Hi Jake, thanks a million for your answer!
It's hard to make any definite diagnoses without seeing the output of the commands you listed, but my suspicion is that you have your "speaker" and "sentence" factors represented in your data frame as numeric objects (e.g., integers 1 to length(speaker)) instead of factor objects. So when you try to test speaker or sentence by putting them in the model as fixed effects, the model tries to fit a non-sensical slope to the arbitrary x-values given by the factor labels.
> Obviously this kind of test doesn't make sense. Yes, I totally agree. But then I did this: speaker <- factor(bt.data$speaker); sentence <- factor(bt.data$sentence); tempo <- factor(bt.data$tempo); From what I read it seems to me that the three variables should then be treated as factor objects; am I right or wrong here? With is.factor(speaker), for example, I receive the response "TRUE" Best wishes, Volker