Distilling grammatically relevant aspects of word meaning is one step in uncovering principles of argument realisation; another important step is to somehow structure and organise these features in a way which allows to develop an efficient syntax/semantics interface which parsi- moniously captures crucial mechanisms of the mapping from semantic arguments of a verb to syntactic functions. One prominent approach involves thematic roles, which “provide one way of relating situations to their participants” (Davis, 2011, p. 400): thematic roles are taken to be discrete and atomic (i.e. semantically unanalysable) concepts/labels which are assigned by a verb to its arguments and define the role a given argument plays in the situation/event denoted by the verb (e.g. Fillmore, 1968, 1970; Jackendoff, 1972) – for overviews, see Davis (2011) and Levin and Rappaport Hovav (2005, chapters 1,2 & 6).1 Typically, a verb’s
lexical entry will list one thematic role for each of its arguments; such lists are termed se- mantic role lists by Levin and Rappaport Hovav (2005) and are also known as ‘case frames’ Fillmore (1968) or ‘theta-grids’ (Stowell, 1981).
Crucially, the set of thematic roles is usually assumed to be fairly small in size – while the exact inventory of thematic roles as well as their definitions have been a matter of intense debate (without any consensus, as pointed out below), in the following example (20), I list a number of frequently encountered roles together with rough characterisations.2
(20) a. Agent : initiator of action, capable of volition; affecting another participant b. Patient : participant being affected/undergoing some change of state c. Theme : participant undergoing some caused change of location/motion
d. Experiencer : participant experiencing some mental effect/undergoing a mental change of state
e. Instrument : intermediary/means (put to use) in performing an action f. Source : the place from which somebody/something moves
g. Goal : the place to which somebody/something moves h. Location : place of object or action
The promise that such thematic roles hold for structuring the syntax/semantics interface is to package semantic similarities of verb meanings reflected in their arguments in a way 1As Manfred Krifka has pointed out to me, the basic concept of thematic roles can be traced back as far as P¯an.ini’s
grammar of Sanskrit (ca. 5thcentury BC), which classified participants of an action in terms of six distinct k¯arakas.
Kiparsky (2009) lists these as Agent (kartr.), Goal (karman), Recipient (sam.prad¯ana), Instrument (karan.a), Locative (adhikaran.a), and Source (ap¯ad¯ana). In functional terms, Kiparsky notes that “the k¯arakas mediate between meaning and morphosyntactic surface structure” (p. 50; emphasis in the original), thus coming quite close to the modern notion of thematic roles.
2The definitions are intended to be roughly representative of frequently found features characterising a given
role and mainly serve illustrative purposes – they areNOTintended to represent the inventory and definitions of a specific approach, however. Some of the definitions are based on those listed by Palmer, Gildea, and Xue (2010, table 1.1 on page 4).
which allows to group verbs into different natural classes whose members systematically share options for argument realisation. To illustrate how thematic roles can be brought to bear upon argument linking patterns, consider again the verbs break and hit. While deviating somewhat in terminology, we may follow Fillmore (1970, p. 131) and assume that these verbs’ lexical entries feature the following role lists:
(21) a. break : Agent, Patient, Instrument b. hit : Agent, Place, Instrument
Note that the two lists differ in listing a Patient and a Place role for their second argument, respectively, i.e. the two verbs differ semantically. The next two examples are in fact (1) and (2) repeated here. The sentence pair in (2) shows again that the two verbs also differ syntactically in their argument realisation options. The differences in argument realisation patterns can now be derived by subject/object selection rules along the following lines (cf. Fillmore, 1970, pp. 128–129): for break, the Patient argument must be expressed, while the Agent and Instrument arguments may be realised. If the Agent is expressed syntactically, it is linked as the subject, as in (1a). If the Instrument argument is expressed, it is realised as the subjectONLYif the Agent isNOT expressed within the sentence (example (22a)) – if
the Agent is realised, however, the Instrument is linked obliquely within a PP (as in example (1a)). IfNEITHERthe Agent nor the Instrument are realised, the Patient becomes the subject,
as in (2a). The Patient is linked as the direct object if either the Agent or Instrument are expressed syntactically (see (1a) and (22a), respectively). For hit, on the other hand, we may assume that the ungrammaticality of (2b) stems from the difference in the composition of the semantic role lists, which leads to the obligatory expression of either the Agent or Instrument arguments, rendering the sole realisation of the Place argument as the subject illicit.
(1) a. The boy broke the window with a rock. b. The boy hit the window with a rock. (2) a. The window broke.
b. * The window hit.
(22) a. The rock broke the window. b. The rock hit the window.
Crucially, the semantic role lists given in (21) as well as the above rules can now be extended to the other verbs patterning together with break (e.g. bent, folded, shattered, cracked – (3)) and hit (such as slapped, struck, bumped, stroked – (4)), respectively. Thus, the correlations between lexical semantics of groups of verbs and their argument linking patterns can be systematised via thematic role lists and linking rules.
