Article

"Thematic Roles Are Not Semantic Roles"

Jan G. van Voorst

Revue québécoise de linguistique, vol. 17, n° 1, 1988, p. 245-259.

Pour citer cet article, utiliser l'information suivante :

URI: http://id.erudit.org/iderudit/602623ar

DOI: 10.7202/602623ar

Note : les règles d'écriture des références bibliographiques peuvent varier selon les différents domaines du savoir.

Ce document est protégé par la loi sur le droit d'auteur. L'utilisation des services d'Érudit (y compris la reproduction) est assujettie à sa politique

d'utilisation que vous pouvez consulter à l'URI https://apropos.erudit.org/fr/usagers/politique-dutilisation/

Érudit est un consortium interuniversitaire sans but lucratif composé de l'Université de Montréal, l'Université Laval et l'Université du Québec à
Montréal. Il a pour mission la promotion et la valorisation de la recherche. Érudit offre des services d'édition numérique de documents

scientifiques depuis 1998.

Pour communiquer avec les responsables d'Érudit : [email protected]

Document téléchargé le 12 février 2017 05:04

 THEMATIC ROLES ARE NOT SEMANTIC
Jan ROLES*
G. van Voorst

1. Introduction
In the model of grammar in (1), as proposed in Jackendoff (1983), we find a
component of conceptual structures instead of a purely semantic component. This
component contains not only all the information covered by language but also the
information covered by other perceptual systems. A set of rules relating the
Conceptual Structure Component and Syntactic Structures, the correspondence rules,
determines which aspects of the conceptual structures are relevant for the semantics of
natural language. Semantics in this model of grammar is the subset of Conceptual
Structures that is mapped onto Syntactic Structures through the component of
Correspondence Rules. In (1), I have omitted all the irrelevant aspects for the present
discussion.
(1)
Phonology Correspondence
Rules

Phonetic Conceptual
representation Structures

I wish to thank two anonymous reviewers, as well as Paul Hirschbuhler, Marisa

Rivero and Co Vet for their comments on earlier versions of this paper.
246 JAN G. VAN VOORST

In Jackendoff (1978) it is argued that the projection rules, the correspondence

rules in (1), can be assumed to be relatively simple. This way, the language
acquisition process is easy on the learner. A methodological assumption in this
context is that there is a correspondence between grammar, Syntactic Structures, and
Conceptual Structures: "Apparent grammatical constraints may reflect conceptual
constraints"1 (page 203). This correspondence is exemplified by the semantic
analysis of, among others, prepositional phrases as denoting locations (2a) and paths
(2b).
(2) a. John was in the house
b. John went into the house
These two notions aie generalized to the abstract domains of indentification (3a) and
possession (3b).
(3) a. David left Nora sick
b. Dick received the money
(3b), for instance, is analyzed as the motion Q£i that is modified by poss, which
identifies the semantic field of possession. TO (DICK) stands for the path along
which this motion takes place.
(4) GO (THE MONEY, TO (DICK))
poss
In this paper I will use the idea of a correspondence between Syntactic and
Conceptual Structures to show that notions such as agent, patient, etc. are conceptual
but not semantic notions in the grammar of Dutch and English. In Fillmore (1968)
these notions are called case roles and in the Government and Binding Framework
they represent the semantic part of thematic (6) roles. Thematic roles in the latter
framework also have a syntactic function in that they determine the well-formedness

1. This idea is not new and it can be found, for instance, in the work of Roman
Jakobson. It is outlined in Waugh (1976) that Jakobson assumes a close relationship between
form and meaning. This relationship is not arbitrary: «Forms are directly connected with a
meaning and both are communicated by the speech signal» (page 48). An extensive analysis
that follows these lines of thinking is given in Jakobson (1936), which is a semantic analysis
of the Russian case system.
 THEMATIC ROLES 247

of syntactic structures. To do this, no reference need to be made to their semantic

content (Styan 1984). However, I will not focus on the syntactic function of
thematic roles, but only on their semantic content when referring to them. I will use
the phenomena of subject selection and middle formation in English to show that a
close relation between thematic roles and the grammatical system is hard to conceive.
The discussion of impersonal passive in section 3 and 4 focuses on one conceptual
structure that ties in more adequately with the grammatical system than thematic
roles.

