Event Construal and Case Role Assignment

                         Scott DeLancey
                      University of Oregon

     While something analogous to, and often a lineal descendant
of, Gruber's and/or Fillmore's original conception of semantic
roles plays a major part in most current approaches to syntax, in
few outside of the cognitive movement do they map directly to
both a cognitive and a syntactic level of representation.  Most
work retreats from full commitment to the original generative
conception of case roles, abandoning the notion of a direct
relationship between them and either semantic or the surface
syntactic representation.  Those working within frameworks in
which there is fairly direct mapping from case roles to surface
structure (e.g. the theta roles of GB theory or the initial GR's
of Relational Grammar) tend to be sceptical as to whether these
roles can be given independent semantic characterization.  On the
other hand, much of the work done in the direct "Case Grammar"
tradition has taken the semantic content of the case roles as
fixed, but has tended to be lax about constraining many-to-many
mappings between these roles and surface syntactic categories.
     A fundamental problem which has prevented the development of
a widely-accepted generative case grammar is the lack of a
principled basis for determining the semantic content of case
roles, and thus for identifying the roles played particular
arguments of clauses.  The root of this problem is the
commitment--usually unarticulated, but sometimes explicit--of
most linguists to an objectivist conception of semantics in
general, and of the semantics of case in particular.  There is a
growing literature (Lakoff 1977, 1987, Hopper and Thompson 1980,
DeLancey 1984, 1985a, b, 1987, 1990a, b, Schlesinger 1989)
showing that the criterial approach to the definition of case
categories which is part of this naive objectivism fails, and
that concepts like Agent or transitive clause have the same sort
of complex and sometimes fuzzy structure as other conceptual
categories.
     The central problem for linguistic semantics is to describe
the relation between the wide and subtle variation found in the
world of experience and the smaller set of categories available
in a language.  My papers cited above for the most part approach
the problem from the world of experience end, and discuss how
various subtly different types of event are sorted into the
categories encoded by different morphosyntactic constructions. 
In this paper I want to approach the problem from the other side,
and present a minimalist account of the semantics of what I will
claim is a set of core case roles, which can be formulated so as
to prevent the objectivist error which lies at the root of much
incorrect analysis.


The objectivist error

     For as long as I have taught I have taught a course in
English grammar, targeted at graduate students and advanced
undergraduates in linguistics and other fields.  The course
includes a brief discussion of case roles and how they can be
thought of as mediating between semantics and syntax.  The first
or second time I taught the course, I gave an exam which included
a strikingly original question in which the students were given
sentences with selected NP's underlined and asked to identify the
case roles of those NP's.  One of the sentences was:

     1)   John threw the ball through the window.

The answers which I intended were John = Agent, the
ball = Patient, and the window = Locative.   But when
the exams came in, nearly half the class had labelled the
ball as Instrument and the window as Patient.  I
marked them all wrong, and wondered how I'd ended up with such a
class full of dodoes.  When I handed the exams back to the
students, there was weeping and wailing and gnashing of teeth.  I
patiently explained that in this case the ball was what underwent
the change of state, and was therefore the Patient, while the
window in this sentence is simply a description of the path, and
thus a Locative.  "But", replied the students, "we were
visualizing the scene with the window closed!"
     Unfortunately, they were right--in terms of everything I had
told them about case, the window in that scenario does indeed
undergo a change of state, and thus qualifies as a Patient, and
the ball is then the Instrument.  But on the other hand, they
were wrong--the ball simply is not an Instrument in that
sentence, nor is the window a Patient.  So, what I had
taught them had led, by perfectly legitimate reasoning, to
conclusions which are clearly wrong.  Therefore there must have
been something wrong, or at least seriously incomplete, about
what I had taught them.
     While this is a more evidently flawed misanalysis than is
commonly found in the linguistic literature, it represents a
fundamental error which also underlies many arguments about the
theory and practice of Case Grammar which do appear in the
literature.  This is the notion that to say that case roles are
semantic means that it should be possible to read them
algorithmically off of an objective description of an incident. 
In this paper I will argue that this misconception is one of the
major obstacles in the development of a generative case grammar,
and offer a few examples of what can be accomplished once it is
eliminated.
     The generative conception of Case Grammar is regarded in
many quarters as discredited on grounds of the apparent
difficulty of developing a limited set of semantically-definable
categories in terms of which morphosyntactic facts can be
systematically explained.  I will argue that many of the
difficulties which have impeded the development of a useful yet
constrained theory of Case Grammar stem from the same
misconception of the semantics of case roles which led my
students to their error--the assumption that case role
assignments for a sentence can be read directly from an
unconstrained description of what an episode would have to be
like for the sentence to refer to it.
     I will also argue for one particular approach to Case
Grammar, in which a core set of case roles is defined with
respect to a small set of cognitive state and event schemas, on
the grounds that it both provides for a principled limitation on
the set of universal case roles, and automatically prevents this
semantic error.  Not much of the outline of localist Case Grammar
which I will present here is entirely new, though parts of it are
likely to seem unfamiliar to some.  My purpose in this paper is
not to defend this particular model, but to show that this kind
of model, in which case roles are defined in terms of a
constrained theory of event structure, automatically prevents the
objectivist error which has been the source of many flawed
analyses in the past.


