Only recently we have proposed highly efficient method for recognizing Japanese sentences containing metaphors by utilizing figurative language examples from a dictionary. The model was trained also on non-figurative language sentences taken from Japanese Wikipedia, local assembly minutes and news articles. After testing three basic text classification methods (Naive Bayes, Support Vector Machines and an Artificial Neural Network) we confirmed very high precision and recall (94-98% F-score depending on the training-testing data size ratio) achieved by the algorithm. Having proven high efficiency of our model, we now want to test it against better balanced data containing more examples of less formal language. In order to do so, we plan to excerpt test data using pieces of literature available at Aozora Bunko digital library.

References:

1. Dybala P., R. Rzepka, K. Sayama, K. Araki. Semantic clues for novel metaphor generator. Proceedings of the Workshop Computational Creativity, Concept Invention, and General Intelligence, pp. 45-50, 2013.
2. Lakoff, G. & M. Johnson. Metaphors we live by. Chicago: University of Chicago Press, 1980.
3. Veale, T., E. Shutova & B. Beigman Klebanov. Metaphor: A Computational Perspective. USA: Morgan & Claypool, 2016.
4. Dybala, P., R. Rzepka, K. Sayama & K. Araki. Detecting false metaphors in Japanese. Proceedings of the 6th Language and Technology Conference LTC, pp. 127-131, 2013.