Verbal Case Frames in English and Tibetan

                         Scott DeLancey
                      University of Oregon


The idea that the syntactic properties of particular verbs and
verb classes is a reflection of their semantic structure, once an
unpopular notion limited to the cognitive fringe of linguistics,
has lately become widely fashionable, and is currently generating
a substantial amount of research in many theoretical frameworks. 
In this paper I start from one of the earliest suggestions along
these lines in the early work of Fillmore.  In the course of the
investigation I hope to expand and improve upon some of
Fillmore's early suggestions, contributing both to our
understanding of one area of the verbal lexicon and to the
development and validation of a particular version of localist
case grammar.  More generally, this paper constitutes a
demonstration that the semantic structures in question must be
understood as construals of particular event types, rather than
direct mappings of some objective reality.
     My argument starts with Fillmore's (1970) identification of
two categories of transitive verb in English, the "surface-
contact" or hit-verbs and the "change-of-state" or
break-verbs.  Fillmore shows that these two classes have
distinct syntactic behaviors, and makes a convincing case that
these distinct behaviors reflect distinct underlying semantic
patterns.  This early paper contains some statements which seem
to assume that these semantic patterns are somehow inherent in
the nature of the event type being lexicalized.  Of course, in
Fillmore's later work we see a much more sophisticated view of
semantics, and in Fillmore 1975, 1977a, 1982 and other papers the
idea that a verb lexicalizes a perspective on an event rather
than some set of inherent "truth conditions" is developed in
great detail (with the implications of this for case grammar
clearly adumbrated in Fillmore 1977b).  But, while this insight
is well-established in Cognitive Grammar argumentation, many
linguists even in the mid-90's have yet to catch up with the
progress in semantics reflected in Fillmore's work of the mid-
and late 1970's.  I will show here that the rather simple level
of syntactic analysis used by Fillmore in 1970, applied to a few
cross-linguistic data, is more than sufficient to demonstrate the
necessity for a relativistic, construal-based understanding of
verbal semantics.
     Briefly, I will show that the same two classes which
Fillmore discussed for English can be identified in Lhasa
Tibetan,(1) but that, while the membership of the two classes is
to a large extent the same in both, there is a substantial body
of vocabulary where translationally equivalent predicates fall
into different classes in the two languages.  I will argue that
this is because, while for some event types there may be a
construal which is sufficiently salient to demand lexicalization
in any language, for others there may be alternate possible
construals, either of which may be lexicalized in any given
language.  (For that matter, as we will see, both may be
lexicalized in the same language).


The grammar of hitting and breaking

Fillmore's original insight is that among English transitive
verbs we can clearly identify two subcategories, distinguished by
syntactic behavior.  (Subsequent research of course has shown, as
Fillmore predicted, that there are many more and more subtly
organized categories than this; see Levin 1993).  Verbs like
break, bend, fold, shatter, and
crack differ conspicuously from other transitive verbs in
being "labile", i.e. functioning both as transitive verbs and as
intransitives with the erstwhile transitive object as subject. 
They are further distinguished by the fact that passives formed
from them are ambiguous between an eventive and a stative
reading:

     1a)  My arm was broken by the force of the blow.

      b)  My arm was broken so I found it hard to shave.

In contrast, verbs like hit, slap, strike,
bump, stroke, kiss, pat,
caress) are not labile; while some of them show (usually
rather marginal) uses in which they lack a direct object, the
subject in such a construction always plays the same role in the
event as the subject of the ordinary transitive construction:

     2)   The rock hit (the roof) with a bang.

Passives derived from these verbs are unambiguously eventive;
examples like (3) have no stative reading, even when one might
seem to be semantically plausible:

     3)   My face was slapped (*so everyone thought I was
          blushing).

Fillmore further notes that verbs of this second category, when
the object is a body part, allow an alternative construction in
which the possessor of the body part is the object, and the body
part occurs in a locative prepositional phrase:

     4)   She kissed my cheek. = She kissed me on the cheek.

