Scott DeLancey

In the strictest traditional sense of the term case refers to the kind of inflectional marking of noun phrases found in German or Latin; it has often been used more broadly in linguistics to refer to any morphological indication of grammatical role, so that we can refer to both German and Japanese as case-marking languages.

Since case is not a universal morphological category, it was not a major topic of research in general linguistic theory during the structuralist and early generative eras, and did not play a prominent role in more modern linguistics until it was reintroduced into linguistic theory by the work of Gruber (1965), Fillmore (1966, 1968), Chafe (1970), and Anderson (1968, 1971). The thread common to the work of these theorists is the conception of a universal syntactic-semantic theory of case roles, of which the morphological case marking found in some languages is only one reflection. In this sense it is possible to talk of "case" in languages like English or even Thai, with impoverished or non-existent systems of morphological case marking. Since then case theory has occupied a rather unsettled place in linguistic theory. A considerable burst of enthusiasm for Case Grammar in the early 1970's faded as it became clear how little agreement existed on both the appropriate form of a theory and the methods for establishing one. While it is clear to most contemporary linguists that some theory of underlying case semantics is a necessary part of an adequate syntactic (not to mention semantic) theory, there is not yet widespread consensus on the appropriate form of such a theory.

A large part of the confusion and controversy in the study of case stems from basic lack of agreement on the scope of case theory and the appropriate methodology for investigating it. It is obvious that any theory of case must be responsible for explaining the case marking of the core arguments of a clause (in languages of the familiar European type the subject, object, and indirect object). But there is a range of opinions on whether case theory needs to provide an account of the semantics of any oblique roles‑‑i.e. what in European languages are expressed in prepositional phrases--and if so, which ones. Case theory has also been invoked as a partial explanation for various syntactic phenomena concerning reflexivization, control of zero anaphora, and other problems with little evident relation to questions of case marking (see for example, papers in Wilkins 1988).

My purpose in this lecture is to outline a theory of the universal basis of case theory. I will take the basic task of case theory as being to account in a coherent way for the surface case marking found on core arguments in languages--concentrating on a specific set of case-marking patterns found in languages around the world. I will show that a handful of innate principles rooted in the structure of perception and cognition determine what is universal about the underlying roles of the core arguments across languages. If we can succeed in this task, it is time enough then to debate what other linguistic phenomena may or may not be illuminated by our understanding of case theory.

On Case Grammar

Let us begin by establishing some fundamental common ground. In all languages, verbs have arguments; a verb and its arguments constitute a clause. It is possible, and in most languages easy, to identify a set of core arguments, or actants (Tesnière 1959). In English these are the NP's in a clause which are not marked by prepositions. There is structure among the arguments: each argument has a distinct syntactic relation to the verb. Each argument also has a distinct semantic role in the situation named by the verb. There is clearly some correlation between an argument's semantic and its syntactic role, but in most languages this correlation is sufficiently indirect that the semantic role cannot be simply read off from the syntactic relation. Determining the semantic role requires additional syntactic tests and/or reference to the semantics of the verb. That is to say, given a set of sentences like:

4) My dog broke/ate/has/needs an egg.

5) My dog likes eggs.

we cannot attribute any constant semantic role to the Subject relation, and identifying the actual semantic roles of the various subjects requires further information of some kind.

The essential problems of clause structure at this level are:

What semantic roles exist, and how should they be characterized?

What kinds of syntactic relations can an argument have to its predicate?

Are the syntactic relations determined by semantic roles? If not, how are they determined?

The last can be rephrased as:

Why are there syntactic relations at all?

That is--suppose we could demonstrate that there are, say, exactly x universal semantic roles which can occur as core arguments in a clause in human language. The most obvious language design would have x case markers, one for each underlying role; every argument would simply be marked for its semantic role, which could then be read directly off the surface morphosyntax. I will argue that we can, in fact, demonstrate that there are exactly 3 such universal roles, and that reasonable approximations of such a language do exist--but the fact remains that in a substantial majority of attested languages, semantic roles are recoverable only indirectly, through a level of syntactic relations which is clearly determined in part by some other factor than semantic roles.

