presented at the 19th International Conference on Sino-Tibetan
Languages and Linguistics, 1986
comments welcome
Relativization as Nominalization in Tibetan and Newari
Scott DeLancey
University of Oregon
We here discuss further data from two Bodic languages concerning
the association between relativization and nominalization noted else-
where in Tibeto-Burman by Matisoff (1972).(1)
Matisoff describes a pattern in Lahu, with parallels in other
TB and other Asian languages, in which a single morpheme functions to
mark genitive NPs, relative clauses, and nominalized verbs and clauses.
Additional data from other TB languages, such as those discussed here,
add to Matisoff's evidence that the syncretic marking of these three
functions(2)
is not accidental, but reflects a structural and/or
functional relationship among these
functions, at least in languages of this type. The somewhat more
complex patterns of relativization cum nominalization which occur in
Newari and Lhasa Tibetan demonstrate that, in these languages, relativ-
ization is simply one function of nominalization, i.e. that relative
clauses are simply dependent or appositive NPs. This provides an
explanation for a pattern found in Tibetan, in which the genitive and
nominalizing morphemes are distinct, and relative clauses are marked
with both, i.e. are NPs marked as dependent by the genitive morpheme.
On the basis of such data we can argue that the nominalization
function is chronologically and systemically prior to relativization,
which is merely one specialized function of nominalization. The rela-
tionship between this and genitivization in a language like Lahu where
there is no separate genitive marking seems less straightforward: while
it is not difficult to understand why nominalization of a clause will
allow it to function as a dependent modifier within an NP, it seems odd
that a dependent noun would be marked as nominalized; thus we probably
cannot derive the genitivization from the nominalization function. We
will, however, examine some Newari data which suggest possible
directions to take in developing both a functional and a diachronic
hypothesis to explain the genitive-nominalization syncretism.
Relativization in Classical Tibetan
Classical Tibetan, and some of the modern dialects(3), show the rather common use of the
genitive postposition to mark relative clauses. Unlike Lahu ve,
however, the Tibetan kyi has no synchronic nominalizing
function(4); the neutral nominalizing
morpheme being the pa which has cognates in a number of other TB
languages. The normal relative clause pattern differs from that found
in Lahu in that the verb is nominalized as well as being marked as
genitive (see Bacot 1946):
1) shing gcod-pa-'i mi 'the man who cuts the tree'
wood cut-NOM-GEN man
2) bcad-pa-'i shing 'cut wood/tree'
Here there is a straightforward syntactic explanation for the use of
the genitive postposition as a relative marker; since the modifying
clause is in fact syntactically an NP, it is marked with the same mor-
pheme as any other dependent NP. The differences between this con-
struction and the Lahu equivalent could perhaps be seen as following
from the fact that in Lahu the genitive ve is also the
nominalizing particle, i.e. has the functions of both Tibetan
kyi and pa.
The most obvious of several unsatisfactory aspects of this line of
argument is that, while accounting for the syncretism of the genitive
and relativizing functions in both Tibetan and Lahu, it leaves the
nominalizing function of Lahu ve out of the account. (I.e. we
can "explain" the genitive/relative syncretism of Lahu by assuming
that, by analogy with Tibetan, Lahu relative clauses can be considered
to be nominalized clauses, and thus to require genitive marking when
used as NP modifiers. Or, alternatively, we can interpret the Lahu
relative ve as a nominalizer, by analogy with the structure of
Tibetan relative clauses. However, the Tibetan pattern provides no
model for the actual trifunctionality of Lahu ve). Since, as
Matisoff argues, and as we shall see in the Newari data, this latter
syncretism is not a simply peculiarity of Lahu, we need a more
satisfactorily integrated account than this.
Relativization and nominalization in Lhasa Tibetan
In contrast to the simplicity of the relative construction in Lahu
and Classical Tibetan(5), spoken Lhasa
Tibetan has an unusually complex system of relativization (cp. Mazaudon
1978). We find in Lhasa four distinct relative markers, the choice
being determined by the semantic role(6) of the head NP in the relative clause, and to some extent also
by the time reference of the relative clause. The four are
mkhan for actor, sa locative/dative, and the Lhasa
default nominalizers yag, used for patients and instruments in
non-perfective relative clauses, and pa, used in perfective
relative clauses when the head noun is not the actor(7). All four are clearly nominalizations;
with all but mkhan the relative clause is also marked as
genitive.
