Weekly Seminar - David Yarowsky

About this event

Talk Title: Machine Translation and Multilingual NLP foer 1600-7000 Languages Abstract: The goals of essentially universal machine translation and natural language processing can be foundationally grounded on important sacred texts such as the Bible and Book of Mormon, which have been carefully translated into a uniquely large number of languages and can form the basis for extensive linguistic knowledge transfer via massively multilingual knowledge projection via verse and word alignments based on their formally analyzed universal meaning. We can then extend translations to new texts and to new languages using statistical, neural and LLM-based techniques incorporating morphological, syntactic, semantic, and word formation models trained on these multilingually aligned texts and fine-grained-feature-based massively multilingual curated translation lexicons derived from them. The talk will also address opportunities and challenges in widely multilingual spoken language translation, dialect translation, and efficient curated community-based translation/NLP resource development for underserved and endangered languages, towards the ultimate goal of universal communication in every human language. Biography David Yarowsky is a 30-year Professor of Computer Science at Johns Hopkins University, and a member of its Center for Language and Speech Processing, specializing in massively multilingual natural language processing and machine translation. He received his PhD from the University of Pennsylvania in 1996. He is an ACL Fellow, NSF CAREER award winner, Rockefeller Fellow, ACL Test-of-time award winner, ACL Treasurer, co-founder of the 30-year EMNLP conference series and longtime ACL/SIGDAT executive committee member. His research has focused on under-resourced languages, cross-lingual information transfer via bilingual word alignments, universal linguistic typology, lexical semantics and morphology, co-training, multi-view machine learning and low-resource bootstrapping.