Event-based characterization of case roles

     The standard approach to describing a case role is, in the
manner of Fillmore 1968, by prose definition.  Such definitional
phrases as Fillmore's "perceived instigator of the action" and
"force or object causally involved in the action or state" imply
a theory of actions and states, but the necessary theory has not
always been perceived as a crucial component of a generative Case
Grammar.  This is in part to blame for the inability of linguists
to agree on a set of case roles.  Such prose definitions have no
automatic constraints; anything can be loaded into them.
     A better approach is to define at least a set of core case
roles strictly in terms of a small set of state and event
schemas.  This seems to be becoming a popular idea (see e.g.
Jackendoff 1990), but was not an explicit part of much work in
Case Grammar until relatively recently.  (Croft (1991) traces the
approach back to Talmy 1976, though something like the idea is
implied in Halliday 1967-8).  If roles are defined strictly in
terms of state and event schemas extra semantic detail is forced
back into the verb, where it belongs.
     We begin with a simple and traditional ontology of states
and events, where event is defined as a change of state or
location.  I will assume a stringently localist hypothesis that
every clause describes a literally or metaphorically locative
relation between a Theme and a Location (Anderson 1971, Diehl
1975).  Note that, as well-established in localist case theory
and the study of lexical metaphor (e.g. Diehl 1975, Lakoff and
Johnson 1980, Langacker 1986), states as well as physical and
temporal locations are Locations.  We can then distinguish states
from events, which describe the Theme coming to be at Loc.  We
neither need nor want to provide any more definition of Theme or
Loc than this; the AT relation which defines its two arguments is
taken to be primitive.  (Cp. the similar discussion in Jackendoff
1990, Langacker 1990ms, and Talmy's (1978) promising approach to
interpreting these in terms of the psychological categories
Figure and Ground).
     Compare this with a prose definition for Theme such as "the
object in motion or being located".  One problem with this
particular definition is its wide applicability, since, at least
once we have identified states as locations, everything that can
be talked about is in motion or is located.  This objection may
seem at first blush like a parody of an objectivist approach to
case semantics, since there is presumably an assumed proviso
along the lines of "which is construed in a given clause as" in
any such definition.  But in fact there are still linguists who
are unable to get this straight.  Huddleston's (1970) familiar
worry about sentences like

     2)   The stove is next to the refrigerator.

     3)   The refrigerator is next to the stove.

exemplifies exactly this error.  The apparent problem with these
sentences can be expressed in our terms as the appearance that
they each have two Themes.  The argument is that since the
stove is clearly a Theme in (2), it must be so also in (3),
and vice versa; so that in each of the sentences each of the two
NP's is Theme.  The same line of argument can then prove that
each is also Loc.
     The correct analysis of these data is given in Gruber
(1965).  Each clause predicates the location of one entity, and
defines the location by a landmark.  The simple fact that we can
infer information about the location of the landmark from the
sentence does not make it a Theme.  Of course each referent is in
a location--just like everything else in the universe--but each
sentence is about the location of only one referent.  (Note that
the same erroneous argument can apply to any locational
predication; after all, if I say that My shoes are on my
feet, does this mean that the NP my feet has the Theme
case role, since its referent must be in my shoes?)
     Unfortunately this error is still alive and well, and can be
found for example in Dowty's worry about:

     the case of predicates that do not have any apparent
     difference at all in their entailments with respect to two
     of their arguments, hence offer no semantic basis for
     assigning distinct role-types to these arguments ...
     (1989:107)

i.e. sentences like

     4)   Mary is as tall as John.

