The WAW Corpus: The First Corpus of Interpreted Speeches and their Translations for English and Arabic

The WAW Corpus: The First Corpus of Interpreted Speeches and their Translations for English and Arabic Ahmed Abdelali, Irina Temnikova, Samy Hedaya, Stephan Vogel Qatar Computing Research Institute & Translation and Interpreting Institute Hamad Bin Khalifa University, Doha, Qatar {aabdelali, itemnikova, shedaya, svogel}@hbku.edu.qa Abstract This article presents the WAW Corpus, an interpreting corpus for English/Arabic, which can be used for teaching interpreters, studying the characteristics of interpreters’ work, as well as to train machine translation systems. The corpus contains recordings of lectures and speeches from international conferences, their interpretations, the transcripts of the original speeches and of their interpretations, as well as human translations of both kinds of transcripts into the opposite language of the language pair. The article presents the corpus curation, statistics, assessment, as well as a case study of the corpus use. Keywords: corpus, interpreting strategies, annotation, translation, Arabic 1. Introduction The Arabic Language Technologies research group at the Qatar Computing Research Institute1 is developing Natu- ral Language Processing (NLP) tools to support Arabic, including an Automatic Speech Recognition (ASR) sys- tem (Khurana and Ali, 2016), and a Speech-to-Text (S2T) Machine Translation (MT) system (Dalvi et al., 2017). De- spite the advances in automatic MT, such systems still lag behind interpreters in accuracy and fluency. One of the planned applications of our S2T translation system is to support the educational domain and provide translations of lectures. As simultaneous interpretation is different from translation, we wanted to better understand the strategies interpreters apply to hopefully implement some of these strategies to improve our system. A first step for this was to build a corpus of interpreted lectures, which can be used to analyze how interpreters deal with the challenges in si- multaneous interpretation. Although a number of Arabic corpora exist (Zaghouani, 2017; Wray and Ali, 2015; Ali et al., 2016a; Ali et al., 2016b; Ali et al., 2017), to the extent of our knowledge there are no publicly available interpret- ing corpora for Arabic. We therefore collected a corpus of lectures by recording talks given at conferences, together with their interpretations done by professional interpreters. After giving a short survey of the related work, we present the WAW Corpus and details about the collection and cu- ration process. Next, we provide a quantitative and quali- tative assessment of the collected data. We also present a pilot/case study of the use of the corpus for extracting in- terpretation strategies used by interpreters. 2. Related Work In (Al-Khanji et al., 2000) interpreting strategies in Arabic have been studied, but no re-usable corpus was released. There are also Arabic speech corpora used for MT2 (Kumar et al., 2014; Zaidan and Callison-Burch, 2014), but they do not include human interpretation of the original speech (only translated speech transcripts are provided). We are 1http://www.qcri.org 2https://catalog.ldc.upenn.edu/LDC97T19. not aware of any other publicly available interpreting cor- pora for Arabic, whereas they exist for Italian, Spanish, English, French, Dutch (Bendazzoli and Sandrelli, 2005; Falbo, 2012), Brazilian Portuguese and German (House et al., 2012), Japanese and Chinese …

Content truncated to 500 words