SP500207 Text Retrieval with the TRW Fast Data Finder chapter M. Mettler National Institute of Standards and Technology Donna K. Harman define ATT `[AT\&[Iamp\;)TIAT and TI\ American Telephone [and I\&) Telegraph)' end The "[lamp;]" notation means to allow an optional "amp;" as was present in the Ziff database. We had the high score on Topic 10 "AIDS Treatments". This may be due to our ability to easily find phrases like "acquired immune deficiency syndrome" or "AIDS related complex" in close proximity to drug names like "ThA", "5-fluorouracil", or "AZT". We had the high score on topic 13 to find documents about Mitsubishi Heavy Industries. Our query that found 111 of the 112 documents the NIST judged relevant, was simply to find the two word phrase "Mitsubishi Heavy". Apparently the other TREC participants had trouble either finding phrases or determining the need to find phrases during query generation. The following sections discuss in detail topic 47 where we achieved the high score, and topic 36 where we achieved a low score. 4.1 Example of Good Performance - Topic 47 Topic 47 was to find documents discussing new contracts for computer systems in excess of $1 million. We found 80 good documents out of 200 submitted, the high score for this topic. We believe we did well on this topic because we were able to look for various numeric representations of $1 million in close proximity to keywords for new contracts and computer systems. To be relevant, a document must identify the selection of a source for the development or delivery of information systems products or services valued at more than $1 million dollars. The PSL query for this topic used three subqueries: one each for the "selection of a source", information systems products or services", and "more than $1 million dollars". The PSL definitions were: define award [3 words -> `[siqnjaward)*I and "contract") end define computer "[computer I communic network phone telecommi mainframe StarlanlpBxlcyber IBM 30901X\-MPIY\-MPISCS\-4olinformation)" end define million [1 word -> "[millionIbillion)[OCRerr]dollar" or I [j [0-9)) [I [0-9)) [0-9] [ j\. ] [[0-9)] [I [0-9)) [I [0-9] ]\ [millionibillion)" or I) [I [0-9)] [1(0-9]) [0-9) [I\,] [0-9] [0-9) [0-9) [ I\,)\ [0-9)[0-9)[0-9]"} end The "award" definition requires the root words "sign" or "award" to be within 3 words of contract" in the text. This word count includes stop words, acronyms, or any other alphanumerics that were in the original text. This definition will find phrases like: a contract was awarded AT&T signed a new contract Bellcore was awarded three new contracts. The "computer" definition looks for any of the root terms shown. Note that looking for alphanumerics such "X-MP" or "IBM 3090", which may include multiple character white 314