Although Huddleston has historical priority in bringing up the
problem, he is explicitly vague on the issue of exactly which of
the then current assumptions about Case Grammar is the primary
impediment to a more satisfactory analysis.  Dowty is quite
explicit in a footnote criticizing Talmy's analysis:

     such pairs are not distinguished by any objective feature of
     the situation described but at best by the "point of view"
     from which it is described. (1989, fn. 14, p. 123)

In other words, case roles are "in the world" (cp. Ladusaw and
Dowty 1988), and are to be read off from the world, not from some
construal of it.
     Some form of this error lies at the base of most of the
problems in the development of Case Grammar.  Dowty and Ladusaw
are indeed correct in their supposition that given this approach
to semantics, a case grammar constrained enough to be interesting
is probably impossible.  (I disagree with them about which of
these must therefore be abandoned).  The objectivist error is
automatically avoided when we define the notions Theme and Loc
strictly in terms of the AT relation; given that (4) must have
the underlying semantic structure Theme AT Loc, there can be no
question about the correct assignment of roles.
     Events are changes of state (or of location); rather than
being depicted as at a state/location the Theme is depicted as
coming to be there.  Events in this sense can be categorized into
simple changes of state and more complex configurations which
include an external cause of this change.  We will take this as
the definition of Agent.  Our grammar so far consists of states
and simple and complex events, or statives, inchoatives, and
causatives (Croft 1991).  We can define three fundamental case
roles, Theme, Location, and Agent, in terms of this simple
grammar of states and events:(1)

     5)   Theme AT Loc

          Theme GOTO Loc

          Agent CAUSE Theme GOTO Loc

The essential point of this approach is that case roles are
defined and assigned in terms of tightly-constrained event
schemas, rather than being assigned with reference to the larger
more amorphous scenarios found in the lexical semantics of verbs.
     In the generally Andersonian model developed here only these
three core cases are defined.  (SOURCE, which appears in the
glosses to a few of the examples below, appears to be an oblique
rather than a core case role, although this remains to be
conclusively argued).  We will take seriously the evidence of
languages with "coverb" serial verb constructions that
Instruments, Benefactives, etc., which in languages like English
are encoded with oblique case forms, can in fact be defined as
playing one of the core roles, but in a distinct event from that
which assigns core case roles in the clause.  The function of
oblique case roles is to allow reference within a clause to
actors from the scenario that do not have one of the core roles
in the event schema which underlies the clause.


The Localist model and traditional Case Grammar

     This model differs from traditional conceptions of Case
Grammar(2) in a number of ways, of varying relevance to my
general argument.  Perhaps most striking is that elements of
surface structure other than NP's--in particular, predicates--are
associated with underlying case roles.  This has to be
accomplished in one way or another in any truly localist Case
Grammar (e.g. Gruber 1965, Anderson 1973, Diehl 1975), but it is
irrelevant to my main argument, which can be established equally
as well in a non-localist model.  (Such a model will require a
richer inventory of underlying state/event schemas; the localist
hypothesis makes it easy to establish a principled constraint on
this inventory).
     Since in this model a verb can have only an underlying State
or Event schema, and the most complex Event schema has only three
arguments, a verb can assign only three core case roles.  Again
this makes my main argument easier, but is not required for it,
as long as the case inventory is read directly off of a
constrained inventory of state/event schemas. (Note that the set
of surface case frames is larger than three because of different
patterns of argument incorporation).  The important aspect of the
model for my present argument is that only those case roles which
are automatically defined by the basic schemas are part of the
inventory of core case roles.  Directly stemming from this is the
abandonment of the notion that all case roles are of equal
status.  There is an explicit distinction here between the core
case roles, defined by the event schemas, and the less-
constrained set of oblique case roles.
     The last difference between this model and the traditional
approach is that all elements of a schema must be referred to in
a clause which encodes that schema.  This point is essential to
my argument at various points, and we will see that it is also
essential to a semantically and syntactically responsible Case
Grammar.


