The inputs required for the metaphorical paraphrase generation approach include a corpus $$C$$ of unlabelled texts, a dataset $$I^{M|L}$$ of texts labelled as literal or metaphorical, and a classification threshold $$h$$.

Google

University of Michigan - Ann Arbor

Literal-to-metaphorical paraphrase generation masks a token in a literal sentence and uses a metaphor-focused language model to replace it with language intended to make the sentence metaphorical. The replacement can involve verbs, nouns, or adjectives rather than being limited to verbs.

Literal-to-Metaphorical Paraphrase Generation by Token Masking

Masked metaphor modeling is a natural language processing task similar to masked language modeling, but the masked tokens are tagged metaphors. Fine-tuning a pre-trained model for this task requires a limited number of texts classified as metaphorical (e.g., by a metaphorical text classification model, $$MC$$) and with the metaphor explicitly tagged within the text.

Masked Metaphor Modeling

Generation of new metaphors requires sentences from a corpus $$C$$ which $$MC$$ has classified as literal. A random token is masked and the masked metaphor model replaces the masked token, effectively trying to create a metaphorical text. Finally, use $$MC$$ to keep only truly transferred metaphorical texts.

Generation of new metaphors

Inputs for Metaphorical Paraphrase Generation

Metaphorical text classification is a binary classification task trained on a dataset of texts labeled as literal or metaphorical, denoted as $$I^{M|L}$$. This classification model, often referred to as $$MC$$, determines whether a given text contains metaphorical language. Common models and algorithms used for this task include Naive Bayes, Random Forests, K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Logistic Regression, Multi-Layer Perceptrons (MLP), BERT, and XLM-RoBERTa (XLM-R).

Learn Before

Related