Parallel Sentence Retrieval From Comparable Corpora for Biomedical Text Simplification

Proceedings of Recent Advances in Natural Language Processing, pages 168–177, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_020 168 Parallel Sentence Retrieval From Comparable Corpora for Biomedical Text Simplification R´emi Cardon UMR CNRS 8163 – STL F-59000 Lille, France remi.cardon@univ-lille.fr Natalia Grabar UMR CNRS 8163 – STL F-59000 Lille, France natalia.grabar@univ-lille.fr Abstract Parallel sentences provide semantically simi- lar information which can vary on a given di- mension, such as language or register. Parallel sentences with register variation (like expert and non-expert documents) can be exploited for the automatic text simplification. The aim of automatic text simplification is to better ac- cess and understand a given information. In the biomedical field, simplification may per- mit patients to understand medical and health texts. Yet, there is currently no such avail- able resources. We propose to exploit compa- rable corpora which are distinguished by their registers (specialized and simplified versions) to detect and align parallel sentences. These corpora are in French and are related to the biomedical area. Manually created reference data show 0.76 inter-annotator agreement. Our purpose is to state whether a given pair of specialized and simplified sentences is paral- lel and can be aligned or not. We treat this task as binary classification (alignment/non- alignment). We perform experiments with a controlled ratio of imbalance and on the highly unbalanced real data. Our results show that the method we present here can be used to au- tomatically generate a corpus of parallel sen- tences from our comparable corpus. 1 Introduction Parallel sentences provide semantically similar in- formation which can vary on a given dimension. Typically, parallel sentences are collected in two languages and correspond to mutual translations. In the general language, the Europarl (Koehn, 2005) corpus provides such sentences in several pairs of languages. Yet, the dimension on which the parallelism is positioned can come from other levels, such as expert and non-expert register of language. The following pair of sentences (first in expert and second in non-expert languages) illus- trates this: • Drugs that inhibit the peristalsis are con- traindicated in that situation • In that case, do not take drugs intended for blocking or slowing down the intestinal tran- sit Pairs of parallel sentences provide useful in- formation on lexicon used, syntactic structures, stylistic features, etc., as well as the correspon- dences between the languages or registers. Hence, pairs built from different languages are widely used in machine translation, while pairs differenti- ated by the register of language can be used for the text simplification. The purpose of text simplifica- tion is to provide simplified versions of texts, in order to remove or replace difficult words or infor- mation. Simplification can be concerned with dif- ferent linguistic aspects, such as lexicon, syntax, semantics, pragmatics and even document struc- ture. Automatic text simplification can be used as a preprocessing step for NLP applications or for producing suitable versions of texts for humans. In this second case, simplified documents are typ- ically created for children (Vu et al., 2014), for people with low …

Content truncated to 500 words