We describe an approach to speed-up inference with latent-variable PCFGs, which have been shown to be highly effective for natural language parsing. Our approach is based on a tensor formulation recently introduced for spectral estimation of latent-variable PCFGs coupled with a tensor decomposition algorithm well-known in the multilinear algebra literature. We also describe an error bound for this approximation, which gives guarantees showing that if the underlying tensors are well approximated, then the probability distribution over trees will also be well approximated. Empirical evaluation on real-world natural language parsing data demonstrates a significant speed-up at minimal cost for parsing performance.

We present a novel technique to remove spurious ambiguity from transition systems for dependency parsing. Our technique chooses a canonical sequence of transition operations (computation) for a given dependency tree. Our technique can be applied to a large class of bottom-up transition systems, including for instance Nivre (2004) and Attardi (2006).

We introduce a spectral learning algorithm for latent-variable PCFGs (Petrov et al., 2006; Matsuzaki et al., 2005). Under a separability (singular value) condition, we prove that the method provides consistent parameter estimates. Our result rests on three theorems: the first gives a tensor form of the inside-outside algorithm for PCFGs; the second shows that the required tensors can be estimated directly from training examples where hidden-variable values are missing; the third gives a PAC-style convergence bound for the estimation method.

Probabilistic grammars are generative statistical models that are useful for compositional and sequential structures. They are used ubiquitously in computational linguistics. We present a framework, reminiscent of structural risk minimization, for empirical risk minimization of probabilistic grammars using the log-loss. We derive sample complexity bounds in this framework that apply both to the supervised setting and the unsupervised setting. By making assumptions about the underlying distribution that are appropriate for natural language scenarios, we are able to derive distribution-dependent sample complexity bounds for probabilistic grammars. We also give simple algorithms for carrying out empirical risk minimization using this framework in both the supervised and unsupervised settings. In the unsupervised case, we show that the problem of minimizing empirical risk is NP-hard. We therefore suggest an approximate algorithm, similar to expectation-maximization, to minimize the empirical risk.

We describe a method for prediction of linguistic structure in a language for which only unlabeled data is available, using annotated data from a set of one or more helper languages. Our approach is based on a model that locally mixes between supervised models from the helper languages. Parallel data is not used, allowing the technique to be applied even in domains where human-translated texts are unavailable. We obtain state-of-the-art performance for two tasks of structure prediction: unsupervised part-of-speech tagging and unsupervised dependency parsing.

We describe a generative model for non-projective dependency parsing based on a simplified version of a transition system that has recently appeared in the literature. We then develop a dynamic programming parsing algorithm for our model, and derive an inside-outside algorithm that can be used for unsupervised learning of non-projective dependency trees.

Weighted logic programming, a generalization of bottom-up logic programming, is a well-suited framework for specifying dynamic programming algorithms. In this setting, proofs correspond to the algorithm's output space, such as a path through a graph or a grammatical derivation, and are given a real-valued score (often interpreted as a probability) that depends on the real weights of the base axioms used in the proof. The desired output is a function over all possible proofs, such as a sum of scores or an optimal score. We describe the PRODUCT transformation, which can merge two weighted logic programs into a new one. The resulting program optimizes a product of proof scores from the original programs, constituting a scoring function known in machine learning as a ``product of experts.'' Through the addition of intuitive constraining side conditions, we show that several important dynamic programming algorithms can be derived by applying PRODUCT to weighted logic programs corresponding to simpler weighted logic programs.

Probabilistic grammars are generative statistical models that are useful for compositional and sequential structures. We present a framework, reminiscent of structural risk minimization, for empirical risk minimization of the parameters of a fixed probabilistic grammar using the log-loss. We derive sample complexity bounds in this framework that apply both to the supervised setting and the unsupervised setting.