One more fact is essential as a prelimary: the fundamental fact of valency, that a clause can have one, two, or three arguments--no more. There appear to be a few languages with no three-argument verbs, but there is no verb in any language with four or more core arguments. The importance of this is clear once we recognize that there are verbs which appear to have four semantic arguments; the standard examples are 'buy' and 'sell' (Jackendoff 1972, cf. Fillmore 1977:72-3). An event does not count as an example of 'buying' or 'selling' unless there are four participants: a seller, a buyer, the merchandise, and the price. Again, the obvious design solution would be for all four to surface as core arguments of the verb, with some surface morphosyntax devoted to indicating which was which. In fact, however, this does not and indeed cannot happen--English sell can have only three core arguments:

6) Some jerk just sold my bozo husband 23 acres of worthless Florida swamp (for 4 million dollars).

and buy only two:

7) I just bought 23 acres of prime Florida real estate (from DeLancey) (for 4 million dollars).

In fact, although the price is, as we have said, an essential part of the semantic content of the verbs buy and sell, there is no way in English to express it as a core argument of either verb. Thus we have one more question to keep in mind: Why are there only three core argument slots in any human language?

Early suggestions

A spate of interest in various versions of these questions in the late 1960's gave us several proposals for a theory of "case", i.e. semantic role (Gruber 1965, Fillmore 1966, 1968, Chafe 1970, Anderson 1971). As we would expect from groundbreaking work in a new field, all of these proposals contain a mix of compelling and important insights, provoking and interesting suggestions, and tentative steps down what turn out in hindsight to be false trails. Unfortunately there has been little systematic research done on case theory since that time--the seminal suggestions of Gruber, Fillmore, and Chafe have each been widely and uncritically adopted, and for the most part any revisions made to the proposals of one's favorite case theoretician tend to be pretty thinly motivated and ad hoc. Rather than repeat the standard unproductive drill of taking Fillmore's or Gruber's original, 30+-year old tentative proposals as a starting point, I will try, with the benefit of 30 years of hindsight, to build from the ground up a theory of semantic roles that will work.

The fundamental requirement for a theory of case is an inventory of underlying case roles. And a basic reason for the failure of case grammar has been the inability of different researchers to agree on such an inventory:

To establish a universal set of semantic roles is a formidable task. Although some roles are demarcated by case or by adpositions in some languages, in many instances they have to be isolated by semantic tests. There are no agreed criteria and there is certainly no consensus on the universal inventory. To a great extent establishing roles and ascribing particular arguments to roles involves an extra-linguistic classification of relationships between entities in the world. There tends to be agreement on salient manifestations of roles like agent, patient, source and instrument, but problems arise with the classification of relationships that fall between the salient ones. There are also problems with determining how fine the classification should be. (Blake 1994:67-8)

In this passage Blake puts his finger on several of the essential problems, but appears not to perceive the problematic nature of one of them. If our purpose is to explain linguistic structure and behavior, we are concerned only with those cognitive categories which are reflected in linguistic structure and behavior--which is what I mean when I say semantic. If there is no linguistic test for a category in any language, then it is not a linguistic category. So, no "classification of relationships between entities in the world" which is in fact "extra-linguistic", i.e. has no linguistic reflection, has any place in our investigations.

In much early work on case, the case roles are defined, in the manner of Fillmore 1968, by prose definition. Such definitional phrases as "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.

Typology and case

Let us begin with our first question: what semantic roles do we find as core arguments? The most obvious path to follow in elucidating this question is an inventory of the surface case distinctions among core arguments. At this juncture, certain case-marking patterns--in particular nominative and canonical ergative constructions--are of little use to us; the fact that such patterns obscure underlying semantic roles is the basis of the problem we are trying to solve. We will return to such non-semantic grammatical relations in subsequent lectures; for the present our interest is in case alternations with a reasonably clear semantic basis.

For example, many languages distinguish some "experiencer" from Agent subjects by case marking, and this is generally interpreted as establishing that these (typically) dative- or locative-marked arguments are not Agents, but have some other semantic role. Typically the case form is the same as that used for recipient arguments of a ditransitive verb:

8) kho-s blo=bzang-la deb cig sprad-song

he-ERG Lobsang-LOC book a give-PERF

'He gave Lobsang a book.'

9) blo=bzang-la deb de dgo=gi

Lobsang-LOC book DEM need-IMPF

'Lobsang needs the book.'

We often find this same case form used to mark the possessor in possessional clauses:

10) blo=bzang-la bod‑gyi deb mang=po 'dug

Lobsang-LOC Tibet-GEN book many have

'Lobsang has a lot of Tibetan books.'

Thus suggests a hypothesis which would group experiencers, recipients, and possessors in possessional clauses together as reflexes of the same underlying role.