2. Subject Selection and Middle Formation

Fillmore (1968) has argued that the semantics of the subject in English can be
explained by using thematic roles or, case roles in his terminology, and a so-called
subject selection hierarchy. This hierarchy is given in (5). Note that the thematic
role of Object has been replaced by the role of patient in later studies.
(5) Subject Hierarchy
If there is an Agent it becomes the subject; otherwise,
If there is an Instrument, it becomes the subject;
otherwise the subject is the Object.
The hierarchy in (5) predicts that if the Agent is left out from the case frame of a verb
like open in (6), the Instrument becomes the subject NP, as in (7b). If the
Instrument is deleted as well, then the Object becomes the subject NP, as in (7c).
(6) Open + LObject (Inst) (Agent)]
(7) a. He opened the door with a key
b. The key opened the door
c. The door opened
That the above hierarchy is descriptively inadequate can be demonstrated quite easily.
The ungrammatical sentence in (8) is predicted to be grammatical by the Subject
248 JAN G. VAN VOORST

Hierarchy, as the Instrument is rightly selected over the Object and the Agent is
absent.
(8) This spoon eats the meat
It follows from this inadequacy that the Subject Hierarchy in (5) does not imply a
complete correspondence between thematic roles and the subject NP. However, there
is another way of looking at a strict correspondence between grammatical categories
and semantic primitives. That is if thematic roles, except for describing the
semantics of a noun phrase, play a role in the functioning of rules of grammar.
Hypothetically this is the case when a rule X applies only to Agents but not to
Instruments and Objects or when a rule Y applies only to Agents and Instruments but
not to Objects, etc. A brief glance at passive and middle formation shows us that
thematic roles do not provide us with this type of explanatory power. Both subjects
in (9) would normally be considered to have the same thematic roles of possessor in a
theory making use of thematic roles 2 . Still (9a) is grammatical but (9b) is not
(9) a. That car is owned by the university
*b. That car is had by the university
A rule like middle formation in English is not sensitive to the thematic role of the
subject either. Patient subjects are allowed in (10) but not in (U)3 .
(10) a. This book sells easily
b. This orange peels easily
(11) *a. This ball throws easily
*b. This story writes easily

2. Note that I simply assign here, and in later examples, thematic roles in an intuitive
way. The absence of a well-worked out theory covering these semantic notion (Grunau 1985)
makes it impossible to determine them in a more motivated way.
3. The data in (10) and (11) indicate that the patient-like character of the subject is not
the essence of this construction. Semantic descriptions that consider the middle from a different
perspective have been given in O'Grady (1980) and Oosten (1977).
 THEMATIC ROLES 249

If we take Jackendoffs ideas seriously, then we can demand from a semantic

notion that it can be distilled from the grammatical system by the language learner
without too much difficulty. But, as far as thematic roles are concerned, we can say
that the language learner is confronted with an ambiguous input that offers no clues
with respect to them. One time an Instrument may be the subject of a transitive
sentence in English (7b); other times it may not (8). Rules like the passive and the
middle do not help the language learner out either, as these roles distinguish between
NPs bearing a similar thematic role.

3. Impersonal Passive
Within the Relational Grammar framework it has been argued that there is a
correspondence between notions from the domain of thematic roles and the
grammatical system. Perlmutter (1978) uses the notion of activity, which I consider
to belong to the domain of thematic roles, as it implies agentivity. Subjects of
verbs expressing activity are said to be nonderived (12a). Subjects of verbs not
expressing activity are derived and originate in direct object position, as in (12b). In
the Relational Grammar framework the direct object position is called a 2.
(12) a. He is sleeping
b. The vasei broke ei

This distinction is said to correlate with the possibilities to apply impersonal

passive in many languages, among which is Dutch. The rule of impersonal passive
in this language inserts passive morphology and replaces the subject NP by the
grammatical formative £r 'there1, as shown in (13). Only verbs that express activity
may passivize, according to Perlmutter.
(13) Er wordt daar geslapen
there is there slept
'People are sleeping there'
250 JAN G. VAN VOORST

