CCG: Funding

Themes | Projects | Funding | Past Projects

Natural Language Processing Technology for Guided Study of Bioinformatics

Period: 2004-2007

Recent advances in Natural Language Processing, in particular the ability to use unstructured data to answer natural language questions, are very exciting from an educational perspective. They offer the promise of systems that can automatically respond to students' questions, thus supporting not only a guided but also and open-ended, exploration based, approach to learning.

Developing software that supports students' learning is all about construtcing the right kind of environment for students, one that facilitates rather than inhibits inquiry through a known knowledge space and provides a jumping-off space for trying to find or generate new knowledge. The goal os this project is to apply research in Computer Science -- particularly Natural Language Processing -- and the Learning Sciences, to developing an intelligent tutor that can provide this needed environment. This tutor will inhabit a human-computer interactive environment in which the computer is able to detect and track the user's cognitive and academic state and act based on this knowledge to aid the student in identifying and accessing relevant knowledge, contribute relevant factual information the student may need and guide the student in selecting potentially relevant subtasks.

The testbed domain in this project involves high school and udergraduate level students studying concepts in Bioinformatics. Bioinformatics is an exceptionally good test bed for development of smart interactive tutoors because the Bioinformatics community has already made enormous amounts of biological data and software freely available on the Web. This provides unparalleled access to cutting-edge scientific resources in an environment where advances in pedagogy can be readily documented, evaluated, and disseminated.

Intellectual Merit

In the context of this project we will attempt to

develop the necessary machine learning, natural language and inference methods that can robustly support a level of natural language understanding that is sufficient to "understand" students and their queries to the level that it can direct it to the right material, make relevant suggestions and develop a meaningful dialog in the context of the subject matter.
create a system that is able to accommodate different student backgrounds and goals and behave appropriately, and learns as it does so, and
study how students learn and how to support students' learning in a computer-aided context.

The problem of developing intelligent tutors for teaching science, we feel, can be best addressed by in interdisciplinary team containing the elements we have gathered; computer scientists, biologists, classroom teachers, and education researchers, server as a prototype for a working partnership among these activities.

Broader Impact

Just giving students sophisticated software does not guarantee that learning will occur. Despite investing an enormous amount of money in computer technology, we are still not sure we have a good model for learning with the aid of computers. For example, these is a pronounced variance in the performance and learning associated with software use. Some students do extraordinarily well, using the software to explore and develop their own understanding in a way that is both efficent and clearly highly motivating. Unfortunately, many others find this open, free form of exploration bewildering.

Our project will contribute to the understanding of how students learning in these environments and use it to develop improved methods for supporting learning in a computer aided environment. This has the potentional for large educational impact for large classes, distance education, and self-paced instruction.

Our computational results in areas such as natural language based human machine interaction, adaptive dialog management, user-sensitive information retrieval and extraction, and machine learning, would be widely aplicable to many other domains, including intelligent information access and interactive support systems for senior citizens and other groups.

From a Bioinformatics perspective, this projects has the potential to go beyond inproving education to lay the foundation for improved computational environments for research in, and research using, Bioinformatics.

Relevant Publications:

A. Klementiev, D. Roth, K. Small, and I. Titov, Unsupervised Rank Aggregation with Domain-Specific Expertise. Proc. of the International Joint Conference on Artificial Intelligence (IJCAI) (2009) [bibitem]
A. Klementiev, D. Roth, and K. Small, A Framework for Unsupervised Rank Aggregation. Proc. of the SIGIR Workshop on Learning to Rank for Information Retrieval (2008) pp. 32-39 [bibitem]
A. Klementiev, D. Roth, and K. Small, Unsupervised Rank Aggregation with Distance-Based Models. Proc. of the International Conference on Machine Learning (ICML) (2008) [bibitem]
M. Connor and D. Roth, Context Sensitive Paraphrasing with a Single Unsupervised Classifier. Proc. of the European Conference on Machine Learning (ECML) (2007) [bibitem]
A. Klementiev, D. Roth, and K. Small, An Unsupervised Learning Algorithm for Rank Aggregation. Proc. of the European Conference on Machine Learning (ECML) (2007) [bibitem]
D. Roth and K. Small, Margin-based Active Learning for Structured Output Spaces. Proc. of the European Conference on Machine Learning (ECML) (2006) [bibitem]
D. Roth and K. Small, Active Learning with Perceptron for Structured Output. ICML-06 Workshop on Learning in Structured Output Spaces (2006) [bibitem]
C. Alm, D. Roth, and R. Sproat, Emotions from text: machine learning for text-based emotion prediction.. Proceedings of HLT/EMNLP 2005 (2005) [bibitem]