This paper presents a probabilistic model for sense disambiguation which chooses the best sense based on the conditional probability of sense paraphrases given a context. We use a topic model to decompose this conditional probability into two conditional probabilities with latent variables. We propose three different instantiations of the model for solving sense disambiguation problems with different degrees of resource availability. The proposed models are tested on three different tasks: coarse-grained word sense disambiguation, fine-grained word sense disambiguation, and detection of literal vs. nonliteral usages of potentially idiomatic expressions. In all three cases, we outperform state-of-the-art systems either quantitatively or statistically significantly. 1

We present a Gaussian Mixture model for detecting different types of figurative language in context. We show that this model performs well when the parameters are estimated in an unsupervised fashion using EM. Performance can be improved further by estimating the parameters from a small annotated data set. 1

We investigate the effectiveness of different linguistic cues for distinguishing literal and non-literal usages of potentially idiomatic expressions. We focus specifically on features that generalize across different target expressions. While idioms on the whole are frequent, instances of each particular expression can be relatively infrequent and it will often not be feasible to extract and annotate a sufficient number of examples for each expression one might want to disambiguate. We experimented with a number of different features and found that features encoding lexical cohesion as well as some syntactic features can generalize well across idioms. 1