Conditional random fields (Lafferty et al., 2001) are quite effective at sequence labeling tasks like shallow parsing (Sha and Pereira, 2003) and namedentity extraction (McCallum and Li, 2003). CRFs are log-linear, allowing the incorporation of arbitrary features into the model. To train on unlabeled data, we require unsupervised estimation methods for log-linear models; few exist. We describe a novel approach, contrastive estimation. We show that the new technique can be intuitively understood as exploiting implicit negative evidence and is computationally efficient. Applied to a sequence labeling problem—POS tagging given a tagging dictionary and unlabeled text—contrastive estimation outperforms EM (with the same feature set), is more robust to degradations of the dictionary, and can largely recover by modeling additional features. 1

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.

We introduce a stochastic graph-based method for computing relative importance of textual units for Natural Language Processing. We test the technique on the problem of Text Summarization (TS). Extractive TS relies on the concept of sentence salience to identify the most important sentences in a document or set of documents. Salience is typically defined in terms of the presence of particular important words or in terms of similarity to a centroid pseudo-sentence. We consider a new approach, LexRank, for computing sentence importance based on the concept of eigenvector centrality in a graph representation of sentences. In this model, a connectivity matrix based on intra-sentence cosine similarity is used as the adjacency matrix of the graph representation of sentences. Our system, based on LexRank ranked in first place in more than one task in the recent DUC 2004 evaluation. In this paper we present a detailed analysis of our approach and apply it to a larger data set including data from earlier DUC evaluations. We discuss several methods to compute centrality using the similarity graph. The results show that degree-based methods (including LexRank) outperform both centroid-based methods and other systems participating in DUC in most of the cases. Furthermore, the LexRank with threshold method outperforms the other degree-based techniques including continuous LexRank. We also show that our approach is quite insensitive to the noise in the data that may result from an imperfect topical clustering of documents. 1.

Statistical language learning research takes the view that many traditional natural language processing tasks can be solved by training probabilistic models of language on a sufficient volume of training data. The design of statistical language learners therefore involves answering two questions: (i) Which of the multitude of possible language models will most accurately reflect the properties necessary to a given task? (ii) What will constitute a sufficient volume of training data? Regarding the first question, though a variety of successful models have been discovered, the space of possible designs remains largely unexplored. Regarding the second, exploration of the design space has so far proceeded without an adequate answer. The goal of this thesis is to advance the exploration of the statistical language learning design space. In pursuit of that goal, the thesis makes two main theoretical contributions: it identifies a new class of designs by providing a novel theory of statistical natural language processing, and it presents the foundations for a predictive theory of data requirements to assist in future design explorations. The first of these contributions is called the meaning distributions theory. This theory

This paper describes an experimental com- parison of three unsupervised learning algorithms that distinguish the sense of an ambiguous word in untagged text.

Word sense disambiguation (WSD) is a computational linguistics task likely to benefit from the tradition of combining different knowledge sources in artificial in telligence research. An important step in the exploration of this hypothesis is to determine which linguistic knowledge sources are most useful and whether their combination leads to improved results. We present a sense tagger which uses several knowledge sources. Tested accuracy exceeds 94 % on our evaluation corpus. Our system attempts to disambiguate all content words in running text rather than limiting itself to treating a restricted vocabulary of words. It is argued that this approach is more likely to assist the creation of practical systems. 1.

We introduce a stochastic graph-based method for computing relative importance of textual units for Natural Language Processing. We test the technique on the problem of Text Summarization (TS). Extractive TS relies on the concept of sentence salience to identify the most important sentences in a document or set of documents. Salience is typically defined in terms of the presence of particular important words or in terms of similarity to a centroid pseudo-sentence. We consider a new approach, LexRank, for computing sentence importance based on the concept of eigenvector centrality in a graph representation of sentences. In this model, a connectivity matrix based on intra-sentence cosine similarity is used as the adjacency matrix of the graph representation of sentences. Our system, based on LexRank ranked in first place in more than one task in the recent DUC 2004 evaluation. In this paper we present a detailed analysis of our approach and apply it to a larger data set including data from earlier DUC evaluations. We discuss several methods to compute centrality using the similarity graph. The results show that degree-based methods (including LexRank) outperform both centroid-based methods and other systems participating in DUC in most of the cases. Furthermore, the LexRank with threshold method outperforms the other degree-based techniques including continuous LexRank. We also show that our approach is quite insensitive to the noise in the data that may result from an imperfect topical clustering of documents. 1.

Word sense disambiguation assumes word senses. Within the lexicography and linguistics literature, they are known to be very slippery entities. The paper looks at problems with existing accounts of `word sense' and describes the various kinds of ways in which a word's meaning can deviate from its core meaning. An analysis is presented in which word senses are abstractions from clusters of corpus citations, in accordance with current lexicographic practice. The corpus citations, not the word senses, are the basic objects in the ontology. The corpus citations will be clustered into senses according to the purposes of whoever or whatever does the clustering. In the absence of such purposes, word senses do not exist. Word sense disambiguation also needs a set of word senses to disambiguate between. In most recent work, the set has been taken from a general-purpose lexical resource, with the assumption that the lexical resource describes the word senses of English/French/. . . , between whi...

. We present two problems for statistically extracting bilingual lexicon: (1) How can noisy parallel corpora be used? (2) How can non-parallel yet comparable corpora be used? We describe our own work and contribution in relaxing the constraint of using only clean parallel corpora. DKvec is a method for extracting bilingual lexicons, from noisy parallel corpora based on arrival distances of words in noisy parallel corpora. Using DKvec on noisy parallel corpora in English/Japanese and English/Chinese, our evaluations show a 55.35% precision from a small corpus and 89.93% precision from a larger corpus. Our major contribution is in the extraction of bilingual lexicon from non-parallel corpora. We present a first such result in this area, from a new method--Convec. Convec is based on context information of a word to be translated. We show a 30% to 76% precision when top-one to top-20 translation candidates are considered. Most of the top-20 candidates are either collocations or words rela...

We describe a method for automatic word sense disambiguation using a text corpus and a machinereadable dictionary (MRD). The method is based on word similarity and context similarity measures. Words are considered similar if they appear in similar contexts; contexts are similar if they contain similar words. The circularity of this definition is resolved by an iterative, converging process, in which the system learns from the corpus a set of typical usages for each of the senses of the polysemous word listed in the MRD. A new instance of a polysemous word is assigned the sense associated with the typical usage most similar to its context. Experiments show that this method can learn even from very sparse training data, achieving over 92 % correct disambiguation performance.