• queries can be posed in natural language to
  extract relevant content from the grid based on
  the underlying ontologies
• Testbed a demonstrator to search university sites

3
Domain areas involved

• Semantic Web
• Ontology mapping
• Knowledge management
• Topic maps
• Text and data mining
• Intelligent agents

4
NL-based, intelligent content search
5
MOSES consortium

• FINSA, Italian software company (agent
  technology, requirements engineering)
• Mondeca, French software company (Knowledge
  management and semantic markup, graph theory)
• Parabots, Dutch software company (text and data
  mining)
• Rome III Univ. (user partner, graph theory)
• Rome II Univ. (language technology, machine
  learning)
• CST (language technology, content-based search)

6
MOSES consortium
7
Planning
8
The semantic web

• The present web is a collection of texts for
  humans to inspect and use.
• On the semantic web, texts are structured (marked
  up) so that programs (agents) can manipulate
  them.
• The semantic structure refers to common
  repositories e.g. ontologies.

9
A scenario

• A student/researcher looking for information on
  university courses or research activities in
  Europe.
• I need a list of institutes offering
  post-graduate courses in computational
  linguistics including corpus linguistics where
  the teaching language is ...
• Which Danish university offers Danish language
  courses for foreign students?

10
Our vision

• Content of web pages is structured according to
  relevant templates and ontologies (the project
  will create those relevant for the domain)
• Help is provided by the system to find the
  templates that best match the pages to be marked
  up
• Search is based not on the words in the text, but
  on the semantic templates
• A linguistic agent processes the results to
  generate relevant answers

11
Our vision

• I need a list of institutes offering ...
• The following institutes offer post-graduate
  courses in computational linguistics including
  corpus linguistics ...
• Which Danish university offers...
• The University of Cph offers Danish language
  courses for foreign students

12
Main work packages

• 1. Requirements and domain analysis
• 2. Architecture design
• 3. Semantic structure and tools
• 4. Implementation of agents
• 5. Content-based engine
• 6. Test and validation

13
Query analyser

• Investigate methods and develop tools to analyse
  user queries and convert them into semantic
  descriptions.
• Based on a realistic corpus of questions/queries.
• Use of shallow linguistic analysis.
• Specific linguistic items, e.g. interrogative
  pronouns.

14
(No Transcript)
15
Topic maps

• Topics and associations

A1
T2
T1
R2
R1
T5
T3
R4
R5
R3
T4
A2
T6
R6
R7
16
Association Example

• This association represents an assertion about
  three topics
• One person Bernard Vatant
• One space-time event Extreme Markup Languages
  2002
• One concept Content Structure Engineering

A
CS
BV
R2
R1
R3
Bernard Vatant is instructor of a tutorial about
Content Structure Engineering hold at Extreme
Markup Languages 2002
EM
17
Example

• List alle lektorerne i italiensk i efteråret
  2003
• (List all associated professors in Italian in the
  Autumn 2003)
• list-all(x) lektor(x),
  subject(italian),
• time(autumn-2003)

18
Example, cont.

• At course-assoc
  
• Rt1 instructor
• Rt2 subject
• Rt3 institution
• Rt4 time
• instructor instructor
• professore professore ricercatore
  professor lektor UA
• ordinario associato

19
Example, cont.

• list-all(x) lektor(x),
  subject(italian),
• time(autumn-2003)
• list-all(x) instructor(x), OR
  list-all(x) prof-associato(x),
• subject(italian), subject(italian),
• time(autumn-2003) time(autumn-2003)

20
Answer generation - example

• kurser i datalingvistik (courses in
  computational linguistics)
• ...educational programme in computational
  linguistics, Göteborg University. A Swedish
  program offering bachelor's and master's degrees
• ...Lund Universitys curriculum 2001-2002.
  Computational linguistics deal with automatic
  analysis of texts and other linguistic
  material...
• (Result of a Google search texts are not tagged
  with concepts!
• Bold face added to relevant information)

21
Answer generation, cont.

• I have found the following courses
• ...educational programme in computational
  linguistics, Göteborg University. A Swedish
  program offering bachelor's and master's degrees
• ...Lund Universitys curriculum 2001-2002.
  Computational linguistics deal with automatic
  analysis of texts and other linguistic
  material...
• The introductory sentence should be in the
  language of the query!

22
Answer generation, cont.

• I have found the following courses
• Göteborg University, bachelors and masters
  degrees
• link
• Lund University link
• Introductory sentence and relevant concepts
  (bachelors and masters degrees) should be in
  the language of the query!

23
More information

• MOSES web site coming up soon
• Link from www.cst.dk
• THANK YOU