Research

My research interests lie in language modelling for computer speech recognition. In particular I am interested in adaptive language modelling, and have spent time investigating both mixture- and cache-based language models. The early work I conducted on these models was based on the British National Corpus, and is presented in this paper:

• P.R. Clarkson and A.J. Robinson. Language Model Adaptation using Mixtures and an Exponentially Decaying Cache. In Proceedings IEEE International Conference on Speech and Signal Processing, Munich, Germany, 1997. Click here to download the postscript of this paper.

Since then, I have been trying to apply language model adaptation techniques to the Hub 4 Broadcast News task. This work is described in the following paper:

• P.R. Clarkson and A.J. Robinson. The Applicability of Language Model Adaptation for the Broadcast News Task. In Proceedings 5th International Conference on Spoken Language Processing, Sydney, Australia, 1998. Click here to download the postscript of this paper.

Most recently, my work has consisted of developing measures of language model quality which correlate better with word error rate than perplexity does. Such measures have also been useful in guiding the development of new language models. This work is described in the following paper:

• P.R. Clarkson and A.J. Robinson. Towards Improved Language Model Evaluation Measures. To appear in Proceedings Eurospeech 1999, Budapest, Hungary. Click here to download the postscript of this paper.

A more complete summary of this work is contained in my PhD thesis. This can be downloaded in the following formats:

• Postscript (1.1MB)
• Gzipped Postscript (355KB)
• PDF (1.2MB)

I am also (at least partially) responsible for the CMU-Cambridge Statistical Language Modeling Toolkit, which is described in the following paper:

• P.R. Clarkson and R. Rosenfeld. Statistical Language Modeling Using the CMU-Cambridge Toolkit In Proceedings ESCA Eurospeech, Rhodes, Greece, 1997. Click here to download the postscript of this paper.

On a non-language modelling note, I spent a summer at Compaq's Cambridge Research Laboratory, investingating the use of Support Vector Machines for Phonetic Classification. This work is described in the following paper:

• P.R. Clarkson and P.J. Moreno. On the use of Support Vector Machines for Phonetic Classification. In Proceedings IEEE International Conference on Speech and Signal Processing, Phoenix, USA, 1999. Click here to download the postscript of this paper.

I am also an author on the following papers

• G.D. Cook, J.D. Christie, P.R. Clarkson, M.M. Hochberg, B.T. Logan, A.J. Robinson, and C.W. Seymour Real-time Recognition of Radio Broadcast Speech. In Proceedings IEEE International Conference on Speech and Signal Processing, Altanta, USA, 1996. Click here to download the postscript of this paper.
• J.F.G. de Freitas, M. Milo, P. Clarkson, M. Niranjan and A.H. Gee Sequential Support Vector Machines. To appear in IEEE International Workshop on Neural Networks for Signal Processing (NNSP99). Winsconsin, USA. Click here to download the postscript of this paper.

Back to my homepage

Last modified 23 April 1999