     There is a fairly obvious semantic basis for these
differences.  Break-type verbs lexicalize a change of
state in an obligatory argument, which is the object if an Agent
is included in the clause and the subject otherwise.  This is why
their passives have the stative reading which others lack.  What
Fillmore calls "surface-contact" verbs, like hit, do not
entail any change of state.(2)  Instead, they lexicalize the
delivery of some force to a particular location.  Since they do
not lexicalize a change of state, there is no defined state for
their passives to refer to, and the passives are thus
unambiguously eventive.  And the fact that they lexicalize
delivery of a force to a location neatly explains their tendency
to allow this location to be encoded as a location rather than a
direct object.
     Surface-contact verbs have one additional characteristic,
not mentioned by Fillmore, which likewise supports this
characterization.  It is precisely this set of verbs which in
English show a peculiar paraphrase with give and the
erstwhile verb used as a noun:

     5)   She gave me a kiss/slap (on the cheek).

While the case analysis of these verbs can be determined from
their other syntactic behavior, as demonstrated by Fillmore, this
construction provides the most direct evidence for correct
analysis.
     While the case-grammar analysis which Fillmore gives of
these categories is essentially correct, as far as illuminating
their syntactic behavior is concerned, it is formulated within a
framework that I think is incorrect (DeLancey 1991); while I do
not stray too far from his understanding of the semantic basis
for these syntactic patterns, I will describe it in terms of a
version of localist case theory (Anderson 1971, 1977, Jackendoff
1972, 1983, Gruber 1976, Diehl 1975).  My version of this theory
recognizes only three case roles which can occur as core
arguments of a clause:  Agent, Theme, and Loc(ative) (which last
subsumes the role separately listed in some systems as Goal).  An
essential element of the theory is the interpretation of states
as metaphorical locations (cf. Jackendoff 1983, Lakoff and
Johnson 1980, Langacker 1987), so that at the level of abstract
case relations a change of state (e.g. from ALIVE to DEAD) is
interpreted as the movement of a Theme to a Loc, just as concrete
motion is.  In its simplest form (which is probably incorrect,
but adequate for our present purpose) the theory claims that
every verb can be interpreted as predicating a relation between a
Theme and a Loc (cf. Lakoff's ubiquitous SOURCE-PATH-GOAL schema
(1987)).
     In this framework change-of-state verbs predicate of a
Theme, which surfaces as the subject of the intransitive or
object of the transitive version, that it moves into a state,
which is lexicalized in the verb.  Surface-contact verbs, in
contrast, predicate of a Location that a Theme, which is
lexicalized in the verb, moves to it.  This interpretation is
most directly inferrable from the English give paraphrase;
a sentence like

     6)   She gave me a kiss.

presents a scenario in which a Theme, the kiss, and a Loc, the
recipient, are related just as are the Theme and Loc in a more
literal give sentence like

     7)   She gave me a doughnut.



Identifying hit and break verbs

Fillmore (1970:130) says that:

     Not every verb having syntactic properties of the kind
     represented by [break-type verbs] is a change-
     of-state verb:  there is a large class of motion verbs
     syntactically like break, including move,
     turn, rotate, spin, etc.

and

     Some change-of-state verbs are understood as affecting
     a place on an object rather than the object as a whole
     ... cut and bite, for example ... show
     paraphrase relations of the type seen ... with body
     part nouns.

