The "direct discourse analysis" introduced by Susumu Kuno is examined and found to be inadequate. To account for the data Kuno discusses, as well as for related data, a new approach to transformations is suggested. By determining the function, rather than the form, of a transformation, certain predictions are possible. Primary is the prediction that two or more grammatical devices may not apply to achieve antagonistic purposes in terms of theme-rheme distinctions. That is, no grammatical device may be used to indicate that an element of a sentence is the theme (or rheme) if another device has already been used to indicate the opposite. (Author/DD)