Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a groundbreaking lecture on machine translation that presents the first successful implementation of large-scale discriminative training. Learn how the violation-fixing perceptron framework was extended to handle latent variables and utilize forced decoding for target derivations, enabling structured learning to process massive training datasets. Discover how this innovative approach incorporates over 20 million sparse features, including lexicalized and non-local elements, resulting in significant BLEU score improvements of more than 2.0 points compared to MERT and PRO baselines. Delivered by CUNY Assistant Professor Liang Huang, an accomplished researcher in computational linguistics and machine learning who has received multiple accolades including a Best Paper Award at ACL 2008 and two Google Faculty Research Awards, this hour-long presentation from November 2013 demonstrates a major breakthrough in scalable machine translation training.
