Towards Adaptive Text Summarization: How Does Compression Rate Affect Summary Readability of L2 Texts?

Proceedings of Recent Advances in Natural Language Processing, pages 1265–1274, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_145 1265 Towards Adaptive Text Summarization: How Does Compression Rate Affect Summary Readability of L2 Texts? Tatiana Vodolazova, Elena Lloret Dept. of Software and Computing Systems University of Alicante Apdo. de Correos 99 E-03080, Alicante, Spain {tvodolazova,elloret}@dlsi.ua.es Abstract This paper addresses the problem of read- ability of automatically generated sum- maries in the context of second language learning. For this we experimented with a new corpus of level-annotated simplified English texts. The texts were summarized using a total of 7 extractive and abstractive summarization systems with compression rates of 20%, 40%, 60% and 80%. We an- alyzed the generated summaries in terms of lexical, syntactic and length-based fea- tures of readability, and concluded that summary complexity depends on the com- pression rate, summarization technique and the nature of the summarized corpus. Our experiments demonstrate the impor- tance of choosing appropriate summariza- tion techniques that align with users needs and language proficiency. 1 Introduction It is often the case that people, at some point in their lives, are incapable of benefiting from avail- able information due to various aspects of text complexity resulting from domain-specific termi- nology and grammatical structure complexity. The literature has identified specific instances of this, such as: diabetes patients with no medical training who try to make sense of recent scientific advances in the treatment of the disease (Ong et al., 2008); elderly people forced to embrace the technical ter- minology of the digital age when using comput- ers and mobile phones (Li and Perkins, 2007); parents trying to decipher the latest slang words used by their teenage children (Vizgirdaite, 2009); people with different degrees of learning difficul- ties such as aphasia (Carroll et al., 1998), dyslexia (Rello et al., 2012) or autism (ˇStajner et al., 2012); and, second languag (L2) learners trying to infer the meaning of idioms from the literal meaning of their constituents (Charteris-Black, 2002). Extensive research has been carried out in the field of automatic text simplification and text en- richment (Rello et al., 2014; Aranzabe et al., 2012; Woodsend and Lapata, 2011; Thomas and Ander- son, 2012; Barbu et al., 2015). But only a few studies integrate them into other applications of natural language processing (NLP), such as, for example, in text summarization for the purpose of improving readability. Text summarization is not considered to be primarily a simplification task. However, it becomes useful when long documents are involved as it aims to reduce text processing time and thus to access quicker the main concepts of the document. In this context most of the stud- ies apply text simplification to reduce redundant or less important information and to increase the informativeness of extractive summaries, but not their readability (Jing, 2000). Lloret et al. (2019) point out that summaries are rarely evaluated for readability. But text summarization can clearly benefit from readability assessment to better serve its purpose of saving reading time and to avoid the generation of …

Content truncated to 500 words