We present a simple and effective semisupervised method for training dependency parsers. We focus on the problem of lexical representation, introducing features that incorporate word clusters derived from a large unannotated corpus. We demonstrate the effectiveness of the approach in a series of dependency parsing experiments on the Penn Treebank and Prague Dependency Treebank, and we show that the cluster-based features yield substantial gains in performance across a wide range of conditions. For example, in the case of English unlabeled second-order parsing, we improve from a baseline accuracy of 92.02 % to 93.16%, and in the case of Czech unlabeled second-order parsing, we improve from a baseline accuracy of 86.13% to 87.13%. In addition, we demonstrate that our method also improves performance when small amounts of training data are available, and can roughly halve the amount of supervised data required to reach a desired level of performance. 1

Abstract not found

Parsing algorithms that process the input from left to right and construct a single derivation have often been considered inadequate for natural language parsing because of the massive ambiguity typically found in natural language grammars. Nevertheless, it has been shown that such algorithms, combined with treebank-induced classifiers, can be used to build highly accurate disambiguating parsers, in particular for dependency-based syntactic representations. In this article, we first present a general framework for describing and analyzing algorithms for deterministic incremental dependency parsing, formalized as transition systems. We then describe and analyze two families of such algorithms: stack-based and list-based algorithms. In the former family, which is restricted to projective dependency structures, we describe an arc-eager and an arc-standard variant; in the latter family, we present a projective and a nonprojective variant. For each of the four algorithms, we give proofs of correctness and complexity. In addition, we perform an experimental evaluation of all algorithms in combination with SVM classifiers for predicting the next parsing action, using data from thirteen languages. We show that all four algorithms give competitive accuracy, although the non-projective list-based algorithm generally outperforms the projective algorithms for languages with a non-negligible proportion of non-projective constructions. However, the projective algorithms often produce comparable results when combined with the technique known as pseudo-projective parsing. The linear time complexity of the stack-based algorithms gives them an advantage with respect to efficiency both in learning and in parsing, but the projective list-based algorithm turns out to be equally efficient in practice. Moreover, when the projective algorithms are used to implement pseudo-projective parsing, they sometimes become less efficient in parsing (but not in learning) than the non-projective list-based algorithm. Although most of the algorithms have been partially described in the literature before, this is the first comprehensive analysis and evaluation of the algorithms within a unified framework. 1.

Previous studies of data-driven dependency parsing have shown that the distribution of parsing errors are correlated with theoretical properties of the models used for learning and inference. In this paper, we show how these results can be exploited to improve parsing accuracy by integrating a graph-based and a transition-based model. By letting one model generate features for the other, we consistently improve accuracy for both models, resulting in a significant improvement of the state of the art when evaluated on data sets from the CoNLL-X shared task. 1

Log-linear and maximum-margin models are two commonly-used methods in supervised machine learning, and are frequently used in structured prediction problems. Efficient learning of parameters in these models is therefore an important problem, and becomes a key factor when learning from very large data sets. This paper describes exponentiated gradient (EG) algorithms for training such models, where EG updates are applied to the convex dual of either the log-linear or maxmargin objective function; the dual in both the log-linear and max-margin cases corresponds to minimizing a convex function with simplex constraints. We study both batch and online variants of the algorithm, and provide rates of convergence for both cases. In the max-margin case, O ( 1 ε) EG updates are required to reach a given accuracy ε in the dual; in contrast, for log-linear models only O(log (1/ε)) updates are required. For both the max-margin and log-linear cases, our bounds suggest that the online EG algorithm requires a factor of n less computation to reach a desired accuracy than the batch EG algorithm, where n is the number of training examples. Our experiments confirm that the online algorithms are much faster than the batch algorithms in practice. We describe how the EG updates factor in a convenient way for structured prediction problems, allowing the algorithms to be

We present a two-stage multilingual dependency parser and evaluate it on 13 diverse languages. The first stage is based on the unlabeled dependency parsing models described by McDonald and Pereira (2006) augmented with morphological features for a subset of the languages. The second stage takes the output from the first and labels all the edges in the dependency graph with appropriate syntactic categories using a globally trained sequence classifier over components of the graph. We report results on the CoNLL-X shared task (Buchholz et al., 2006) data sets and present an error analysis. 1

We present a comparative error analysis of the two dominant approaches in datadriven dependency parsing: global, exhaustive, graph-based models, and local, greedy, transition-based models. We show that, in spite of similar performance overall, the two models produce different types of errors, in a way that can be explained by theoretical properties of the two models. This analysis leads to new directions for parser development. 1

We present a data-driven variant of the LR algorithm for dependency parsing, and extend it with a best-first search for probabilistic generalized LR dependency parsing. Parser actions are determined by a classifier, based on features that represent the current state of the parser. We apply this parsing framework to both tracks of the CoNLL 2007 shared task, in each case taking advantage of multiple models trained with different learners. In the multilingual track, we train three LR models for each of the ten languages, and combine the analyses obtained with each individual model with a maximum spanning tree voting scheme. In the domain adaptation track, we use two models to parse unlabeled data in the target domain to supplement the labeled out-ofdomain training set, in a scheme similar to one iteration of co-training. 1

We explore a stacked framework for learning to predict dependency structures for natural language sentences. A typical approach in graph-based dependency parsing has been to assume a factorized model, where local features are used but a global function is optimized (McDonald et al., 2005b). Recently Nivre and McDonald (2008) used the output of one dependency parser to provide features for another. We show that this is an example of stacked learning, in which a second predictor is trained to improve the performance of the first. Further, we argue that this technique is a novel way of approximating rich non-local features in the second parser, without sacrificing efficient, model-optimal prediction. Experiments on twelve languages show that stacking transition-based and graphbased parsers improves performance over existing state-of-the-art dependency parsers. 1

We first show how a structural locality bias can improve the accuracy of state-of-the-art dependency grammar induction models trained by EM from unannotated examples (Klein and Manning, 2004). Next, by annealing the free parameter that controls this bias, we achieve further improvements. We then describe an alternative kind of structural bias, toward “broken ” hypotheses consisting of partial structures over segmented sentences, and show a similar pattern of improvement. We relate this approach to contrastive estimation (Smith and Eisner, 2005a), apply the latter to grammar induction in six languages, and show that our new approach improves accuracy by 1–17 % (absolute) over CE (and 8–30% over EM), achieving to our knowledge the best results on this task to date. Our method, structural annealing, is a general technique with broad applicability to hidden-structure discovery problems. 1