Transcript aied09_panel - University of Pittsburgh

The Evolution of Natural Language
Processing in AIED:
Successes and Challenges
Diane Litman
Learning Research and Development Center
&
Computer Science Department
University of Pittsburgh
Pittsburgh, PA USA
What have we learnt from the past what works?
 Tutorial
dialogue (human and computer) "works"
– Significant learning gains
– … even with noisy Natural Language Understanding
– … and noisier Speech Recognition
What have we learnt from the past what does not (at least as expected)?
 Dialogue
doesn't necessarily work "better"
– Mixed results on the "Interaction Hypothesis"
» What type of system?
» What type of student?

“Dialogue” can be broadly construed
– Vicarious learning results
What have we learnt from the past what works?
 Automated
Scoring/Evaluation of student responses
– “works” enough to be commercialized
What have we learnt from the past what works and what does not?
 Automated
Scoring/Evaluation of student responses
– “works” enough to be commercialized
– focused applications
» Analysis of Meaning
» Analysis of Form

Grammatical Error Detection, Discourse Analysis, Spoken Language
What are the next challenges to NLP for
AIED (and who will benefit)?
 Data
Mining of online text and speech
– For teachers and/or students
» Automatic identification or generation of materials at a particular readability
or grade level
» Automatic generation of test questions
» Processing of and access to online lecture materials
 Accessibility/Globalization
– Machine translation
 Data
Mining of human tutorial interactions
– For researchers
» eliminate manual authoring of systems!
How shall AIED respond to rapid
technological advancements
Value-added,
technology
not gratuitous, use of new
The broad view of systems that care – what
ways of caring can we provide

Natural language provides many signals to (nonobtrusively) detect user states, and to convey system
“caring”
– What is said
– How it is said

Increasing theoretical and empirical work