Academia continues to ramp up its research into neural machine translation (NMT). Five months into the year, the number of papers published in the open-access science archive, arXiv.org, nearly equals the research output for the entire year 2016. The spike confirms a trend Slator reported in late 2016, when we pointed out how NMT steamrolls SMT.
As of May 7, 2017, the Cornell University-run arXiv.org had a total of 137 papers in its repository, which had NMT either in their titles or abstracts. From only seven documents published in 2014, output went up to 11 in 2015. But the breakthrough year was 2016, with research output hitting 67 contributions.
NMT, or an approach to machine translation based on neural networks, is seen as the next evolution after phrase-based statistical machine translation (SMT) and the previous rules-based approach.
While many studies and comparative evaluations have pointed to NMT’s advantages in achieving more fluent translations, the technology is still in its nascent stage and interesting developments in the research space continue to unfold.
At press time, NMT papers submitted in 2017 were authored by 173 researchers from across the world, majority of them (63 researchers) being affiliated with universities and research institutes in the US.
The most prolific contributor is Kyunghyun Cho, Assistant Professor at the Department of Computer Science, Courant Institute of Mathematical Sciences Center for Data Science, New York University. Cho logged 14 citations last year.
He has, so far, co-authored three papers this year — “Nematus: a Toolkit for Neural Machine Translation,” “Learning to Parse and Translate Improves Neural Machine Translation,” and “Trainable Greedy Decoding for Neural Machine Translation” — in collaboration with researchers from the University of Edinburgh, Heidelberg University, and the University of Zurich in Europe; the University of Tokyo and the University of Hong Kong in Asia; and the Middle East Technical University in Turkey.
Aside from Cho, 62 other researchers with interest in NMT have published their work on arXiv under the auspices of eight American universities: UC Berkeley, Carnegie Mellon, NYU, MIT Computer Science and Artificial Intelligence Laboratory, Cambridge, Stanford, Georgia Institute of Technology Atlanta, Johns Hopkins University, and Harvard.
Sixty-one researchers from Europe have also substantially contributed to the collection, with authors from the UK (18), Germany (11), Ireland (13), and the Netherlands (7) submitting the most papers.