SENCORPUS: A French-Wolof Parallel Corpus

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2803–2811 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2803 SENCORPUS: A French-Wolof Parallel Corpus Elhadji Mamadou Nguer, Alla Lo, Cheikh Bamba Dione, Sileye Oumar Ba, Moussa Lo UVS, UGB, UiB, Dailymotion, UVS Senegal, Senegal, Norway, France, Senegal {elhadjimamadou.nguer,moussa.lo}@uvs.edu.sn, lo.alla@ugb.edu.sn, sileye.ba@dailymotion.com, dione.bamba@uib.no Abstract In this paper, we report efforts towards the acquisition and construction of a bilingual parallel corpus between French and Wolof, a Niger-Congo language belonging to the Northern branch of the Atlantic group. The corpus is constructed as part of the SYSNET3LOc project. It currently contains about 70,000 French-Wolof parallel sentences drawn on various sources from different domains. The paper discusses the data collection procedure, conversion, and alignment of the corpus as well as it’s application as training data for neural machine translation. In fact, using this corpus, we were able to create word embedding models for Wolof with relatively good results. Currently, the corpus is being used to develop a neural machine translation model to translate French sentences into Wolof. Keywords: Parallel corpus, low-resource language, neural machine translation, Wolof, word embeddings. 1. Introduction Parallel corpora are valuable resources for bilingual lex- icography and natural language processing (NLP) appli- cations such as statistical (Brown et al., 1990) and neu- ral (Sutskever et al., 2014a) machine translation (NMT). Using machine learning techniques, computer models can learn from parallel corpus data to translate texts with rela- tively good quality, as illustrated by Google Translate. For most of the resource-rich languages, there are already many parallel corpora such as Europarl (Koehn, 2005), the Bible translations collected and annotated by Resnik et al. (1999) and the OPUS corpus (Tiedemann, 2012). However, for low-resource languages such as Wolof and Fula, such cor- pora are virtually nonexistent and, to a certain extent, this limits NLP research on these languages. In fact, Google Translate does not currently provide modules for these lan- guages. For Wolof, an LFG-based computational gram- mar (Dione, 2014) and a Universal Dependency treebank (Dione, 2019) have been recently developed. Besides, lan- guage resources and tools are scarce for that language. To promote the development of low-resource African lan- guages, the SYSNET3LOC project has set up a team, which comprises researchers in NLP and Artificial Intel- ligence from the Virtual University of Senegal (UVS), Gas- ton Berger University (UGB), Dailymotion, and the Uni- versity of Bergen (UiB). The project aims to effectively use machine learning methods to implement automatic transla- tion systems between local languages in Senegal and West- ern languages. To the speakers of these languages, this will open the door to great levels of knowledge about the world which they may not otherwise have access. Achievement of this objective requires the setting-up of multilingual corpora of about several thousand or million parallel sentences between local languages in Senegal and Western languages. However, building such linguistic re- sources presents a number of challenges. First, parallel texts between Senegalese languages and Western languages are …

Content truncated to 500 words