diathesis alternations are alternate ways in which the arguments of a verb are expressed syntactically. the syntactic changes are sometimes accompanied by slight changes in the meaning of the verb. an example of the causative alternation is given in (1) below. in this alternation, the object of the transitive variant can also appear as the subject of the intransitive variant. in the conative alternation, the transitive form alternates with a prepositional phrase construction involving either at or on. an example of the conative alternation is given in (2). we refer to alternations where a particular semantic role appears in different grammatical roles in alternate realisations as "role switching alternations" (rsas). it is these alternations that our method applies to. recently, there has been interest in corpus-based methods to identify alternations (mccarthy and korhonen, 1998; lapata, 1999), and associated verb classifications (stevenson and merlo, 1999). these have either relied on a priori knowledge specified for the alternations in advance, or are not suitable for a wide range of alternations. the fully automatic method outlined here is applied to the causative and conative alternations, but is applicable to other rsas.however, a considerably larger corpus would be required to overcome the sparse data problem for other rsa alternations. we have discovered a significant relationship between the similarity of selectional preferences at the target slots, and participation in the causative and conative alternations. diathesis alternations are alternate ways in which the arguments of a verb are expressed syntactically. the fully automatic method outlined here is applied to the causative and conative alternations, but is applicable to other rsas. we propose a method to acquire knowledge of alternation participation directly from corpora, with frequency information available as a by-product. notably, only one negative decision was made because of the disparate frame frequencies, which reduces the cost of combining the argument head data. diathesis alternations have been proposed for a number of nlp tasks. earlier work by resnik (1993) demonstrated a link between selectional preference strength and participation in alternations where the direct object is omitted. the syntactic changes are sometimes accompanied by slight changes in the meaning of the verb. these have either relied on a priori knowledge specified for the alternations in advance, or are not suitable for a wide range of alternations. for the conative, a sample of 16 verbs was used and this time accuracy was only 56%.