Although he goes on to suggest that such apparent mismatches
between semantics and syntax may yield to subtler and more
sophisticated lexical analysis, there is nevertheless a dangerous
retreat here from the general program presented in that paper of
explaining syntax in terms of lexical semantics.  If we want to
regard syntactic behavior as motivated, rather than autonomous,
then our semantic analysis must be strictly accountable to
syntactic facts.  Thus, to whatever extent turn has the
same syntactic behavior as break, it must be because it is
semantically the same kind of verb.  If cut and
bite behave syntactically like surface-contact rather than
change-of-state verbs, in spite of our intuition that they belong
semantically to the change-of-state class, it can only be because
they are in fact surface-contact rather than change-of-state
verbs.  When there is conflict, we must abandon our intuitions
and remain true to the objective evidence, not the other way
around.
     So what is really going on with these verbs?  The example of
motion verbs behaving like change-of-state verbs is not
problematic; this is exactly what localist theory predicts. 
Considerable discussion over the last 20 years has established
the linguistic evidence for equating states with locations, so
that breaking something is linguistically (and, presumably,
conceptually) construed as "moving" it from one state to another
(cf. DeLancey 1991).  The sort of parallel syntactic behavior
that Fillmore notes is part of the abundant evidence for this
equation.
     Cut and bite are more interesting.  Fillmore
notes these as anomalous because, although semantically they do
bring about a change of state in the object, syntactically they
show the same "possessor raising" alternation as hit-
verbs:

     8)   Noseferatu bit my neck.

     9)   Noseferatu bit me on the neck.

In fact, these verbs meet every syntactic test for surface-
contact verbs.  They show the give paraphrase:

     10)  Noseferatu gave me a bite on the neck.

and have unambiguous passives.  And, while both of these verbs,
like break but unlike hit, have monovalent uses,
they do not show the "ergative" case-role alternation of change-
of-state verbs; i.e. in examples like (11-12) the subject is
still the Agent:(3)

     11)  That dog bites.

     12)  These scissors don't cut.

     These facts establish that these are surface-contact rather
than change-of-state verbs.  That is, regardless of what our
intuitions tell us about someone who has been cut or bitten being
an affected patient, the grammar of English seems rather to
construe events of this kind as involving the delivery of
something to a recipient.  The "something", of course, is the cut
or bite, just as with other hit-type verbs it is the blow,
caress, or whatever.  And notice that all of our natural ways of
speaking in English about cutting and biting are consistent with
this construal.  We say

     13)  I have a cut/bite (on my neck).

rather than

     14)  ?? I am bitten/cut.

     This is an excellent illustration of an extremely important
principle.  Our confusion here arises from the indisputible fact
that biting, and certainly cutting, do cause a change of state in
their object.  In this they (or cut, at least) differ from
other events lexicalized by change-of-state verbs, like hitting
or kicking, which may but do not necessarily bring about a change
of state, or touching and kissing, which ordinarily do not.  But
to infer from these facts about the world that bite and
cut must be change-of-state verbs is to seriously
misunderstand the nature of semantics.  The meanings of words and
sentences relate not to some objective reality, but to ways of
construing and presenting a representation of reality.  The
actual event of cutting someone involves both a metaphorical
transfer of an action (like slap or kiss) and an
ensuing effect on the recipient of the action.  Thus in some
sense the one who is cut is both the Loc with respect to the
transfer and the Theme with respect to the ensuing change of
state.  But a single clause can predicate only a single Theme-Loc
relation, and thus a given use of a verb can lexicalize only one
of these two aspects of the event.  The grammatical facts of
English tell us that bite, and cut when it takes an
animate object, lexicalize the surface-contact rather than the
change-of-state scenario.
     We should note in passing (though the problem deserves more
attention than I can give it here) that with inanimate objects
and certain body parts cut shows most of the behavior of a
change-of-state rather than a surface-contact verb.  It shows the
characteristic ambiguous passive:

     15a) The rope was cut with a hatchet.

       b) The rope was cut, so it was useless to us.

In this use cut does not show the lability of true
break-verbs.  But it fails the test for surface-contact
verbs--except on a very forced reading, (16) is not synonymous
with (17):

     16)  They cut his throat.

     17)  They cut him on the throat.

Thus which of the two Theme-Loc relations entailed by an event of
cutting is lexicalized by the verb cut may be affected by
other semantic or pragmatic factors.  The essential point,
however, is that only one can be lexicalized at a time, and
whether in a particular use the verb represents a surface-contact
or a change-of-state predicate can always be determined from its
syntactic behavior.

Surface-contact and change-of-state verbs in Tibetan

A problem in Tibetan grammar which has captured the interest of
Tibetan grammarians for the last millenium is the fact that some
transitive verbs take an unmarked object, while others require
locative marking on the object.  A classic pair of examples,
still used in school grammars, is:

     18)  shing*(-la) sta=re-s gzhus-pa
          tree-LOC    axe-ERG  hit
          hit the tree with an axe

     19)  sta=re-s shing(*-la) 'chad-pa
          axe-ERG  tree        cut
          cut the tree with an axe

There is nothing optional about the alternation; LOC-marking is
obligatory in all circumstances on the object of gzhus
'hit', and impossible under any circumstances on the object of
'chad 'cut'.
     The distribution of LOC vs. zero marking on objects is not
amenable to any of the usual functional explanations.  It is
entirely determined by the verb, and has nothing to do with the
animacy or topicality of the object.  Its patterning with verbs
cannot be described in terms of gradient models of transitivity
such as that of Hopper and Thompson (1980).  The distribution
does, however, correlate fairly well with Fillmore's surface-
contact vs. change-of-state distinction; indubitably change-of-
state verbs like bcag 'break' and bsad 'kill' do
not permit LOC marking, while surface-contact verbs like
gzhus 'hit', rdog=rgyag gzhus 'kick', and kha
bskyal 'kiss',  require it.
     These examples already show an interesting aspect of how
Lhasa Tibetan handles this distinction.  Note that both 'kick'
and 'kiss' are light verb constructions, consist of a noun in
construction with a semantically fairly empty verb, rather than
simplex roots.  Modern Tibetan has a very large set of these
light verbs (there is a small set of extremely common verbs which
account for most of these).  Most surface-contact verbs are of
this type.  The result is that the case frame which they
instantiate is the same as that verbs like 'put':(4)

     20)  nga-s blo=bzang-la rdog=rgyag gzhus-pa yin
          I-ERG Lobsang-LOC  kick       hit/throw-PERF/CONJUNCT
          I kicked Lobsang.

     21)  nga-s snyu=gu yig=cog sgang-la bzhag-pa yin
          I-ERG pen     table   on-LOC   put-PERF/CONJUNCT
          I put the pen on the table.

Thus the surface morphosyntax of these predicates directly
reflects the case analysis in which the lexically specified
argument of the verb is a Theme, with the zero-marking
characteristic of Themes in Tibetan, and the NP corresponding to
the English object is a Loc, and case-marked as such.


Tibetan and English

Both English and Lhasa Tibetan distinguish syntactically
identifiable categories of change-of-state and surface-contact
verbs, and the two languages demonstrate considerable agreement
about the membership of these categories.  However, they also
disagree in a number of cases; specifically, there is a
significant number of Tibetan surface-contact verbs whose closest
English translational equivalent belongs to the change-of-state
category.  For example, compare English polish the table
with Tibetan cog=tse-r rtsi btang, English saddle a
horse with Tibetan rta-la rta=sga brgyab.  The
evidence shows that the English verbs take Theme rather than Loc
objects.  Note the ambiguity of (22-23), cp. the naturalness of
(24-25):

     22)  The table was polished.

     23)  The horse was saddled.

     24)  The table is nicely polished.

     25)  The horse is saddled.

And these verbs cannot take the locative extensions
characteristic of LOC-object verbs:

     26)  *Polish the table on its legs.
     27)  *Saddle the horse on its back.

In contrast, the surface morphosyntax of the Tibetan verbs shows
transparently that the NP's corresponding to these English Themes
are LOCs, since they are marked with Locative case:

     28)  cog=tse-r rtsi   btang
          table-LOC polish emit

     29)  rta-la    rta=sga brgyab
          horse-LOC saddle  throw

One could object at this point that these examples do not show a
clear difference between Tibetan and English, in that one can
also say in English things like put (some) polish on the
table, put a saddle on the horse.  However, these are
structurally but not translationally equivalent; put some
polish on the table neither asserts nor entails anything
about rubbing the polish in or trying to produce a shine in the
table.  Tibetan cog=tse-r rtsi btang does; this is the way
(and the only way) to refer to the complex of actions involving
putting polish on the table and rubbing it in to produce a
polished surface.(5)
     Consider one more similar example.  The English verb
brand is a change of state verb, as evidenced by the
ambiguity of its passive; (30) can have either a stative or an
eventive interpretation:

     30)  The sheep are branded.

Again, the nearest translational equivalent in Tibetan, rtags
rgyab 'mark'(6), is a surface-contact predicate, which
requires LOC marking on its recipient:

     31)  lug-la    rtags rgyab
          sheep-LOC mark
          'brand the sheep'

The basis for the variation between the two languages is quite
evident here.  To brand an animal effects a change of state--both
physically, in the obvious sense that the animal now has a
(slightly) different physical appearance, and conceptually, in
that the animal is now in a new social/legal category.  But this
change of state is accomplished by means of the application of a
symbol onto the body of the animal.  Thus the nature of the event
involves both a surface-contact and a change-of-state schema. 
But a predicate can reflect only one schema, one set of Theme-Loc
relations.  The English verbs brand and mark
lexicalize the change-of-state schema.  The Tibetan predicate
rtags rgyab, like the English locution put one's brand
on, lexicalizes the surface-contact schema.  The nature of
branding is such that in the case of this event category each of
these schemas entails the other, so that the two lexicalizations
are not only truth-conditionally but even image-schematically
equivalent, but the actual lexicalizations are not semantically
identical.


Meaning and syntax

Any approach to these data within a functionalist or Cognitive
Grammar framework requires that the different syntactic patterns
associated with different verbs in a language be explained in
terms of their meaning.  (And, more and more, more "formal"
approaches to syntax are coming to acknowledge this necessity). 
Presumably we should also be able to give a semantic account of
cross-linguistic variation of the sort we have seen here.
     Let us consider, first, verbal notions which behave
consistently in the two languages, like 'break' or 'kick'. 
Simply thinking of both of these as transitive verbs masks an
important difference in their semantics.  Verbs like 'break'
refer explicitly and exclusively to a change of state in their
object (or subject in their intransitive use).  A sentence like

     32)  Sam broke the computer.

is vague about exactly what happened to the computer, but is
completely mute about what Sam did.  It states that whatever Sam
did (or, conceivably, didn't do) resulted in a change of state in
the computer, but the verb in no way specifies anything about
that action or inaction.  Indeed, Sam's role is not even a
necessary part of the scenario; hence the intransitive use of
break.  In contrast,

     33)  Sam kicked a tree.

is very explicit about exactly what action Sam carried out--but
says nothing whatever about effects on the tree.  The tree's role
is very simple; it defines the end point of Sam's kick, and
nothing more.
     This characterizes the difference between canonical change-
of-state and surface-contact verbs.  Those predicates whose
behavior is less consistent, intra- or cross-linguistically, are
precisely those with more complex meanings, involving scenarios
with fairly specific roles for both participants.  The scenario
invoked by the verbal notion 'saddle', for example, involves
something quite specific happening to the horse, as a result of a
fairly specific action on the part of an Agent, and it can be
lexicalized either way.

Notes

1)   Tibetan data used here come primarily from work with
Yundrung Manang; I have also consulted Goldstein 1975, 1984.

2)   This is a point which still seems to cause some confusion to
some students of case grammar who take seriously the notion of
Patient as a case role, since the typical definition of Patient
as the argument undergoing a change of state effectively implies
that most transitive verbs refer to a change of state.  The
source of this confusion lies, not in some mystery about verbal
semantics, but in the inappropriate notion of Patient.

3)   Not an Instrument, pace the standard analysis (Schlesinger
1989, DeLancey 1984, 1991).

4)   Note that rdog=rgyag in ex. (20) does not mean
'foot', i.e. this is not interpretable as something like 'hit
(with) the foot'.  In fact the root rdog originally meant
'step, footstep, kick', and rgyag 'throw' is one of the
common verbal elements in these composite verbs.  Thus
rdog=rgyag is a nominal derived from an older light verb
construction meaning 'to kick'.

5)   There is, however, a difference between the English and
Tibetan locutions; English polish can be used for any
process which imparts a smooth shiny surface to an object,
whether or not any substance identifiable as 'polish' is
involved, as in polishing stones.  Tibetan rtsi btang
cannot; the action of grinding or abrading something so as to
produce a smooth shiny surface requires a different locution,
'jam bton or 'jam=po bzos 'make smooth' or 'od
bton 'produce a shine'.  All of these are change-of-state
rather than surface-contact verbs.

6)   This has a broader semantic range than English brand,
as it can refer to other types of marks (e.g. clipped ears,
etc.).


References

Anderson, John.  1971. The grammar of case:  towards a
     localistic theory.  Cambridge:  Cambridge University
     Press.
Anderson, John.  1977. On case grammar.  London:  Croom
     Helm.
DeLancey, Scott. 1984. Notes on agentivity and causation. 
     Studies in Language 8.2:181-213.
DeLancey, Scott. 1991. Event construal and case role assignment. 
     BLS 17:338-353.
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.
Dixon, R.M.W.  1994.  Ergativity.  Cambridge:  Cambridge
     University Press.
Fillmore, Charles. 1970.  The grammar of hitting and
     breaking.  in R. Jacobs and P. Rosenbaum, eds.,
     Readings in English Transformational Grammar, pp.
     120-133.  Waltham, MA:  Ginn.
Fillmore, Charles. 1975.  An alternative to checklist theories of
     meaning.  BLS 1:155-59.
Fillmore, Charles.  1977a. Topics in lexical semantics.  pp. 76-
     138 in Roger Cole, ed., Current issues in linguistic
     theory.  Bloomington: Indiana University Press.
Fillmore, Charles.  1977b. The case for case reopened.  pp. 59-81
     in Peter Cole and Jerrold Sadock, eds., Syntax and
     semantics 8:  Grammatical relations.  NY: Academic
     Press.
Fillmore, Charles. 1982. Frame semantics.  pp. 111-137 in
     Linguistic Society of Korea, ed., Linguistics in the
     Morning Calm.  Seoul:  Hanshin Publishing Co.
Goldstein, Melvyn.  1975.  Tibetan-English dictionary of
     Modern Tibetan.  Kathmandu:  Ratna Pustak Bhandar.
Goldstein, Melvyn. 1984.  English-Tibetan dictionary of Modern
     Tibetan.  Berkeley:  University of California Press.
Gruber, Jeffrey.  1976.  Lexical relations in syntax and
     semantics.  Amsterdam:  North Holland.
Hopper, Paul, and Sandra Thompson.  1980.  Transitivity in
     grammar and discourse.  Language 56:251-99.
Jackendoff, Ray. 1972.  Semantic interpretation in Generative
     Grammar.  Cambridge:  MIT Press.
Jackendoff, Ray. 1983. Semantics and cognition. 
     Cambridge: MIT Press.
Lakoff, George.  1987.  Women, fire, and dangerous things. 
     Chicago:  University of Chicago Press.
Lakoff, George, and Mark Johnson.  1980.  Metaphors we live
     by. Chicago:  University of Chicago Press.
Langacker, Ronald.  1987.  Foundations of Cognitive Grammar,
     vol. I:  Theoretical prerequisites.  Stanford:  Stanford
     University Press.
Levin, Beth. 1993.  English verb classes and alternations:  A
     preliminary investigation.  Chicago:  University of
     Chicago Press.
Schlesinger, I.M. 1989.  Instruments as agents:  on the nature of
     semantic relations.  J. of Linguistics 25:189-210.