The inherited CT construction with pa + genitive can be
used in Lhasa only in relative clauses with perfective time reference
where the head noun is coreferential with a non-actor NP in the
relative clause, as in:
3) kho-s bsad-pa-'i stag pha=gi red
he-ERG kill(PF)-NOM-GEN tiger that be
That is the tiger which he killed.
Compare kho-s bsad-pa 'what he killed', as in:
4) kho-s bsad-pa stag red
he-ERG kill-NOM tiger be
'What he killed is a tiger.'
If the relative clause has other than perfective reference, a different
nominalizer must be used; with patient head noun this is yag:
5) kho-s gsod-yag-gi stag pha=gi red
he-ERG kill(FT)-NOM-GEN tiger that be
'That is the tiger that he will kill.'
(This aspectual distinction also characterizes the use of pa and
yag as nominalizers, although the semantics of the distinction
involves intricacies which I am at present unable to explicate; see
Goldstein 1973).
When the NP head is coreferential with the actor of the relative
clause, the relative clause is marked with the agentive nominalizer
mkhan, without genitive marking:
6) stag gsod-mkhan mi pha=gi red
tiger kill-NOM man that be
'That is the man who killed/kills/will kill the tiger.'
(Compare stag gsod-mkhan 'the one who killed the tiger, tiger-
killer').
Locative, dative, and benefactive nominalizations and corres-
ponding relative clauses are formed with sa(8):
7) kho sdod-sa-'i khang=pa pha-gi red
he stay(IP)-NOM-GEN house that be
'That's the house where he lives.'
(Compare kho sdod-sa '(the place) where he lives').
8) nga-s deb sprod-sa-'i mi pha=gi red
I-ERG book give-NOM-GEN man that be
'That is the man who I gave the book to.'
9) nga-s kha=lag bzo-sa-'i mi pha=gi red
I-ERG food cook-NOM-GEN man that be
'That is the man who I cooked food for.'
(the nominalizer sa requires the imperfective stem; the
resulting relative clause is neutral as to time reference).
Relativization off an instrumental(9) requires yag:
10) kho-s stag gsod-yag-(gi) me=mda pha=gi red
he-ERG tiger kill-NOM-(GEN) gun that be
'That is the gun he shot the tiger with.'
Here (although not, as far as I know, in examples like (3)) the gen-
itive marking, while prescriptively required, is optional in colloquial
speech.
Lhasa relativization strategies in diachronic perspective
The Lhasa data just outlined constitute further strong evidence
that there is a principled basis to the nominalization-relativization
syncretism, for here it is not one, conceivably coincidental morpheme
occurring in both types of construction, but four; every relativizer is
also a nominalizer. The Tibetan data also allow us to argue that the
nominalizing function of such syncretic morphemes is primary, and the
relativizing function a secondary development, i.e. that every
relativizer originates as a nominalizer.
As we have seen above, the Classical Tibetan relativization con-
struction utilizes only one of the four nominalizers occurring in that
function in Lhasa(10). The
differentiation which has developed in the Lhasa system represents the
utilization of already existing nominalization constructions to
introduce new syntactic distinctions into the relativization function;
the grammar obviously recognizes relativization as a species of
nominalization.
Classical Tibetan, and presumably all other varieties, (and pre-
sumably all TB languages; note in connection with this discussion the
list for Lahu given by Matisoff), have available semantically more
specific nominalizers than pa, one of which is the agentive
nominalizer mkhan which is the Lhasa agentive nominalizer/
relativizer. According to Jaschke (1881), in earlier texts this tends
to have a clearly derivational force, occurring both with nouns and
verbs and producing items like shing-mkhan 'carpenter
(shing 'wood'), while in later texts it is used productively to
nominalize clauses, as in Jaschke's example:
11) nga-'i bu=mo 'dod-mkhan
I-GEN daughter desire-NOM
'such as are courting my daughter'
While the historical documentation for the development of sa is
not yet as clear, the general path of development is straightforward.
