"Deep" Learning : Detecting Metaphoricity in Adjective-Noun Pairs

Abstract

Metaphor is one of the most studied and widespread figures of speech and an essential element of individual style. In this paper we look at metaphor identification in Adjective-Noun pairs. We show that using a single neural network combined with pre-trained vector embeddings can outperform the state of the art in terms of accuracy. In specific, the approach presented in this paper is based on two ideas: a) transfer learning via using pre-trained vectors representing adjective noun pairs, and b) a neural network as a model of composition that predicts a metaphoricity score as output. We present several different architectures for our system and evaluate their performances. Variations on dataset size and on the kinds of embeddings are also investigated. We show considerable improvement over the previous approaches both in terms of accuracy and w.r.t the size of annotated training data.

Literal, Metphorical or Both? Detecting Metaphoricity in Isolated Adjective-Noun Phrases

“…The results of our experiments are reported in Table 10. In this case, the accuracy obtained by the network with one hidden dense layer was equal to 0.969 (between the results given in (Bizzoni et al, 2017)). This significant increase is due to the much smaller number of different adjectives and the larger number of phrases with the same adjective in this data set.…”

Bigrams and BiLSTMs Two Neural Networks for Sequential Metaphor Detection

“…This model is a generalization of neural architectures for bigram phrase compositions as tested on Adjective-Noun phrases in Bizzoni et al (2017). While a similar approach is already attempted in Do Dinh and Gurevych (2016), we introduce a recursive variant which can make the compositions deeper and while allowing wide window sizes.…”

“…Köper and im Walde (2017) try detecting all metaphoric verbs in the Amsterdam corpus using this single feature. Bizzoni et al (2017) show how a network trained for metaphor detection on pairs of word embeddings can "side-learn" noun abstractness.…”

“…A recent approach is that of using neural networks for metaphor detection with pretrained word embeddings initialization. Bizzoni et al (2017) and Rei et al (2017) proved that this is a valuable strategy to predict metaphoricity in datasets of bigrams without any extra contextual or explicit world knowledge representations. While Bizzoni et al (2017) show how a simple fully connected neural network is able to learn pre-existing a dataset of metaphoric bigrams with high accuracy and to achieve a better performance than previous approaches, Rei et al (2017) present an ad-hoc neural design able to compose and detect metaphoric bigrams in two different datasets.…”

Multi-Module Recurrent Neural Networks with Transfer Learning

“…The idea of transfer learning has not been widely explored in the context of predicting the metaphoricity, especially in the context of verbs. We do not consider the method described in Bizzoni et al (2017) to be fully transfer learning.…”