We will return to the question of dative subject predicates soon, but I want to begin with a less well-known distinction, between two semantically different types of "object". An age-old problem of Tibetan grammar is that some transitive verbs require a case postposition on their non-Agent argument, while others forbid it:

11) thub=bstan‑gyis blo=bzang‑la gzhus‑song

Thubten-ERG Lobsang-LOC hit-PERF

'Thubten hit Lobsang.'

12) *thub=bstan‑gyis blo=bzang gzhus‑song

13) thub=bstan‑gyis blo=bzang(*-la) bsad‑pa red

Thubten-ERG Lobsang(*-LOC) kill-PERF

'Thubten killed Lobsang.'

This is not the familiar pattern of pragmatic object marking found, for example, in Romance and Indic languages, where the presence or absence of dative/locative marking on the object reflects its degree of inherent or discourse-based topicality (Comrie 1979, Genetti 1997, inter alia). In Tibetan, any given verb either requires -la-marking on its object, or forbids it; nothing about the object NP itself has any effect on case marking.

The traditional explanation (as explained to me by Tibetans who learned it in school) is that there is a difference in the relation of the argument in question to verbs of these two classes. Some, like gzhus 'hit', are construed as describing conveyance of something to the object. Others, like gsod 'kill', describe the object as undergoing a change of state. That is, the traditional explanation is, fairly literally, that gzhus 'hit'-type verbs have the core argument structure (AGENT, THEME, LOC), where the THEME may be inherent in the verb itself (as it is in (11, 48)), and gsod 'kill'-type verbs have the actant structure (AGENT, THEME).

As far as I know, explicit case marking of this distinction, as in Tibetan, is not common across languages. But it is found as a covert category in other languages, such as English, where it was discovered by Fillmore (1970), who neatly identifies the underlying semantic distinction--as I will show later in this lecture, the object of a "change-of-state" verb like break has some sort of patient or undergoer role, which we will call Theme, while the object of "surface contact" verbs like hit is a Locative.

My aim in this and the next lecture is to present a theory of case roles which explains this sort of phenomenon--one which explains such cross-linguistically widespread phenomena as the fact that the first argument of (9) is in a different surface case than the first argument of (8), and in the same case as the second argument of (9), and both of these are in the same case as the first argument of (10). We also want to explain patterns like that illustrated in (11-13); although as far as I know that case marking pattern is not particularly widespread, the distinction which it marks is covertly present, and syntactically diagnosable, in English and other languages.

The grammar of THEME and LOC

We will discuss the Agent category in the next lecture. In this lecture I want to argue that all other core argument roles are instantions of two underlying relations, THEME and LOCATION, which correspond very directly to the perceptual constructs FIGURE and GROUND. Theme and Loc are adopted from the work of Gruber (1976, cf. Jackendoff 1972, 1983), and I will follow the Gruber-Jackendoff tradition of referring to them as thematic relations. Neither Theme nor Loc can be defined independently; they define one another--the Theme is that argument which is predicated as being located or moving with respect to the Location, which is that argument with respect to which the Theme is predicated as being located or moving. The most concrete and transparent instantiation of the Theme and Loc roles is in a simple locational clause:

14) The money's in the drawer.

THEME LOC

In the rest of this section I will show that all non-Agent core argument roles can be reduced to Theme and Location.

The Semantic Structure of Ditransitive Clauses

Ditransitive verbs offer the most direct insight into underlying case. While the case roles associated with intransitive subject, or with each of the two argument positions of a bivalent predicate, may have at least two possible roles associated with it, the semantics of the three arguments of a trivalent verb are invariant, across different verbs and different languages. The nominative or ergative argument is always Agent, the accusative or absolutive argument is Theme, and the recipient is Loc.

The nuclear (in the sense of Dixon 1971) verb of this class is 'give', which in its most concrete sense involves actual movement of an object from the physical position of one individual to another:

15) Just give me that gun.

And this is literally true for many other ditransitive clauses:

16) She handed me the book.

In English, as in many other languages, the ditransitive construction alternates with a nearly-synonymous construction syntactically identical to that which expresses caused motion:

17) He left his papers to the library.

18) He sent the kids to the library.

19) He put the dishes in the sink.

Though there are subtle semantic differences between the "dative-shifted" and prepositional constructions with give-verbs (Goldsmith 1980), the differences are primarily pragmatic (see e.g. Goldberg 1995:89-95), which is to say that the semantic structure of the trivalent construction is exactly parallel to that of clearly spatial predications like (18-19).