The original function of sa is as a noun meaning 'earth,
ground', with a secondary sense of 'place', well-attested in
derivational nominalizations such as yod-sa 'place of residence'
(cp. sdod-sa in ex. 4 above). The direction of development is
thus parallel to that of mkhan, from a derivational function with
specific semantic content to a semantically bleached general
nominalizing function which is either immediately or very soon used
also in a relativizing function.
In considering the diachronic development in particular it is
necessary keep in mind the artificiality of speaking of the development
of the relativizing from the nominalizing structure, since, as noted
above, they are at least at first the same thing: nominalization
creates an NP which can then be used like any other in an attributive
function, then constituting what we want to call a relative clause. It
is, indeed, not uncommon to find in the less linguistically oriented
literature on Tibetan and other TB languages the statement that they
lack relative clauses, or at least, in the cautious phrasing of Hahn
(1974), relative clauses "in unserem Sinn", and there is certainly a
sense in which this is true; there are languages, like Lahu and Clas-
sical Tibetan, in which relative clauses are structurally simply depen-
dent NPs, so that there is no distinct relative clause construction.
There remains, however, the relativization function, which apparently
does develop over time as a given nominalizer loses some of its seman-
tic specificity and becomes more productive (e.g. the development of
mkhan).
Dependent and appositive relatives
An obvious problem in the Lhasa data, with clear relevance to the
general problem we are interested in here, is the difference in gen-
itive marking between sa and pa, which require genitive
marking, mkhan, which cannot have it, and yag which is
apparently something of a transitional case. There is no problem,
given the nominalized construction of the relative clause, in
accounting for the presence of genitive marking, but for that very
reason its absence in other cases is puzzling. Since we have accepted
the evidence for a unitary interpretation of sa and pa as
nominalizers in all of the constructions in which they occur, it would
be awkward to have to concoct a different analysis for mkhan.
The alternative is to view mkhan relative clauses, and
yag relative clauses which lack genitive marking, as NPs
standing in an appositive relation to their head. This provides a
straightforward structural analysis of all of the Lhasa relative
structures we have seen: NPs with pa relative clauses have the
structure:
12) [NP [MOD [NP [S kho-s 0 bsad ] pa ] 'i] [NP stag ] ]
while clausal modifiers without genitive marking have the structure:
13) [NP [NP [S 0 stag gsod ] mkhan ] [NP mi ] ]
This analysis of Lhasa, however, still does not provide us with a
clear avenue to the analysis of the Lahu ve. In Lhasa we can
identify the 'i as unequivocally genitive, marking an NP as a
dependent modifier, and mkhan and any similarly-behaving
morpheme as unequivocally a nominalizer, which creates an NP which can
then stand in either a modifying or an appositive relation to another.
A satisfactory analysis of Lahu ve must contrive to unite these
two functions as one. The Newari nominalization/subordination complex,
in several respects more similar to that of Lahu than the Tibetan
system, provides some further clues as to the nature of the
genitivization-nominalization relationship.
Relativization as nominalization in Newari
The Newari nominalizer gu(li)(11) has a range of functions strikingly similar to that of Lahu
ve (cf. Kölver 1977)(12). Among other things it functions as a nominalizer and a marker
of relative clauses, and marks genitive NPs, relative clauses, and some
other dependents within a NP(13):
14) ji~-i~ khun-a-gu
I-ERG cook-PART-NOM
'what I cooked', 'my cooking'
15) ji~-i~ khun-a-gu la 'the meat which I cooked'
meat
16) ji-gu la 'my meat'
I meat
Newari gu(li) is thus a better parallel to Lahu ve than
any of the Tibetan morphemes we have examined. Like Tibetan, however,
Newari provides strong evidence for the non-arbitrariness of this
complex of functions associated with one morpheme in the form of other
nominalizers with virtually the exact same range of functions(14). The choice of nominalizer in
Newari is determined not by semantic/ syntactic role, but by the
animacy and number of the referent of the NP: singular animates
requiring mha rather than gu(li), and plural animates
pi~:
17) ji~-i~ nyan-a-mha nya
I-ERG buy-PART fish
'the [live] fish that I bought'
18) ji~-i~ nyan-a-mha
'the [living] thing that I bought'
19) ji-mha nya 'my [live] fish'
Thus, while the distinctions encoded by the different nominalizers
in Newari and Tibetan are quite distinct, the existence in both lan-
guages of several morphemes with exactly the same set of functions
further demonstrates that the overlap of these apparently distinct
functions is not accidental, but integral to the grammar of the lan-
guage. A comparison of exx. 14 with 15, and 17 with 18, clearly shows
the appositive nature of the Newari relative construction; ex. 17 is
transparently 'the [live] thing I bought, the fish'.
