Ralf Brown

Timeline

1. 1987–2000

   Maintaining the Interrupt List

   By October 1987, he maintains the MS-DOS Interrupt List, a reference to x86 interrupts, calls, data structures and memory addresses that the community comes to call Ralf Brown's Interrupt List. More than 650 listed contributors worldwide help build it over 15 years, and it reaches revision 61 in 2000. Addison-Wesley publishes PC Interrupts in 1993 and Network Interrupts in 1994, both written by him and Jim Kyle.

2. 1988

   Anaphora resolution at COLING

   With Jaime Carbonell, he presents "Anaphora Resolution: A Multi-Strategy Approach" at the COLING conference in Budapest in 1988. The paper argues for combining several strategies, each exploiting a different knowledge source, to resolve anaphora.

3. 1993

   Joining the Language Technologies Institute

   With a Ph.D. in computer science from Carnegie Mellon, he becomes a member of the university's Language Technologies Institute in 1993.

4. 1996

   PanEBMT in the Pangloss system

   In 1996, he describes PanEBMT, the example-based machine translation engine of the Pangloss system, which needs a large parallel corpus of example sentences and a bilingual dictionary. In an internal evaluation it covers 70.2% of unrestricted Spanish news-wire text.

5. 1997–2001

   Generalized example-based translation

   From 1997 to 2001, he is co-principal investigator, with Jaime Carbonell, of the National Science Foundation's Generalized Example-Based Machine Translation project. He is the creator of CMU-EBMT, an open-source example-based machine translation system.

6. 2001

   Diplomat and Tongues

   He works with Robert Frederking, Alexander Rudnicky and Alan Black on the Diplomat and Tongues projects, which build translation systems between two languages quickly. Their handheld, voice-enabled speech-to-speech system supports U.S. Army chaplains, first for English and Croatian, and work in Arabic leads to a real-time demonstration for the U.S. Army in 2001.

7. 2012–2014

   Identifying over 1,300 languages

   In 2012, he publishes "Finding and Identifying Text in 900+ Languages" in Digital Investigation. At Text, Speech, and Dialogue in September 2013, he presents an n-gram language identifier for 1100 languages that reaches over 99.2% accuracy with smoothing on strings of at most 65 characters. His 2014 EMNLP paper, "Non-linear Mapping for Improved Identification of 1300+ Languages", appears in Doha.

8. As of October 2026

   Principal Systems Scientist at LTI

   He is a Principal Systems Scientist at Carnegie Mellon's Language Technologies Institute. He is the creator of the LTI LangID Corpus, a resource for training language identifiers in over 2,000 languages, whose fifth release holds training data for 1266 languages.