Citation Context

... that most relevant knowledge is retained and stored in a quickly accessible form, and redundant knowledge is removed. Examples of such algorithms are the decision-tree algorithms igtree [9] and c4.5 =-=[12]-=-. Another popular inductive algorithm is the connectionist Back-propagation (bp) [13] learning algorithm. We provide a summary of the basic functions of these learning algorithms. 1. ib1 [1] construct...

Citation Context

... the weighting function W(fi) (cf. equation 1). This function computes for each feature, over the full instance base, its information gain, a function from information theory that is also used in id3 =-=[11]-=- and c4.5 [12] (for more details, cf. Daelemans and Van den Bosch [6]). In short, the information gain of a feature expresses its relative importance compared to the other features in performing the m...

Citation Context

...d redundant knowledge is removed. Examples of such algorithms are the decision-tree algorithms igtree [9] and c4.5 [12]. Another popular inductive algorithm is the connectionist Back-propagation (bp) =-=[13]-=- learning algorithm. We provide a summary of the basic functions of these learning algorithms. 1. ib1 [1] constructs a data base of instances (the instance base) during learning. An instance consists ...

Citation Context

...riments reported in this paper. 1.2 Algorithms and Methods for Inductive Learning Inductive learning in its most straightforward form is exhibited by memory-based lazy learning algorithms such as ib1 =-=[1]-=- and variations (e.g., ib1-ig [6, 9]), in which all instances are fully stored in memory, and in which classification involves a pass along all stored instances. To optimise memory lookup and minimise...

Citation Context

...learned by an inductive-learning algorithm. Differences exist in the ways inductive-learning algorithms extract knowledge from the available instances. In lazy learning (such as memory-based learning =-=[14, 5]-=-), there is no abstraction of higher-level data structures such as rules or decision trees at learning time; learning consists of simply storing the instances in memory. A new instance of the same pro...

Citation Context

...d instances of which the middle letter is mapped to a class denoting a morpheme boundary decision. To generate fixed-sized instances, we adopt the windowing scheme proposed by Sejnowski and Rosenberg =-=[15]-=- which generates fixed-sized snapshots of words. In its most basic form, the classification of each instance denotes whether the focus letter of the instance maps to a morpheme boundary (‘yes’, or ‘1’...

Citation Context

...ssify new instances that were not in the training material. A method that gives a good estimate of the generalisation performance of an algorithm on a given instance base, is 10-fold cross-validation =-=[19]-=-. This method generates on the basis of an instance base 10 partitionings into a training set (90%) and a test set (10%), resulting in 10 experiments and 10 results per learning algorithm and instance...

Citation Context

...cessing, and being language-independent. 1 Introduction Morphological analysis is often deemed to be an important, if not essential subtask in linguistic modular systems for text-to-speech processing =-=[2]-=- and hyphenation [4]. In text-to-speech processing, it serves to prevent the wrong application of grapheme-phoneme conversion rules across morpheme boundaries (e.g., preventing carelessly from being p...

Citation Context

....2 Algorithms and Methods for Inductive Learning Inductive learning in its most straightforward form is exhibited by memory-based lazy learning algorithms such as ib1 [1] and variations (e.g., ib1-ig =-=[6, 9]-=-), in which all instances are fully stored in memory, and in which classification involves a pass along all stored instances. To optimise memory lookup and minimise memory usage, more eager learning a...

Citation Context

...arning. In previous research we have demonstrated the application of the memory-based (lazy) learning approach to several linguistic problems, e.g., segmentation as in hyphenation and syllabification =-=[6, 17]-=-, and identification as in grapheme-phoneme conversion [18, 16, 7], and stress assignment [8]. In most cases, the memory-based (lazy) approach outdid the more eager inductive algorithms. We believe th...

Citation Context

...d as classification tasks, i.e., given a description of an input in terms of a number of feature-values, a classification of the input is performed. Two types of classification tasks can be discerned =-=[5]-=-: – Identification: given a set of possible classifications and an input of feature values, determine the correct classification for this input. For example, given a letter surrounded by a number of n...

Citation Context

...ion of the memory-based (lazy) learning approach to several linguistic problems, e.g., segmentation as in hyphenation and syllabification [6, 17], and identification as in grapheme-phoneme conversion =-=[18, 16, 7]-=-, and stress assignment [8]. In most cases, the memory-based (lazy) approach outdid the more eager inductive algorithms. We believe that in a ‘noisy’ domain such as natural language, abstractingfrom ...

Citation Context

...ion of the memory-based (lazy) learning approach to several linguistic problems, e.g., segmentation as in hyphenation and syllabification [6, 17], and identification as in grapheme-phoneme conversion =-=[18, 16, 7]-=-, and stress assignment [8]. In most cases, the memory-based (lazy) approach outdid the more eager inductive algorithms. We believe that in a ‘noisy’ domain such as natural language, abstractingfrom ...

Citation Context

...onger words. Morphological analysis on a probabilistic basis, using only a morpheme lexicon, an analyses generator, and a probabilistic function to determine the analysis with the highest probability =-=[10]-=- does not suffer from the disadvantageous knowledge acquisition and fine-tuning phase, but is nevertheless also confronted with an explosion of the number of generated analyses.2.2 Inductive-Learning...

Citation Context

...ion of the memory-based (lazy) learning approach to several linguistic problems, e.g., segmentation as in hyphenation and syllabification [6, 17], and identification as in grapheme-phoneme conversion =-=[18, 16, 7]-=-, and stress assignment [8]. In most cases, the memory-based (lazy) approach outdid the more eager inductive algorithms. We believe that in a ‘noisy’ domain such as natural language, abstractingfrom ...

Citation Context

...anguage-independent. 1 Introduction Morphological analysis is often deemed to be an important, if not essential subtask in linguistic modular systems for text-to-speech processing [2] and hyphenation =-=[4]-=-. In text-to-speech processing, it serves to prevent the wrong application of grapheme-phoneme conversion rules across morpheme boundaries (e.g., preventing carelessly from being pronounced as /k�’r�l...

Citation Context

...arning. In previous research we have demonstrated the application of the memory-based (lazy) learning approach to several linguistic problems, e.g., segmentation as in hyphenation and syllabification =-=[6, 17]-=-, and identification as in grapheme-phoneme conversion [18, 16, 7], and stress assignment [8]. In most cases, the memory-based (lazy) approach outdid the more eager inductive algorithms. We believe th...