The traditional picture: thematic roles
The problem of understanding the relations between a verb and its arguments is often talked about as a problem of the syntax–semantics inter face. This is a term used mainly in generative grammar (Chomsky 1965, Carnie 2007), the theoretical tradition in linguistics in which these questions have mostly been explored. Generative grammar postulates a strict division between different components of the grammar: syntax, phonology, semantics, and so on, each of which is assumed to have its own explanatory principles and structure (cf. 7.2.1). The interfaces between each of the components are the areas where these different explanatory principles interact: in the present case, there is an interaction between semantic principles concerning the meaning conveyed by the clause, and the morphosyntactic principles governing such phenomena as case-marking and grammatical relations. For reasons that will become apparent, the questions we asked above are referred to in the generative tradition as the linking problem, the argument mapping, argument selection or argument realization problem, or the theta-role problem. We’ll explore this mainly in relation to English, specifically concentrating on the question of the principles determining subject and object choice in the clause.
The standard explanation of the linking problem in generative syntax is due to Gruber (1965) and Fillmore (1968), later developed and extended by a host of investigators. This approach has two parts. First, the lexical entries (the representations of linguistic information associated with a lexeme) for verbs are assumed to include a specification of the types of argument they have associated with them. It was assumed that the possible arguments of all verbs could be classified into a small number of classes, called thematic roles, participant roles, semantic roles, or theta roles. The following list (based on the one in Carnie 2007) shows a commonly assumed set of roles:
In the words of Fillmore (1968: 24–25), these roles ‘comprise a set of universal, presumably innate, concepts which identify certain types of judgments human beings are capable of making about the events that are going on around them, judgments about such matters as who did it, who it happened to, and what got changed’. It was assumed that the arguments of all verbs could be assigned to one of these roles.
Take kill, for example. Killing involves someone who kills (the ‘killer’), and someone who is killed (the ‘killee’). Obviously, the killer is ‘the initiator or doer of the action’, and the killee is the ‘entity that undergoes the action’. This means that kill is associated with agent and theme participants. This information about kill is stored in long-term memory as a distinct part of the verb’s lexical entry:
We will talk of this situation as kill subcategorizing agent and theme roles. Other verbs that subcategorize these roles are hit, drive as in Wotan drove the chariot, and hug.
Now consider fear. The ‘fearer’ isn’t considered to be ‘initiating’ or ‘doing’ the action of fearing, since this implies a greater degree of agency and control than actually exists. We’re not really in control of whether we fear something or not. The first participant of fear therefore isn’t an agent, but an ‘experiencer’. The thing feared, however, is still a theme:
In receive we have a recipient and a theme as participants. The ‘receiver’ doesn’t ‘initiate’ the action of receiving, which isn’t under their control:
Put, as in put the book on the table, subcategorizes three roles, since it is characteristically ditransitive. The putter is an agent, the thing that is put is the theme, and the place where it is put is the goal.
Die, by contrast is intransitive, and consequently subcategorizes only a theme:
The idea here is that by matching up the specific semantics of the arguments of individual verbs with the wider classes of ‘agent’, ‘theme’, ‘experiencer’ and so on, it would be possible to classify the entire verbal lexicon using a finite set of participant roles. The list of the subcategorized arguments of each verb was assumed to constitute a separate aspect of the verb’s lexical entry, its theta-grid or subcategorization frame. Such grids or frames are shown for five verbs in (5).
These grids were taken to be a distinct part of a verb’s lexical entry, separate from all other aspects of its semantic representation.
The identification of verbs’ theta-grids constitutes the first part of the standard generativist explanation of the linking problem. The second part explains how the various labelled participant roles are linked or mapped onto morphosyntactic positions like subject and object. As the examples we’ve looked at show, subjects aren’t always agents, and objects aren’t always themes. This means that it’s not possible to propose any invariant linking rules associating a particular thematic role or (set of thematic roles) with either subject or object position. Instead, the basic insight behind the proposed solution is that the different thematic roles are not equivalent: some are more likely to be coded as subject, and others as object. It was suggested that it is possible to rank the different roles in an order which shows their relative accessibility to subject position.
Many versions of this ranking have been suggested. For English, an appropriate ranking might be something like this, with ‘>’ read as ‘outranks for subject’:
This ranking says that if there is an Agent in the situation being referred to, it will automatically be coded as subject. In the absence of an agent, any Beneficiary or Experiencer will be given subject status, and so on. In a transitive clause, the other participant will be coded as object, and any other participants as adjuncts (obliques). We won’t attempt to give evidence for the whole of the hierarchy here. Instead, we will illustrate various parts of it. In each case, the subject of the clause is the participant that is ranked higher. We will start with evidence showing that Agent outranks all other roles.
If there is no Agent, the Beneficiary is the next highest-ranked participant:
Facts like these lead to the postulation of the hierarchy. Researchers initially assumed that it would be possible to discover a single, cross-linguistically valid hierarchy of participant roles, specified by Universal Grammar. This allowed evidence from other languages to be used to fill in the gaps left in a single language.