Agents

     The characterization of Theme in this approach is probably
not controversial, as the Theme role has never been given the
sort of elaborate and highly specific definition that notions
such as Dative and Agent have.  The characterization of Agent in
these terms is more contentious.  Most work that makes any
reference to case roles has insisted on some version or other of
an a priori definition of Agent which includes bulky semantic
baggage--in particular animacy and volition--for which there is
no obvious place in this simple event schema.  While there is an
idealized model of event causation, with explicit reflections in
morphosyntactic structure, which incorporates these
parameters,(3) it is clear that if we intend case roles to play
an explanatory role in morphosyntax, we must recognize a broader
and simpler conception of Agent, which is simply the causal
argument in the expanded event schema.(4)
     A simple definition of Agent as first cause (cp. Croft's
(1991) "autonomous cause") eliminates the notorious problems of
natural forces, and of similar problematic data such as these:


     6)   His attitude infuriates me.

     7)   This mess really bothers him.

     8)   The beauty of this vista has inspired many artists.

     9)   The look on her face would curdle milk.


As far as I know no substantial body of grammatical evidence has
ever surfaced that the subjects of these clauses have a different
case role from those of indisputably Agentive examples, nor is
there any satisfactory account of what their case role should be
if they are not Agents.  Thus we have no justification for
supposing that they are anything but Agents, which is just what
our event model requires.


Event construal and case roles

     A properly-constrained Case Grammar will prevent the version
of the objectivist error which my students made.  The problem is
that the sentence which I gave them could be imagined as
referring to a scenario in which both the ball and the window
underwent changes of state, and therefore in terms of an
argument-centered definition, are Themes (or Patients, as I
taught them at the time).  The intuitively obvious explanation of
why the students' answer is wrong is that the sentence is "about"
what happened to the ball, not what happened to the window, in
exactly the same way as (2) is about the location of the stove,
not that of the refrigerator.  This is explicitly represented in
our event grammar by the fact that throw, like any other
verb, can have only one Theme.  Throw is about the change
of location of the thrown object.  One could also construct a
sentence about the change of state of the window, but this would
need a verb appropriate to that kind of change of state, such as
break or smash.  The window is Theme with
respect to its own change of state, but is not a Theme with
respect to the change of state of the ball, and thus is not a
Theme (or a Patient) in the example.
     My students' error is unexpected because the predicates that
could be chosen to describe the ball event and the window event
in this scenario are so different.  In parallel examples,
however, where predicates have as part of their meaning a
scenario such as this which involves more than one event, even we
linguists seem to get confused.  The students' fundamental error
is one which, sometimes but not always in subtler forms,
underlies a great many arguments concerning the analysis of
putatively problematic data in Case Grammar.  That is the
assumption that anything and everything that can be said about a
situation is part of the semantic representation of any sentence
which describes that situation.
     Consider the traditional hay/wagon, etc., examples:

     10)  He loaded the wagon (with hay).

     11)  He loaded the hay on the wagon.

A standard approach to a Case Grammar analysis of these data
starts from the assumption that because the wagon is Loc
(or Goal) in (11), it must have that role in (10).  But this is
the objectivist error again.  Taking a pile of hay and moving it
all into a previously empty wagon involves two events in our
sense--the hay changes its location, from being on the ground to
being in the wagon, and the wagon changes its state, from being
empty to having hay in it.  Under the first construal the
hay is Theme, under the second the wagon is.  Only one
of these construals at a time can be encoded in a clause.  In
each case the mapping of case roles to surface relations is
direct:  the Agent is subject, and the Theme object.
     An important distinction to maintain here is between what I
am referring to as the "scenario" which the verb describes, and
the constrained notion of "event" which I have defined.  The
loading scenario involves two events, but only one event at a
time can be encoded in a clause.  More evidence for such a
constrained version of Case Grammar is found in the syntax of
transactional predicates such as  buy and sell.  It
has been widely observed in several different contexts that these
lexical items have in some sense four semantic arguments: two
individuals, the merchandise, and the payment.  But neither
English nor any other language has a clause type with four core
arguments.  What English does with these scenarios is to encode
them as two- and three-argument clauses.
     It has been argued (Dowty 1989:106, Jackendoff 1990:59-60),
that we must analyze single arguments of clauses with these verbs
as having multiple case roles.(5)  As in the case of the
hay/wagon examples, the assumption underlying this analysis is
that in the sentences

     12)  Esau sold Jacob his birthright.