Passive of intransitive verbs not expressing activity is excluded in the Relational

Grammar framework for reasons we need not go into here.
It can be shown that the relation between the semantic notion of activity and
possibilities of passive is ill-conceived. Intransitive motion verbs in Dutch can
passivize, as in (14a). Daar in (14a) is a locative adverbial. When these verbs
subcategorize a directional PP, as in (14b), passive cannot take place. The directional
PP in (14b) is the phrase d& deur uil which contains the postposition nil.
(14) a. Erwordtdaargelopen
there is there walked
'People are walking there'
*b. Erwordt daar de deur uitgelopen
there is there the door out walked
The corresponding active sentences are grammatical, as shown in (15).
(15) a. Zeliependaar
'They were walking there'
b. Ze liepen de deur out
'They walked out the door'
In terms of the notion of activity, it is hard to make a distinction between the subject
in (15a) and (15b) and the verbs of the two sentences. The entities denoted by the
subjects in both sentences are actively involved in an event of walking and, as such,
they are both agents.
We can conclude that Perlmutter's approach does not show convincingly that
there is a correspondence between notions from the domain of thematic roles and the
grammatical system. This means in Jackendoff s model of grammar in (1) that
notions covered by thematic roles do not figure in the grammatical subsystem, if,
again, we assume that there must be a correspondence between semantic primitives
and the functioning of a grammar. In (1) this means that these roles are not mapped
from Conceptual Structure onto syntactic structures by correspondence rules.
 THEMATIC ROLES 251

4. Event Structure
In the preceding sections I argued that thematic roles are conceptual rather than
semantic notions. These roles are not mapped onto syntactic structures. We can ask
now which conceptual notions are mapped onto Syntactic Structures so that we can
explain phenomena like subject selection, middle formation and the limited
possibilities of impersonal passive in Dutch. In this section, I will outline briefly
what one such a conceptual notion may be. I will wholly concentrate on the Dutch
impersonal passive data in doing this. I will not give an in depth account of this
conceptual notion. I will only indicate the nature of one of the primitives that is
more relevant to a semantic theory under our assumptions.
The two sentences in (15) in section 3 do not differ with respect to the
thematic role of the subject NP. Still the sentence with the directional PP cannot
passivize, but the sentence without this PP can. The semantic difference between
these two sentences is describable using Vendler's (1967) notion of accomplishment.
A sentence containing a directional PP implies the accomplishment or the terminal
point of the event expressed by it. A sentence without such a PP does not. A way
to show this difference is the insertion of an adverbial denoting a stretch of time like
for five minutes or vijf minuten. in Dutch. An accomplishment cannot take place
over a period of time (16a); a nonaccomplishment can take place for a certain number
of minutes, as shown in (16b) 4 .
(16) *a. Hij liep vijf minuten de deur uit
he walked five minutes the door out
b. Hij liep daar vijf minuten
'He was walking there for five minutes'
Accomplishment of intransitive constructions in Dutch is directly expressed by the
grammatical system, because all intransitive accomplishment verbs select the
auxiliary zijn 'to be' instead of the usual hebben 'to have'. A motion verb with a
directional PP selects zijn (17a); when this PP is absent hebben is selected (17b).

4. Co Vet (personal communication) pointed out to me that (16a) is grammatical under

an iterative interpretation. This fact is not counter to what I argue here, if the iterative reading
is seen as the iteration of a certain number of accomplishments.
252 JAN G. VAN VOORST

(17) a. Zij zijn de deur uitgelopen

they are the door out walked
'They walked out the door'
b. Zij hebben daar gelopen
they have there walked
'They have been walking there'
The notion of accomplishment is more directly reflected in the grammatical
system than any of the thematic roles are. This means that it is quite easy for the
Dutch language learner to set off intransitive accomplishment verbs from other
verbs 5 .
We may conclude on the basis of (16a) and (16b) that Dutch impersonal
passive is sensitive to the distinction between accomplishment and nonaccomplish-
ment semantics: intransitive constructions expressing accomplishment may not but
the ones expressing nonaccomplishment may passivize. That this is not true is easy
to show. There is another type of intransitive construction that cannot passivize
although it does not express accomplishment. The intransitive verbs occurring in
these constructions do not select the auxiliary zijn 'to be' but hebben 'to have'. In
Perlmutter (1978) these verbs are classified along with verbs like breken 'to break'
and vallen 'to fall' as belonging to the class of nonactivity verbs. The subject of
these verbs is derived under Perlmutter's approach. (18a) and (18b) are examples of
these constructions.

