Transcript rose_dzikovska_moore_quad_2008

Expanding the Accessibility and Impact of Language Technologies for Supporting
Education (TFlex): Edinburgh Effort
Dr. Myroslava Dzikovska, Prof. Johanna Moore, University of Edinburgh
September 2004 – December 2010
Screen shot of the TFlex Extractor Tool
Technical Approach
•
Summary of extraction results
Improve tools for educators through the use of more sophisticated
computational linguistics techniques
•
•
Make language analysis tools more accessible through userfriendly interfaces (more detail in CMU effort)
Improve language processing in tutorial dialogue systems by
developing tools to rapidly expand their lexicons
•
Unknown words cause language interpretation failures
• New words have to be added manually to each system, requiring
significant computational linguistics expertise
Issues detected and
marked for cleanup
•
Extract information from large-scale public domain lexical
databases and dictionaries
• Much of the necessary information is available from various
sources, but requires merging and cleaning up to be usable
Project Objective
•
Support large-scale deployment of language technology in learning
environments
•
•
Allow educational experts to use language understanding
capabilities in computer-supported learning environments
•
•
Accomplishments
•
•
e.g., automatically alert group facilitator when chat goes off
topic
Allow dialogue system developers to quickly and efficiently
deploy tutorial dialogue systems in new domains
support complex tasks like understanding student explanations
Navy Relevance
•
Support advanced computer-based training systems and improve
training effectiveness
•
Allow for cost-effective deployment of language technology in a wide
range of teaching and learning tools
First 3 years of funding resulted in tools that speed up development of
language interpretation components, and improve their accuracy and
efficiency
•
•
•
E.g., increased the verb lexicon of the BEETLEII dialogue system from
1,000 to about 2,500 verbs entries by extracting 1,196 lexical entries
from VerbNet lexical database
Published 5 peer-reviewed papers
Currently improving TFlex extraction tools by taking into account
additional information available from the FrameNet lexicon
Published 2 conference/workshop papers in 2008
Future plans
•
•
•
Develop tools for defining domain-specific terminology not found in
general databases
Integrate TFlex resources into BEETLE II tutorial dialogue system
Integrate TFlex tools and resources into computer-supported
collaborative learning environments
Expanding the Accessibility and Impact of Language Technologies for Supporting
Education (TFlex extension): CMU Effort
Dr. Carolyn Rosé, Carnegie Mellon University
September 2004 – December 2010
Screen shots from SIDE Summarization Integrated Development Environment
•
Technical Approach
–
Extend the tools and techniques developed in Edinburgh
and CMU during FY2005-2007
–
Specifically, building an infrastructure around basic tools
developed during FY2005-2007 to facilitate rapid prototyping
of dialogue-based learning environments
Boolean
Expression
Tree
–
Development effort so far has proceeded in two directions:
–
Development of SIDE Summarization Integrated
Development Environment, building on basic
technology developed during FY2005-2007, to facilitate
rapid development of reporting interfaces for facilitators
of collaborative learning interactions
–
Development of the Basilica framework for coordinating
a conversation between multiple humans and multiple
dialogue agents
Ranker
Limiter
•
Project Objective
–
–
First 3 years of funding produced publically available tools
with a current user base of 219 in 49 countries, 10 peer
reviewed publications, and 6 invited talks
•
Language is crucial in many learning tasks, so improving
language understanding will help improve learning
–
Completed version 1.0 of SIDE and Basilica, currently
piloting
•
Issues that need to be addressed to make language
technology more accessible
–
3 conference publications so far for 2008
–
Existing tools require computational linguistics
expertise, and are not accessible to educators or
domain experts
Systems in new domains are difficult to deploy
because of limited coverage in existing tools
Navy Relevance
–
Accomplishments
Support large-scale deployment of language technology in
learning environments
–
•
•
Improved computer-based training tools will improve the
training effectiveness for the Navy
•
Future Plans
–
Preparing for a tutorial at CMU in July
–
Collaborating with Dan Suthers, Nancy Law, and Nathan
Dwyer on an workshop related to developing a community
resource for analysis of collaborative learning data
–
Further develop and disseminate our tools, work towards
greater integration with Edinburgh technology