SP500215 Recent Developments in Natural Language Text Retrieval chapter T. Strzalkowski J. Carballo National Institute of Standards and Technology D. K. Harman representations of sentences. Introduction of compound terms also complicates the task of discovery of various semantic relationships among them, including synonymy and subsumption. For example, the term natural language can he considered, in certain domains at least, to subsume any term denoting a specific human language, such as English. Therefore, a query containing the former may be expected to retrieve documents containing the latter. The same can he said about language and English, unless language is in fact a part of the compound term programming language in which case the association language - Fortran is appropriate. This is a problem because (a) it is a standard practice to include both simple and compound terms in document representation, and (1,) term associations have thus far been computed primarily at word level (includ- ing fixed phrases) and therefore care must he taken when such associations are used in term matching. This may prove particularly troublesome for systems that attempt term clustering in order to create "meta-terms" to he used in document representation. The system presented here computes term associa- tions from text at word and fixed phrase level and then uses these associations in query expansion. A fairly primitive filter is employed to separate synonymy and subsumption relationships from others including anto- nymy and complementation, some of which are strongly domain-dependent. This process has led to an increased retrieval precision in experiments with both ad-hoc and routing queries for IREC-1 and ThEC-2 experiments. However, the actual improvement levels can vary sub- stantially hetween different databases, types of runs (ad- hoc vs. routing), as well as the degree of prior processing of the queries. We continue to study more advanced clustering methods along with the changes in interpreta- tion of resulting associations, as signaled in the previous paragraph. In the remainder of this paper we discuss particu- lars of the present system and some of the observations made while processing TREC-2 data. The above coin- ments will provide the background for situating our present effort and the state-of-the-art with respect to where we should he in the future. OVERALL DESIGN Our information retrieval system consists of a trad- itional statistical backbone (NIST's PRISE system; Har- man and Candela, 1989) auginented with various natural language processing components that assist the system in database processing (steiming, indexing, word and phrase clustering, selectional restrictions), and translate a user's information request into an effective query. This design is a careful compromise hetween purely statistical non-linguistic approaches and those requiring rather 124 accomplished (and expensive) semantic analysis of data, often referred to as `conceptual retrieval'. In our system the database text is first processed with a fast syntactic parser. Subsequendy certain types of phrases are extracted from the parse trees and used as compound indexing terms in addition to single-word terms. The extracted phrases are statistically analyzed as syntactic contexts in order to discover a variety of simi- larity links hetween smaller subphrases and words occur- ring in them. A further filtering process maps these simi- larity links onto semantic relations [OCRerr]eneralization, spe- cialization, synonymy, etc.) after which they are used to transform a user's request into a search query. The user's natural language request is also parsed, and all indexing terms occurring in it are identilied. Cer- tain highly ambiguous, usually single-word terms may he dropped, provided that they also occur as elements in some compound terms. For example, "natural" is deleted from a query already containing "natural language" because "natural" occurs in many unrelated contexts: "natural numher", "natural logarithin", "natural approach", etc. At the same time, other terms may he added, namely those which are linked to some query term through admissible similarity relations. For exam- ple, "unlawful activity" is added to a query [OCRerr]EC topic 055) containing the compound term "illegal activity via a synonymy link hetween "illegal" and "unlawful". One of the striking observations made during the course of ThEC-2 was to note that removing low-quality terms from the queries is at least as important (and often more so) as adding synonyms and specializations. In some instances (e.g., routing runs) low-quality terms had to he removed (or inhibited) before sin:ular terms could he added to the query or else the effect of query expan- sion was all but drowned out by the increased noise.' Mter the final query is constructed, the database search follows, and a ranked list of documents is returned. It should he noted that all the processing steps, those performed by the backbone system, and those per- formed by the natural language processing components, are fully automated, and no human intervention or manual encoding is required. FAST PARSING WITH TTP PARSER TIP Cragged Text Parser) is based on the Linguis- tic String Grammar developed by Sager (1981). The parser currendy encompasses some 400 grammar pro- ductions, but it is by no means complete. The parser's output is a regularized parse tree representation of each `We would like to thank Donna Hannan for tunling our attention to the importance of term weighting achemes, including term deletion.