     13)  Jacob bought Esau's birthright (from him).

the case roles are constant, that is, Jacob in both is
Goal (in our grammar it will be Loc), Esau in the first
example and him in the second both Source,(6) with
birthright and the mess of pottage--which note can only be
included as an oblique in either clause--both Themes.  But
nothing whatever in the syntax of either example bears out this
analysis.  Ditransitive predicates are easily described in terms
of our simple schemata; they can be described as complex events
in which the verb lexicalizes neither the Theme nor the Location,
so that both are left as arguments of the verb.  Thus a simple-
minded assignment of case roles to these examples--one based only
on the syntax, with no reference to the meanings of sell
and buy--would give us the following:

     14)  Esau sold Jacob his birthright.
          AGENT     LOC   THEME

     15)  Jacob bought Esau's birthright (from him).
          AGENT        THEME             (SOURCE)

     The more elaborate structure attributed to such clauses by
Jackendoff and others is based on the intuition that since Jacob
and Esau play the same role in the transaction in both clauses,
Jacob must be Loc (or Dative, or whatever) in (14) as in
(15), and Esau Source in (15) as in (14).  Thus it is worth
pointing out that the subject of sell is not necessarily a
Source in any sense, nor that of buy a Loc (or Goal):

     16)  My realtor sold my house.

     17)  My agent bought some property for me.

Note that brokers, agents, and suchlike are not normally encoded
as Source; one buys through, not from, a realtor. 
All that is necessary to qualify as subject of either verb is
Agenthood, i.e. being the cause (as far as the clause reports) of
the change of (possessional) location on the part of the Theme.
     On the objectivist view of case roles it is surprising that
these verbs do not manage to carry their full set of apparent
arguments with them.  In our grammar, on the other hand, they
could not do so.  If the scenario which the verbs describe
involves change in location of two different entities, then it
involves two distinct events.  Only one can be reflected in the
clause structure.  Both buy and sell refer to the
event of which the merchandise is Theme, and the payment has no
direct role in this event.  A different verb, e.g. pay,
must be used to construct a clause describing the event of which
the payment is Theme.


Staging and event construal

     Essentially the same error, seeking case role assignments in
some imagined "real world", lies at the root of another set of
errors which have bedeviled Case Grammar since the beginning.  A
prime example is the identification of a spurious class of
Instrument subjects in English and some other languages, which
leads inevitably to debilitating and ultimately fatal weakness in
the grammar.  DeLancey (1984) and Schlesinger (1989) present
arguments against any notion that the inanimate subjects of
transitive verbs which have generally been so analyzed are
anything but Agents.
     There is a range of types of data involved here.  We have
already discussed the category of inanimate forces such as
lightning, which have long been identified as Instruments.  The
source of this error, as already mentioned, is the packing of the
case roles with irrelevant semantic detail.  An event-based Case
Grammar could not permit such an analysis:  the complex event
schema has a slot for an Agent, and there must be an Agent in a
clause encoding that schema.  Obviously in events in which forces
like lightning and wind are the causal agents, no other possible
Agent is conceptually or even imaginably present.  Schlesinger
points out that the same argument is applicable to clauses like
the subordinate clause of (18):

     18)  The clock was ticking so loudly that it woke the baby.

Although a clock, as an artifact, seems intrinsically much more
Instrument-like than natural forces, nevertheless it is hard to
come up with a principled way of finding another Agent for this
clause than the clock.
     Another subcategory is the more controversial class of data
represented by exx. such as these:

     19)  The janitor opened the lock with a key.

     20)  The key opened the lock.

     21)  The assassin's poison killed its victim.

     22)  The axe made a satisfying "chunk" when it hit the wood.

