| there has been a dramatic increase in the application of probabilistic models to natural language processing over the last few years. the appeal of stochastic techniques over traditional rule-based techniques comes from the ease with which the necessary statistics can be automatically acquired and the fact that very little handcrafted knowledge need be built into the system. in contrast, the rules in rule-based systems are usually difficult to construct and are typically not very robust. one area in which the statistical approach has done particularly well is automatic part of speech tagging, assigning each word in an input sentence its proper part of speech [church 88; cutting et al. 92; derose 88; deroualt and merialdo 86; garside et al. 87; jelinek 85; kupiec 89; meteer et al. 911. stochastic taggers have obtained a high degree of accuracy without performing any syntactic analysis on the input. these stochastic part of speech taggers make use of a markov model which captures lexical and contextual information. the parameters of the model can be estimated from tagged ([church 88; derose 88; deroualt and merialdo 86; garside et al. 87; meteer et al. 91]) or untag,ged ([cutting et al. 92; jelinek 85; kupiec 89]) text. once the parameters of the model are estimated, a sentence can then be automatically tagged by assigning it the tag sequence which is assigned the highest probability by the model. performance is often enhanced with the aid of various higher level pre- and postprocessing procedures or by manually tuning the model. a number of rule-based taggers have been built [klein and simmons 63; green and rubin 71; hindle 89]. [klein and simmons 63] and [green and rubin 71] both have error rates substantially higher than state of the art stochastic taggers. [hindle 89] disambiguates words within a deterministic parser. we wanted to determine whether a simple rule-based tagger without any knowledge of syntax can perform as well as a stochastic tagger, or if part of speech tagging really is a domain to which stochastic techniques are better suited. in this paper we describe a rule-based tagger which performs as well as taggers based upon probabilistic models. the rule-based tagger overcomes the limitations common in rule-based approaches to language processing: it is robust, and the rules are automatically acquired. in addition, the tagger has many advantages over stochastic taggers, including: a vast reduction in stored information required, the perspicuity of a small set of meaningful rules as opposed to the large tables of statistics needed for stochastic taggers, ease of finding and implementing improvements to the tagger, and better portability from one tag set or corpus genre to another.we have presented a simple part of speech tagger which performs as well as existing stochastic taggers, but has significant advantages over these taggers. there has been a dramatic increase in the application of probabilistic models to natural language processing over the last few years. the fact that the simple rule-based tagger can perform so well should offer encouragement for researchers to further explore rule-based tagging, searching for a better and more expressive set of patch templates and other variations on this simple but effective theme. in addition, the tagger has many advantages over stochastic taggers, including: a vast reduction in stored information required, the perspicuity of a small set of meaningful rules as opposed to the large tables of statistics needed for stochastic taggers, ease of finding and implementing improvements to the tagger, and better portability from one tag set or corpus genre to another. the rule-based tagger overcomes the limitations common in rule-based approaches to language processing: it is robust, and the rules are automatically acquired. perhaps the biggest contribution of this work is in demonstrating that the stochastic method is not the only viable approach for part of speech tagging. the tagger is extremely portable. the appeal of stochastic techniques over traditional rule-based techniques comes from the ease with which the necessary statistics can be automatically acquired and the fact that very little handcrafted knowledge need be built into the system. |