In other ditransitive clauses, there is no actual movement of a physical object. In one category, what changes is not the physical location of something, but its socially (e.g. legally) defined ownership:

20) My grandfather left me his farm.

21) Fred gave me his seat.

Still, the semantic relations are the same here as in (15-16). There is no reason to think otherwise, since no language will mark a different set of surface case relations in these and in (20-21). And the metaphorical extension from physical location to social ownership is both intuitively natural and robustly attested.

The Semantic Structure of Possessional Clauses

This interpretation neatly unites the semantic interpretation of possessives and ditransitives, which (as has long been noted, cf. Lyons 1967, 1969) can be easily interpreted as causatives of possessive constructions:

22) He gave me a wrench.

AG LOC THEME

23) I have a wrench

LOC THEME

As Lyons puts it:

It is clearly not by chance that the case of the indirect object (the 'dative') and the directional of 'motion towards' fall together in many languages. In the 'concrete' situations in which the child first learns his language, it would seem that the causative ... Give me the book is indeterminate as between possessive and locative ('Make me have the book' and 'Make the book come to/be at me'): note that Give it here is frequently used in such situations and is eqivalent to Give it to me. The distinction of locatives and posessives would be a subsequent language-specific development, resting largely upon the syntactic recognition of a distinction between animate and inanimate nouns in various languages. Indeed, is there any other way of saying what is meant by 'possessive'? (1967:392)

The confusion which Lyons is referring to concerns the distinction between actual physical possession and socially- or legally-defined ownership, which can hardly be an innate category. Clearly physical possession is the primitive concept here, and clearly, as Lyons notes, it is closely related to the concept of spatial location.

The strongest evidence for this interpretation of possession comes from languages in which possessional clauses are identical, or nearly so, to simple locational clauses, as in Tibetan:

24) blo=bzang bod-la 'dug

Lobsang Tibet-LOC exist

'Lobsang's in Tibet.'

25) blo=bzang-la ngul-tsam 'dug

Lobsang-LOC money-some exist

'Lobsang has some money.'

Note that the location in (24) and the possessor in (25) have the same locative -la postposition, while the possessum in (25), like the Theme argument in (24), is in the unmarked absolutive case. The only formal difference between the two constructions is in the order of the two arguments. Exactly the same argument structure occurs in existential clauses:

26) bod-la g.yag mang-po 'dug

Tibet-LOC yak many exist

'There are a lot of yaks in Tibet.'

Of course, this function can also be carried out by a possessional construction in English:

27) My kitchen table has ants all over it.

This is a very widespread pattern; a particularly interesting example is the Mixean language Olutec, where the two constructions differ only in the presence or absence of the inverse marker ‑ü‑ (cf. Lecture 7), which elsewhere in the language marks a transitive verb in which the object rather than the subject argument is more topical (Zavala 2000):

28) ?it-pa-k pixtü?k xu?ni-jem

exist-IMPF-ANIM fleas dog-LOC

'There are fleas on the dog.'

29) ?it-ü-pa-k pixtü?k xu?ni

exist-INV-IMPF-ANIM fleas dog

'The dog has fleas.'

In English, we do find the locational metaphor for possession in certain constructions:

30) You got any money on you?

But this construction retains the sense of literal location, i.e. if I have money in the bank, or at home, it is not on me; this construction can only refer to physical possession. In languages in which the locational construction is the basic possessional construction, it has grammaticalized to the point where any such semantic tie to physical location is lost.

Such data suggest that possessive and existential/locational constructions have the same underlying structure, and differ only in the relative salience, inherent or contextually-determined, of the two arguments. The interpretation of possessum and possessor as Theme and Loc (under one set of terms or another) is an old idea whose introduction into contemporary linguistic thought owes much to Allen (1964) and especially Lyons (1967, 1969); it is a fundamental part of the localist case theories of Gruber, Anderson, and Diehl. As Jackendoff puts it:

Being alienably possessed plays the role of location; that is, "y has/possesses x" is the conceptual parallel to spatial "x is at y." (1983:192)

Gruber, and Jackendoff following him, argue for this interpretation primarily on the basis of gross parallels in the organization of the syntactic and lexical expression of the two concepts in English, but the strongest evidence for it is the large number of languages in which it is grammatically explicit. (This is also not a new observation; Benveniste (1960/1971:170) notes the cross-linguistic prevalence of "the 'mihi est' type over the "habeo" type" of possessional construction (cf. Lyons 1967, 1969:392)).

"Experiencers" as locatives

Besides abstract motion, the Theme in a ditransitive clause may also be quite abstract:

31) That gives me an idea!