The nature of the Newari genitive construction
The example sets above suggest that Newari, unlike Tibetan but
like Lahu, shows the further extension of the nominalizing function
into genitivization which I have noted as the most problematic aspect
of the complex. However, genitive constructions in Newari show an
interesting set of complications which may ultimately offer some help
in understanding the overall pattern. The examples of genitive mark-
ing given above were deliberately chosen to obscure the fact that
genitive full nouns are inflected for case(15), with the gu(li) series nominalizers suffixed onto the
inflected form:
20a) ram-ya-gu tasbir
Ram-GEN-NOM picture
b) ram-ya-mha khica
dog
As discussed at length in Kölver 1977, use of the gu(li)
series morphemes in genitive constructions is variable, in that NP's
occur in which a dependent NP is inflected as genitive but no
gu(li) morpheme occurs:
21) ram-ya kala 'Ram's wife'
It turns out, not surprisingly, to be impossible to specify in formal
terms the precise conditions under which gu(li)-series marking
varies; but Kölver (whose extended and detailed argument and
documentation we cannot reproduce here) demonstrates a set of semantic
conditioning factors which can be subsumed under a general statement
that the presence of a gu(li) morpheme indicates a greater, and
the absence a lesser, degree of conceptual independence between the
dependent and the head NP. On the continuum of junction the end point
of which is actual noun-compounding, gu(li)-marked junction is
farther from, and gu(li)-less junction closer to, compounding.
(Newari marks a third degree of junction, just short of true
compounding, in which the dependent noun is not marked either by
gu(li) or by genitive inflection). Kölver cites a number
of illustrative pairs, perhaps the most illuminating being the contrast
between 17a, above, which refers to a picture which belongs to Ram, and
the contrasting Ram-ya tasbir 'a picture of Ram', in which the
Ram-ness is an intrinsic quality of the picture rather than contingent
upon some external relation to a distinct entity.
Genitivization and nominalization
The Newari data analyzed by Kölver are instructive in two
respects. In the first place, they demonstrate a specific function of
nominalizers used to mark genitive NPs: since we find their presence
or absence in the genitival construction marking a functional dis-
tinction, we can provisionally identify the function in which they
occur as the general function of such marking. Thus it is encouraging
to return to Lahu and discover that there is an apparently very similar
distinction between ve-ful and ve-less genitive
constructions in Lahu (Matisoff 1973:148-9), so that it may well be
possible to identify a more specific function for ve than pure
dependent marking.
However, the real problem that we have been worrying about is not
whether a nominalizer used as a genitive marker, or in a genitive
construction, has a function -- it surely does, if only to mark depen-
dence -- but why that particular morpheme is used in that function. It
is quite interesting that, according to the analysis suggested here, it
is in some sense the pragmatically unmarked genitive relation which
receives formal marking, and the pragmatically more marked which is
formally unmarked. This is, in fact, quite in keeping with the
function which I am associating with ve and gu(li),
which, by explicitly marking a dependent NP as an NP, reflect its
relative referential independence of the head NP.
The Newari data also offer us at least the possibility of a dia-
chronic explanation, although its applicability to other cases such as
that of Lahu requires further investigation. Kölver points out
that a comparison of the genitive construction with and without a
gu(li) morpheme suggests that these morphemes might be
considered (and probably originally were unambiguously) noun heads,
thus explaining the genitive marking on the noun to which they are
suffixed. (Note that this ordering makes the syntactic analysis of the
Newari genitive more problematic than that of the Tibetan relative
clause, where the genitive marking occupies its expected position
following the nominalizer, marking the relation between two NP's).
