Automated Text Simplification as a Preprocessing Step for Machine Translation into an Under-resourced Language

Proceedings of Recent Advances in Natural Language Processing, pages 1141–1150, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_131 1141 Automated Text Simplification as a Preprocessing Step for Machine Translation into an Under-resourced Language Sanja ˇStajner Symanto Research Nuremberg, Germany sanja.stajner@symanto.net Maja Popovi´c ADAPT Centre, DCU Dublin, Ireland maja.popovic@adaptcentre.ie Abstract In this work, we investigate the possibil- ity of using fully automatic text simpli- fication system on the English source in machine translation (MT) for improving its translation into an under-resourced lan- guage. We use the state-of-the-art au- tomatic text simplification (ATS) system for lexically and syntactically simplifying source sentences, which are then trans- lated with two state-of-the-art English-to- Serbian MT systems, the phrase-based MT (PBMT) and the neural MT (NMT). We explore three different scenarios for using the ATS in MT: (1) using the raw out- put of the ATS; (2) automatically filtering out the sentences with low grammaticality and meaning preservation scores; and (3) performing a minimal manual correction of the ATS output. Our results show im- provement in fluency of the translation re- gardless of the chosen scenario, and differ- ence in success of the three scenarios de- pending on the MT approach used (PBMT or NMT) with regards to improving trans- lation fluency and post-editing effort. 1 Introduction In spite of recent advances in machine translation (MT), the MT into under-resourced languages is still facing a number of problems. First, there is not enough parallel data to build robust phrase- based and neural systems. Second, the major- ity of those languages (including Serbian) have a very rich morphology and suffer from data spar- sity when it comes to less frequently used cases, tenses, etc. Third, there is a number of syntac- tic differences which are difficult to capture. For English-to-Serbian phrase-based system, a num- ber of language-related problems has been iden- tified so far (Popovi´c and Arˇcan, 2015). Most of them are related to syntactic differences, e.g. missing verb parts due to distinct structure of cer- tain verb tenses, incorrect prepositions, or incor- rect translations of English sequences of nouns. Although the neural approach better handles some grammatical aspects, it still often fails to generate correct inflections, prepositions and translations of the English noun phrases (Popovi´c, 2017). Text simplification (TS) has the goal of trans- forming given text or sentence into its simpler variant, while preserving the original meaning. What is considered to be a simpler variant de- pends on the target application, or the target reader in mind. In the case of simplifying texts for hu- mans, a simpler variant is the one that requires a shorter reading time and leads to better text com- prehension scores. In the case of text or sentence simplification used as a preprocessing step for a given natural language processing (NLP) task, e.g. machine translation (MT), information extraction (IE), summarization, and semantic role labeling (SRL), a simpler variant is the one that leads to better performances of that NLP system. Text simplification was originally proposed as a pre-processing step for machine translation …

Content truncated to 500 words