Hermann Ney is a professor of computer science at RWTH Aachen University whose research group has been one of the most productive and influential in statistical natural language processing. His contributions span speech recognition, language modelling, and machine translation, with particular impact on smoothing methods, alignment models, and phrase-based translation systems.
Early Life and Education
Born in Germany in 1950, Ney studied electrical engineering and received his doctoral degree from the University of Goettingen. He joined RWTH Aachen University, where he established the Human Language Technology and Pattern Recognition group, building it into one of Europe's leading NLP research laboratories.
Born in Germany
Completed doctoral studies at the University of Goettingen
Co-developed Kneser-Ney smoothing with Reinhard Kneser
Developed RWTH Aachen's phrase-based and neural MT systems
Received the ISCA Medal for Scientific Achievement
Key Contributions
Kneser-Ney smoothing, developed with Reinhard Kneser in 1995, is widely regarded as the most effective smoothing method for n-gram language models. Unlike simple discounting methods, Kneser-Ney uses the number of different contexts in which a word appears (its continuation count) as the basis for lower-order distributions, providing better estimates for unseen n-grams. Modified Kneser-Ney smoothing, which uses multiple discount values, is the standard baseline for language model evaluation.
Ney's group at RWTH Aachen developed highly competitive phrase-based statistical machine translation systems and contributed key innovations in word alignment (including HMM-based alignment models as alternatives to IBM Models), minimum error rate training, and log-linear model combination. They were also early adopters of neural methods for MT, producing state-of-the-art attention-based translation systems.
"The success of statistical methods in speech and language processing demonstrates that well-founded mathematical models, combined with sufficient training data, can capture the regularities of natural language." — Hermann Ney
Legacy
Kneser-Ney smoothing remains the gold standard for n-gram language models and is implemented in every major language modelling toolkit. Ney's RWTH group has trained dozens of PhD students who have gone on to lead research at major companies and universities, and their open-source MT systems have been widely used in the research community.