Steven Lytinen's Research Projects

Currently I am involved in the following research projects:

Question Answering Systems. There are 2 thrusts to this work:

TREC-style Question Answering. Each year, the Text Retrieval Conference sponsors a Question Answering track, in which computer systems attempt to find the answers to a collection of questions in a corpus of documents. The system must identify not just the document which contains the answer to the question, but must identify the answer itself.
While we have not written a stand-alone question-answering system to compete in TREC, recently we developed an answer verification system, which attempts to judge the likelihood that a candidate answer to a question is a plausible answer. The goal of the research is to improve TREC-style question-answering systems by providing a back-end to filter out implausible candidate answers.
A paper describing this work is
Smith, T., Repede, T., and Lytinen, S. (2005). Determining the plausibility of answers to questions. In Proceedings of the Workshop on Inference for Textual Question Answering, Pittsburgh PA, July 2005. AAAI Press, Technical Report WS-05-05, p. 52-58.
FAQFinder. This project is in collaboration with Noriko Tomuro, Robin Burke, and Stan Mlynarczyk (DePaul), Kristian Hammond (Northwestern University), and Vladimir Kulyukin (Utah State University). The goal of this project is to develop a natural language-based browser of Internet Frequently Asked Questions (FAQ) files. Each of these files contain questions about a particular topic that is frequently asked by participants of a news group, along with answers to these questions. To use FAQFinder, a user simply types a question to the system, about any topic at all. The system then searches FAQ files for a frequenty-asked question which is similar to the user's, and retrieves the answer to this question in the FAQ file. We have implemented a prototype of the system which works quite well using a limited number of FAQ files.
We have incoporated an automated sense-tagging algorithm into FAQFinder, with improved results. More recently, we have investigated using question category as a metric to improve matching of questions.
Papers describing this work include:
Mlynarczyk, S., and Lytinen, S. (2005). FAQFinder question answering improvements using question/answer matching. In Proceedings of L&T-2005 - Human Language Technologies as a Challenge for Computer Science and Linguistics, Poznan Poland, April 2005.
Lytinen, S. and Tomuro, N. (2002). The use of question types to match questions in FAQFinder. In Proceedings of the 2002 AAAI Spring Symposium on Mining Answers from Texts and Knowledge Bases, p. 46-53.
Lytinen, S., Tomuro, N. and Repede, T. (2000). The use of WordNet sense tagging in FAQFinder. . In Proceedings of the AAAI-2000 workshop on AI and Web Search, Austin TX, July 2000.
Burke, R., Hammond, K., Kulyukin, V., Lytinen, S., Tomuro, N. and Schoenberg, S. (1997). Question answering from Frequently Asked Question files: Experiences with the FAQFinder system. AI Magazine, 18(2), p. 57-66.

Agent-based Modeling of Ecological Systems. I am working with Steven Railsback of Humboldt State University's Environmental Modeling program, and Volker Grimm of the UFZ Center for Environmental Research in Leipzig, Germany. The main goal of the project so far has been to compare the many different software platforms that exist to facilitate the development of Agent-based Models.

The following paper describes a comparison study we have undertaken of 5 ABM platforms:

Railsback, S., Lytinen, S., and Jackson, S. (2006). Agent-based Simulation Platforms: Review and Development Recommendations. Simulation 82(9), p. 609 - 623.

ARCH: An Adaptive Agent for Retrieval based on Concept Hierarchies. This project is in collaboration with Bamshad Mobasher, Robin Burke and Ahu Sieg. ARCH is a tool which uses a concept hierarchy (currently the Yahoo hierarachy) to assist a user in constructing a query for searching the Web. The user initially types a keyword query, similar to most search engines. Then, the system displays a portion of a concept hierarchy which is most likely to be relevant to the user's query. The user can then select nodes in the hierarchy which are most relevant to his or her query. The system uses this information to modify the query, using Rocchio's method. ARCH is on the Web, at http://arch.ahusieg.com, although the current prototype uses only a small subset of the Yahoo hierarchy and has a very small number of Web pages indexed.

Papers describing this work include:

Sieg, A., Mobasher, B., Burke, R., Prabu, G., and Lytinen, S. (2005). Representing User Information Context with Ontologies. In Proceedings of the 3rd International Conference on Universal Access in Human-Computer Interaction, Las Vegas, NV, July 2005.
Sieg, A., Mobasher, B., Lytinen, S., and Burke, R. (2003). Concept based query enhancement in ARCH. In Proceedings of the International Conference on Internet Computing 2003 (IC'2003), Las Vegas, June 2003.
Parent, S., Mobasher, B., and Lytinen, S. (2001). An adaptive agent for web exploration based on concept hierarchies. In Proceedings of the Ninth International Conference on Human-Computer Interaction, New Orleans LA, August 2001. Published in Smith, M., Salvendy, G, Harris, D., and Koubek, R. (eds), Usability Evaluation and Interface Design. Lawrence Erlbaum Associates, Mahwah NJ, p. 903-907.

LINK: A Unification-based Parser. This work is in collaboration with Noriko Tomuro. LINK is a unification-based parser which uses an extension of left-corner parsing (originally from context-free grammars). The notion of a reachability table, which provides information as to whether or not a top-down expectation can be realized by a potential subconstituent, is extended for unification grammar, thus allowing for all types of grammatical information to be used, including semantics, rather than just phrase structure information, in parser predictions.

As part of this work, we formalized LINK's parsing algorithm, and related it to Shieber's abstract parsing algorithm for unification grammars. In the process, we discovered that Shieber's parser produces what we call nonminimal derivations, which contain more features than they ought to.

Papers describing this work include:

Lauterburg, S. (2004). On building a unification-based parser. DePaul CTI Technical Report 04-004.
Tomuro, N. and Lytinen, S. (2001). Abstract left-corner parsing for unification grammars. In Proceedings of the Natural Language Processing Pacific Rim Symposium (NLPRS 2001), Tokyo, Japan
Tomuro, N. and Lytinen, S. (2001). Nonminimal derivations in unification-based parsing. Computational Linguistics, 27(2), p. 277-285. A longer version of this paper is available as a technical report.

Also relevant is Noriko Tomuro's dissertation.