Salim Roukos is a Lebanese-American computer scientist who has spent his career at IBM Research, where he has led teams developing some of the most influential systems in speech recognition, machine translation, and natural language understanding. His work bridges the gap between theoretical advances and practical, large-scale NLP systems deployed in commercial products.
Early Life and Education
Born in Lebanon in 1953, Roukos earned his PhD in electrical engineering from the University of Florida. He joined IBM Research in the 1980s and rose to lead the Department of Multilingual NLP, overseeing projects in machine translation, information extraction, and question answering.
Born in Lebanon
Joined IBM Research and contributed to speech recognition systems
Co-authored the BLEU metric for machine translation evaluation
Led IBM's multilingual NLP research and Watson NLU components
Key Contributions
Roukos was a co-author of the BLEU (Bilingual Evaluation Understudy) metric, published in 2002, which became the standard automatic evaluation measure for machine translation. BLEU computes a modified precision over n-grams between system output and reference translations, combined with a brevity penalty. Despite its known limitations, BLEU's simplicity and reproducibility made it the de facto standard for MT evaluation for over two decades.
At IBM, Roukos led the development of large-scale statistical and neural machine translation systems that were deployed commercially, including Arabic-English and Chinese-English translation systems that achieved top results in NIST MT evaluations. He also contributed to IBM's work on question answering and information extraction systems, including components used in IBM Watson.
"The ability to evaluate machine translation automatically and reliably changed the pace of research — what gets measured gets improved." — Salim Roukos, on the impact of the BLEU metric
Legacy
The BLEU metric transformed MT research by enabling rapid, automatic evaluation of translation quality, accelerating the pace of experimentation. Roukos's leadership in building production-scale NLP systems demonstrated that statistical methods could deliver real-world value, helping establish NLP as a core capability in technology companies. He has been recognised as an IBM Fellow for his contributions.