Neural Metaphor Detecting with CNN-LSTM Model

Abstract

Metaphors are figurative languages widely used in daily life and literatures. It's an important task to detect the metaphors evoked by texts. Thus, the metaphor shared task is aimed to extract metaphors from plain texts at word level. We propose to use a CNN-LSTM model for this task. Our model combines CNN and LSTM layers to utilize both local and long-range contextual information for identifying metaphorical information. In addition, we compare the performance of the softmax classifier and conditional random field (CRF) for sequential labeling in this task. We also incorporated some additional features such as part of speech (POS) tags and word cluster to improve the performance of model. Our best model achieved 65.06% F-score in the all POS testing subtask and 67.15% in the verbs testing subtask.

Learning Outside the Box: Discourse-level Features Improve Metaphor Identification

“…In addition, for our model specifically, Conversation genre contexts are much shorter on average (23.8 vs. 97.3). Our best performing model (ELMo LAC) is within 0.4 F1 score of the first-place model in the VUA shared task (Wu et al, 2018). The GloVe LAC model would also have obtained second place at 65.2 F1, yet is considerably simpler than the systems used in the shared task, which employed ensembles of deep neural architectures and hand-engineered, metaphor-specific features.…”

“…At the recent VU Amsterdam (VUA) metaphor identification shared task (Leong et al, 2018), neural approaches dominated, with most teams using LSTMs trained on word embeddings and additional linguistic features, such as semantic classes and part of speech tags (Wu et al, 2018;Stemle and Onysko, 2018;Mykowiecka et al, 2018;Swarnkar and Singh, 2018). Most recently, Gao et al (2018) revisited this task, reporting state-of-the-art results with BiLSTMs and contextualized word embeddings (Peters et al, 2018).…”

“…We first compare our models to the baselines of the VUA shared task (Leong et al, 2018): Baseline 1, a logistic regression classifier trained only on onehot encodings of verb lemmas; and Baseline 2, the same classifier with additional WordNet class and concreteness features. We also compare to the best systems submitted to the VUA shared task: Wu et al (2018), an ensemble of 20 CNN-BiLSTMs trained on word2vec embeddings, part-of-speech tags, and word embedding clusters; and Stemle and Onysko (2018), a BiLSTM trained on embeddings from English language learner corpora.…”

Neural Metaphor Detection in Context

“…It assigns the metaphor label if the word is annotated metaphorically more frequently than as literally in the training set, and the literal label otherwise. We also compare our (2) a neural similarity network with skip-gram word embeddings (Rei et al, 2017), (3) a balanced logistic regression classifier on target verb lemma that uses a set of features based on multisense abstractness rating (Köper and im Walde, 2017), and (4) a CNN-LSTM ensemble model with weighted-softmax classifier which incorporates pre-trained word2vec, POS tags, and word cluster features (Wu et al, 2018). 2 We experiment with both sequence labeling model (SEQ) and classification model (CLS) for the verb classification task, and the sequence labeling model (SEQ) for the sequence labeling task.…”

End-to-End Sequential Metaphor Identification Inspired by Linguistic Theories

“…The most recent approaches (Wu et al, 2018;Gao et al, 2018) treat this as a sequence tagging task: the classified labels are only conditioned on BiLSTM (Graves and Schmidhuber, 2005) hidden states of target words. This approach is not tailormade for metaphors; it is the same procedure to that used in other sequence tagging tasks, such as Part-of-Speech (PoS) tagging (Plank et al, 2016) and Named Entity Recognition (NER) (Lample et al, 2016).…”