Depósito Legal: B-48039-2000
188.8.131.52. Semantic Role Lists
Semantic role lists constitute one of the most common
and simplest forms of lexical semantic representation adopted. They are
also referred to as "case frames" (Fillmore 1968) or "theta-grids"(Stowell
1981). They have acquired a prominent role in linguistic theory thanks
to the pioneering work of Gruber (1965), Fillmore’s (1968, 1977) seminal
papers, and Jackendoff’s early work (1972, 1976). Nevertheless, they have
a long history going back to Panini’s karakas (cf. Cardona 1974,
cited in Dowty 1991).
Semantic roles (also known as thematic roles or theta
roles) attempt to capture similarities and differences in verb meaning
that are reflected in argument expression, with emergent generalizations
that will contribute to the mapping from semantics to syntax. They belong,
then, to the semantics/syntax interface. Some of the characteristics that
theories of thematic roles strive for, in order to fulfill their function
are the following, according to Dowty:
i. Completeness: Every argument of every verb
is assigned some thematic role or other.
ii. Uniqueness: Every argument of every verb is assigned only one
iii. Distinctness: Every argument of every verb is distinguished
from the other arguments by the role it is assigned. Two levels can be
distinguished: strong distinctness, if Uniqueness also holds, and
distinctness, if it does not. In this last case, each argument is assigned
a different set of roles from other arguments of the same verb.
iv. Independence: Each role is given a consistent semantic definition
that applies to all verbs and all situations. Thus, role definitions do
not depend on the meaning of the particular verb or on the other thematic
roles it assigns.
For instance, (ii) and (iii) are the ones that give
GB’s theta-criterion its identity. However, it is doubtful whether they
can be maintained. Uniqueness does not seem to hold with animate
subjects of verbs of motion in sentences such as the following:
(5.13) John ran into the house
Gruber (1965) and Jackendoff (1972) claimed that John
both agent, since it initiates and sustains the movement, and theme, since
it is the object that moves. It is also very difficult to maintain
if examples like the ones below are considered:
(5.14) John met with Mary.
(5.15) John resembles his mother.
(5.16) A is similar to B.
Both participants seem to be playing the same role in
The goal of semantic role theories is to obtain a set
of semantic roles that can apply to any argument of any verb. Their function
is to make possible the unique identification of the arguments of the verb
so that the mapping to syntax can be carried out. Dowty calls this the
argument-indexing function of thematic roles. There have been a large number
of proposals with regard to the number and nature of the list of semantic
roles needed. Below are two lists from recent Syntax textbooks:
(5.17) Thematic Roles
Agent: The ‘doer’ or instigator of the action denoted by the predicate.
Patient: The ‘undergoer’ of the action or event denoted by the predicate.
Theme: The entity that is moved by the action or event denoted by the
Experiencer: The living entity that experiences the action or event
denoted by the predicate.
Goal: The location or entity in the direction of which something moves.
Benefactive: The entity that benefits from the action or event denoted
by the predicate.
Source: The location or entity from which something moves
Instrument: The medium by which the action or event denoted by the
predicate is carried out.
Locative: The specification of the place where the action or event
denoted by the predicate in situated. (Aarts 1997: 88)
(5.18) Theme (or Patient): Entity undergoing the effect
of some action.
(Mary fell over)
Agent/Causer: Instigator of some action.
(John killed Harry)
Experiencer: Entity experiencing some psychological state.
(John felt happy)
Recipient/Possessor: Entity receiving/ possessing some entity.
(John got Mary a present)
Goal: Entity towards which something moves.
(John went home) (Radford 1997: 326)
It should be noticed that there is no agreement about
which and how many roles are needed. This is precisely one of the major
drawbacks of the semantic role list approach. Although Croft (1991: 156)
has pointed out that most theories of thematic roles assume that there
is only a small finite number of them, no consensus has been reached on
the number or nature of the roles. Proposals range from just a few to hundreds
of them. As two extreme cases, consider Anderson’s (1971) localist approach
with just three semantic roles (Source, Location and Goal) from which all
non-local values would derive, contrasted with the view in HPSG where each
verb would assign its own peculiar semantic roles, different from the semantic
roles of any other verb (cf. Pollard and Sag 1994). So, the verb love,
for example, would assign two semantic roles: ‘lover’ and ‘lovee’. In this
case, there would be no thematic role types but individual thematic roles,
using Dowty’s (1989) terms. But with them, the semantic generalizations
that make the notion interesting are lost.
A related problem with semantic role lists is what Dowty
(1991:553) calls "role fragmentation", that is, how finely thematic roles
should be divided. He illustrates it with Agent, which authors like Jackendoff
(1983) divide into Agent and Actor, others such as van Valin (1990), into
Agent and Effector, Cruse (1973), into volitive, effective, initiative,
and agentive, and Lakoff (1977) proposing up to fourteen different characteristics.
Another role that can be divided is Theme, as Dowty (1991) does in distinguishing
between Incremental, Non-incremental and Holistic Themes.
Another problem is how and where to establish the boundary
between role types. Roles that are problematic in this respect are Instrumental/Theme,
as in the sentences below:
(5.19) a. Load the truck with these rocks.
Load these rocks onto the truck.
Also, Instrument vs. Comitative are similarly problematic:
(5.20) a. John cut the meat with a knife.
b. John burgled the house with an accomplice.
c. John won the appeal with a highly-paid lawyer.