This is consistent not only with the structure of genitive NP's per se,
with the "optionality" of gu(li) marking on genitive NP's, and
with the unsurprising fact that gu(li)-morpheme-headed
constructions (like ve constructions in Lahu) can function as
independent NPs (e.g. Ram-ya-gu 'Ram's (inanimate)', Ram-ya-
mho, 'Ram's (animate thing)') but also with the fact that the
gu(li) morphemes inflect for case just like nouns (albeit with
some morphological irregularities), the lexical noun status of
mha, and the participation of all three of the gu(li)
morphemes in the nominal classifier system. Thus for a NP like Ram-
ya-mha khica 'Ram's dog', we can reconstruct, and possibly even
synchronically infer, an appositive structure rather than a genitive,
i.e. a structure like 19 rather than 20:
22) [NP [NP [MOD [ Ram ] ya ] mha ] [NP khica ]]
23) [NP [MOD [ Ram-ya ] mha ] khica ]
There is no immediately obvious structural problem with applying
such an analysis to Lahu genitive constructions, arguing that the
structural marking of the genitive relation in Lahu is simply appo-
sition, so that NP ve has a head, ve, and a dependent NP,
just like a ve-less genitive NP NP construction. As far as I
know, however, there is no independent structural evidence in Lahu in
favor of such an analysis, as there is in Newari.
There are, moreover, other problems associated with such an ana-
lysis of ve, or of Newari gu(li). It is not hard to
reconstruct a semantic content for mha sufficient to justify
analyzing it at least etymological as a noun (i.e. we could claim that
the "real", or at least etymological, structure of Ram-ya-mha
khica is 'Ram's animate thing, a dog'), and its synchronic status
as a independent NP (meaning 'body') lends further credibility to the
hypothesis In the case of the default nominalizer, however, the
analysis seems less credible. In the first place, it seems less likely
that a language would require a semantically virtually empty noun as a
head in all genitive constructions (though certainly not inconceivable;
cp. the Thai genitive marker khOOng, which has precisely this
origin). In the second place, long histories can be reconstructed for
ve, and very possibly gu(li), according to which they
have been nominalizers -- not nouns -- since PTB times. (Indeed,
Thurgood (1981) and Matisoff (1983) have suggested that the origin of
ve and many other TB nominalizers is with a copula, rather than
a noun).
It may well be that the Newari facts which point toward this
interpretation are Newari-specific developments of various kinds, prob-
ably involving the somewhat idiosyncratic development of noun inflec-
tion (probably an areal feature diffusing northward from Indic), the
development of the classifier system, and the expansion of the nominal-
ization system from an earlier stage using only gu(li) or some
analogue in something much like the Lahu pattern, by the introduction
of a true nominal reflected in modern mha. In that case we must
continue to seek a synchronic explanation for the Lahu-type pattern.
If, as I have tentatively suggested, the avenue along which to seek the
explanation is that marked by the functional distinction in Newari and
apparently Lahu by the ve/0/ alternation in genitives, these and
any additional TB data which may be brought to bear on the problem
will, as Matisoff suggested, provide possibly unique insight into the
relation between the structure and the semantic-pragmatic function of
dependent nominals.
Notes
(1) The work reported here was supported in part by
a grant from the Joint Committee on South Asia of the Social Science
Research Council and the American Council of Learned Societies with
funds provided by the Ford Foundation and the National Endowment for
the Humanities, and by the National Science Foundation under grant #
BNS-831502. I am grateful to my language consultants, Harsha Dha-
ubhadel, Ngawanthondup Narkyid, Yungdrung Manang, and Rajendra
Shrestha, for their insightful help in the analysis as well as for the
data. back to text
(2) A fourth extension of the complex, in which the
nominalized clause construction invades the finite verb system,
eventually giving rise to new tense/aspect marking, is not of direct
interest here, although it is of considerable general interest, and
there are synchronic Newari data (cf. Kölver 1977) and apparently
diachronic Tibetan data which bear importantly on it. back to text
(3) E.g. Balti, as described by Read (1934). It
should be noted that an adequate description of the relativization
function in either Classical Tibetan or Lhasa, the two varieties of
Tibetan for which extensive data are available, will be considerably
more complex than what is presented here, an it is virtually certain
that the same is true for relatively unknown dialects such as Balti.
back to text
(4) Although Thurgood (1981) reconstructs an
original nominalizing function for a ST etymon reflected in Tibetan
kyi, and there is some evidence for this in the tense/aspect
paradigms of some modern dialects. back to
text
(5) As noted above, we are here ignoring both some
complexities of the CT relative genitive construction, and the
existence in both CT and Lhasa, and almost certainly other dialects, of
another construction, using demonstratives as markers of definiteness,
which has a similar function to familiar relative constructions.
