It has become common, even necessary in some theories, to postulate an ordered set of named semantic roles: the "thematic role hierarchy", usually presented (albeit implicitly) as reflecting some kind of universal extralinguistic knowledge about participant roles and their relationships. It is commonly assumed that the thematic role hierarchy determines the hierarchical order of the (syntactic) arguments which denote the roles (e.g. Larson, 1990:597), and some authors postulate a relationship between a semantic role's "prominence" on its hierarchy and the corresponding argument's grammatical status (Grimshaw, 1992:33ff.).

Thematic role theory has (at least) two major limitations: (i) lack of agreement as to which thematic roles exist, and (ii) the lack of any effective way to independently justify the assignment of noun phrases to thematic roles in particular sentences (Dowty, 1989:70). A third limitation is the lack of agreement as to the order of the roles in the hierarchy.

I argue that the ordered set of role names can be replaced by a simple measure of the "engagement" of each role-player in a situation under review. In principle, role engagement is a one-dimensional gradable quantity, varying from initiation (highest level) to mere presence in the situation (lowest level of engagement). Four zones on the scale can be defined in terms of two features which are commonly found, as terms, if not as features, in the literature: [Control] and [Affected]. The combination [+Control, -Affected] corresponds to initiation, [+Control, +Affected] to active response, [-Control, +Affected] to passive response, and [-Control, -Affected] to presence. A third feature, tentatively termed [Active], discriminates between [-Affected] roles associated with certain verbs. The only names which need be given to the roles are those particular to the situation ("massager" and "massagee" in a massage situation, for example).

A number of English argument structure alternations, and constraints on these alternations, are explained by the feature system proposed. For example "We sent Tom /*London the parcel" reflects the fact that Tom, but not London, can bear the feature [+Control]. "Tom pierced /*cut the knife through the cloth" reflects the fact that "cut", but not "pierce", requires a [+Affected] object. The two [-Control, -Affected] arguments in "They loaded the hay onto the wagon" are discriminated by the feature [Active], whereas the "same" two arguments in "They loaded the wagon with hay" are discriminated by [Affected], as the wagon here is [+Affected]. The feature [Active] also distinguishes the two [+Control] agents in a causative situation.

The role engagement scale does away with the major problem of the thematic role hierarchy: the fact that it depends on assigning names to roles, and it may well be that this simple scale is adequate for the purposes to which the thematic role hierarchy is most commonly put.