5. Except for intransitive accomplishment verbs, also copula verbs select zijn 'to be' as
their auxiliary. This class of verbs in Dutch is easily distinguishable from the accomplishment
verbs because they are always accompanied by a predicate, (a) and (b) are examples of two
Dutch copula constructions.
(a) Hij is ziek geweest
he is sick been
He has been sick'
(b) Die jas is mij te groot gebleken
that coat is me too big seemed
That coat turned out to be too big for me'
 THEMATIC ROLES 253

(18) a. Deze sigaren hebben nooit gestonken

these cigars have never stunk
These cigars never stank1
b. Deze stofheeft tot nu toe zacht aangevoeld
this tissue has till now up soft felt
This tissue felt soft up to now'
The constructions in (18) cannot be passivized, as shown in (19).
(19) *a. Er werd door deze sigaren nooit gestonken
there was by these cigars never stunk
*b. Er werd door deze stof tot nu toe zacht
there was by this tissue till now up soft
aangevoeld
felt
The constructions in (18) are different from constructions that express
accomplishment or nonaccomplishment They do not denote an event like these two
constructions but a state. If states are defined as not taking place in reality, no
reference can be made to their being accomplished or nonaccomplished. Under this
definition states only imply the predication of a property, which is expressed by the
VP in (18), over an entity denoted by the subject NP. This can be shown, among
other ways, through the insertion of adverbials like zonder veel inspanning 'without
much effort*. These adverbials describe the way the entity denoted by the subject NP
is involved in keeping an event going, as shown in (20a). When used in a stative
construction (20b), the sentence becomes ungrammatical. States do not take place,
which means that they do not imply the involvement of the entity denoted by the
subject NP 6 .

6. Vendler defines states as expressing qualities. My definition includes Vendler's

(1967) definition of facts. Facts do not take place. I think that the same is true of states, (a) is
an example of a fact Vendler gives.
(a) For many years it was a fact that Africa was dominated by European powers.
254 JAN G. VAN VOORST

(20) a. Deze man werkt altijd zonder veel inspanning aan zijn boek
This man always works on his book without much effort'
*b. Deze man ruikt altijd zonder veel inspanning naar
this man smells always without much effort of
eau de cologne
eau de cologne
The grammatical phenomena of impersonal passive and auxiliary selection
together make it possible to divide Dutch intransitive verbs into four classes, as
shown in (21). When using the notions of accomplishment, nonaccomplishment and
our definition of states, it is possible to create a correspondence between these classes
and a semantic primitive.
(21)
I n m IV
auxiliary hebben hebben zijn zijn
selection
impersonal yes no yes no
passive
semantics nonaccom- stative ? accom-
plishment plishment
example werken ruiken ? lopen+dirPP
(20a) (20b) (17a)
The language learner can derive the semantic classes in (21) directly from the input
data. One possible class in (21) contains no verbs. This is the class of verbs that
can undergo impersonal passive and select the auxiliary zjjn 'to be'. The reason for
the nonexistence of this class will be understood better once we have explained why
only verbs of the first class can be passivized. The fourth class in (21) contains only
accomplishment verbs. Besides motion verbs that subcategorize a directional PP,
many other verbs belong to this class. Among these we find verbs like breken 'to
break' and sterven 'to die'. In this class we find most of the nonactivity verbs of
Perlmutter (1978) except for the verbs of our class II, the statives.
To understand better why only verbs of class I can undergo impersonal
passive, we have to take a closer look at the semantics of class I, class II and class IV
 THEMATIC ROLES 255