     23)  When the first stone hit a policeman, it provoked a
          violent reaction.

In these, as in the previous examples, there is an NP with the
morphosyntax appropriate to an Agent, which nevertheless most
linguists would want to analyze as an Instrument.  Like the
supposed Instruments in the previous examples, these share with
true Instruments the irrelevant factor of inanimacy.  But unlike
lightning and ticking clocks, these represent entities which our
real-world knowledge tells us could not have played the role they
are described in the clause as playing without initially being
manipulated by someone.  That is, these cannot easily be
conceptualized as autonomous causes, occupying the initial node
in a causal chain.
     Thus the source of the analysis of (20), for example, as
having Instrument subject is the same objectivist approach to
case semantics that we have been examining.  The underlying
argument is that since the key in (19) is an Instrument,
and since (19) and (20) could refer to the same scenario, the
key must be Instrument in (20) as well.  An event-based Case
Grammar imposes a very different analysis.  The requirement that
a clause reflect in its structure one of the state or event
schemas means that all elements of the schema must be present in
the clause.  Nothing in this approach to event semantics gives us
warrant to go wandering through the world which the clause
suggests to our imagination looking for arguments.  (20), since
it is a perfectly good transitive clause, must encode the complex
event schema, with an Agent that causes the change of state of
the lock.  That Agent can only be the key; there is no
other candidate in the clause.
     Schlesinger points out that many such examples manifest
semantic behavior more consistent with this analysis than with
the traditional one.  For example, in spite of the widely
asserted hospitality of English to supposed instrumental subject
constructions, a sentence like (24) is odd:

     24)  This pencil draws lines.

In contrast, (25) is perfectly ordinary:

     25)  This pencil draws very thin lines.

This reflects the fact that while there is nothing odd about
depicting a particular pencil as playing the causal role in
effecting thin lines (as opposed to just any lines), any
pencil presumably can draw lines, and it is thus hard to conceive
of a particular pencil as playing any unusual causal role with
respect to line-drawing in general.(7)  Similarly, outside of the
popular folklore of linguists, sentences like (20) are not freely
usable in English.  Speakers presented with this sentence in
isolation generally have a clear intuition that the key is
being given some contrastive force--that the sentence evokes a
context in which a particular key, rather than any other, or the
key rather than some other means, was essential to the successful
opening of the door.
     In general, clauses like (20) and (25) are acceptable in
English precisely to the extent that there is a plausible
construal of the described scenario in which the supposed
"Instrument" can be seen as playing an essential causal role in
the complex event schema.  But, by our constrained definition, to
be an Agent is to be identified as the causal argument in a
complex event schema--so in such clauses as these the subjects
are Agents, not Instruments.
     But it is not clear that any such interpretation will
explain the syntax of exx. (21-23).  Let us stipulate that in the
most natural context for these examples the inclusion in the
clause of reference to the inanimate proximal cause, but not the
animate ultimate cause whose existence is inferable, is a matter
of "staging", rather than of attribution to the inanimate actor
of sufficient causal force to qualify it as an Agent.  This is a
phenomenon that is widely attributed to "pragmatics" (e.g. in
Fillmore 1977), and thus considered to be non-semantic and
irrelevant to the study of case roles per se.
     Again, the intuition that the assassin's poison in
(21), for example, must be analyzed as an Instrument depends on
the objectivist analysis which says that we know from the
sentence that the poison was administered by an assassin.  Then
the poison, considering its role in the overall scenario, clearly
best fits the definition of Instrument, while the assassin best
fits the definition of Agent.  But the assassin in fact
has no case role in the clause (note that it is more natural in
this clause to refer to its than to his victim). In
(23) the putative Agent is not mentioned anywhere in the clause,
sentence, or any necessary context, and indeed it is difficult to
cram one into the sentence.
     The error of the Instrument subject analysis here is again
the error of trying to incorporate into the semantic
representation of the clause inferences which, however legitimate
they may be, are not in fact part of the event representation of
the clause.  (21) describes a complex event, of which poison is
represented as the cause.  As soon as we open the door to
admitting into the semantic representation of a clause entities
with no grammatical role in the clause (as in (21)), and even
some not actually available within the discourse (as in (23)),
our grammar will inevitably have the potential to allow any
number of clearly undesirable analyses.  What, for example, would
prevent us from analyzing a sentence like

     26)  He finally talked under relentless interrogation.

as having a higher Agent--the interrogator(s)--as part of its
semantic representation, and then analyzing he as playing
a causee role of some sort?
     In effect, the Instrument subject analysis of a sentence
like (21) depends upon the claim that the sentence is synonymous
with one like:

     27)  The assassin killed his victim with poison.

