Some experiments with a Naive Bayes WSD system

Deniz Yuret. In Proceedings of SENSEVAL-3, the Third International Workshop on the Evaluation of Systems for the Semantic Analysis of Text



Abstract: This document describes the architecture of a WSD system that participated in the SENSEVAL-3 English all words evaluation exercise. The system uses two independent statistical models, one based on local collocations and another based on a bag of words around the target. The model with the higher confidence provides the final answer for each instance. Both models use Naive Bayes and supervised training with different feature sets. The experiments using this system indicate that the specific smoothing parameters used for Naive Bayes make a big impact on the performance, smaller context sizes give better accuracy, and that the bag of words model adds little to the performance.

Full post... Related link

Relationships Between Amino Acid Sequence and Backbone Torsion Angle Preferences

Özlem Keskin, Deniz Yuret, Attila Gürsoy, Metin Türkay and Burak Erman. In Proteins: Structure, Function, and Bioinformatics. 55(4):992-8. (PDF)

Abstract: Statistical averages and correlations for backbone torsion angles of chymotrypsin inhibitor 2 are calculated by using the Rotational Isomeric States model of chain statistics. Statistical weights of torsional states of phi-psi pairs, needed for the statistics of the full chain, are obtained in two different ways: 1) by using knowledge-based pairwise dependent phi-psi energy maps from Protein Data Bank (PDB) and 2) by collecting torsion angle data from a large number of random coil configurations of an all-atom protein model with volume exclusion. Results obtained by using PDB data show strong correlations between adjacent torsion angle pairs belonging to both the same and different residues. These correlations favor the choice of the nativestate torsion angles, and they are strongly context dependent, determined by the specific amino acid sequence of the protein. Excluded volume or steric clashes, only, do not introduce context-dependent phi-psi correlations into the chain that would affect the choice of native-state torsional angles.

Full post... Related link

Omnibase: Uniform Access to Heterogeneous Data for Question Answering

Boris Katz, Sue Felshin, Deniz Yuret, Ali Ibrahim, Jimmy J. Lin, Gregory Marton, Alton Jerome McFarland, Baris Temelkuran. In Birger Andersson, Maria Bergholtz, Paul Johannesson (Eds.): Natural Language Processing and Information Systems, 6th International Conference on Applications of Natural Language to Information Systems, NLDB 2002, Stockholm, Sweden, June 27-28, 2002, Revised Papers. Lecture Notes in Computer Science 2553 Springer 2002, ISBN 3-540-00307-X.

Abstract: Although the World Wide Web contains a tremendous amount of information, the lack of uniform structure makes finding the right knowledge difficult. A solution is to turn the Web into a virtual database and to access it through natural language. We built Omnibase, a system that integrates heterogeneous data sources using an object-property-value model. With the help of Omnibase, our Start natural language system can now access numerous heterogeneous data sources on the Web in a uniform manner, and answers millions of user questions with high precision.

Download PDF.

Full post... Related link

Alpha-beta-conspiracy search

David McAllester and Deniz Yuret. ICGA Journal Vol. 25, No. 1 - March 2002 (PDF)

Abstract: We introduce a variant of α-β search in which each node is associated with two depths rather than one. The purpose of α-β search is to find strategies for each player that together establish a value for the root position. A max strategy establishes a lower bound and the min strategy establishes an upper bound. It has long been observed that forced moves should be searched more deeply. Here we make the observation that in the max strategy we are only concerned with the forcedness of max moves and in the min strategy we are only concerned with the forcedness of min moves. This leads to two measures of depth --- one for each strategy --- and to a two-depth variant of α-β called ABC search. The two-depth approach can be formally derived from conspiracy theory and the structure of the ABC procedure is justified by two theorems relating ABC search and conspiracy numbers.

Full post... Related link

Discovery of Linguistic Relations Using Lexical Attraction

Deniz Yuret, PhD Thesis, MIT, May 1998. (HTML, PDF, PS.GZ, Slides, Summary paper)

Abstract
This work has been motivated by two long term goals: to understand how humans learn language and to build programs that can understand language. Using a representation that makes the relevant features explicit is a prerequisite for successful learning and understanding. Therefore, I chose to represent relations between individual words explicitly in my model. Lexical attraction is defined as the likelihood of such relations. I introduce a new class of probabilistic language models named lexical attraction models which can represent long distance relations between words and I formalize this new class of models using information theory.

Within the framework of lexical attraction, I developed an unsupervised language acquisition program that learns to identify linguistic relations in a given sentence. The only explicitly represented linguistic knowledge in the program is lexical attraction. There is no initial grammar or lexicon built in and the only input is raw text. Learning and processing are interdigitated. The processor uses the regularities detected by the learner to impose structure on the input. This structure enables the learner to detect higher level regularities. Using this bootstrapping procedure, the program was trained on 100 million words of Associated Press material and was able to achieve 60% precision and 50% recall in finding relations between content-words. Using knowledge of lexical attraction, the program can identify the correct relations in syntactically ambiguous sentences such as ``I saw the Statue of Liberty flying over New York.''

Full post...