The hassle is at the mo you have to train the program to your voice, so,
with Dragon Naturally Speaking, I got 6 NSs to first train for about an
hour (about 4,000 words), and then to do a sight unseen text (about 1,000)
words which I analysed in terms of t-unit, clauses, groups and words.
Results as you might expect degrade with the complexity of the unit of
analysis. the means for the categories (for NSs) are:
t-units clauses groups words
64.6% 68.7% 78.1% 80.9%
The paper's coming out later this year in TEXT Technology.
Dave Coniam
Chinese U of HK