Fortunately, since I am not yet in a position to explicate either the
structure or function of such clauses or their relation to the relative
constructions discussed here, this construction is not particularly
relevant to the present discussion. back to
text
(6) Or perhaps syntactic function. I am skeptical
about the existence in Tibetan, or TB languages generally, of syntactic
relations such as subject and object, but believers in such things
should have little difficulty in translating my description into
appropriately syntactic terms. back to text
(7) A complete description of the relevance of
aspect to the choice of relative construction in Lhasa (and apparently
also in CT, where there is data suggesting that it is already relevant)
will require more data than I have at present; the present account
should be taken as provisional. In Lhasa each nominalizer, at least
when used in the relative construction, requires a specific aspectual
form of the verb; thus pa requires the perfective stem,
mkhan and sa the imperfective stem. back
to text
(8) At least with locatives the relative clause can
also be formed with pa, with perfective reference; I do not know
at present whether this is also the case with datives and benefactives.
back to text
(9) I suspect that such relative clauses, at least
with this construction, are rather marginal in the grammar; one of my
consultants simply refused to attempt to construct one. back to text
(10) Actually it would be premature to assert
without an extensive examination of Classical texts that analogues to
the sa and mkhan relative clauses never occur in earlier
stages of the language, although, as discussed below, Jaschke points
out a clear diachronic development from derivational nominalizer to
relative marker in the case of mkhan. In any case there is no
doubt that the normal construction in CT is the pa
nominalization, which was used in place of all of the Lhasa
constructions (cp. ex. 1 with ex. 3). back to
text
(11) Newari has undergone considerable attrition of
final consonants, and in a few cases final syllables; however the
eroded segments are retained in certain case forms in the nominal
declension. gu(li) has the nominative form gu, but the
second syllable surfaces in the irregular locative and
ablative/instrumental forms guli and guli~.
back to text
(12) Among these is an extension into the finite
verb paradigm, which we will not examine here. back
to text
(13) An additional complication in the Newari
system, which we will not discuss here, is the fact that gu(li)
and the other two morphemes in the set (see below) also function as
part of the classifier system. While this function is clearly related
to the others, and the relationship is certainly one which must be
explored in an adequate analysis of Newari, such an exploration is
probably not crucial to our present enterprise; my current belief is
that an understanding of the nominalization system will contribute more
to an understanding of the growth and function of the Newari classifier
system than the latter will to the former. back to
text
(14) The other members of the gu(li) set
lack only the independent clause function of gu(li) and
ve, a fact which is easily predictable from their obligatorily
animate reference. back to text
(15) I am ignoring here the diachronic problem
posed by the occurrence of the ya here identified as the
genitive inflection as part of the dative and locative forms, which
poses the choice of whether to analyze these latter as built on the
genitive form, or to consider the ya as marking simply a general
oblique form of the noun. back to text
References
Bacot, Jacques. 1946. Grammaire du tibétain
littéraire. Paris: Librairie d'Amérique et
d'Orient.
Goldstein, Melvyn. 1973. Modern Literary Tibetan. Occ. Papers
of the Wolfenden Society on Tibeto-Burman Linguistics, 5. Urbana,
IL: Center for Asian Studies, University of Illinois.
Hahn, Michael. 1974. Lehrbuch der klassischen tibetischen
Schriftsprache. Bonn: Michael Hahn.
Kölver, Ulrike. 1977. Nominalization and lexicalization in
Newari. Arbeiten des kölner Universalien-Projekts, no.
30.
Matisoff, James. 1972. Lahu nominalization, relativization, and
genitivization. pp. 237-57 in J. Kimball, ed., Syntax and
Semantics I. New York: Seminar Press.
Mazaudon, Martine. 1978. La formation des propositions relatives en
tibétain. BSLP 73:401-414.
Read, A.F.C. 1934. Balti grammar. London: Royal Asiatic Society.
Thurgood, Graham. 1981. Notes on the origins of Burmese creaky
tone. (Monumenta Serindica no. 9). Tokyo: ISLCAA.