32) Fred rents me space in his garage.

33) He willed the Church his copyrights.

34) That'd sure give you the heebie-jeebies!

Again, there is no linguistic evidence whatever to analyze such sentences as having a different array of case roles from ditransitives with more concrete Themes and paths. By analogy with examples like (15), the heebie-jeebies in (34) is Theme and you is Loc. By the same logic that we used in the last section to argue that recipients and possessors have the same underlying role, we can equate the recipient object of (31) and the subject of (35):

35) I have an idea.

As far as any linguistic facts which might be adduced are concerned, both of these arguments are Locatives, and an idea in both sentences is Theme.

This suggests, intuitively, that the subject of (36) might likewise be a Locative:

36) I'm thinking of an idea.

In English there is no direct grammatical evidence, at least of the straightforward sort that we are looking for, to support this analysis--subject formation in English obscures any such differences in underlying case role. And we require linguistic evidence; we are not entitled to assign case roles purely on intuition. But, as is well-known, there is ample cross-linguistic evidence for exactly this analysis. The syntactic and semantic problems posed by "dative subject" or, in the more contemporary locution, "experiencer subject" constructions have been and continue to be the focus of considerable research by syntacticians of all persuasions (see e.g. Verma and Mohanan 1990, Pesetsky 1995, Filip 1996, inter alia), as well as psychological research into the cognitive basis for the distinction (see Brown and Fish XXX).

As with the analysis of possessional constructions, so here the strongest evidence is the wide range of languages in which experiencer subjects--i.e. the experiencer arguments of some verbs of cognition and emotional state--are case-marked in the same way as recipients. Again, we can illustrate this with Tibetan, where the experiencer argument of certain verbs like 'need/want', 'dream', etc., is marked as Locative:

37) khong‑la snyu=gu cig dgo=gi

he-LOC pen a want-IMPF

He needs/wants a pen.

In many languages, the lexical encoding of situations of this type may be even more explicit in identifying the experiencer as a location, as in Newari (a Tibeto-Burman language of Nepal):

38) j sw~-ya-gu bas khaya

I.ERG flower-GEN-CLS smell(N) took

'I smelled the flower.' (Agentive)

39) ji-ta sw~-ya-gu bas wala

I-DAT flower-GEN-CLS smell(N) came

'I smelled the flower.' (non-agentive)

(lit. 'The smell of the flower came to me.')

And parallel evidence can be found even in languages with no distinct dative subject construction; cf. English sentences like:

40) A while ago a crazy dream came to me.

Locative and Theme objects

In a seminal paper (which has not received the attention that it merits) Fillmore (1970) elegantly demonstrates that not all English direct objects have the same underlying role. The object of a "change-of-state" verb like break has an undergoer type role which Fillmore calls "object"; this is our Theme. The object of what Fillmore calls "surface contact" verbs--generally, verbs of affectionate or hostile physical contact like hit, hug, kick, kiss--is some sort of locative.

Fillmore notes several syntactic differences between these two types of transitive clause. Change-of-state verbs have passives which are ambiguous between a state and an event reading:

41) The window was broken (by some kids playing ball).

42) The window was broken (so we froze all night).

Other transitive verbs have only eventive passives. Many change-of-state verbs characterized by the "ergative" alternation, i.e. they occur transitively with the Theme argument as object, and intransitively with the Theme as subject:

43) The window broke.

44) I broke the window.

Surface-contact verbs are characterized in English by a peculiar use of a locative prepositional phrase which is unique to this class of verbs. With any other kind of clause an oblique locative can only denote the place where the overall event occurred:

45) I broke the glass in the sink.

(The reading in which the PP belongs to the object NP is irrelevant here). With hit-class verbs, however, an oblique locative can be added which specifies more precisely the part of the object toward which the action is directed:

46) I kissed her on the lips.

Another piece of evidence which can be added to Fillmore's case is that this class of verbs in English is uniquely eligible for a productive light verb construction with the verb stem used as a noun and give used as the verb:

47) I gave her a kiss.

As we have already seen, the recipient argument of a trivalent verb is underlyingly a Locative. Thus her in (47) is transparently a Locative argument, lending indirect semantic support to Fillmore's suggestion that it is likewise in (46).

