Hackerman Hall B17 @ 3400 N. Charles Street, Baltimore, MD 21218
In my talk I will overview five research contributions in the field of neural machine translation recently conducted with my colleagues at Amazon AI and my students at FBK. I will start presenting work on (i) recurrent neural MT to cope with morphologically rich languages and (ii) enhancements to the transformer model to improve training under low resourced settings. Then, I will discuss two data augmentation and training methods to, respectively, (iii) integrate terminology handling intoNMT and (iv) to improve robustness of NMT against speech recognition errors. Finally, I will present (v) simple techniques that permit to bias NMT to produce shorter or longer translations and discuss our main use case.
Marcello Federico is a Principal Applied Scientist at Amazon AI, East Palo Alto, California, United States, since 2018. He received the Laurea degree in Information Sciences, summa cum laude, from the University of Milan, Italy, in 1987. From 1989 to 2007 he was researcher and head of research at the Istituto per la Ricerca Scientifica e Tecnologica (IRST), Trento, Italy. From 2008 to 2017 he directed the HLT-MT research unit at Fondazione Bruno Kessler, in Trento. Since 2018, he is an Affiliated Fellow of FBK, where he supervises PhD students and mentors researchers. Since 2002, he has been member of the ICT International Doctoral School’s Committee of the University of Trento where he lectures a course on machine translation . He is co-founder and scientific advisor of MateCat Srl and co-founder and former CEO of MMT Srl, the first company offering a real-time adaptive neural machine translation technology. His research expertise is in machine translation, spoken language translation, language modeling, information retrieval, and speech recognition. In these areas, he has co-authored over 200 scientific publications, contributed in 20 international and national projects, and co-developed open source software packages for machine translation and language modeling used worldwide byresearch and industry. He has served on the program committees of all major international conferences in the field of human language technology. Since 2004, he has been co-organising the International Workshop on Spoken Language Translation workshop series, and was chair of the EAMT 2012 conference. He has also been editor-in-chief of the ACM Transactions on Audio, Speech and Language Processing; associate editor for Foundations and Trends in Information Retrieval, and a senior editor for the IEEE/ACM Transactions on Audio, Speech, and Language Processing. He has been a board member of the Cross Lingual Information Forum, the European Association for Machine Translation, and founding officer of the ACL Special Interest Group on Machine Translation. He is a senior member of the IEEE and of the ACM.