Linguistically Motivated Features for Classifying Shorter Text into Fiction and Non-Fiction Genre

Proceedings of the 29th International Conference on Computational Linguistics, pages 922–937 October 12–17, 2022. 922 Linguistically Motivated Features for Classifying Shorter Text into Fiction and Non-Fiction Genre Arman Kazmi1 Sidharth Ranjan2 Arpit Sharma1 Rajakrishnan Rajkumar1 IISER Bhopal1 IIT Delhi2 {kazmiarman85, sidharth.ranjan03}@gmail.com {arpit, rajak}@iiserb.ac.in Abstract This work deploys linguistically motivated fea- tures to classify paragraph-level text into fic- tion and non-fiction genre using a logistic re- gression model and infers lexical and syntactic properties that help distinguish the two genres. Previous works have focused on classifying document-level text into fiction and non-fiction genres, while in this work, we deal with shorter texts which are closer to real-world applications like sentiment analysis of tweets. For the task of short-text classification on the Brown corpus, a model containing linguistically motivated fea- tures confers a substantial accuracy jump over a baseline model consisting of simple POS-ratio features found effective in previous work. The efficacy of the above model containing a lin- guistically motivated feature set also transfers over to another dataset viz, Baby BNC corpus. Subsequently, we compared the classification accuracy of the logistic regression model with two deep-learning models. A 1D-CNN model gives an increase of 2% accuracy over the lo- gistic regression classifier on both datasets. A BERT-based model gives state-of-the art clas- sification accuracies of 97% on Brown corpus and 98% on Baby BNC corpus. Although, both these deep learning models give better results in terms of classification accuracy, the prob- lem of interpreting these models remains an open question. In contrast, regression model coefficients revealed that fiction texts tend to have more character-level diversity and have lower lexical density (quantified using content- function word ratios) compared to non-fiction texts. Moreover, subtle differences in word or- der exist between the two genres, i.e., in fiction texts Verbs precede Adverbs in contrast to the opposite pattern in non-fiction texts (inter-alia). 1 Introduction Written text can be classified into various cate- gories based on its content or writing style. This pa- per focuses on classifying shorter texts into fiction and non-fiction genres based on their writing style. In general, the fiction writing has an imaginative writing style and involves non-factual prose content. In contrast, the non-fiction writing deals with actual events, places, and persons and is written purely based on the facts. In some cases, distinguishing between these two writing categories is easy due to the content of the text. However, classification be- comes challenging in many instances due to blurry boundaries between them. For example, a short story may contain imaginary characters situated in real-life settings. Therefore it is essential to factor in the writing style of texts while classifying them into fiction and non-fiction genres. The problem of genre identification using lin- guistically motivated features has been extensively investigated in NLP (see references in the rest of this section). However, the particular problem of fiction vs. non-fiction genre classification has started receiving serious attention only in recent years (Vicente et al., 2021; Qureshi et al., 2019). In the cited works, different …

Content truncated to 500 words