Fillmore's evidence for this distinction is linguistic, and thus legitimate, but by itself does not meet the criterion which I want to insist on of considering only distinctions reflected in some language in case marking distinctions. But, once again, Tibetan provides exactly that evidence. We have already seen exactly that, in exx. (11-13), repeated here:

48) thub=bstan‑gyis blo=bzang‑la gzhus‑song

Thubten-ERG Lobsang-LOC hit-PERF

'Thubten hit Lobsang.'

49) *thub=bstan‑gyis blo=bzang gzhus‑song

50) thub=bstan‑gyis blo=bzang(*-la) bsad‑pa red

Thubten-ERG Lobsang(*-LOC) kill-PERF

'Thubten killed Lobsang.'

Verbs which require a Locative-marked argument are the likes of 'hit', 'kiss', 'kick', etc.--that is, Fillmore's surface-contact category. And verbs which take an absolutive argument are verbs like 'break', 'boil', 'kill', etc., corresponding quite neatly to Fillmore's change-of-state category. So, in Tibetan, the arguments which Fillmore analyzes as "objects" have the zero case-marking which in Tibetan marks Themes, and the arguments which he shows are locatives are case-marked as Locatives.

The Syntax and Semantics of Theme and Loc

This analysis of transitive verbs and their argument structure raises both syntactic and semantic issues. To take the most evident semantic question first, we need to provide some semantic support for the case role idenfications which we are making. For many theorists, the object argument of a verb like break (and, for some, the object of hit as well) belongs to a distinct category, Patient (a role which has no place in the system I am expounding here). Calling these arguments Themes entails identifying objects of transitive verbs, or subjects of intransitives, which describe them as undergoing a change of state, with the corresponding arguments of verbs like 'send' or 'go' which predicate concrete motion or location of their arguments.

The necessary conceptual basis for this analysis is the localist interpretation of existence in and change of state which has been argued for by a number of scholars (see Chafe 1970, Anderson 1971, Jackendoff 1972, 1983, 1990, Diehl 1975, Lakoff and Johnson 1980, Lakoff 1987, 1993, etc.), in which states are seen as abstract or metaphorical locations which entities occupy and move in and out of. Thus an entity which is described as in a state is a Theme just as is one described as being in a location. Likewise a change of state is metaphorically motion from one condition to another, and the entity conceptualized as thus changing its metaphorical location is a Theme just as much as when it is conceptualized as moving through space. Note that the ordinary ways of talking about such questions are unambiguously localist: one can be in a state, fly into a rage, get out of a depressed state, change or turn into something, etc. Even terms dealing with states and changes of state which are not transparently localist (e.g. become) are more often than not etymologically so. As one more illustration of the semantic plausibility of this analysis, consider one more ditransitive example, this time with a concrete Theme argument but a very abstract Loc:

51) Your "logic" is gonna drive me crazy.

Once more, we have no evidence to argue that the thematic relations here are any different from those of any other distransitive, which means that me is Theme, and crazy is Loc.

Indeed, the native Tibetan grammatical tradition has, since its beginning, noted the fact that with some transitive verbs the non-ergative argument is unmarked, while with others it is marked as Locative. The classic pair of examples used in grammars and schools is:

52) shing‑la sta=re gzhus‑pa

tree-LOC axe hit

hit the tree with an axe

53) sta=re‑s shing 'chad‑pa

axe-INSTR tree cut

cut down the tree with an axe

The native analysis of the pattern is the same as ours. In (52) the axe moves so as to come in contact with the tree, and these arguments are case marked exactly as in any other clause depicting an object moving to a location. In (53), on the other hand, the verb does not explicitly refer to the movment of the axe or its contact with the tree, but rather to the change of state of the tree, which is brought about through the medium of the axe.

The syntactic issue requires a more significant departure from traditional conceptions of case grammar. In introducing the categories Theme and Locative, I claimed that they are mutually dependent--an argument can only be a Theme relative to Location, or a Location relative to a Theme. But then both hit- and break-type clauses have a missing argument. If the glass in the glass broke is a Theme argument, where is the Loc? And if her in I kissed her is Loc, where is the Theme?

The semantic justification which I have just presented for this analysis identifies the intransitive subject or transitive object of a change-of-state verb as a Theme because it is changing state, metaphorically moving from one state to another. Then the Location with respect to which it is a Theme is the state to which it is moving. This state is, in fact, named by the verb--is, in fact, the essential part of the meaning of the verb. Thus we must recognize the possibility that one of the two fundamental thematic relations may be lexicalized in the verb. So the definition of a change-of-state verb is one which lexicalizes a state, which represents a thematic Location, and takes a separate Theme argument.

