Neural Text Simplification of Clinical Letters with a Domain Specific Phrase Table

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 380–389 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 380 Neural Text Simplification of Clinical Letters with a Domain Specific Phrase Table Matthew Shardlow Department of Computing and Mathematics Manchester Metropolitan University m.shardlow@mmu.ac.uk Raheel Nawaz Department of Operations, Technology, Events and Hospitality Management Manchester Metropolitan University r.nawaz@mmu.ac.uk Abstract Clinical letters are infamously impenetrable for the lay patient. This work uses neural text simplification methods to automatically improve the understandability of clinical let- ters for patients. We take existing neural text simplification software and augment it with a new phrase table that links complex medi- cal terminology to simpler vocabulary by min- ing SNOMED-CT. In an evaluation task us- ing crowdsourcing, we show that the results of our new system are ranked easier to under- stand (average rank 1.93) than using the origi- nal system (2.34) without our phrase table. We also show improvement against baselines in- cluding the original text (2.79) and using the phrase table without the neural text simplifica- tion software (2.94). Our methods can easily be transferred outside of the clinical domain by using domain-appropriate resources to pro- vide effective neural text simplification for any domain without the need for costly annotation. 1 Introduction Text Simplification is the process of automatically improving the understandability of a text for an end user. In this paper, we use text simplification methods to improve the understandability of clin- ical letters. Clinical letters are written by doctors and typically contain complex medical language that is beyond the scope of the lay reader. A pa- tient may see these if they are addressed directly, or via online electronic health records. If a patient does not understand the text that they are reading, this may cause them to be confused about their di- agnosis, prognosis and clinical findings. Recently, the UK Academy of Medical Royal Colleges in- troduced the “Please Write to Me” Campaign, which encouraged clinicians to write directly to patients, avoid latin-phrases and acronyms, ditch redundant words and generally write in a man- ner that is accessible to a non-expert (Academy of Medical Royal Colleges, 2018). Inspired by this document, we took data from publicly avail- able datasets of clinical letters (Section 3), used state of the art Neural Text Simplification software to improve the understandability of these docu- ments (Section 4) analysed the results and iden- tified errors (Section 5), built a parallel vocabu- lary of complex and simple terms (Section 6), inte- grated this into the simplification system and eval- uated this with human judges, showing an overall improvement (Section 7). 2 Related Work The idea of simplifying texts through machine translation has been around some time (Wubben et al., 2012; Xu et al., 2016), however with re- cent advances in machine translation leveraging deep learning (Wu et al., 2016), text simplifica- tion using neural networks (Wang et al., 2016; Nisioi et al., 2017; Sulem et al., 2018) has be- come …

Content truncated to 500 words