verbs. States, as we saw above, never allow an accomplishment reading. This

means that states make no reference to an entity that can be used to identify their end.
States only stop occurring, but they do not get to an end like events do. In contrast,
the termination of the event of walking to a certain destination, as in (17), can be
identified as soon as the zij have reached that destination. In addition to speaking
about an entity used to identify the termination of an event, we can equally well
speak of an entity used to identify the event's beginning. I will call the former
entity the point of termination and the latter the point of origin. Accomplishment
constructions imply an event that has an origin. This origin is left unexpressed in
intransitive accomplishments. Only the point of termination of the event of
breaking is expressed in (22a) but not its point of origin, as is the case in (22b). In
(22b) the causer Ian is the point of origin of the event as this entity initiates the
event.
(22) a. Devaasbrak
' The vase broke'
b. Jan brak de vaas
'John broke the vase'
The point of origin of an event is always expressed in constructions containing verbs
of class I in (21) but now the point of termination is not mentioned. In (23) the
entity denoted by the subject NP d£ man is the origin of the event, as he initiates it.
(23) De man werkte
' The man was working'
Concluding we can say that constructions containing verbs of class I in (21)
always contain an NP denoting the point of origin of an event; when containing a
verb of class IV only the point of termination is part of the construction; when
containing a stative verb of class II a point of origin nor a point of termination is
included in it.
The generalizatipn is that Dutch intransitive constructions can only be
passivized when the point of origin of an event is present. This is the case for (23)
but not for (18) and (22a). Under this interpretation, the preposition door T)y' in the
door-phrase that contains the NP that is the subject in the corresponding active
256 JAN G. VAN VOORST

sentence subcategorizes a "point of origin". Only verbs from class I passivize:

statives do not imply an origin and constructions containing intransitive
accomplishments only express the point of termination. If we assume that the
selection of the auxiliary zijn automatically implies that a verb expresses
accomplishment, it can be understood why class III in (21) does not contain any
verbs. Passive requires a point of origin, but this point is not expressed in
constructions containing an intransitive verb selecting zijn.
The conceptual structure that is mapped onto syntactic Structure in a
Jackendoff model of grammar is called Event Structure. Events consist of a point of
origin and a point of termination. These two aspects of events are related through a
correspondence rule to the subject and the direct object NP in Syntactic Structure.
The correspondence rule is shown in (24)7 .
(24) Event Structure point of point of
origin termination
o o
Syntactic Structure subject NP direct object NP
The direct object NP comes into play in transitive constructions. Here the
accomplishment of the event is always related to the direct object NP rather than to
the subject NP. This is shown in (25). This sentence only means that the entity
denoted by the direct object NP gets tired but not the one denoted by the subject NP.
The person denoted by the NP haar Tier1 is the point of termination of the event it is
its tiredness that makes the termination of the event identifiable.
(25) Hij danste haar moe
'He danced her tired'

7. The notions of point of origin and point of termination can represent the semantic
content of thematic roles in the Chomskian framework. Then, NPs are not considered as
participants in events but as representing constituents of Event Structure. Other NPs than the
subject NP and the direct object NP can be analyzed as constituents of Event Structure as well.
At this moment, I can only speculate as to the nature of their constituency. One possibility is
to think in terms of the notions used to describe the semantics of case morphemes in studies
like Jakobson (1936). Unfortunately, these notions have not been developed with the aspectual
notion of Event Structure in mind.
 THEMATIC ROLES 257

The subject in transitive constructions denotes the point of origin of the event. The
subject in (26a) stands for the point of origin of the event of seeing; the subject in
(26b) for the origin of the event of eating.
(26) a. Hij heeft gisteren de koningin gezien
'He saw the queen yesterday'
b. Hij heeft een snoepje gegeten
'He ate a candy'
That the direct object NP is the only NP that can denote the point of
termination can be generalized to intransitive constructions. This makes it necessary
to assume that a number of intransitive verbs have a derived subject (27a), but not
(27b), has a derived subject if we follow this line of thinking. (27a) is an
accomplishment and contains a point of termination; (27b) is a nonaccomplishment,
which means that it only contains a point of origin8 .
(27) a. Hiji liep e^ dedeuruit
t i
'He walked out the door'
b. Hij liep de hele dag op en neer
'He was walking up and down all day'
8. These assumptions are radically different from Perlmutter's. In the Relational
Grammar framework the derivation of the subject of certain intransitive verbs is linked to the
notion of activity. We exclude the statives from having a derived subject by our approach, but
we include other cases in Dutch that are absent from Perlmutter's enumeration of nonactivity
verbs. These are constructions containing a motion verb and a directional PP and constructions
like (a). The verb werken 'to work' subcategorizes the perfective particle yit 'out' and selects the
auxiliary zijn. When not subcategorizing this particle hebben is the auxiliary of the verb (b).
(a) Hij is uitgewerkt i
he is out worked
'He has finished working'
(b) Hij heeft gewerkt
'He worked'
Copula constructions remain a separate problem as they do not express accomplishment
but still select zijn like accomplishment verbs. See for this note 5.
258 JAN G. VAN VOORST

