Classifying Idiomatic and Literal Expressions Using Topic Models and Intensity of Emotions

Abstract

We describe an algorithm for automatic classification of idiomatic and literal expressions. Our starting point is that words in a given text segment, such as a paragraph, that are highranking representatives of a common topic of discussion are less likely to be a part of an idiomatic expression. Our additional hypothesis is that contexts in which idioms occur, typically, are more affective and therefore, we incorporate a simple analysis of the intensity of the emotions expressed by the contexts. We investigate the bag of words topic representation of one to three paragraphs containing an expression that should be classified as idiomatic or literal (a target phrase). We extract topics from paragraphs containing idioms and from paragraphs containing literals using an unsupervised clustering method, Latent Dirichlet Allocation (LDA) (Blei et al., 2003). Since idiomatic expressions exhibit the property of non-compositionality, we assume that they usually present different semantics than the words used in the local topic. We treat idioms as semantic outliers, and the identification of a semantic shift as outlier detection. Thus, this topic representation allows us to differentiate idioms from literals using local semantic contexts. Our results are encouraging.

Idiom Token Classification using Sentential Distributed Semantics

“…We first evaluate these representations using a "per expression" study design (i.e., one classifier per expression) and compare our results to those of Peng et al (2014) who applied multiparagraphs contexts to generate best results. We also experiment with a "general" classifier trained and tested on a set of mixed expressions.…”

“…Skip-Thought Vectors (Sent2Vec) (Kiros et al, 1 However, it is not possible for us to reproduce their results directly as they "apply the (modified) Google stop list before extracting the topics" (Peng et al, 2014(Peng et al, , p. 2023 and, to date, we do not have access to the modified list. So in our experiments we compare our results with the results they report on the same data.…”

“…Feldman and Peng (2013) explore two different approaches to outlier detection based on principle component analysis (PCA) and linear discriminant analysis (LDA) respectively. Building on this work, Peng et al (2014) assume that phrases within a given text segment (e.g., a paragraph) that are semantically similar to the main topic of discussion in the segment are likely to be literal usages. They use Latent Dirichlet Allocation (LDA) (Blei et al, 2003) to extract a topic representation, defined as a topic term document matrix, of each text segment within a corpus.…”

“…They then trained a number of models that classify a phrase in a given text segment as a literal or idiomatic usage by using the topic term document matrix to project the phrase into a topic space representation and label outliers within the topic space as idiomatic. To the best of our knowledge, Peng et al (2014) is currently the best performing approach to idiom token classification and we use their models as our baseline 1 .…”

“…Previous work on idiom token classification, such as (Sporleder and Li, 2009) and (Peng et al, 2014), often frame the problem in terms of modelling the global lexical context. For example, these models try to capture the fact that the idiomatic expression break the ice is likely to have a literal meaning in a context containing words such as cold, frozen or water and an idiomatic meaning in a context containing words such as meet or discuss (Li and Sporleder, 2010a).…”