Published online by Cambridge University Press: 28 November 2008
In our study of concord phenomena in spoken Brazilian Portuguese we found disfluencies, including apparent corrections, in about 15% of the relevant tokens in our corpus of recorded speech data. Disfluencies have very little effect on the rate of marking. When fluency is included in a variable rule analysis as a factor group containing categories for different types of disfluency, as well as for data without any disfluency, it is not selected as statistically significant. Furthermore, separate analyses of all data, only fluent data, and only disfluent data revealed no significant changes in the numerical results obtained for other factor groups. We conclude that, at least insofar as the variable phenomena we studied are concerned, speech is not overly laden with errors; there is nothing in the data to mislead the language learner.