Introduction

The ability of the computers to handle larger amounts of texts and the availability of more texts in electronic form led to the rise of data driven research in computational linguistics. In the case of Machine Translation (MT) and other kinds of multilingual Natural Language Processing (NLP), the first source of large data came from collections of translations, initially in the Statistical MT (SMT) approach from IBM [1], which was based on the Proceedings of the Canadian Parliament in English and French as their data source. This research direction was followed by a proliferation of SMT models, which relied on larger and larger collections of parallel data, which consist of exact translations between a pair of languages or several languages at the same time.