Research in Natural Language Processing (NLP) at UIC focuses on semantics, and discourse and dialogue processing. Our goal is to use NLP to support both education and instruction, and collaboration between human or artificial agents (for those readers who are not familiar with NLP, NLP studies the computational models that underlie the processing of human languages, and develops key technology that makes it possible for users to interact with a computer system using English, Italian or Japanese rather than a programming language).
Our group focuses on the computational modeling of extended text
(discourse) and conversations between two or more agents
(dialogue). The theoretical aspects of our research concern the linguistic
analysis, and the knowledge representation and reasoning that support
the understanding and generation of NL discourse and dialogue.
The intended applications range from
automatically producing instructional manuals (e.g., those that
accompany any piece of equipment such as a stereo), to providing
dialogue capabilities for Intelligent Tutoring Systems (ITSs),
computer based tutors that can help students master a subject. The
methodology we employ blends empirical and symbolic approaches, and
consists of: data mining from text corpora; development of
computational frameworks based on the information extracted from the
corpus; and rigorous evaluation of the computational models via user
studies.
A full list of publications is available
here .
Our major areas of interest right now are:
Computational models of
tutorial dialogue (supported by the Office of Naval Research).
This work concerns building
ITSs that can participate in a dialogue with their
users. The project DIAG-NLP has been one of the first to show that
a language interface to an ITS does engender more learning in
students (please see the following two papers, AIED05, and ACL05).
Currently, we are exploring what distinguishes expert from novice
tutors, in order to model the more effective tutors in an interface
to an ITS that tutors students on basic data structures and
algorithms. Read more
Lexical semantics
and inductive logic programming to learn discourse relations and
domain knowledge (supported by an NSF CAREER award). We employ a
novel methodology that couples a corpus parsed to obtain rich
semantic representations (HLT-NAACL03)
and annotated with discourse relations to
learn a first order model for discourse relations via inductive logic
programming. The final goal is to (semi)automatically acquire domain
knowledge about action verbs and rhetorical knowledge about how those
actions are related in instructional discourse. Read more
Modeling
collaboration in human-human and computer-human dialogues
(supported by NSF, Advanced Learning Technologies program).
We are extending our model of commitment in dialogue, developed under
the project Coconut project at the
University of Pittsburgh (see
IJHCS 00, [.ps.gz]). We are modeling peer interactions in
learning, and developing a peer dialogue agent in the domain of basic
data structures and algorithms. Read more
We are also active in other areas of research, including:
Collaborators:
Alumni: