FrenLyS: A Tool for the Automatic Simplification of French General Language Texts

Proceedings of Recent Advances in Natural Language Processing, pages 1196–1205 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_135 1196 FrenLyS: A Tool for the Automatic Simplification of French General Language Texts Eva Rolin UCLouvain CENTAL Quentin Langlois UCLouvain Patrick Watrin UCLouvain CENTAL Thomas Franc¸ois UCLouvain CENTAL Abstract Lexical simplification (LS) aims at replacing words considered complex in a sentence by simpler equivalents. In this paper, we present the first automatic LS service for French, FrenLyS, which offers different techniques to generate, select and rank substitutes. The pa- per describes the different methods proposed by our tool, which includes both classical ap- proaches (e.g. generation of candidates from lexical resources, frequency filter, etc.) and more innovative approaches such as the ex- ploitation of CamemBERT, a model for French based on the RoBERTa architecture. To eval- uate the different methods, a new evaluation dataset for French is introduced. 1 Introduction It is widely acknowledged that reading difficulties, either due to insufficient education or to mental deficiencies for example, can hinder access to in- formation, which is likely to result in a loss of autonomy and freedom (Mutabazi and Wallenhorst, 2020). Faced with this challenge, researchers imag- ined applying natural language processing (NLP) to automatically transform sentences in a text in order to make it more readable, thus facilitating access to information. This is the objective pursued in the field of Automatic Text Simplification (ATS), in which the main goal is to preserve grammat- icality and meaning while carrying out effective transformations to make the text simpler. ATS is generally investigated at the sentence level (Alva-Manchego et al., 2020) and has there- fore mostly focused on two subtasks: on the one hand, lexical simplification, described by Saggion (2017) as the task of ”replacing difficult words with easy-to-read (or understand) expressions while pre- serving the meaning of the original text segments”, on the other hand, syntactic simplification, that consists in simplifying syntactic structures in a sen- tence by carrying out various transformations (split- ting, clause deletion, etc.). Both tasks have been the subject of a great deal of research, as synthe- sized in Shardlow (2014); Siddharthan (2014); Sag- gion (2017); Paetzold and Specia (2017b); Alva- Manchego et al. (2020); Al-Thanyyan and Azmi (2021), but most of it has been carried out on En- glish or, to a lesser extent, on Spanish. Other lan- guages are hardly represented, which is also the case for French (Gala et al., 2018; Elguendouze, 2020). This why we have chosen to focus on this one. In French, ATS was addressed at first at the syntactic level (Seretan, 2012; Brouwers et al., 2014) using rule-based systems. In parallel, lexi- cal simplification was also investigated, based on lexicons or resources (Billami et al., 2018; Car- don, 2018; Hmida et al., 2018). Due to the lack of training data, machine translation approaches - which are standard for English - were applied to French (Rauf et al., 2020) only very recently, with mixed results. As a result, the situation of ATS for French is clearly lagging behind …

Content truncated to 500 words