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.

Inducing a grammar from text has proven to be a notoriously challenging learning task despite decades of research. The primary reason for its difficulty is that in order to induce plausible grammars, the underlying model must be capable of representing the intricacies of language while also ensuring that it can be readily learned from data. The majority of existing work on grammar induction has favoured model simplicity (and thus learnability) over representational capacity by using context free grammars and first order dependency grammars, which are not sufficiently expressive to model many common linguistic constructions. We propose a novel compromise by inferring a probabilistic tree substitution grammar, a formalism which allows for arbitrarily large tree fragments and thereby better represent complex linguistic structures. To limit the model’s complexity we employ a Bayesian non-parametric prior which biases the model towards a sparse grammar with shallow productions. We demonstrate the model’s efficacy on supervised phrase-structure parsing, where we induce a latent segmentation of the training treebank, and on unsupervised dependency grammar induction. In both cases the model uncovers interesting latent linguistic structures while producing competitive results.

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, generalpurpose 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.

In this paper, we develop multilingual supervised latent Dirichlet allocation (MLSLDA), a probabilistic generative model that allows insights gleaned from one language’s data to inform how the model captures properties of other languages. MLSLDA accomplishes this by jointly modeling two aspects of text: how multilingual concepts are clustered into thematically coherent topics and how topics associated with text connect to an observed regression variable (such as ratings on a sentiment scale). Concepts are represented in a general hierarchical framework that is flexible enough to express semantic ontologies, dictionaries,

We present an empirical study of two very simple approaches to unsupervised grammar induction. Both are based on Klein and Manning’s Dependency Model with Valence. The first, Baby Steps, requires no initialization and bootstraps itself via iterated learning of increasingly longer sentences. This method substantially exceeds Klein and Manning’s published numbers and achieves 39.4 % accuracy on Section 23 of the Wall Street Journal corpus — a result that is already competitive with the recent state-of-the-art. The second, Less is More, is based on the observation that there is sometimes a trade-off between the quantity and complexity of training data. Using the standard linguistically-informed prior but training at the “sweet spot ” — sentences up to length 15, it attains 44.1 % accuracy, beating state-of-the-art. Both results generalize to the Brown corpus and shed light on opportunities in the present state of unsupervised dependency parsing. 1

We investigate the task of unsupervised constituency parsing from bilingual parallel corpora. Our goal is to use bilingual cues to learn improved parsing models for each language and to evaluate these models on held-out monolingual test data. We formulate a generative Bayesian model which seeks to explain the observed parallel data through a combination of bilingual and monolingual parameters. To this end, we adapt a formalism known as unordered tree alignment to our probabilistic setting. Using this formalism, our model loosely binds parallel trees while allowing language-specific syntactic structure. We perform inference under this model using Markov Chain Monte Carlo and dynamic programming. Applying this model to three parallel corpora (Korean-English, Urdu-English, and Chinese-English) we find substantial performance gains over the CCM model, a strong monolingual baseline. On average, across a variety of testing scenarios, our model achieves an 8.8 absolute gain in F-measure. 1 1

We introduce a novel training algorithm for unsupervised grammar induction, called Zoomed Learning. Given a training set T and a test set S, the goal of our algorithm is to identify subset pairs Ti, Si of T and S such that when the unsupervised parser is trained on a training subset Ti its results on its paired test subset Si are better than when it is trained on the entire training set T. A successful application of zoomed learning improves overall performance on the full test set S. We study our algorithm’s effect on the leading algorithm for the task of fully unsupervised parsing (Seginer, 2007) in three different English domains, WSJ, BROWN and GENIA, and show that it improves the parser F-score by up to 4.47%. 1

We consider a new subproblem of unsupervised parsing from raw text, unsupervised partial parsing—the unsupervised version of text chunking. We show that addressing this task directly, using probabilistic finite-state methods, produces better results than relying on the local predictions of a current best unsupervised parser, Seginer’s (2007) CCL. These finite-state models are combined in a cascade to produce more general (full-sentence) constituent structures; doing so outperforms CCL by a wide margin in unlabeled PARSEVAL scores for English, German and Chinese. Finally, we address the use of phrasal punctuation

Abstract—We present an approach to unsupervised partial parsing: the identification of low-level constituents (which we dub clumps) in unannotated text. We begin by showing that CCLParser [1], an unsupervised parsing model, is particularly adept at identifying clumps, and that, surprisingly, building a simple right-branching structure above its clumps actually outperforms the full parser itself. This indicates that much of the CCLParser’s performance comes from good local predictions. Based on this observation, we define a simple bigram model that is competitive with CCLParser for clumping, which further illustrates how important this level of representation is for unsupervised parsing. I.

We present new training methods that aim to mitigate local optima and slow convergence in unsupervised training by using additional imperfect objectives. In its simplest form, lateen EM alternates between the two objectives of ordinary “soft ” and “hard ” expectation maximization (EM) algorithms. Switching objectives when stuck can help escape local optima. We find that applying a single such alternation already yields state-of-the-art results for English dependency grammar induction. More elaborate lateen strategies track both objectives, with each validating the moves proposed by the other. Disagreements can signal earlier opportunities to switch or terminate, saving iterations. De-emphasizing fixed points in these ways eliminates some guesswork from tuning EM. An evaluation against a suite of unsupervised dependency parsing tasks, for a variety of languages, showed that lateen strategies significantly speed up training of both EM algorithms, and improve accuracy for hard EM. 1