Gerard Salton

1927-1995

In Memory of:

Gerard Salton
Professor
PhD Harvard University, 1958

For information about Professor Salton's research, please contact Chris Buckley

Natural-language text processing is a rapidly expanding field of research and development. Large masses of machine-readable text now exist that can be cheaply stored on high-density optical storage media and rapidly retrieved on demand. Furthermore, sophisticated methods are available for analyzing document texts, formulating appropriate user queries, conducting rapid file searches, and ranking the retrieved items in decreasing order of importance to the users.

At Cornell, we design and operate large, general-purpose text processing environments where texts can be handled without restrictions on size or subject matter. In the absence of knowledge bases that would be useful for unrestricted text databases, we use corpus-based text analysis systems that determine the meaning of words and expressions by a refined context analysis using statistical and probabilistic criteria. Using the corpus-based approaches, we are able to determine text similarity with a high degree of accuracy. There are two main applications:

The automatic generation of structured text collections (hypertext) where semantically similar pieces of text are automatically linked. Hypertext representations of large databases provide flexible browsing capabilities for general-purpose text access.
The automatic retrieval of interesting text excerpts in response to available search queries.

We have done extensive work with an automated encyclopedia containing about 25,000 encyclopedia articles (the Funk and Wagnalls New Encyclopedia). In addition, we are processing the TREC collection of about 800,000 full-text documents covering a number of different subject areas (over 2 gigabytes of text).

A sophisticated search and retrieval service exists, as well as a text linking system capable of relating different text sections, paragraphs, and sentences. The main test vehicle continues to be the current version of the SMART text analysis and retrieval system, operating under UNIX on Sun Sparc Stations and Sun-4 terminal equipment.

University Activities

Member, Engineering College Library Committee

Professional Activities

Associate Editor, ACM Transactions on Information Systems
Program Committee: SIGIR-95 Eighteenth International Conference on Research and Development in Information Retrieval, Seattle 1995; SIGIR-96 Nineteenth International Conference on Research and Development in Information Retrieval Zurich, Switzerland, 1996

Awards

Fellow, Association for Computing Machinery

Lectures

Automatic hypertext construction for structured texts. Swiss Federal Institute of Technology (ETH), Zurich, Switzerland, June 6, 1994.
Automatic text structuring and decomposition. Computer Science Colloquium, University of Maryland, College Park, MD, November 1, 1994.
Automatic passage retrieval. TREC-3 Conference Workshop, Gaithersburg, MD, November 2, 1994.
Modern Information Retrieval. Computer Science Colloquium, Dartmouth College, Hanover, NH, April 1995.

Publications

Automatic analysis, theme generation and summarization of machine-readable texts. Science, 264, 3 (June 1994), 1421-1426 (with J. Allan, C. Buckley, and A. Singhal).
The effect of adding relevance information in a relevance feedback environment. Proceedings SIGIR-94, W.B. Croft and C.J. van Rijsbergen, editors. Springer-Verlag, London, (July 1994), 292-301 (with C. Buckley and J. Allan).
Automatic text decomposition and structuring. Proceedings RIAO-94 Conference, CID, Paris, October 1994, 6-20 (with J. Allan).
Automatic query expansion using SMART - TREC-3. An Overview of the Third Text Retrieval Conference (TREC 3), D.K. Harman, editor. National Institute of Science and Technology. Special Publication 500-225, Gaithersburg, MD, (1995), 69-80 (with C. Buckley, J. Allan, A. Singhal).

Software

The SMART text analysis and retrieval system is made available free of charge for research purposes. Several hundred copies of SMART (version 11) have been distributed and are used around the world.

Return to:

1994-1995 Annual Report Home Page
Departmental Home Page If you have questions or comments please contact: www@cs.cornell.edu. Last modified: 16 October 1998 by Elly Cramer, Cornell University, ADM (elly@cs.cornell.edu).