Hunter MT: A Course for Young Researchers in WMT17

Proceedings of the Conference on Machine Translation (WMT), Volume 2: Shared Task Papers, pages 422–427 Copenhagen, Denmark, September 711, 2017. c⃝2017 Association for Computational Linguistics Hunter MT: A Course for Young Researchers in WMT17 Jia Xu and Yi Zong Kuang and Shondell Baijoo and Hyun Lee and Uman Shahzad and Mir Ahmed and Meredith Lancaster and Chris Carlan Hunter College, City University of New York Jia.Xu@hunter.cuny.edu, {YiZong.Kuang38, shondell.baijoo41, hyun.lee16, Uman.Shahzad34, Mir.Ahmed57, Meredith.Lancaster88, Chris.Carlan70}@myhunter.cuny.edu Abstract This paper documents an undergraduate course at Hunter College, in which one instructor, six undergraduates, and one high school student built 17 machine translation systems in six months from scratch. The team successfully partici- pated in the second Conference on Ma- chine Translation (WMT17) evaluation on the news task in Finnish-English and Latvian-English and on the bio-medical task in French-English, English-French, English-German, English-Romanian, and English-Polish. 1 Introduction Machine learning has advanced the state-of-the-art of artificial intelligence at a rapid speed. There has been an increasing amount of related courses introduced. However, hands-on experience is of vital importance to novices (Lopez et al., 2013). Through conventional education it may take many years for beginners to find a research direc- tion in the field of machine translation. Introduc- tory courses in this field can be either too theoret- ical or too detailed, leaving students lost in cod- ing. Therefore, our goals were to make the ma- terial both detailed and comprehensive and also bring novelty and excitement into this course. We propose teaching methods that are centered around the machine translation competition. With this idea in mind we were able to achieve our goals be- cause of three key factors. First, we were able to focus on the pragmatical aspects of the teaching material; second, the study was comprehensive, since we covered all the components of a machine translation system; and third, student motivation was enhanced, because the results were directly available in the MT community through the WMT evaluation. These key factors helped us to attain our goals optimally and efficiently. We will discuss the teaching methods to intro- duce undergraduates to topics of advanced ma- chine translation technology. We will describe the outcome, the machine translation systems in 17 languages, in particular the ones that were suc- cessfully submitted in the WMT17, as well as the research revised in this course. In conclusion, we will evaluate our course with feedback of students. 2 Backgrounds At Hunter college, we experimentally designed an applied machine translation course to provide an opportunity for novice students to learn by com- peting against the best senior teams around the world in one of the most significant machine trans- lation competitions, the WMT17. The students’ levels ranged from high school to college senior, and none of them had any prior knowledge of ma- chine learning or machine translation. Hunter college offered supportive facilities for the course. Two students took an introductory C++ programming class (Software Analysis and Design I) (Hunter, 2016). Some students took an one-week “Linux introductory class” (Hunter, 2016) at …

Content truncated to 500 words