Although it is likely that the NP hij in (27a) is also the point of origin of the event,
we are forced to say that this is not expressed by the sentence meaning. The
accomplishment is linked to this NP: the event is accomplished when the hij is out
the door. This analysis of intransitive accomplishments makes it possible to
maintain a one-to-one mapping between Event Structure and Syntactic Structures.
When the language learner encounters a transitive construction, it is either a
state and cannot passivize (28) or it denotes an event (29). When it denotes an event
a point of origin is present. Passive may apply in this case because the rule is
sensitive to the presence of the point of origin of an event, as is also the case with
impersonal passive.
(28) *a. This book was had by him
*b. Dit boekwerd door hem gehad
(29) a. This pie was eaten by him
b. Deze taart werd door hem opgegeten
Again the language learner is faced with a set of syntactic input data from which he
can easily discern semantic primitives.

5. Conclusion
We have seen that notions covered by case or thematic roles are not reflected
unambiguously in the grammatical system. It is hard, for instance, to explain
phenomena like subject selection and middle formation in English using thematic
roles. The discussion of the Dutch impersonal passive showed that one of the
semantic primitives that is directly reflected by the grammatical system belongs to
the domain of aspect. This primitive is Event Structure. As such Event Structure is
a more likely candidate than thematic roles for mapping from the conceptual onto the
grammatical system.
Jan G. van Voorst
University of Ottawa
 THEMATIC ROLES 259

Références

CHOMSKY, Noam (1981) Lectures on Government and Binding, Hollande, Dordrecht

FILLMORE, C.J. (1968) «The case for Case» dans Bach, E. and R.T. Harms (red), Universals
in Linguistic Theory, New York, pp. 1-91.
GRUNAU, J.M. (1985) «Towards a Systematic Theory of the Semantic Role Inventory*,
Proceedings of the Chicago Linguistic Society; EÛfort et al. (red.), Chicago.
JACKENDOFF, Ray (1978) «Grammar as Evidence for Conceptual Structure» dans Linguistic
Theory and Psychological Reality; Halle, M. J. Bresnan, G A. Miller (red.); Cambridge,
MTT, pp.201-229.
JACKENDOFF, Ray (1983) Semantics and Cognition, Cambridge, MTT.
JAKOBSON, R. (1936) «Beitrag zur allgemeinen Kasuslehre: Gesamtbedeutungen der
russischen Kasus» dans Jakobson, R. Selected Writings II, Word and Language;
Mouton, the Hague; 1971, pp. 23-72.
O'GRADY, W.D. (1980) «The Derived Intransitive Construction in English», Lingua 52, pp.
57-72.
OOSTEN, J. van (1977) «Subject and Agenthood in English» dans Proceedings of the XHIth
Chicago Linguistic Society Meeting, Chicago, pp. 459-471.
PERLMUTTER, D.M. (1978) «Impersonal Passive and the Unaccusative Hypothesis»,
Berkeley Linguistic Society 4, pp. 157-189.
STYAN, E.M. (1984) «Theta-Roles in the Lexicon: Linguistic Evidence» dans McGill Working
Papers in Linguistics, Montréal, pp.118-47.
VENDLER, Z. (1967) Linguistics in Philosophy, Cornell Press, Ithaca.
WAUGH, L.R. (1976) Roman Jakobson's Science ofLanguage, Hollande, Lisse.