In this sense, with can introduce a wide range
of roles, such as Theme, Instrument and Comitative, making it very difficult
to identify them.
A further handicap is the problem posed by arguments
of Measure or Extent. Dowty illustrates this with the following set of
(5.21) a. I walked a mile.
I swam 30 meters.
I slept twelve hours.
b. This weighs five pounds.
The piano measures 6’5’’.
It took me an hour to grade
The book cost me $5.
c. I paid $5 (this amount) (?this $5-bill) for the book.
The book cost me $5 (?this amount) (#this $5-bill).
I bought the book for $5 (this amount) (#this $5-bill)
d. I paid for the book with ?$5 (#this amount) (this $5-bill).
I bought the book with ? $5 (#this amount) (this
e. I’ll trade this record for the book.
The Measure phrases in the sentences in (a) are usually
recognized as adjuncts rather than as subcategorized elements. Fillmore
(1968) considered that adjuncts can also be assigned theta roles, but the
problem, as Dowty points out, is where to stop. Thus, adjuncts are not
generally thought of as receiving a theta role from the verb. By contrast,
the Measure or Extent phrases in the (b) examples are required by the verbs.
They should be assigned a thematic role. But then the drawback is how to
justify a different treatment for both sets of sentences when they are
very similar semantically. The problem with (c), (d) and (e) is how to
distinguish Extent from Theme. This $5-bill refers to a concrete
object and should not be assigned an Extent role, but rather some other
role like Theme, in parallel with (e). Carol Tenny (1995a) discusses this
same issue for motion verbs like walk and makes a distinction between
referential total-path NPs and distance NPs:
(5.22) a. Laura walked the Long Trail/ the course/ the
b. Mary walked the length of the field/ the entire distance.
Whereas the (b) examples incorporate Extent or Measure
phrases, the (a) examples illustrate a concrete traversed path.
Another problem for most thematic role theories is the
lack of any internal organization. Since they usually take the form of
unstructured lists, they are unable to capture the commonalities and differences
across roles. A Patient, for example has much more in common with the Goal
role than with Agent. Also, as Croft (1991: 157) points out, there are
languages in which goals and benefactives share the same morphological
case. There seems to be a structure organizing the semantic roles, but
the typical inventory with a small set of unanalyzable roles fails to account
for this. Jackendoff’s (1983) discussion of verbs of motion is given by
Levin and Rappaport Hovav (1996) as further evidence of the need for internal
organization within the set of semantic roles. As the following sentences
show, a motion verb like come can appear with different complements:
(5.23) a. Pat came to the library.
b. Pat came from the cafeteria.
c. Pat came from the cafeteria to the library.
d. Pat came towards us.
e.Pat came through the woods.
In a typical semantic role list theory with Goal, Source,
Path, and perhaps even Medium as theta roles, nothing would unify these
uses of come. But, as has been seen throughout this whole dissertation,
and as Jackendoff himself points out, the verb come appears with
a path, with many different types of paths and components of paths. Thus,
unstructured lists of semantic roles are inadequate.
Dowty (1991) offers one of the approaches to semantic
roles that is most influential nowadays. It is based on two basic prototypical
role types that Dowty calls Proto-Agent and Proto-Patient characterized
by a list of entailments (1991: 572):
(5.24) Contributing properties for the Agent Proto-Role:
a. volitional involvement in the event or state
b. sent[i]ence (and/or perception)
c. causing an event or change of state in another participant
d. movement (relative to the position of another participant)
e. exists independently of the event named by the verb)
(5.25) Contributing properties for the Patient Proto-Role:
a. undergoes change of state
c. causally affected by another participant
d. stationary relative to movement of another participant
e. does not exist independently of the event, or not at all)
Arguments of the verbs will be more agent-like or patient-like
according to the number of Agent or Patient Proto-role properties they
fulfill. The argument with the largest number of Agent-role properties
will be the subject, and the other the object. In case they fulfill the
same number of Proto-agent entailments, either one can be the subject.
This explains, for example, the varying nature of psychological predicates,
as illustrated by the contrast between Spanish gustar and English
And within English, the theory accounts for the difference between what
Croft (1991: 214) calls "experiencer-subject" verbs (like, admire, detest,
fear, despise, enjoy, hate, honor, love, esteem) and "experiencer-object"
verbs (please, scare, frighten, amuse, bore, astonish, surprise, terrify,
thrill). The two arguments of these verbs, the Experiencer and the
Stimulus, have the same number of Proto-Agent entailments. This is why
some of the verbs require as subject the experiencer, and others, the stimulus
that causes the psychological state. It also explains diachronic variability.
Verbs such as like and think, for instance, had the stimulus
as subject in earlier stages of the history of English.
The advantage of Dowty’s proposal is that thematic roles
are not viewed as discrete categories, but as prototypical concepts, in
line with Wittgenstein’s family resemblances and the prototypes of Rosch
and her followers (cf. Wittgenstein 1953, Rosch and Mervis 1975). This
overcomes the problems of role fragmentation and boundaries discussed above.
Nevertheless, as Levin and Rappaport Hovav (1996) point out, the main weakness
of Dowty’s account is the origin and exact nature of the lexical entailments
relevant for proto-role assignment.
capítulo 5 I
Depósito Legal: B-48039-2000