The amount of readily available online text has reached hundreds of billions of words and continues to grow. Yet for most core natural language tasks, algorithms continue to be optimized, tested and compared after training on corpora consisting of only one million words or less. In this paper, we evaluate the performance of different learning methods on a prototypical natural language disambiguation task, confusion set disambiguation, when trained on orders of magnitude more labeled data than has previously been used. We are fortunate that for this particular application, correctly labeled training data is free. Since this will often not be the case, we examine methods for effectively exploiting very large corpora when labeled data comes at a cost.

Levin’s (1993) study of verb classes is a widely used resource for lexical semantics. In her framework, some verbs, such as give, exhibit no class ambiguity. But other verbs, such as write, have several alternative classes. We extend Levin’s inventory to a simple statistical model of verb class ambiguity. Using this model we are able to generate preferences for ambiguous verbs without the use of a disambiguated corpus. We additionally show that these preferences are useful as priors for a verb sense disambiguator.

This paper proposes a new method for word translation disambiguation using a machine learning technique called `Bilingual Bootstrapping'. Bilingual Bootstrapping makes use of # in learning# a small number of classified data and a large number of unclassified data in the source and the target languages in translation. It constructs classifiers in the two languages in parallel and repeatedly boosts the performances of the classifiers by further classifying data in each of the two languages and by exchanging between the two languages information regarding the classified data. Experimental results indicate that word translation disambiguation based on Bilingual Bootstrapping consistently and significantly outperforms the existing methods based on `Monolingual Bootstrapping'.

Classifier combination is an effective and broadly useful method of improving system performance. This article investigates in depth a large number of both well-established and novel classifier combination approaches for the word sense disambiguation task, studied over a diverse classifier pool which includes feature-enhanced Naïve Bayes, Cosine, Decision List, Transformation-based Learning and MMVC classifiers. Each classifier has access to the same rich feature space, comprised of distance weighted bag-of-lemmas, local ngram context and specific syntactic relations, such as Verb-Object and Noun-Modifier.

A popular method for creating an accurate classifier from a set of training data is to build several classifiers, and then to combine their predictions. The ensembles of simple Bayesian classifiers have traditionally not been a focus of research. One way to generate an ensemble of accurate and diverse simple Bayesian classifiers is to use different feature subsets generated with the random subspace method. In this case, the ensemble consists of multiple classifiers constructed by randomly selecting feature subsets, that is, classifiers constructed in randomly chosen subspaces. In this paper, we present an algorithm for building ensembles of simple Bayesian classifiers in random subspaces...

Dictionary-based protein name recognition is often a first step in extracting in-formation from biomedical documents because it can provide ID information on recognized terms. However, dictionary-based approaches present two fundamental difficulties: (1) false recognition mainly caused by short names; (2) low recall due to spelling variations. In this paper, we tackle the former problem using machine learning to filter out false positives and present two alternative methods for alle-viating the latter problem with spelling variations. The first is achieved by using approximate string searching, and the second by expanding the dictionary with a probabilistic variant generator, which we propose in this paper. Experimental re-sults using the GENIA corpus revealed that filtering using a naive Bayes classifier greatly improved precision with only a slight loss of recall, resulting in 10.8 % im-provement in F-measure, and dictionary expansion with the variant generator gave further 1.6 % improvement and achieved an F-measure of 66.6%. Key words: protein name recognition, naive Bayes classifier, approximate string search, spelling variant generator 2 1

This paper describes a methodology for supervised word sense disambiguation that relies on standard machine learning algorithms to induce classifiers from sense-tagged training examples where the context in which ambiguous words occur are represented by simple lexical features. This constitutes a baseline approach since it produces classifiers based on easy to identify features that result in accurate disambiguation across a variety of languages. This paper reviews several systems based on this methodology that participated in the Spanish and English lexical sample tasks of the Senseval-2 comparative exercise among word sense disambiguation systems. These systems fared much better than standard baselines, and were within seven to ten percentage points of accuracy of the mostly highly ranked systems.

Abstract. A popular method for creating an accurate classifier from a set of training data is to train several classifiers, and then to combine their predictions. The ensembles of simple Bayesian classifiers have traditionally not been a focus of research. However, the simple Bayesian classifier has much broader applicability than previously thought. Besides its high classification accuracy, it also has advantages in terms of simplicity, learning speed, classification speed, storage space, and incrementality. One way to generate an ensemble of simple Bayesian classifiers is to use different feature subsets as in the random subspace method. In this paper we present a technique for building ensembles of simple Bayesian classifiers in random subspaces. We consider also a hill-climbing-based refinement cycle, which improves accuracy and diversity of the base classifiers. We conduct a number of experiments on a collection of real-world and synthetic data sets. In many cases the ensembles of simple Bayesian classifiers have significantly higher accuracy than the single “global ” simple Bayesian classifier. We consider several methods for integration of simple Bayesian classifiers. The dynamic integration better utilizes ensemble diversity than the static integration. 1

This paper explores two directions tbr the next step beyoud the state of the art of statistical parsing: probabilistic partial parsing and committee-based decision making. Probabilistic partial parsing is a probabilistic extension of the existing notion of partial parsing, which ellables fine-grained arbitrary choice on the tradeoff between accuracy and coverage. Committeebased decision making is to combine the outputs fi'om ditibrent systems to make a better decision. While various committee-based tech.- niques for NLP have recently been investigated, they would need to be flirther extended so as to be applicable to probabilistic partial pars- ing. Aiming at this coupling, this palmr gives a general fi'amcwork to committee-based decision making, which consists of a set of weighting fimctions and a combination timetlon, and discusses how it can be coupled with probabilistic partial parsing. Our experiments have so far been producing prolnising rcsult, s

Word Sense Disambiguation (WSD) is the task of choosing the right sense of a polysemous word given a context. It is obviously essential for many natural language processing applications such as human-computer communication, machine translation, and information retrieval. In recent years, much attention have been paid to improve the performance of WSD systems by using combination of classifiers. In (Kittler, Hatef, Duin, and Matas 1998), six combination rules including product, sum, max, min, median, and majority voting were derived with a number of strong assumptions, that are unrealistic in many situations and especially in text-related applications. This paper considers a framework of combination strategies based on different representations of context in WSD resulting in these combination rules as well, but without the unrealistic assumptions mentioned above. The experiment was done on four words interest, line, hard, serve; on the DSO dataset it showed high accuracies with median and min combination rules.