comments welcome:  delancey@darkwing.uoregon.edu




        Diachronic Notes on the Klamath Verb Suffixes

                        Scott DeLancey
                     University of Oregon


     The Klamath verb, as described by Barker, presents a very
complicated picture, with a number of suffixal position classes
following the verb stem.  The purpose of this paper is to present
language-internal evidence for the origins of some of this
suffixal morphology.  Several of the inflectional classes can be
shown to be "escapes" from the locative suffix class, i.e.
represent development of inflectional morphemes from earlier
derivational elements.
     While there are a few points where the analysis presented
here may be useful for synchronic purposes in highlighting
certain aspects of the verbal structure other than those which
motivate Barker's analysis, this is intended as a set of
historical inferences from the facts as laid out by Barker, not
as a synchronic reanalysis of his system.


The Klamath verb:

Barker identifies 25 position classes in the verb, which he
subsumes under seven headings:  Prefixes (valence-changing
constructions and reduplication), Stems (classificatory morphemes
and verb stems), Stem Formants (a miscellaneous set of
derivational suffixes), Locatives (these are discussed as
"dependent motion stems" in DeLancey 1989), Aspects (Barker's
classes 15-22), Modals (class 23), and Derivationals (classes 24-
25).  (The labels should not be taken too literally).  The
Locatives (Barker's classes 10-13) can be considered constituents
of the stem (DeLancey 1989); this is probably also appropriate
synchronically for the class 14 "directives", which Barker also
lists in his section on Locatives, although it is not clear that
these belong with the other locatives historically.  Except for
three valence-changing prefixes, which appear to be old, all
modification of a lexical verb stem is suffixal.
     Klamath verb stems can be oversimply described as consisting
of one or more of three types of stem constituent.  Free stems
can on their own take inflectional suffixes.  There are two large
position classes of bound stems (DeLancey 1988; in Barker's more
detailed analysis there are minor subclasses of each), those
which precede and those which follow the other element of the
stem.  The prefixed bound stem elements can be roughly
characterized as absolutive/instrumental noun classifying
prefixes and the suffixed elements as motion/location morphemes
"dependent motional stems" (DMS's).  This other element can be a
free stem, but more common are stems consisting of one
classifying prefix and one DMS.  This means that, among the stem-
constituent classes, only free stems (Barker's position class 7)
and DMS's (classes 11-14) can take the inflectional suffixes
which mark tense/aspect/modality categories--which can be taken
as evidence for positing a verbal origin for the DMS category
(DeLancey 1988).
     Our interest here is in the remaining 11 classes, Barker's
Aspects, Modals, and Derivationals.  There are clearly
differences among the various morphemes in these classes in the
relative antiquity of their incorporation into the verb complex,
and our present purpose is to begin to separate the synchronic
forms out into chronological layers.  The largest and most
heterogenous of Barker's categories is the Aspects; I will
suggest below that this is because the Modals and Derivationals
for the most part represent much older morphology than the
Aspects, and thus diverge sufficiently from the Aspects, both
morphologically and semantically, to justify their analysis as
separate categories.
     The members of the Aspects category are listed, by position
class:

       Class     Members

          15        o:l  'completive, finishing an action,
                         undoing an action'

          16        cn'  'along, action while moving'
                    c'n  'just did and no more, just finished'

          17        ebg  CISLOCATIVE
                    ye:g INCEPTIVE

          18        ebli 'back, toward, returning, behind
                         oneself'

          19        obg  DURATIVE
                    odg  PERFECTIVE

          20        all  'pejorative, ruined forever, done in a
                         bad sense, wickedly, immorally''
                    aksg    'almost, came close to doing'
                    damn 'habitual, doing over and over'
                    nannwi  'right away, at once'
                    n'apg     'feel like, intend to, about to'
                    o:t  'while'
                         'instrumental nominalizer (w. {-s})
                    samni     'intend, plan to'

          21        i:   BENEFACTIVE

          22        ang  polite sg. imperative
                    astg    'tried to'
                    dgi  'want s.o. to do'
                    Wi:  'almost, used to, nearly

                    k'   instrumental nominalizer
                    w    perfective nominalizer

Examples illustrating the semantic range of each morpheme can be
found under these headings in KD.