And the problem of surface-contact verbs is solved in the same way. If her in I kissed her is Location, then the Theme can only be the kiss, which is lexicalized in the verb. This is clearly the correct analysis of the give-paraphrase:

54) I gave her a kiss.

where her and a kiss are transparently Loc and Theme. This interpretation receives further support from Tibetan, where the number of ordinarily transitive surface-contact verbs like gzhus 'hit' is very small. The equivalents of the vast majority of English 'hit' verbs, e.g. punch, kick, hug, kiss, etc., in Tibetan are light verb constructions, consisting of a semantically almost empty verb stem and a noun carrying the specific semantics of the predicate, e.g. so rgyab 'bite' (='tooth throw'), kha skyal 'kiss' (='mouth deliver'):

55) thub=bstan‑gyis blo=bzang‑la kha bskyal‑song

Thubten-ERG Lobsang-LOC mouth delivered-PERF

'Thubten kissed Lobsang.'

56) thub=bstan‑gyis blo=bzang‑la mur=rdzog gzhus‑song

Thubten-ERG Lobsang-LOC fist(ABS) hit-PERF

'Thubten punched Lobsang.'

In this construction the Theme, lexicalized in the verb in English, is given its syntactic independence as an argument, and the motivation for the locative marking is clear.

This interpretation of underlying semantic relations differs from many conceptions of case theory, where case roles are, by definition, the roles played by NP arguments. In such a framework, an intransitive clause, since it has only one argument, has only one case role represented. In my analysis, every clause, in every language, has a Theme-Loc structure--both relations are present in every proposition. One of the two may be lexicalized in the verb, as is the case with intransitive and ordinary transitive predicates. Or both may be nominal arguments as with ditransitive, possessional, and "experiencer subject" clauses. This is sufficiently different from the usual use of the term case role that it is probably better to use a different term; I will refer instead to thematic relations.

The Theoretical Importance of Thematic Relations

Defining Relations in Terms of Event Structure

We can develop the same hypothesis which I have just presented by a different route, beginning with a simple and traditional ontology of states and events, where stative predicates describe an argument as being in a state, and event is defined as a change of state or location, with 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. Talmy's (1978) seminal paper introducing the psychological categories Figure and Ground, and similar discussion in Jackendoff 1990, Langacker 1991).

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

57) The stove is next to the refrigerator.

58) 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 (57), it must be so also in (58), 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 by 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

59) 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 (59) 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. (I will discuss the nature of agentivity in more detail in the next lecture). 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:

60) 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. Recall our observation of the typological fact that languages universally have no more than three core arguments. This universal constraint falls directly out from the theory--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.

Theme, Loc, and Innateness

I have not, of course, provided sufficient syntactic and typological evidence here to establish the superiority of this account of case marking in core argument positions over other possibilities. But, assuming for the sake of argument that this superiority can be established (note, among other things, that many of the most useful insights and analyses of Relational Grammar fall out fairly directly from the scheme presented here), how should we explain the universality of this model of clause structure? If it is true that every clause, in every language, can be analyzed as representing a Theme-Loc configuration, why should this be? If it is truly universal, there is good warrant to consider the possibility that it reflects innate structure, but, given the warnings expressed at the beginning of this paper, how should such a hypothesis be pursued?

It turns out that this theory looks very much like the fundamental structural construct of perception--Figure and Ground:

One of the simplest and most basic of the perceptual processes involves what the Gestalt psychologists call figure-ground segregation. Every meaningful perceptual experience seems to require in its description the property of "figuredness." That is, phenomenally, perception is more than a collection of unrelated, unintegrated, sensory elements. The units of perception are, rather, figures, or things, segregated from their backgrounds. (Dember 1963:145-6)

In their concrete spatial use, Theme and Loc correspond directly to Figure and Ground. Nothing is intrinsically Theme or Loc; these are relational notions. A speaker presents one referent in relation to another; the first we call Theme, and the second Loc. Thus, despite some argument to the contrary in early literature on Case Grammar (see Huddleston 1970), (61) and (62) are by no means synonymous:

61) The bank is next to the Post Office.

62) The Post Office is next to the bank.

(61) describes the location of the bank, using the Post Office as a reference point; (62) describes the location of the Post Office, using the bank as a reference point. Thus the subject of each sentence denotes the referent to which the speaker wishes to draw the addressee's attention, and the oblique NP denotes a referent used as a background against which the subject can be identified.

