Memory-Based Named Entity Recognition using Unannotated Data

Memory-Based Named Entity Recognition using Unannotated Data Fien De Meulder and Walter Daelemans CNTS - Language Technology Group University of Antwerp {Fien.DeMeulder,Walter.Daelemans}@ua.ac.be Abstract We used the memory-based learner Timbl (Daelemans et al., 2002) to find names in En- glish and German newspaper text. A first sys- tem used only the training data, and a number of gazetteers. The results show that gazetteers are not beneficial in the English case, while they are for the German data. Type-token gen- eralization was applied, but also reduced per- formance. The second system used gazetteers derived from the unannotated corpus, as well as the ratio of capitalized versus uncapitalized use of each word. These strategies gave an increase in performance. 1 Introduction This paper describes a memory-based approach to learn- ing names in English and German newspaper text. The first system used no unannotated data - only the provided training material, and a number of gazetteers. It was shown that the gazetteers made for a better per- formance in the German task, but not in the English task. Type-token generalization was helpful for neither English nor German. The second system used unannotated data, but only for the English task. The extra data were used in two ways: first, more gazetteers were derived from the corpus by exploiting conjunctions: if in a conjunction of capitalized strings one string is recognized as being a certain type of name, the other strings are assumed to be of the same type and stored in a new gazetteer. This list was then used to construct an additional feature for training the machine learning algorithm. The second approach counts how of- ten each word form in the additional corpus is capitalized, and how often it is not. This is used as another feature for the learning algorithm. 2 Memory Based Learning We used Timbl (Daelemans et al., 2002), a memory- based learner. When presented with training instances, the learner stores them all, and then classifies new data on the basis of its k nearest neighbours in the training set. Before classification, the learner assigns weights to each of the features, marking their importance for the learning task. Features with higher weights are treated as more important in classification as those with lower weights. Timbl has some parameters which can be adjusted in order to improve learning. For the NER system described in this paper, we varied the parameters k and m. k is the number of nearest neighbours Timbl looks at. m deter- mines the feature metrics, i.e. the importance weights given to each feature, and the way similarity between values of the same feature is computed. This parame- ter can be adjusted separately for each feature. The two metrics used were weighted overlap and modified value difference. 3 System 1: Description 3.1 Features For the basic English system, 37 features were used. The first seven features were the lowercase versions of the fo- cus word, and a context of three words to the left and the right. The next …

Content truncated to 500 words