But this is simply not true (even leaving aside the interesting
semantic effects that we get by introducing into both sentences
other material, e.g. adverbials like slowly).  In order to
explain the syntactic reflexes of a case role analysis of a
clause such as (21), we must see the "staging" phenomenon as one
more example of event construal.  The "real world", as we
perceive it, is a complex web of causal relations.  Any
transitive clause presents one single cause-effect relation. 
There are no case roles, as a conceptual or a linguistic
phenomenon, until a complex event representation has been
constructed.  Whatever is included in the causal slot of that
representation is the Agent, for that is what an Agent is.


On Case Grammar

     In the original conception of case roles in the various
versions of case grammar developed in the sixties (those of
Gruber, Fillmore, and John Anderson) they were to mediate between
semantic and syntactic representations.  Some level of semantic
representation from which case roles are recoverable is thus
generative, and these approaches would be as legitimately called
"generative semantics" as the logic-based approach which bore
that name.  The arguments criticized in this paper are retreats
from a strong theory of Case Grammar, in which there is a direct
relation between underlying semantic case roles (whatever those
be) and surface morphosyntax.  The data represent variations in
syntactic behavior in what are claimed to be the same underlying
case.  These and other data are more easily understood under a
strong version of Case Grammar, once we understand that case
roles, like any other semantic categories, encode construals of
events rather than objective facts.


Notes

(1)  This classification of predicate types has a long and broad
history; my thinking here most directly reflects the lexical
decomposition approach of Generative Semantics and the Vendlerian
approach developed by Dowty (1979) and Foley and Van Valin
(1984).  For present purposes differences in formalization and
terminology between this and other proposals along the same lines
are more expository than substantive.  For example, I use GOTO
instead of the BECOME function often used here (e.g. in Dowty
1979) simply to call attention to the fact that this schema
represents both literal spatial motion and metaphorically
motional change of state.

(2)  I use the vague reference to "traditional" conceptions
advisedly throughout the paper; though most linguists who discuss
case explicitly link their work to the foundational frameworks of
Fillmore (1968, 1977) and/or Gruber (1976), there has grown up a
traditional account of the semantics and syntax of case roles,
such that we find papers such as Rosen 1984 and Holisky 1987
making crucial use of demonstrably incorrect conceptions of the
semantics of particular case roles without explicit reference to
any work in Case Grammar (see DeLancey 1985a).

(3)  DeLancey 1981, 1984, 1987, 1990a, b.

(4)  DeLancey 1984, 1985, 1990a, b.  In DeLancey 1990a I show how
volition can be represented in more complex event schemata.

(5)  This is a different argument from Gruber's (1965) more
plausible suggestion that a single argument of a clause may have
two case roles, if and only if one of them is Agent.

(6)  The place of the Source relation in the localist schema that
I have outlined here is not entirely clear, but for the present I
will assume on the basis of its syntactic behavior that it is an
oblique rather than a core case role.

(7)  Cp. the argumentation with respect to a different phenomenon
in Lakoff 1977.References