To arrive at a predictive and parsimonious theory of argument realisation using thematic roles, it is generally agreed that the set of roles must be limited in size, which – in turn – implies
that each role must allow a large number of related arguments to fit its definition. Thus, finding the right ‘grain size’ in choosing the roles and their definitions is critical for any model of argument linking making reference to thematic roles – however, as discussed by several authors (e.g. Ackerman & Moore, 2001; Dowty, 1991; Levin & Rappaport Hovav, 2005), no consensus on these aspects has emerged yet.
While the notion of ‘being an Agent’ is relatively intuitive, for instance, defining an Agent’s properties has proven surprisingly difficult. As a case in point, Dowty (1991), Ackerman and Moore (2001) and Levin and Rappaport Hovav (2005) all cite Cruse (1973), who splits ‘the’ Agent role into the four ‘Agent-like’ subroles Volitive, Effective, Initiative and Agent- ive. According to this four-way split, since John in (23a) is wilfully involved in the event of tearing, he counts as a Volitive, while it would bear the Agent role following more standard definitions; the machine in (23b), on the other hand, counts as Agentive according to the definition of Cruse, but may qualify as Instrument according to the rough role characterisa- tions provided in (20).
(23) a. John tore the document to pieces.
b. The machine shreddered the document to pieces.
The lack of widely accepted role inventories and definitions has led to a rather loose applica- tion of role labels in many cases and some of them are often used as defaults for arguments whose role eludes straightforward classification – as Jackendoff (1987, p. 381) remarks, “[s]ome have taken Theme or Patient to be such a default role: if one can’t think of anything else to call an NP, call it Theme or Patient (and some have treated the two terms as inter- changeable)”. Also, as pointed out by Levin and Rappaport Hovav (2005, pp. 27–28), there is no easily definable relation between the syntactic function of direct object and its semantic complement in terms of thematic roles. While ‘true’ Patients (and/or Themes) are cross- linguistically linked as direct objects with syntactically transitive verbs, a whole number of other roles may be realised as direct objects in many languages. Levin and Rappaport Hovav (2005, p. 28) list the examples in (24) (the references are provided in the original).
(24) a. The engineer cracked the bridge. Patient
b. The engineer destroyed the bridge. Patient/consumed object c. The engineer painted the bridge. incremental theme ; Dowty (1991)
d. The engineer moved the bridge. theme
e. The engineer built the bridge. effected object/factitive ; Fillmore (1968)
f. The engineer washed the bridge. location/surface
g. The engineer hit the bridge. location ; Fillmore (1970)
h. The engineer crossed the bridge. path
i. The engineer reached the bridge. goal
j. The engineer left the bridge. source
l. The engineer hated the bridge. stimulus/target or object of emotion Parallel issues arise with the use of the label ‘Agent’ for subjects and the range of semantic roles that may be linked as subjects; Ackerman and Moore (2001, p. 23) note that “rather than a label such as AGENT being empirically motivated as the result of employing [a prin-
cipled] algorithm, it is applied post-hoc to class together entities that are known to share some important behaviors” (emphasis in the original). The following examples (Ackerman & Moore, 2001, p. 23) in (25) all have subject arguments, but “it is not evident that these arguments share any semantic property. Labeling them all AGENTis tantamount to simply stipulating that the subject of active clauses is equivalent to being an AGENT” (Ackerman &
Moore, 2001, p. 23). One strategy might be to fragment thematic roles into subtypes along the lines of Cruse (1973), but if carried through, one runs the risk of ending up with what Dowty (1991) calls ‘individual thematic role types’, which amounts to assigning to the sub- ject of a verb such as break a ‘breaker’ role and to that of hit the ‘hitter’ role. Obviously, however, such a fragmentation means to miss the chance of identifying those semantic com- monalities present across arguments which allow to make generalisations about their linking behaviour.
(25) a. The duck is swimming. b. The duck is dying.
c. The duck swallowed the frog. d. The duck saw the frog.
Another problem with thematic roles discussed by Dowty (1991) arises with the roles as- signed to argument of symmetric predicates, such as resemble (26) or border (27). In such cases, the two arguments appear to bear the same thematic role and it is not clear what (unique) event-based role to assign to each of the verbs’ arguments in a way consistent with principles of thematic role-based linking principles.3
(26) a. Nella resembles Luk. = b. Luk resembles Nella.
(27) a. The field borders the wood. = b. The wood borders the field.
While thematic roles have been the predominant means of addressing issues related to ar- gument realisation throughout the last decades, the above discussion points out some of the unresolved issues which have been plaguing this approach, the foremost of which being the question about the right grain size in choosing the set of relevant thematic roles and their definitions. Other problems inherent to thematic role-based theories of argument realisation are discussed by Dowty (1991), Levin and Rappaport Hovav (2005), Ackerman and Moore (2001), Davis and Koenig (2000) and Davis (2011), amongst others.
3Note that a similar issue arises with the verb pair buying/selling, since here “both buyer and seller must act
Agentively (voluntarily) [...] and there is no obvious reason why either is entailed to act ‘more Agentively’ than the other” (Dowty, 1991, p. 556).