Incoherent Sentence Detection in Scientific Articles in Russian and English
Quang Huy Nguyen, Mark Zaslavskiy · 2021
Text coherence is an important factor that often gets overlooked by novice writers. Incoherence in academic writing directly affects both the reading experience and the comprehensibility of the articles. This paper introduces and describes a method for detecting incoherence in academic writing. The method utilized a fine-tuned BERT model in conjunction with a graph clustering algorithm. We benchmarked the method against baseline models on Discordant Sentence Detection using Time-travel dataset, and the results showed that the proposed method outperformed baseline models in terms of F1-score. Afterwards, the method was tested on corpora of Russian and English scientific articles in order to assess its proficiency in Narrative Incoherence Detection when applied to the paper's main research subject: academic writing. The paper's proposed method achieved a decent F1 of over 0.65 in Discordant Sentence Detection. For future work, our biggest goal is to further refine the method and be able to effectively deploy it on existing systems for reviewing academic corpora.