Now, figure-ground organization is, self-evidently, not a feature of the physical universe; rather, it is a pattern imposed on a stimulus by the process of perception. Much work in perception has been concerned with what we might think of as prewired determinants of figure-ground identification. All other things being equal (e.g. in a properly designed experimental context), humans will make a moving stimulus a figure, and the stable environment against which it moves the ground. Other factors which increase the eligibility of some part of the visual percept for figure status include defined boundaries, brightness, color, centrality in the visual field, and, of course, lack of competition from other areas of the perceptual field sharing these characteristics.

But in ordinary life other things are not often equal; any perceiver in any real-life circumstance is predisposed by her existing cognitive structures, and long-term and transient "interests", to focus on certain types of structure as opposed to others. A universal pattern, which is probably innate, is that a percept interpretable as a human figure has a higher eligibility for figurehood than anything else, and a human face the highest of all. There is abundant evidence for what is sometimes called a motivation effect in perception, i.e. the fact that a perceiver, being more interested in some types of information than others, will tend to organize the perceptual field so that relevant information counts as figure.

As any introduction to perceptual psychology will point out, beyond the simple neurophysiology of edge detection, color perception, etc., perception is a cognitive process. In fact, it is common in perceptual psychology to distinguish between sensation and perception--the former applying to the simple physiological response of the perceptual organs, and the latter to the cognitively-constructed interpretation of those data.

Thus perception cannot be considered in isolation from cognition. But the reverse is also true; cognition at the most basic level involves mental manipulation of representations of objects (or, at the next higher level, categories of objects), and the discrimination of objects is the basic task of perception. Indeed, the figure-ground opposition is fundamental to--we could even say, is--object discrimination. The process of discerning an object is the process of perceiving it as figure.

It thus makes eminent sense that the evolution of cognition should work from preadapted perceptual structure, and that the opposition of figure and ground should be carried over from its origins in the perceptual system to higher-order cognitive structures which evolved to process, store, and manipulate information obtained from the perceptual system. If these higher-order structures then were the preadaptative ground on which grew the language faculty, there would be no surprise in seeing the same basic structural principle retained.

Indeed, if we think of language functionally, in the most basic sense, it is almost inevitable that fundamental aspects of its structure should mirror the structure of perception. The same philosophical tradition which gives us the peculiar conception of intelligence as information-processing, inclines us to imagine that what is passed from one mind to another in the course of communication is some sort of pure information. It is, of course, no such thing. In its communicative function, language is a set of tools with which we attempt to guide another mind to create within itself a mental representation which approximates one which we have. In the simplest case, where we are attempting to communicate some perceived reality, the goal is to help the addressee to construct a representation of the same sort that he would have if he had directly perceived what we are trying to describe (cf. DeLancey 1987). Clearly all of the necessary circuits and connections will be much simpler if that input, which is thus in a very real sense an artificial percept, is organized in the same way as an actual percept. This involves many other aspects which are also conspicuous in linguistic structure--deixis, to take one striking example--but must, fundamentally, involve figure-ground organization, since that is fundamental to perception.

Thus the hypothesis that Figure-Ground structure might inform the basic structure of syntax has exactly the sort of biological plausibility that any innatist hypothesis must have. We can identify the preexisting structure from which it might have evolved, and construct a scenario by which it might have evolved from that preexisting structure. The availability of such a story does not, of course, by itself establish the correctness of either the evolutionary scenario or the linguistic hypothesis itself. This or any other account of case roles and clause structure must established on the basis of valid induction from linguistic facts. But the fact that there is a readily-available, biologically plausible account of how such an innate linguistic structure might come to be gives this hypothesis a kind of legitimacy lacking in many contemporary proposals about the nature of "Universal Grammar".

It is possible in some languages to derive causatives of trivalent verbs, producing a clause with four arguments. I will argue, though, along with most other syntacticians, that we in such cases we must recognize the fourth argument as actually introduced in a distinct clause represented by the causative derivation.

Buy can have three arguments, as in:

I bought Jerry his own toad.

But the third argument is an added benefactive, not one of the underlying semantic arguments of buy. We will return to problems like dative shift, benefactive advancement, and related issues of applicatives later.

It can be a core argument of pay:

1) Poor Fred paid some shark 4 mill (for a bunch of Florida mud).

But pay is not as tightly tied to the commercial transaction frame as are buy and sell; there are many other things for which one can pay besides merchandise that is bought and sold.

The case marker, la (-r after vowel-final monosyllables), is the locative (and allative) marker, and also marks dative arguments (recipients, possessors, experiencer subjects).