In this survey, we review work in machine learning on methods for handling data sets containing large amounts of irrelevant information. We focus on two key issues: the problem of selecting relevant features, and the problem of selecting relevant examples. We describe the advances that have been made on these topics in both empirical and theoretical work in machine learning, and we present a general framework that we use to compare different methods. We close with some challenges for future work in this area.

Current approaches to computational lexicology in language technology are knowledge-based (competence-oriented) and try to abstract away from specific formalisms, domains, and applications. This results in severe complexity, acquisition and reusability bottlenecks. As an alternative, we propose a particular performance-oriented approach to Natural Language Processing based on automatic memory-based learning of linguistic (lexical) tasks. The consequences of the approach for computational lexicology are discussed, and the application of the approach on a number of lexical acquisition and disambiguation tasks in phonology, morphology and syntax is described.

We present a general architecture for efficient and deterministic morphological analysis based on memory-based learning, and apply it to morphological analysis of Dutch. The system makes direct mappings from letters in context to rich categories that encode morphological boundaries, syntactic class labels, and spelling changes. Both precision and recall of labeled morphemes are over 84% on held-out dictionary test words and estimated to be over 93% in free text.

We describe an approach to grapheme-to-phoneme conversion which is both language-independent and data-oriented. Given a set of examples (spelling words with their associated phonetic representation) in a language, a grapheme-to-phoneme conversion system is automatically produced for that language which takes as its input the spelling of words, and produces as its output the phonetic transcription according to the rules implicit in the training data. We describe the design of the system, and compare its performance to knowledge-based and alternative data-oriented approaches.

This paper addresses the issue of "algorithm vs. representation" for case-based learning of linguistic knowledge. We first present empirical evidence that the success of case-based learning methods for natural language processing tasks depends to a large degree on the feature set used to describe the training instances. Next, we present a technique for automating feature set selection for case-based learning of linguistic knowledge. Given as input a baseline case representation, the method modifies the representation in response to a number of predefined linguistic biases by adding, deleting, and weighting features appropriately. We apply the linguistic bias approach to feature set selection to the problem of relative pronoun disambiguation and show that the casebased learning agorithm improves as relevant biases are incorporated into the underlying instance representation. Finally, we argue that the linguistic bias approach to feature set selection offers new possibilities for case-based learning of natural language: it simplifies the process of instance representation design and, in theory, obviates the need for separate instance representations for each linguistic knowledge acquisition task. More importantly, the approach offers a mechanism for explicitly combining the frequency information available from corpus-based techniques with linguistic bias information employed in traditional linguistic and knowledge-based approaches to natural language processing.

The availability of on-line corpora is rapidly changing the field of natural language processing (NLP) from one dominated by theoretical models of often very specific linguistic phenomena to one guided by computational models that simultaneously account for a wide variety of phenomena that occur in real-world text. Thus far, among the best-performing and most robust systems for reading and summarizing large amounts of real-world text are knowledge-based natural language systems. These systems rely heavily on domain-specific, handcrafted knowledge to handle the myriad syntactic, semantic, and pragmatic ambiguities that pervade virtually all aspects of sentence analysis. Not surprisingly, however, generating this knowledge for new domain...

A fundamental debate in the machine learning of language has been the role of prior knowledge in the learning process. Purely nativist approaches, such as the Principles and Parameters model, build parameterized linguistic generalizations directly into the learning system. Purely empirical approaches use a general, domain-independent learning rule (Error Back-Propagation, Instance-based Generalization, Minimum Description Length) to learn linguistic generalizations directly from the data. In this paper we suggest that an alternative to the purely nativist or purely empiricist learning paradigms is to represent the prior knowledge of language as a set of abstract learning biases, which guide an empirical inductive learning algorithm. We test our idea by examining the machine learning of simple Sound Pattern of English ( S P E)-style phonological rules. We represent phonological rules as finite-state transducers that accept underlying forms as input and generate surface forms as output. We show that OSTIA, a general-purpose transducer induction algorithm, was incapable of learning simple phonological rules like flapping. We then augmented OSTIA with three kinds of learning biases that are specific to natural language phonology, and that are assumed explicitly or implicitly by every theory of phonology: faithfulness (underlying segments

Pronunciation by analogy (PbA) is a data-driven method for relating letters to sound, with potential application to next-generation text-to-speech systems. This paper extends previous work on PbA in several directions. First, we have included `full' pattern matching between input letter string and dictionary entries, as well as including lexical stress in letter-to-phoneme conversion. Second, we have extended the method to phonemeto -letter conversion. Third, and most important, we have experimented with multiple, different strategies for scoring the candidate pronunciations. Individual scores for each strategy are obtained on the basis of rank and either multiplied or summed to produce a final, overall score. Five strategies have been studied and results obtained from all 31 possible combinations. The two combination methods perform comparably, with the product rule only very marginally superior to the sum rule. Nonparametric statistical analysis reveals that performance improves as more strategies are included in the combination: this trend is very highly significant ( p 0 0005). Accordingly for letter-to-phoneme conversion, best results are obtained when all five strategies are combined: word accuracy is raised to 65.5% relative to 61.7% for our best previous result and 63.0% for the best-performing single strategy. These improvements are very highly significant ( p 0 and p 0 00011 respectively). Similar results were found for phoneme-to-letter and letter-to-stress conversion, although the former was an easier problem for PbA than letter-to-phoneme conversion and the latter was harder. The main sources of error for the multi-strategy approach are very similar to those for the best single strategy, and mostly involve vowel letters and phonemes. 1

Morphological analysis is an important subtask in text-to-speech conversion, hyphenation, and other language engineering tasks. The traditional approach to performing morphological analysis is to combine a morpheme lexicon, sets of (linguistic) rules, and heuristics to find a most probable analysis. In contrast we present an inductive learning approach in which morphological analysis is reformulated as a segmentation task. We report on a number of experiments in which five inductive learning algorithms are applied to three variations of the task of morphological analysis. Results show (i) that the generalisation performance of the algorithms is good, and (ii) that the lazy learning algorithm ib1-ig performs best on all three tasks. We conclude that lazy learning of morphological analysis as a classification task is indeed a viable approach; moreover, it has the strong advantages of avoiding the knowledge-acquisition bottleneck, being fast and deterministic in learning and pr...

. Nearest-neighbor algorithms are known to depend heavily on their distance metric. In this paper, we investigate the use of a weighted Euclidean metric in which the weight for each feature comes from a small set of options. We describe Diet, an algorithm that directs search through a space of discrete weights using cross-validation error as its evaluation function. Although a large set of possible weights can reduce the learner's bias, it can also lead to increased variance and overfitting. Our empirical study shows that, for many data sets, there is an advantage to weighting features, but that increasing the number of possible weights beyond two (zero and one) has very little benefit and sometimes degrades performance. 1 Introduction In recent years, instance-based methods [5, 13, 1] have emerged as a promising approach to machine learning, with researchers reporting excellent results on many real-world induction tasks. The basic approach involves storing training cases and their as...