Probabilistic grammars offer great flexibility in modeling discrete sequential data like natural language text. Their symbolic component is amenable to inspection by humans, while their probabilistic component helps resolve ambiguity. They also permit the use of well-understood, general-purpose learning algorithms. There has been an increased interest in using probabilistic grammars in the Bayesian setting. To date, most of the literature has focused on using a Dirichlet prior. The Dirichlet prior has several limitations, including that it cannot directly model covariance between the probabilistic grammar's parameters. Yet, various grammar parameters are expected to be correlated because the elements in language they represent share linguistic properties. In this paper, we suggest an alternative to the Dirichlet prior, a family of logistic normal distributions. We derive an inference algorithm for this family of distributions and experiment with the task of dependency grammar induction, demonstrating performance improvements with our priors on a set of six treebanks in different natural languages. Our covariance framework permits soft parameter tying within grammars and across grammars for text in different languages, and we show empirical gains in a novel learning setting using bilingual, non-parallel data.

We consider the search for a maximum likelihood assignment of hidden derivations and grammar weights for a probabilistic context-free grammar, the problem approximately solved by ``Viterbi training.'' We show that solving and even approximating Viterbi training for PCFGs is NP-hard. We motivate the use of uniform-at-random initialization for Viterbi EM as an optimal initializer in absence of further information about the correct model parameters, providing an approximate bound on the log-likelihood.

Adaptor grammars extend probabilistic context-free grammars to define prior distributions over trees with ``rich get richer'' dynamics. Inference for adaptor grammars seeks to find parse trees for raw text. This paper describes a variational inference algorithm for adaptor grammars, providing an alternative to Markov chain Monte Carlo methods. To derive this method, we develop a stick-breaking representation of adaptor grammars, a representation that enables us to define adaptor grammars with recursion. We report experimental results on a word segmentation task, showing that variational inference performs comparably to MCMC. Further, we show a significant speed-up when parallelizing the algorithm. Finally, we report promising results for a new application for adaptor grammars, dependency grammar induction.

Variational EM has become a popular technique in probabilistic NLP with hidden variables. Commonly, for computational tractability, we make strong independence assumptions, such as the mean-field assumption, in approximating posterior distributions over hidden variables. We show how a looser restriction on the approximate posterior, requiring it to be a mixture, can help inject prior knowledge to exploit soft constraints during the variational E-step.

We present a family of priors over probabilistic grammar weights, called the shared logistic normal distribution. This family extends the partitioned logistic normal distribution, enabling factored covariance between the probabilities of different derivation events in the probabilistic grammar, providing a new way to encode prior knowledge about an unknown grammar. We describe a variational EM algorithm for learning a probabilistic grammar based on this family of priors. We then experiment with unsupervised dependency grammar induction and show significant improvements using our model for both monolingual learning and bilingual learning with a non-parallel, multilingual corpus.

We explore a new Bayesian model for probabilistic grammars, a family of distributions over discrete structures that includes hidden Markov models and probabilistic context-free grammars. Our model extends the correlated topic model framework to probabilistic grammars, exploiting the logistic normal distribution as a prior over the grammar parameters. We derive a variational EM algorithm for that model, and then experiment with the task of unsupervised grammar induction for natural language dependency parsing. We show that our model achieves superior results over previous models that use different priors.

Weighted logic programming, a generalization of bottom-up logic programming, is a successful framework for specifying dynamic programming algorithms. In this setting, proofs correspond to the algorithm's output space, such as a path through a graph or a grammatical derivation, and are given a weighted score, often interpreted as a probability, that depends on the score of the base axioms used in the proof. The desired output is a function over all possible proofs, such as a sum of scores or an optimal score. We describe the PRODUCT transformation, which can merge two weighted logic programs into a new one. The resulting program optimizes a product of proof scores from the original programs, constituting a scoring function known in machine learning as a ``product of experts.'' Through the addition of intuitive constraining side conditions, we show that several important dynamic programming algorithms can be derived by applying PRODUCT to weighted logic programs corresponding to simpler weighted logic programs.

In morphologically rich languages, should morphological and syntactic disambiguation be treated sequentially or as a single problem? We describe several efficient, probabilistically-interpretable ways to apply joint inference to morphological and syntactic disambiguation using lattice parsing. Joint inference is shown to compare favorably to pipeline parsing methods across a variety of component models. State-of-the-art performance on Hebrew Treebank parsing is demonstrated using the new method. The benefits of joint inference are modest with the current component models, but appear to increase as components themselves improve.