next up previous contents
Next: Conclusions Up: Standardising Subcategorisation Previous: Semantic class

Preliminary Recommendations


Thematic Roles

Thematic relations were introduced in generative grammar during the mid-1960s and early 1970s (Gruber, 1976; Fillmore, 1968; Jackendoff, 1972) as a way of classifying the arguments of natural language predicates into a closed set of participant types which were thought to have a special status in grammar. A list of the most popular roles and the properties usually associated with them is given below.

-- A participant which the meaning of the verb specifies as doing or causing something, possibly intentionally. Examples: subjects of kill, eat, hit, smash, kick and em watch.

-- A participant which the verb characterises as having something happen to it, and as being affected by what happens to it. Examples: objects of kill, eat and smash but not those of watch, hear and love.

-- A participant who is characterised as aware of something. Examples: subject of love or object of annoy.

-- A participant which is characterised as changing its position or condition, or as being in a state or position. Examples: objects of give and hand, subjects of walk and die.

-- The thematic role associated with the NP expressing the location in a sentence with a verb of location. Examples: subjects of keep, own, retain and know and locative PPs.

-- Object from which motion proceeds. Examples: subjects of buy and promise, objects of deprive, free and cure.

-- Object to which motion proceeds. Examples: subject of receive and buy, dative objects of tell and give. (Adapted from Dowty (1989))

Since its inception, the classification of argument positions into role types was meant to be carried out in terms of primitive semantic properties of predicates. Jackendoff (1972) suggested that thematic relations should be defined in terms of the three semantic subfunctions CAUSE, CHANGE and BE which constitute some of the primitive building blocks of lexical meanings. For example, the semantic representation of a transitive verb like open would be that of (154) where, according to Jackendoff's characterisation of roles in terms of semantic subfunctions, NPtex2html_wrap_inline8553 is agent and NPtex2html_wrap_inline8555 theme.

(154)displaymath7522  tex2html_wrap_inline8557

An analogous proposal was developed by Dowty (1979) within a Montague Grammar framework and later adopted and extended by Foley & van Valin (1984).

Dowty (1989) assumes that there are only two `thematic-role-like concepts' for verbal predicates: the proto-agent and proto-patient role. Proto-roles are conceived of as `cluster-concepts' which are determined for each choice of predicate with respect to a given set of semantic properties. The properties which contribute to the definition of the proto-agent and proto-patient roles are listed below.

Contributing properties for the proto-agent role
-- Volition; sentience (and/or perception); causes event; movement.
Contributing properties for the proto-patient role
-- Change of state (including coming-to-being, going-out-of-being); incremental theme (i.e. determinant of aspect); causally affected by event; stationary (relative to movement of proto-agent).

According to Dowty, proto-roles are essentially meant for argument selection, e.g. lexical assignment of grammatical functions to subcategorised arguments. Proto-roles are related to argument selection through the Argument Selection Principle and two corollaries which govern the lexicalisation of GRs.
Argument Selection Principle: The argument of a predicate having the greatest number of proto-agent properties entailed by the meaning of the predicate will, all else being equal, be lexicalised as the subject of the predicate; the argument having the greatest number of proto-patient properties will, all else being equal, be lexicalised as the direct object of the predicate.

Corollary 1: If two arguments of a relation have (approximately) equal numbers of entailed proto-agent and proto-patient properties, then either may be lexicalised as the subject (and similarly for objects).

Corollary 2: With a three-place predicate, the non-subject argument having the greater number of entailed proto-patient properties will be lexicalised as the direct object, the non-subject argument having fewer entailed proto-patient properties will be lexicalised as an oblique or prepositional object (and if two non-subject arguments have approximately equal entailed proto-patient properties, either may be lexicalised as direct object). (Dowty, 1987, 20)

