1 . 2 Trigram Statistical Language Model

First and foremost, I would like to thank Salim Roukos at IBM’s Thomas J. Watson Research Center and Ronald Rosenfeld at Carnegie Mellon University. The work presented here was conducted while I was a visitor at TJ Watson during the summer and fall of 1993 and it was an extension of my previous work on trigger language models done at TJ Watson the preceding summer. Salim was my supervisor and mentor throughout my stays at TJ Watson. Roni Rosenfeld and I worked together on my earlier trigger language model exploration at IBM. Without Roni, many of the implementation issues related to maximum entropy would have not been resolved as efficiently. I would also like to thank Peter Brown and Robert Mercer presently at Renaissance Technology and Stephen Della Pietra and Vincent Della Pietra presently at TJ Watson. Peter, Bob, Stephen, Vincent and Salim all jointly conceived and developed the maximum entropy framework for building statistical models at TJ Watson. Finally, I cannot forget Victor Zue, my thesis supervisor, who took time out of his busy schedule to offer his valuable input.