Bikel, D. M., Miller, S., Schwartz, R., and Weischedel, R. Nymble: a high-performance learning name- nder. In Fifth Conference on Applied Natural Language Processing (1997).  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... improvements Find the best rule Update NE labels in training data Generated rules Rule templates Rule generation Preprocessing Figure 1: Procedures for preprocessing and rulegeneration Characteristics of a word, called the word features, sometimes give important clues for NE recognition [4]. For example, the capitalisation of the rst character of a word, except for the rst word of a sentence, gives that word a higher possibility of being a proper noun named entity word. Table 1 lists some possible word features. A deterministic computation must be able to be performed to obtain ....

D. Bikel, S. Miller, and R. Schwartz. Nymble: a HighPerformance Learning Name-nder. In Proc. Applied Natural Language Processing, pages 194-201, 1997.

....Although the assumption that a word s part of speech or name class can be predicted by the previous n 1 words and their classes is counter intuitive to our understanding of linguistic structures and long distance dependencies, this simple method does seem to be highly e ective in practice. Nymble (Bikel et al. 1997), a system which uses HMMs is one of the most successful such systems and trains on a corpus of marked up text, using only character features in addition to word bigrams. Although it is still early days for the use of HMMs for IE, we can see a number of trends in the research. Systems can be ....

D. Bikel, S. Miller, R. Schwartz, and R. Wesichedel. 1997. Nymble: a highperformance learning name-nder. In Proceedings of the Fifth Confererence on Applied Natural Language Processing, pages 194-201.

....part of speech (or semantic class) label or word frequency in the target domain. For cases where the observation o i contains only alphabetic characters but is labeled as an NSW because it is out of vocabulary, it may be useful to compute features of the word to predict the tag probability, as in (Bikel et al. 1997). Observation Model. For purposes of simplifying the discussion, assume that the hypothesized word sequence can be reliably parsed so that there is a one to one mapping between observation tokens o i and words w i . Next, we assume that the observed realization of a word will be conditionally ....

Bikel, D, S. Miller, Richard Schwartz, and Ralph Weischedel. 1997. Nymble: a high-performance learning name-nder. In Applied Natural Language Processing Conference, pages 194-201.

....in bibliographies These conventions strive to give a collection of articles (e.g. journals, edited volumes) a uniform look. Some styles di er from each other signi cantly while others only have minor di erences. Consider a reference for the same article in IEEE, AAAI, and NEWAPA, respectively: [1] S. Lawrence, C.L. Giles, K. Bollacker, Digital libraries and autonomous citation indexing , IEEE Computer, vol. 6, no.4, pp. 67 71, 1999. Lawrence, Giles, Bollacker1999] Lawrence, S. Giles, C.L. and Bollacker, K. 1999. Digital libraries and autonomous citation indexing ing, IEEE ....

.... Traditionally, HMMs have been used for statistical pattern recognition and signal processing, e.g. speech recognition [9] topic spotting [10] and part of speech tagging [5] More recently, they have also been used for protein sequence analysis [3] and extraction of information from texts [1, 8]. These approaches typically use hand crafted models assembled by inspecting training data [2] A few approaches also learn the unknown model (e.g. 4] A hidden Markov model probabilistically links the observation to the state transitions in the system. The theory provides a means by which ....

R. S. D.M. Bikel, S. Miller and Weischedel, \Nymble: A high-performance learning name nder," in Proceedings of ANLP-97, pp. 194-201, 1997. of bibliographic entries, and  is a signcance level. 10

....or large lists, such systems can be extremely expensive to develop and maintain. While a variety of name detection algorithms have been proposed in the literature, in this paper, we mention only those that incorporate a strong machine learning component. The best known of these systems is nymble [ Bikel et al. 1997 ] a statistical system based on a Hidden Markov Model (HMM) Rabiner, 1993 ] nymble is reported to have an F 1 score of 93. While nymble s approach is e ective, it requires large computational resources. Another system using machine learning techniques with similar, but slightly lower ....

D. Bikel, S. Miller, R. Schwartz, and R. Weischedel. nymble: a high-performance learning name-nder. In Proceedings of the Fifth Conference on Applied Natural Language Processing, pages 194-201, Washington, D.C., 1997. ACL.

....and robustness to sparse data through shrinkage (see Section 4) 4 Related Work and Conclusions Various related work has used mixture models for modeling text. The Nymble system from BBN uses an HMM with uniform transition probabilities to perform information extraction of named entities [ Bikel et al. 1997 ] Similar models have been used for text segmentation and topic tracking [ Yamron et al. 1998 ] The most related work is that of Imai et al. 1997 ] in which a mixture model is used for multi label document classi cation. However, their work does not have a class set conditional mixture ....

Daniel M. Bikel, Scott Miller, Richard Schwartz, and Ralph Weischedel. Nymble: a high-performance learning name-nder. In Proceedings of ANLP97, pages 194-201, 1997.

....learning the appropriate model structure (the number of states and transitions) automatically from data. Other systems using HMMs for information extraction include that by Leek [ 1997 ] which extracts information about gene names and locations from scienti c abstracts, and the Nymble system [ Bikel et al. 1997 ] for named entity extraction. These systems do not consider automatically determining model structure from data; they either use one state per class, or use hand built models assembled by inspecting training examples. 4.1 Experiments The goal of our information extraction experiments is to ....

D. Bikel, S. Miller, R. Schwartz, and R. Weischedel. Nymble: a high-performance learning name-nder. In ANLP-97, 1997.

....by means of the Viterbi algorithm, an e cient method for nding the most probable sequence of model states corresponding to a given document. A few previous projects have used HMMs for information extraction, although with di ering structures, and none with shrinkage over HMM states (Leek 1997; Bikel et al. 1997). A related project focuses on the quite di erent task of learning HMM state transition structure for information extraction (Seymore, McCallum, Rosenfeld 1999) We describe experiments on two real world data sets: on line seminar announcements and Reuters newswire articles on company ....

....states. Leek uses network topology to model natural language syntax and trims training examples for presentation to the model. States are unigram language models, as in our work, but it is unclear what smoothing policy is used. Unknown tokens are handled by special gap states. The Nymble system (Bikel et al. 1997) uses HMMs to perform named entity extraction as de ned by MUC 6. All di erent elds to be extracted are modeled in a single HMM, but to avoid the resulting di cult structure learning problem, there is a single state per target and the state transition structure is completely connected. ....

Bikel, D. M.; Miller, S.; Schwartz, R.; and Weischedel, R. 1997. Nymble: a high-performance learning name-nder.

No context found.

Bikel, D. M., Miller, S., Schwartz, R., and Weischedel, R. Nymble: a high-performance learning name- nder. In Fifth Conference on Applied Natural Language Processing (1997).

No context found.

D. Bikel, S. Miller, R. Schwartz, and R. Wesichedel. Nymble: a high-performance learning name- nder. In Proceedings of the Fifth Confererence on Applied Natural Language Processing, pages 194-201, 1997.

No context found.

Daniel M. Bikel, Scott Miller, Richard Schwartz, and Ralph Weischedel. Nymble: a high-performance learning name-nder. In Proceedings of ANLP-97, 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC