Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.1 (151)
On grammatical relations as constraints
on referent identification
Randy J. LaPolla
La Trobe University
Based on a Relevance Theory-informed view of language development, this
paper argues that grammatical relations are construction-specific
conventionalizations (grammaticalizations) of implicatures which arise out
of repeated patterns of reference to particular types of referents. Once
conventionalized, these structures function to constrain the hearer’s
identification of referents in discourse. As they are construction-specific, and
hence language-specific, there is no category “subject” across languages;
different languages will either show this type of grammaticalization or not,
and if they do, may show it or not in different constructions. Any
cross-linguistic use of terms such as “subject” (and “S”, as in “SOV”) should
then be avoided.
. Introduction*
In a 1991 paper, Prof. Shibatani discussed the grammaticalization of topic into
subject. He argued for the following view of subjects in that paper (p. 103):
(a) it is a syntactic category resulting from the generalization of an agent over
other semantic roles, (b) languages vary as to how far this generalization has
taken place; i.e. the grammatical status of subject differs from one language to
another, and therefore, (c) the subject is not necessarily a universal category.
Prof. Shibatani goes on to show how languages can differ in terms of the degree
to which the reference-related properties of subject and the (semantic) rolerelated
properties of subject are conflated on a single NP, and also in terms of
which NP is grammaticalized into the subject. In this paper I would like to try
to take the next step, to explain what it means for a language to grammaticalize
phenomena we relate to the concept of “subject”, and show the cognitive
motivation for such a grammaticalization. I will also go a step further than
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.2 (152)
 Randy J. LaPolla
Prof. Shibatani and argue that not only is “subject” not a universal category,
it is not a category at all, but the lumping together of individual phenomena
that function to constrain the interpretation of the nature of referents in discourse.
In order to do that, though, I will need to first discuss the nature of
communication and the role of language in communication.
. Ostension and inference
Sperber and Wilson (1996; see also LaPolla 2003) argue that a coding-decoding
model of communication cannot explain how communication is achieved between
two people. Building on the blueprint model of Reddy (1979), they argue
that what happens in communication is that one person makes some sort of ostensive
act (a stimulus, an act obviously done for some purpose), and that, as a
request for attention, this act directly communicates the intention of the person
to communicate by means of that ostensive act, and at the same time is direct
or indirect evidence of what the person intents to communicate. The ostensive
act may be linguistic, but it may not be; it could be a simple movement of the
hand, a raising of the eyebrow, or any other obviously purposely done action.
Upon seeing or hearing the ostensive act, the addressee will first infer that the
communicator intends to communicate some information using that ostensive
act, and then will further attempt to infer the communicator’s informative
intention (the information that the communicator intends to communicate).
Underlying all of this is the assumption on the part of both communicator
and addressee that both are rational people (and so act with goals in mind – see
Grice 1975), and that the ostensive act automatically conveys a presumption of
relevance, that is, that the information the communicator wishes to communicate
is worth processing (i.e., it will improve the quality, quantity, and/or
organization of the individual’s knowledge), and that the effort to process it
will not be unjustifiably great. Because the ostensive act communicates the
presumption of its own relevance, the addressee will attempt to determine the
reason why the communicator made that ostensive act by constructing a context
of interpretation (a set of assumptions) in which the particular ostensive
act the communicator produced will be relevant (i.e. will “make sense”), and
thereby infer the informative intention of the communicator. Communication
then relies on this one principle, the principle of relevance: the speaker assumes
that the addressee will recognize the ostensive act as an attempt to communicate,
and that the addressee will assume that the ostensive act is relevant in
some way, and will then attempt to infer what that relevance is.
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.3 (153)
On grammatical relations as constraints 
Above I mentioned that there is also an assumption that the effort to process
the ostensive act will not be unjustifiably great, and because of this, the
addressee will also assume that the ostensive act is the most appropriate one
that the speaker could use in that context with that addressee. Therefore, the
addressee will choose the first accessible interpretation of the ostensive act (the
reason for the speaker making the ostensive act) that seems relevant. The ostensive
act may give direct evidence for what the communicator wishes to
communicate (it may “show something”), such as when someone ostensibly
rises out of a wheelchair and walks across the room to communicate that he
can (now) walk, or it may give indirect evidence by saying something about
it, such as if the same person just mentioned simply said I can walk. There is
a continuum of cases between the two extremes of direct evidence and indirect
evidence, and, more important for this paper, there is also a continuum
of explicitness in terms of the indirect evidence given. The communicator will
tailor the ostensive act in such a way that it will allow the interpreter to be able
to infer the intended information, and without requiring the speaker or hearer
to expend any unnecessary energy. The speaker then also must use inference,
in this case infer what assumptions are manifest (known or could be known)
to the hearer and which could be used in the context of interpretation. The
greater the number of relevant assumptions available to the hearer, the less explicit
the ostensive act need be. Consider the following example. My wife and
I were rushing to clean up our living room, as guests were on their way over.
She stretched out her arm and wagged her finger towards the far side of the
room. I looked in that direction and noticed that my coat was on the couch,
and I inferred that her communicative intention was for me to notice that the
coat was on the couch and realize that it should not be on the couch, and so I
should go get the coat and put it away. Rather than simply wagging her finger,
she could have said any of the utterances given in (1):
(1) a. Your coat!
b. Get your coat!
c. Your coat is on the couch.
d. You should put your coat away.
e. Your coat is on the couch, and it shouldn’t be there, so you had better go
over and get it and put it away.
Because she just wagged her finger, I had to infer that I should look in the direction
she was wagging towards, then had to try to infer that it was the coat she
was wagging her finger at, and then infer that the coat being on the couch was
significant, and then infer what the significance of it being there was, and then
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.4 (154)
 Randy J. LaPolla
infer that she intended for me to do something about the coat being on the
couch. I was able to do all of that inferential work successfully because of certain
contextual assumptions that I could add to the context of interpretation,
such as the semantic frame of “cleaning up”, and the assumption that whatever
was communicated within that frame would be relevant to that frame. If these
assumptions were not manifest to me, such as if I had been sitting working on
my computer and she wagged her finger the same way, I would not be able to
understand her intention with such a simple ostensive act. She would have to
use one of the more explicit ostensive acts given in (1). She could use (1a) if
she thought I would have trouble trying to infer what referent she intended
me to notice. She could use (1b) if she thought I would have trouble inferring
that she wanted me to get the coat. She could use (1c) if she thought I would
have trouble inferring that the coat being on the couch was somehow significant.
She could use (1d) if she thought I would have trouble inferring what
I should do with the coat. And she could use (1e) if she wanted to be sure I
would have little trouble inferring all aspects of her intention. Greater explicitness
can be achieved through the addition of more lexical items, the addition of
grammatical marking, or, more commonly, through the addition of both. Here
grammatical marking includes both grammatical forms, such as the genitive
marking and the word so in (1e), and also grammatical constructions that constrain
the interpretation of some aspect of the speaker’s intention, such as the
constructions we will be talking about here (relevant to grammatical relations)
for constraining the identification of referents.
Languages differ in terms of what functional domains the speakers must
obligatorily constrain. Obligatory marking develops when speakers constrain
the interpretation in a particular way so often that it becomes conventionalized
and thereby recognized as unmarked. Failure to use the marking will not necessarily
lead to misinterpretation, but will be seen as marked, “wrong”, or “ungrammatical”.
In the case of (1d), Should put coat away would not be an acceptable
utterance, as English conventions require the speaker to overtly mention a
Subject and to overtly constrain the identification of the coat with a genitive or
demonstrative pronoun. This is not true of all languages. Languages also differ
in the degree to which the interpretation of a particular functional domain
is constrained. Two languages may, for example, both obligatorily constrain
the interpretation of the time of an action relative to the speech act time (i.e.
have grammaticalized tense marking), but one language may only distinguish
one past tense, while the other language distinguishes four past tenses (e.g.
English and Rawang, a Tibeto-Burman language of northern Burma, respectively).
Languages that constrain the interpretation of a particular functional
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.5 (155)
On grammatical relations as constraints 
domain may also differ in terms of the particular formal mechanism used to
constrain the interpretation. (See LaPolla 2003 for examples and discussion.)
. Grammatical relations
In this paper we are interested in conventionalized patterns and marking which
are related to grammatical relations and their function in constraining the creation
of the context of interpretation. For example, in English, an NP that
immediately precedes a verb will be understood to be representing the actor
of the action represented by the verb, if the verb is an action-type verb and is in
the active voice, and the NP will be seen as representing the patient or recipient
of the action represented by the verb if the verb is in the passive voice.1
A noun
phrase that immediately follows a verb will be understood as representing the
patient, benefactive, or recipient of the action represented by the verb. That is,
word order in the clause is obligatorily used to constrain the interpretation of
who is doing what to whom. In example (1d), we were forced to assume that
you was representing the actor and your coat was representing the patient because
of their positions relative to the verb, and this information was added
to our context of interpretation. The NP in preverbal position has to be the A
(agentive argument) of a transitive verb or the S (single direct argument) of an
intransitive verb when the verb is in the active voice. This restricted neutralization
of roles in that position gives that particular NP a special status we will
call “pivot”.2
Not all languages use word order to constrain the identification of
referents. For example, word order in Chinese does not constrain the interpretation
of the role of the referent of an NP; the hearer must rely on real-world
knowledge and expectations to infer whether an NP is representing an actor or
something else. Consider the following examples:3
(2) a. Xúesh¯eng
student
f¯a-le
distribute-asp
chéngj¯ıd¯an.
grade.report
‘The students were given their grade reports.’
b. Lˇaosh¯ı
teacher
f¯a-le
distribute-asp
chéngj¯ıd¯an.
grade.report
‘The teacher distributed the grade reports.’
The interpretation of the initial NP of both clauses could be either that it represents
an actor, or it represents a recipient, but in most cases assumptions
from real-world knowledge, such the fact that usually teachers distribute grade
reports and usually students receive them, would be added to the context of
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.6 (156)
 Randy J. LaPolla
interpretation, and so the meanings would most likely those given in the free
translations. The important point here is that the position of the NP does not
constrain the identification of the referent of the NP as actor or recipient the
way it does in English.
There are several other constructions in English that also work in a similar
way, that is, constrain the identification and roles of the referents mentioned,
and the majority of these all identify an [A,S] pivot. This pivot is what has come
to be known as “subject” in English. It isn’t the case that all constructions and
marking that constrain the identification of referents are related to what we
think of as “subject” (e.g. gender systems, possessor marking), but I would like
to argue that all of the phenomena we lump together as “subject” phenomena
are involved in referentidentification, and they develop as speakers consistently
use a particular pattern of coreference or word order often enough that it becomes
conventionalized (grammaticalized), and so then obligatorily forces a
particular interpretation.
Let us look now at other constructions and marking which are seen as “subject”
phenomenato see how they constrain the identification of referents.In (3)
we have a conjoined structure.
(3) John hit Bill and cried.
In this example, even if we have no contextual information, we must assume
that the referent of the initial noun phrase, John, is the one who cried (cf.
Comrie 1988). This is because the speakers of English have conventionalized
a cross-clause coreference pattern in conjoined clauses where the second clause
does not have an overt referent mentioned in the clause. The pivot of this construction
is also [A,S], that is, the zero pronoun in the second clause must be
coreferential with an A or S in the first clause. This is not true of all languages.
There are two possibilities, either a different pivot might be identified, or the
language may not have any pivot for this construction. Dyirbal (Dixon 1972,
1980) is a famous example of a language that has grammaticalized a different
pivot in this construction; the pivot is a [P,S] pivot, not an [A,S] pivot. That
is, the construction applies to a patient (P) and the single argument of an intransitive
verb (S), but not to an Actor (A). Consider the example in (4) (from
Dixon 1980:462).
(4) Balan guda
she+abs
ba]gul
dog+abs
yara-]gu
he+erg
bura-n
man-erg see-past
bu]a-n.
descend-past
‘The man saw the dog and (the dog) went downhill.’
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.7 (157)
On grammatical relations as constraints 
This example has a similar structure to the English example in (3), but the
interpretation of the zero pronoun in the second clause is obligatorily coreferential
with the P in the first clause, that is, it must be that the dog is the one
that went downhill.
An example of a language that has not grammaticalized a pivot for this
construction is Rawang, a Tibeto-Burman language of northern Burma:
(5) a. Vp¯ungí Vd´øs`vng vdip b´øà n`ø ng´øa:pmì
Vp¯ung-í
ypung-agt
Vd´ø-s`vng
Adeu-loc
vdip
hit
b´ø-à
pfv-Trans.past
n`ø
ps
ng´ø-ap-ì
cry-TMdys-Intrans.past
‘Apung hit Adeu and cried.’ (Adeu cried)
b. Vp¯ungí Vd´øs`vng vdip b´øà n`ø vh¯øshì a:pmì
Vp¯ung-í
Apung-agt
Vd´ø-s`vng
Adeu-loc
vdip
hit
b´ø-à
pfv-Trans.past
n`ø
ps
vh¯ø-shì
laugh-r/m
ap-ì
TMdys-Intrans.past
‘Apung hit Adeu and laughed.’ (Apung laughed)
Here the structures in (5a) and (5b) are exactly the same, though the actor
of the second clause is interpreted differently due to real world expectations
of who would be more likely to cry or laugh after an act of hitting. There is
no grammatical constraint on the interpretation of the referent of the zero in
the second clause. I’ve written ‘Adeu cried’ and ‘Apung laughed’ after the free
translations, but the interpretation could be that the one who cried or laughed
was either one of these two people, or even a third person, such as someone
standing nearby watching what was happening between Adeu and Apung. Most
Sino-Tibetan languages are similar to Rawang and Chinese (LaPolla 1993) in
not having syntactic constraints that force particular interpretations of crossclause
coreference.
What is important here is how a conventionalized pattern of coreference
constrains the hearer’s creation of the context of interpretation. In English, the
hearer cannot create a context of interpretation that includes an assumption
that would lead to the P argument being seen as coreferential with the zero of
the second clause, for example, in terms of (3), that someone who gets hit is
more likely to cry (and so the interpretation of (3) might be that Bill cried).
In Dyirbal the hearer cannot create a context of interpretation that includes
an assumption that would lead to the A argument being seen as coreferential
with the zero of the second clause, for example, in terms of (4), some real-
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.8 (158)
 Randy J. LaPolla
world consideration that might lead to the interpretation that the man went
downhill. In Rawang there isn’t any conventionalized coreference pattern, and
so the hearer is not constrained in the creation of the context of interpretation.
The structures conventionalized4
in English and Dyirbal force particular interpretations
(or at least limit the possible interpretations), while in Rawang the
interpretation remains unconstrained.
In English, verb agreement is another aspect of the grammar associated
with subject that constrains the identification of the relevant referent to an A
or S. The number marking in the agreement paradigm also helps constrain the
identification of the referent. In (6) the use of singular vs. plural agreement
(both of the verb and the predicate NP) in (6a) and (6b) forces two different
analyses of the structures. In (6a) pine nuts and basil must be interpreted as
two different items about which the same predication is made, while in (6b)
they must be interpreted as one item (a dish with two things combined) about
which a predication is made.
(6) a. Pine nuts and basil are my favorites.
b. Pine nuts and basil is my favorite.
In a language without this sort of agreement, the interpretation of the referent
could not be constrained in this way. The hearer would be left to guess (infer)
the nature of the referent(s) based on non-linguistic assumptions.
Case distinctions, such as those related to the pronouns in English, are also
a way of constraining the identification of referents and their roles in the action
being described, marking the NP as A/S (nominative) or non-A/S (oblique).
For example, use of the nominative form of the 3sg pronoun in (7) by the
speaker prevents the hearer from including any assumptions in the context of
interpretation that would lead to a non-agentive interpretation of the referent
of the pronoun.
(7) He left me at the train station.
In the case of Modern English, the case marking is redundant, in that grammatical
relations are also marked by word order, and in fact case marking has come
to be determined by word order, but this was not the case in older varieties of
English, which did not use word order consistently to mark grammatical relations.
What is important here is that in languages with such case marking the
identification of the referent is more constrained (by the use of the marking)
than in a language without such marking. For example, in Chinese there is no
such case marking, and word order does not distinguish grammatical relations
(see above, also LaPolla 1995, to appear), and so the hearer’s identification of
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.9 (159)
On grammatical relations as constraints 
the roles of the 1sg referent and the interrogative/indefinite pronoun in (8) is
unconstrained by the morphosyntactic structure.
(8) Wˇo
1sg
shéi
who
d¯ou
all
bù
neg
rènshi.
know
(a) I don’t know anyone / (b) No one knows me.
The construction called “raising” in the generative literature, where an argument
of an embedded clause appears in the matrix clause, is a construction
which in English has an [A,S] pivot: only the A or S of an embedded clause can
be “raised” to the subject of a verb such as seem. In (9b) the A argument of the
embedded clause in (9a) is “raised” to the subject of the matrix clause. This
cannot happen with the P argument of the embedded clause, as testified by the
unacceptability of (9c).
(9) a. It seems that man has broken the vase.
b. That man seems to have broken the vase.
c. *The vase seems that man has broken.
This construction would have developed as speakers attempted to constrain the
identification of the relevant topic by placing it in initial position. The effect of
this constraint on referent identification (aside from making the A or S of the
embedded clause the theme) is similar to that of the cross-clause coreference
construction discussed above: the construction forces a coreference interpretation
of the theme (the subject of seem) and the A or S of the embedded clause.
In languages without such a constraint on “raising”, such as Chinese (LaPolla
1993), where the equivalent of (9c) is possible, the interpretation of the relationship
between the main topic and the embedded clause is not constrained
grammatically.
In some languages there is a pivot for relativization, such that only certain
arguments in a clause can be relativized on. For example, in Malagasy (Western
Austronesian; Madagascar) there is an [A,S] pivot, that is, the head of the relative
clause must function as the single argument of an intransitive clause or the
actor of an active voice transitive verb in the relative clause (Keenan 1976:265).
In (10) there is an example of a simple clause and an example of a relative
construction (Keenan 1976:265).
(10) a. Man-asa
atv-wash
ny
det
lamba
clothes
ny
det
zazavavy.
girl
‘The girl is washing the clothes.’
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.10 (160)
 Randy J. LaPolla
b. ny
det
zazavavy
girl
(izay)
(that)
man-asa
atv-wash
ny
det
lamba
clothes
‘the girl who is washing the clothes’
In general restrictive relative clauses are used to help the hearer identify referents,
but having a constraint on the head of a relative clause constrains the
hearer’s interpretation of the semantic role of the referent being identified. English
does not have such a constraint on (finite) relativization; just about any
argument can be relativized on, but the head in a Malagasy relative clause must
have a specific syntactic relation within the relative clause. In (11a) we can see
that simple relativization of the undergoer of the clause is not acceptable. If the
head functions as the undergoer (or instrument) of the transitive verb in the
relative clause, then a special voice form must be used in the relative clause,
such as the passive in (11b), which can then be relativized, as in (11c) (all
examples from Keenan 1976:265).
(11) a. *ny
det
lamba
clothes
(izay)
(that)
man-asa
atv-wash
ny
det
zazavavy
girl
Intended: ‘the clothes that the girl washed’
b. Sasa-n’ny
wash-pass-det
zazavavy
girl
ny
det
lamba.
clothes
‘The clothes are washed by the girl.’
c. ny
det
lamba
clothes
(izay)
(that)
sasa-n
wash-pass
ny
det
zazavavy
girl
‘the clothes that are washed by the girl’
We’ve seen that some languages have grammaticalized pivots for certain constructions,
while others have not, and we have seen in languages that have
grammaticalized pivots that we often can find the same pivot type for several
constructions in the language, but which constructions have the pivot and
which do not is not predictable. For example, Malagasy and Dyirbal have pivots
for relativization (though the former has [A, S, derived-S(passive)] while the
latter has [P, S, derived-S(antipassive)]), while English does not, even though
English has an [A, S, derived-S(passive)] pivot for a number of other constructions.
There are other languages, though, that have grammaticalized different
pivot types for different constructions. One of these is Jacaltec (a morphologically
ergative Mayan language of Guatemala; Van Valin 1981; using data from
Craig 1977 and Datz 1980). According to Van Valin (1981:375; citing Larsen &
Norman 1979), several groups within the Mayan family have a [P, S, derivedS]
pivot for relativization, WH-question formation, and clefting, as in (12a–c)
(Craig 1977:212–213). The only way an A can be relativized, questioned, or
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.11 (161)
On grammatical relations as constraints 
clefted is for it to appear in an antipassive derived-S form, as in (13a–c) (Craig
1977:213). In the antipassive, the verb is marked by the suffix -n(i), and the
ergative marker does not appear on the verb.5
(12) a. ha’
cleft
ix
cl/she
x-Ø-y-’il
asp-3abs-3erg-see
naj
cl/he
‘It is her that he saw.’ (*‘It is she who saw him.’)
b. mac
wh
x-Ø-y-’il
asp-3abs-3erg-see
naj
cl/he
‘Whom did she see?’ (*‘Who saw him?’)
c. w-ohtaj
1sgerg-know
ix
cl/she
x-Ø-y-’il
asp-3abs-3erg-see
naj
cl/he
‘I know the woman that he saw.’ (*‘I know the woman who saw him.’)
(13) a. ha’
cleft
naj
cl/he
x-Ø-’il-ni
asp-3abs-see-anti
ix
cl/she
‘It is he that saw her.’ (*‘It is him that she saw.’)
b. mac
wh
x-Ø-’il-ni
asp-3abs-see-anti
ix
cl/she
‘Who saw her?’ (*‘Whom did she see?’)
c. w-ohtaj
1sgerg-know
naj
cl/he
x-Ø-’il-ni
asp-3abs-see-anti
ix
cl/she
‘I know the man that saw her.’ (*‘I know the man that she saw.’)
While there is this clear [P, S, derived-S] pivot for these constructions, there is
an [A, S, derived-S] pivot for co-reference of pronouns in sequences of clauses
in discourse. That is, the topical referent of a discourse will generally be S or A,
and if it is P will generally appear in a passive construction as a derived-S, as in
the following two examples (Van Valin 1981:380, adapted from Datz 1980:103,
105 – the glossing differs somewhat from the examples from Craig 1977):
(14) a. scawilal
near
tu’
that
xil
saw
naj
hei
hu¨ne’
one
nach’en
cave
tz’ulik;
small
betu’
there
x’octoj
entered
naj
hei
yeba’
hide
sba.
self
saylax
was.looked.for
naj
hei
yaj
but
ma¨xtic’a
never
ch’ilcha
was.seen
naj
hei
‘Nearby he saw a small cave; there he entered to hide. He was looked
for but he was never seen.’
b. xinitij
brought
naj
hei
yanma
his.heart
yulyin,
in.him
stzabi
grasped
naj
hei
swixie’
their.shoulders
heb
pl
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.12 (162)
 Randy J. LaPolla
ix;
cl
x’ilaxtoj
was.carried
naj
hei
xol
through
cake’;
air
hatxca
there
yilaxtoj
was.carried
naj
hei
xtxam
edge
co¨nob
town
‘Gathering his courage, he grasped the women’s shoulders, and he was
carried through the air, and there he was carried to the edge of town.’
In these examples the pronouns referring to the topical referent and the passivized
verbs are italicized. The verbs are passivized because of the preference
for the pronoun referring to the topical animate referentto be in A or S (including
derived-S) function. So in this language we have the grammaticalization of
pivots, but more than one type within the same language.
. Conclusion
We have seen that languages differ in terms of what constraints on referent
identification they have conventionalized. There may be differences in how
strong the constraints are as well; since they are conventionalized, and conventionalization
is a gradual process, even when two languages both conventionalize
a particular type of constraint, the constraints may hold to different
degrees in different languages. Grammatical relations are then a constructionspecific,
and therefore language-specific phenomenon, not a cross-linguistic
category (see also Van Valin 1977, 1981; Van Valin & LaPolla 1997:Ch. 6; Dryer
1997; LaPolla to appear). Given this fact, and the fact that the function of constraining
the interpretation of referents is not unique to the types of pivots
associated with grammatical relations, the logical conclusion is that there is no
category of “subject” to talk about. Use of this term in language descriptions is
problematic, particularly when it is not defined properly for the language being
described, as the reader does not know which particular constraints have been
conventionalized in the language. Even closely related languages long talked
about as having a category of “subject”, such as Italian and English, differ
in terms of which constraints on referent identification have been conventionalized.
For example, Italian has not conventionalized a strong cross-clause
coreference constraint in conjoined clauses the way English has (ex. (3)). Very
few linguists are aware of this difference because of the assumption of “subject”
as a category, and the assumption that subject is comparable across languages.
The nature of being a language-specific phenomenon is in fact not specific
to the constructions traditionally associated with grammatical relations. I
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.13 (163)
On grammatical relations as constraints 
would argue that we should treat all morphosyntactic phenomena as languagespecific
phenomena. In analyzing a language, then, we cannot take any categories
or constructions as “given”; we need to look at what constructions and
marking have become conventionalized in the language, and then look at each
construction or marking found to see what functional domain it is constraining
the interpretation of, how and to what degree it constrains the interpretation
of that domain, and with what morphosyntactic form(s) it constrains the
interpretation of the domain.
Notes
* Although I was not a student of Prof. Shibatani, I have been learning from him since my
student days through his many writings. I have also since been fortunate to have gotten to
know Prof. Shibatani personally, and have benefited greatly from our discussions. I would
like to offer this paper to Prof. Shibatani as an expression of my respect, gratitude, and
affection.
. Prof. Shibatani has recently (2002) argued that passive, inverse, and split-ergativity all are
manifestations of voice, as all ‘refer to the origin and terminal point of an action’ (2002:16),
where natural event configurations (e.g. speech act participant acts on 3rd person) receive
the unmarked voice expression and unnatural event configurations (e.g. 3rd person acting
on speech act participant) receive the marked voice expression. In the terms of this paper I
would say that they all constrain the identification of the roles of the referents involved.
. It is a neutralization because either A or S can be the pivot, but it is restricted because
the P (patient) argument of a transitive cannot appear as pivot. See Van Valin and LaPolla
(1997:Ch. 6) for a full discussion of pivots and their use in cross-clause referent tracking.
Here I am using the term pivot in a somewhat broader sense, to include both pivot and
controller, and am talking about referent identification in a more general sense than just
cross-clause referent tracking in discourse.
. Abbreviations used in the examples: abs absolutive; agt agentive; anti antipassive
marker; asp aspect marker; atv active voice; cl classifier; det determiner; dem demonstrative
pronoun; erg ergative; Intrans.past third person intransitive past; loc locative/dative;
pass passive voice; pl plural; inst instrumental voice; pfv perfective; ps predicate sequence;
r/m reflexive/middle; TMdys past tense, 1 day–1 year ago; Trans.past 3rd person transitive
past; wh wh-question marker.
. Grammaticalization and lexicalization are both forms of conventionalization, and differ
only in the generality of the elements produced.
. Due to the effect of certain morphophonemic rules, the verbs in (12a–c) would be pronounced
[xil]. See Craig (1977:213–214).
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.14 (164)
 Randy J. LaPolla
References
Comrie, B. (1988). “Coreference and conjunction reduction in grammar and discourse”. In J.
Hawkins (Ed.), Explaining Language Universals (pp. 186–208). Oxford: Basil Blackwell.
Craig, C. G. (1977). The Structure of Jacaltec. Austin and London: University of Texas Press.
Datz, M. (1980). Jacaltec Syntactic Structures and the Demands of Discourse. PhD
dissertation, University of Colorado.
Dixon, R. M. W. (1972). The Dyirbal Language of North Queensland. Cambridge: Cambridge
University Press.
Dixon, R. M. W. (1980). The Languages of Australia. Cambridge, London, and New York:
Cambridge University Press.
Dryer, M. S. (1997). “Are grammatical relations universal?” In J. Bybee, J. Haiman, & S.
A. Thompson (Eds.), Essays on Language Function and Language Type (pp. 115–143).
Amsterdam and Philadelphia: John Benjamins.
Grice, H. P. (1975). “Logic and conversation”. In P. Cole & J. L. Morgan (Eds.), Speech Acts
(pp. 41–58). New York: Academic Press.
Keenan, E. L. (1976). Remarkable subjects in Malagasy. In C. N. Li (Ed.), Subject and Topic
(pp. 247–301). New York: Academic Press.
LaPolla, R. J. (1993). “Arguments against ‘subject’ and ‘direct object’ as viable concepts in
Chinese”. Bulletin of the Institute of History and Philology, 63, 759–813.
LaPolla, R. J. (1995). “Pragmatic relations and word order in Chinese.” In P. Downing & M.
Noonan (Eds.), Word Order in Discourse (pp. 297–329). Amsterdam and Philadelphia:
John Benjamins.
LaPolla, R. J. (to appear). “On describing word order”. In F. Ameka, A. Dench, & N.
Evans (Eds.), Catching Grammar. Berlin: Mouton de Gruyter. (downloadable from
http://www.latrobe.edu.au/linguistics/stlapolla_data/PublicationItems/Papers/
describingwo.pdf)
LaPolla, R. J. (2003). “Why languages differ: Variation in the conventionalization of
constraints on inference.” In D. Bradley, R. J. LaPolla, B. Michailovsky, & G. Thurgood
(Eds.), Language Variation: Papers on Variation and Change in the Sinosphere and in the
Indosphere in Honour of James A. Matisoff (pp. 113–144). Canberra: Pacific Linguistics,
Australian National University.
Larsen, T. & Norman, W. (1979). “Correlates of ergativity in Mayan grammar”. In F. Plank
(Ed.), Ergativity: Towards a Theory of Grammatical Relations (pp. 348–370). London:
Academic Press.
Reddy, M. J. (1979). “The conduit metaphor – a case of frame conflict in our language
about language”. In A. Ortony (Ed.), Metaphor and Thought (pp. 284–324). Cambridge:
Cambridge University Press.
Shibatani, M. (1991). “Grammaticization of topic into subject”. In E. C. Traugott & B. Heine
(Eds.), Approaches to Grammaticalization, Vol. II (pp. 93–133). Amsterdam and
Philadelphia: John Benjamins.
Shibatani, M. (2002). “On the conceptual framework for voice phenomena”. Paper presented
at City University of Hong Kong, May 8, 2002.
Sperber, D. & Wilson, D. (1986/1996). Relevance: Communication and Cognition (2nd
edition, 1996). Cambridge, MA: Harvard University Press.
Uncorrectedproofs-JohnBenjaminsPublishingCompany
TSL[v.20020404] Prn:17/10/2005; 15:54 F: TSL6506.tex / p.15 (165)
On grammatical relations as constraints 
Van Valin, R. D., Jr. (1977). “Ergativity and the universality of subjects”. In W. A. Beach, S.
E. Fox, & S. Philosoph (Eds.), Papers from the Thirteenth Regional Meeting[:] Chicago
Linguistics Society (pp. 689–705). Chicago: Chicago Linguistics Society.
Van Valin, R. D., Jr. (1981). “Grammatical relations in ergative languages”. Studies in
Language, 5, 361–394.
Van Valin, R. D., Jr. & LaPolla, R. J. (1997). Syntax: Structure, Meaning, and Function.
Cambridge: Cambridge University Press.