The idea underlying this approach to argument selection is that the clustering of semantic properties such as those above provide a ranking according to which the arguments of a verb compete with one another for subjecthood and objecthood. For example, the subjects of a ditransitive verb such as write correspond to the arguments for which the properties volition, sentience, causation and motion are entailed, while the direct object argument is generally understood as being an incremental themegif, causally affected and stationary as well as undergoing change. The indirect object in turn has fewer entailed proto-agent properties than the subject argument (e.g. it lacks volition and causation) and fewer proto-patient properties than the direct object arguments (it does not undergo change and is not causally affected). At parity of ranking, alternative lexicalisation patterns may arise. According to Dowty, this is what happens with lexical `doublets' such as buy and sell:

Consider first the case of buy vs. sell, lend vs. borrow. A sale transaction requires both a buyer and a seller to be sentient, to act volitionally, causally and -- normally -- with some movement (so that the buyer gets the sold object and the seller gets the buyer's money). Both these participants qualify well for subjecthood according to the selection principle, but moreover they qualify equally well, so [the selection principle] licences both lexicalisations.
In some cases, the determination of grammatical relations is more subtle. Consider the case of psychological verbs such as like and please where the syntactic realisation of the experiencer and stimulus arguments differ in spite of meaning similarities. Dowty observes that with respect to properties which promote agentivity (e.g. volition, sentience, causation, motion) either the stimulus or experiencer role can be realised as a subject.
(i) the predicate entails that the experiencer has some perception of the stimulus -- thus the experiencer is entailed to be sentient/perceiving though the stimulus is not -- and (ii) the stimulus causes some emotional reaction or cognitive judgement in the experiencer. The first of these is a property that counts licensing the experiencer as subject, while the second is one that counts as licensing the stimulus as subject.
What tips the scale in favour of the stimulus argument with verbs such as please is the possibility of an inchoative interpretation which implies a change of state in the Experiencer as shown in (155). Argument roles which have the property of undergoing change of state are canonically more suitable to be realised as objects.

(155)displaymath7522   The birthday party is pleasing Mary (right now)

All else being equal, psychological verbs which may express change of state (e.g. amuse, please, frighten and irritate) will thus realise the stimulus argument as subject and the experiencer as object. Interestingly enough, verbs such as like, as in (156)where the experiencer surfaces as the subject, do not seem to be able to give rise to an inchoative interpretation, i.e. may not be construed as expressing change of state (cf. Dowty (1987)):

(156)displaymath7522  *Mary is liking the birthday Party (right now)

Sanfilippo & Poznański (1992), Sanfilippo (1993b) and Sanfilippo (1993a) propose to extend the functionality of Dowty's prototype roles by including, in the defining clusters, properties which are instrumental for the identification of semantic verb (sub)classes. For example, it is well known that at least six subtypes of psychological verbs can be distinguished according to semantic properties of the stimulus and experiencer arguments (see Jackendoff (1972) and references therein), as in table 4.21.


non-causative source neutralreactive emotive experience
" positive " admire
" negative " fear
neutral caus. source neutral affected emotive interest
positive " positive " delight
negative " negative " scare
Table 4.21: Six subtypes of psychological verbs 

This characterisation of psychological verbs can be rendered by defining a lattice of thematic sorts relative to the stimulus and experiencer arguments which extend prototype agent and patient roles, as in figure 4.3.

Figure 4.3: Lattice of thematic sorts 

This approach can be further extended taking into account properties pertaining to other semantic classes of verbs (Austin, 1962; Searle, 1969; Talmy, 1985; Pustejovsky, 1991; Pustejovsky, in press; Asher & Lascarides, 1993; Sanfilippo, 1995a). For example, Sanfilippo (1995b) proposes to treat thematic roles as complex predicates which specify

This is shown in figure 4.4.

Figure 4.4: Thematic roles as complex predicates 

The specification of basic properties is done in terms of features which are instrumental in identifying semantic verb subclasses, as in figure 4.5.

Figure 4.5: Specification of basic properties 

The examples in figures 4.6-4.9 illustrate a few specific applications of the ensuing classification.

Figure 4.6: Example 1 

Figure 4.7: Example 2 

Figure 4.8: Example 3 

Figure 4.9: Example 4 

next up previous contents
Next: Conclusions Up: Standardising Subcategorisation Previous: Semantic class