We are aware of two other groups of researchers currently investigating similar approaches to DF language modeling. In [4] a cleanup-style model for filled pauses is described. Ries and Qui at CMU [8] have experimented with models for repetitions and certain types of sentence deletion that incorporate the cleanup assumption. Overall, their results are consistent with ours (higher perplexity for filled pauses, lower perplexity for repetitions and deletions), but the overall effects are small, as in our case.