Jointly Learning Author and Annotated Character N-gram Embeddings: A Case Study in Literary Text

Proceedings of Recent Advances in Natural Language Processing, pages 684–692, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_080 684 Jointly Learning Author and Annotated Character N-gram Embeddings: A Case Study in Literary Text Suraj Maharjan⋆ Deepthi Mave⋆ Prasha Shrestha⋆ Manuel Montes-y-G´omez† Fabio A. Gonz´alez‡ Thamar Solorio⋆ ⋆Department of Computer Science, University of Houston †Instituto Nacional de Astrofısica Optica y Electronica, Puebla, Mexico ‡Systems and Computer Engineering Department, Universidad Nacional de Colombia {smaharjan2, dmave, pshrestha3, tsolorio}@uh.edu mmontesg@ccc.inoep.mx, fagonzalezo@unal.edu.co Abstract An author’s way of presenting a story through his/her writing style has a great impact on whether the story will be liked by readers or not. In this paper, we learn representations for authors of literary texts together with representations for character n-grams annotated with their functional roles. We train a neural character n- gram based language model using an ex- ternal corpus of literary texts and transfer learned representations for use in down- stream tasks. We show that augment- ing the knowledge from external works of authors produces results competitive with other style-based methods for book lika- bility prediction, genre classification, and authorship attribution. 1 Introduction Literary texts have been computationally modelled by extracting stylistic traits such as readability and writing density, flow of emotions, and even by cover images of books (Maharjan et al., 2018b,a, 2017; Ashok et al., 2013). However, modelling of authors through their work has not been explored until now. An author’s style of presenting stories has a great influence on whether a book will be liked by readers or not. We can find evidence of the effect that an au- thor’s style has on readers in book reviews and through readers’ comments left on Goodreads1 shown in Table 1. The readers talk about the im- pact of the author’s writing style on their reading experience. In the first two examples, it left a pos- itive impact on the readers, while in the last it had a negative impact. These examples provide fur- ther evidence for the need for modeling authors’ 1https://www.goodreads.com This author’s writing style is just perfect in every way, you will feel everything one should experi- ence when you read a genre such as this. The author’s writing style is straightforward which made it easy to understand. I think that the writing is very uneven. Over- whelmingly episodic, not terribly consistent, and largely as dimensionless as the characters. Table 1: Readers comments showing the impor- tance of authors’ writing style writing style for the task of likability prediction of books. In this paper we propose a new approach to capture style in text by jointly learning author specific embeddings and character based n-gram embeddings. The idea of using author embed- dings is motivated by reader comments as dis- cussed above. The use of character n-gram em- beddings comes from previous work on author- ship attribution (AA) that has shown character n-grams to have strong prediction value for the task (Keˇselj et al., 2003; Peng et al., 2003; Kop- pel et al., 2009; Stamatatos, 2009; Sapkota et …

Content truncated to 500 words