A fully Bayesian approach to unsupervised part-of-speech tagging

By Sharon Goldwater, T. Griffiths. Annual Meeting of the Association for Computational Linguistics, 2007

422 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/469d720411f8d8d75d2352b170dbe4611b508cff

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Rethinking LDA: Why Priors Matter (2009, methodology): “In order to measure similarity between two sets of topic assignments Z and Z ′ forW , we can compute the distance between these partitions using variation of information (VI) [11, 6] (see suppl.”
Bayesian Naïve Bayes classifiers to text classification (2018, background): “Inspired by the success of Bayesian counterparts of many classical methods, such as HMM, PCA, SVM and MDS, this study proposes three Bayesian counterpart classifiers, where it turns out that classical NB classifier with Bernoulli event model is equivalent to Bayesian counterpart.”
Learning Syntactic Categories Using Paradigmatic Representations of Word Context (2012, methodology): “The following default settings were used: (i) each word Distributional Models MTO VM (Lamar et al., 2010a) .708 - (Brown et al., 1992)* .678 .630 (Goldwater et al., 2007) .632 .562 (Ganchev et al., 2010)* .625 .548 (Maron et al., 2010) .688 (.0016) - Bigrams (Sec.”
Analyzing the Errors of Unsupervised Learning (2008, background): “Bayesian approaches can also improve performance (Goldwater and Grifﬁths, 2007; Johnson, 2007; Kurihara and Sato, 2006).”
A segmental framework for fully-unsupervised large-vocabulary speech recognition (2016, methodology): “As in their model, we also follow a Bayesian approach, which is useful for incorporating prior knowledge and for ﬁnding sparser solutions [34].”
Unsupervised Word Segmentation and Lexicon Discovery Using Acoustic Word Embeddings (2016, background): “have been shown to prefer sparser solutions than maximumlikelihood models with the same structure [37].”
The effects of part–of–speech tagging on text–to–speech synthesis for resource–scarce languages (2010, background): “Bayesian HMM tagging [34] advocates that, instead of estimating a single set of optimal model parameters such as in EM, a distribution over the latent variables, given the observed data, may be obtained by integrating over all possible model parameter values.”
Collapsed Variational Bayesian Inference for Hidden Markov Models (2013, background): “(1997) expressed HMMs as Bayesian networks, which promoted the development of a number of Bayesian approaches (MacKay, 1997; Beal, 2003; Goldwater and Griffiths, 2007).”
Distributional Representations for Handling Sparsity in Supervised Sequence-Labeling (2009, methodology): “HMMs have been used many times for POS tagging and chunking, in supervised, semi-supervised, and in unsupervised settings (Banko and Moore, 2004; Goldwater and Grifﬁths, 2007; Johnson, 2007; Zhou, 2004).”
Unsupervised learning for text-to-speech synthesis (2013, background): “Recent work has shown the possibility of doing without any dictionary information, and inducing POS-like tags from scratch (Goldwater and Griffiths, 2007; Gao and Johnson, 2008).”
Posterior Regularization for Structured Latent Variable Models (2010, methodology): “Previous work has attempted to solve this problem by applying the Bayesian approach, using a prior to encourage sparsity in the model parameters (Gao and Johnson, 2008; Johnson, 2007; Goldwater and Griffiths, 2007).”
On the Theory and Practice of Privacy-Preserving Bayesian Data Analysis (2016, methodology): “Probabilistic models trained via Bayesian inference are widely and successfully used in application domains where privacy is invaluable, from text analysis (Blei et al., 2003; Goldwater and Griffiths, 2007), to personalization (Salakhutdinov and Mnih, 2008), to medical informatics (Husmeier et al.”
Unsupervised Modeling of Twitter Conversations (2010, methodology): “The model structure in Figure 2 is 5The short length of Twitter posts makes this assumption reasonable. similar to previous HMMs for supervised dialogue act recognition (Stolcke et al., 2000), but our model is trained unsupervised.”