Anderson, John M. 1971. The grammar of case:  towards a
     localistic theory.  Cambridge:  C.U.P.
     . 1973.  An essay concerning aspect.  The Hague:  Mouton.
Croft, William. 1991.  Syntactic categories and grammatical
     relations:  The cognitive organization of information. 
     Chicago:  University of Chicago Press.
DeLancey, Scott. 1981. An interpretation of split ergativity and
     related patterns. Language 57.3:626-57.
     . 1984.  Notes on agentivity and causation.  Studies in
     Language 8.2:181-213.
     . 1985a. Agentivity in syntax. CLS Parasession on causatives
     and agentivity, pp. 1-12.
     . 1985b.  On active typology and the nature of agentivity.
     F. Plank, ed., Relational Typology.  Mouton.
     . 1987. Transitivity in cognitive grammar. R. Tomlin, ed.,
     Discourse Relations and Cognitive Units, pp. 53-68.
     Benjamins.
     . 1990a. Ergativity and the cognitive model of event
     structure in Lhasa Tibetan.  Cognitive Linguistics 1.3:289-
     321.
     . 1990b. Cross-linguistic evidence for the structure of the
     Agent prototype. Proc. of the 29th Child Language Research
     Forum.  Stanford.
Diehl, Lon. 1975.  Space case:  Some principles and their
     applications concerning linear order in natural languages. 
     Working Papers of the Summer Institute of Linguistics,
     University of North Dakota Session 19.93-150.
Dowty, David. 1979.  Word meaning and Montague Grammar. 
     Dordrecht:  Reidel.
     .  1989. On the semantic content of the notion of 'thematic
     role'.  pp. 69-129 in G. Chierchia, et. al., eds.,
     Properties, types, and meaning, vol. II:  Semantic issues. 
     Dordrecht:  Kluwer.
Fillmore, Charles. 1968.  The case for case.  pp. 1-90 in E. Bach
     and R. Harms, eds., Universals in linguistic theory.  New
     York:  Holt, Rinehart.
     .  1977. The case for case reopened.  pp. 59-82 in P. Cole
     and J. Sadock, eds., Syntax and Semantics 8:  Grammatical
     relations.  New York:  Academic Press.
Foley, William, and Robert Van Valin. 1984. Functional syntax and
     universal grammar.  New York:  Cambridge University Press.
Gruber, Jeffrey. 1965. Studies in lexical relations.  Ph.D.
     dissertation, M.I.T.  circulated by the Indiana University
     Linguistics Club (1970), and included in Gruber 1976.
     . 1976.  Lexical structures in syntax and semantics. 
     Amsterdam:  North Holland.
Halliday, M.A.K. 1967-8.  Notes on transitivity and theme.  J.
     Linguistics 3:37-81, 199-244, 4:179-215.
Holisky, Dee Ann. 1987. The case of the intransitive subject in
     Tsova-tush (Batsbi).  Lingua 71:103-132.
Hopper, Paul, and Sandra A. Thompson.  1980.  Transitivity in
     grammar and discourse.  lg. 56:251-99.
Huddleston, Rodney. 1970. Some remarks on case grammar. 
     Linguistic Inquiry 1:501-11.
Jackendoff, Ray. 1990.  Semantic structures.  Cambridge:  MIT
     Press.
Ladusaw, William, and David Dowty. 1988. Toward a nongrammatical
     account of thematic roles.  pp. 62-73 in W. Wilkins, ed.,
     Syntax and semantics 21:  Thematic relations.  New York: 
     Academic Press.
Lakoff, George.  1977. Linguistic gestalts.  CLS 13:236-87.
     . 1987.  Women, fire, and dangerous things.  Chicago: 
     Chicago University Press.
     , and Mark Johnson. 1980.  Metaphors we live by.  Chicago: 
     University of Chicago Press.
Langacker, Ronald. 1986.  Abstract motion.  BLS 12:455-71.
     .  1990ms.  Clause structure.  (Part II of Foundations of
     cognitive grammar, volume II:  Descriptive application). 
     Ms., University of California, San Diego.
Rosen, Carol. 1984. The interface between semantic roles and
     initial grammatical relations.  pp. 38-77 in D. Perlmutter
     and C. Rosen, eds., Studies in Relational Grammar 2. 
     Chicago:  University of Chicago Press.
Schlesinger, I. M. 1989.  Instruments as agents:  on the nature
     of semantic relations.  J. Linguistics 25:189-210.
Talmy, Leonard. 1976.  Semantic causative types.  pp. 43-116 in
     M. Shibatani, ed., Syntax and semantics 6:  The grammar of
     causative constructions.  New York: Academic Press.
     . 1978.  Figure and ground in complex sentences. pp. 625-49
     in J. Greenberg, et. al., eds., Universals of human
     language, vol. 4: Syntax.  Stanford:  Stanford UP.