Classes 15-16

The sole member of class 15, the completive {o:l}, has no evident
etymology; for the time being it must remain as an established
member of the verb complex.
     In class 16, {c'n} 'just did and no more, just
finished' can be shown by its irregular allomorphy to contain a
root related to {en} (class 10) 'action away, taking away,
going', i.e. TRANSLOCATIVE1.  {c'n} has an allomorph,
|c'a|, which occurs only before {ebg} 'toward, action
toward the speaker', i.e. CISLOCATIVE and {obg} 'durative,
continuous action occurring over a period of time', which I will
gloss DURATIVE.  Now, {ebg} and {en} cannot co-occur, for
obvious semantic reasons.  In fact, though they belong to
different position classes synchronically (see below), pairs such
as {gv-en-a} /gena/ 'go', {gv-ebg-a} /gepga/ 'come' suggest that these 
may once have been contrasting members of a single morphological series.
Then the morpheme which Barker lists as |c'n|, |c'a|, is etymologically 
*c'a obligatorily followed by either a cislocative or translocative
particle.  This is quite consistent with the semantics of {c'n};
the sense of 'just did and no more' is an easy extension from 'finish
doing and go/come'.  Thus we can postulate a sense 'finish' for this
*c'a-.
     Unlike {ebg}, {obg} can occur after {en}, as {gv-en-obg-a} 
'go-TRANS-DUR-INDIC' /gempga/ 'is going' (KD:129).  Thus we could
in principle expect forms such as */spiwc'ampga/, from *|spiw-c'n-bg-a|, 
for the durative form of this completive; instead, we get a form
homophonous with the cislocative, /spiwc'apga/, which is either 'just came
dragging an obj. through the water', with cislocative, or 'is just
dragging an obj. through the water', with durative, sense.  The durative
sense can only reflect |spiw-c'a-obg-a|, with the same |ca| allomorph 
as occurs with {ebg}. It would seem that the form /c'apg-/ etymologically
reflects only *c'a-ebg, and that the durative sense available for a
form like /spiwc'apga/ represents a later reinterpretation of /c'apg-/,
probably facilitated by the increasing opacity of {c'n}, and the 
concomitantly increasing degree of arbitrariness of its allomorphy.  (One
problem with this analysis is that synchronically {en}, a class 10
DMS, cannot follow our hypothetical *c'a, which comes later in the
verb complex.  However, various grammaticalizations which we will discuss
below show that at an earlier stage the DMS's had more freedom of
positional occurrence than they do now, so this objection is not in
principle insurmountable).
     For the other member of this class, {cn'} 'along, action
while moving', I can suggest only a partial etymology.  There is
good semantic motivation for trying to link the first segment
with class 14 {ca} 'motion away from for a purpose, going
to go and do what the verb stem says'.  The remaining piece,
|n'|, is identical in form to the obscure class 11 morpheme {n'}
'action onto, repeatedly, again and again', and some sort of
relation between these is semantically plausible.  But as the
latter occurs only following |n|, it is likely to be some
sort of fossilized reduplication rather than a distinct morph
that could be used in constructing other etymologies.  Other than
this there is no ready etymological analysis for {cn'}.
     Overall the categories in these classes may very well have
deep roots in the language, and with, for the time being, no
other evidence bearing on their origins, we must reconstruct a
corresponding positional category for our early Klamath verb
formula.  In the reconstruction scheme which I will suggest later
in this paper, I will call this the DEICTIC category, for no very
compelling reason except an intuitive hunch that that was its
original function.

Classes 17-19

My next, and easiest, demonstration will be that categories 17-
19, and all of their members, represent grammaticalizations of
Dependent Motional Stems, and by all appearances of no great
antiquity.  We might anticipate this result from the proximity
within the verb complex of the two categories and from the
semantics of the morphemes, all of which have motional and/or
aspectual senses of a sort which commonly develop from verbal or
adverbial lexemes of motional/locational sense.  In Klamath,
however, the inference is unnecessary, as the structural and
etymological relations between the two classes are such as to
make the historical relationship obvious almost on inspection.
     The most important clue to the history of this set is the
fact that, like independent stems and DMS's, but unlike members
of any other position class, these can occur with classificatory
prefixes to create verb stems.  For four of the five members of
classes 17-19, examples in KD show them functioning exactly
like dependent stems in combining with classificatory prefixes:

     17        {ebg}, 'toward, action toward the speaker'
                    |?v-ebg| /?epg-/ 'bring a long
                    obj.'; {?v} Class 4 'act upon a
                    long obj.'
          
     18        {ebli}, 'back, again, returning, behind'
                    |ksv-ebli| /ksebli/ 'take a living
                    obj. back, turns a living obj. over;
                    {ksv}, Class 4, 'act upon a living
                    object'

     19        {obg}, 'durative, continuous action occurring over
               a period of time'.
                    |kt-obg| /ktopg-/ 'is hitting, striking';
                    {kt}, Class 4, 'hit with the hand, fist'

               {odg} 'been, been doing, in a past state now
               completed'
                    |tgv-odg| /tgotg-/ 'is standing';
                    {tgv}, Class 4, 'stand' 
          
The remaining member of this set, {ye:g} (17) 'starting, beginning', 
is particularly instructive.  Barker (KD:468) notes the resemblance
of this to the dependent stem {oye:g} 'up, raising', which of course does
occur with Class 4 prefixes to form stems (e.g. |ksv-oye:g| /ksoy:g-/ 
'raises, picks up a living obj. (intr. also)').  There is no doubt that
these are synchronically distinct morphemes; Barker (KG:149) cites
a form in which they co-occur:

          |ngat'-ye:g-kang-ye:g| /ngatye:kkangye:g-/ 'starts to
          jump up here and there'
               {ngat'} 7 'jump', {oye:g} 11 'up', {okang} 12
               'here and there', {ye:g} 17 'starting'

Diachronically, however, they are clearly the same etymon;  the
differences between them--phonological shape, position class, and
meaning--all show {ye:g} to be a more grammaticalized, and hence
phonologically reduced and semantically more abstract, development 
of {oye:g}.
     The reason why {ye:g} doesn't occur directly after prefixes,
therefore, is because in fact it does, but when it does it is
still identifiable as the original {oye:g}.   We can infer such a
stage for the other motionals, i.e. where, for example, {ebli}
had positionally distinct behaviors corresponding to the more
concrete motional sense of 'back, returning', and the more
abstract sense of 'again', with the collapse of the positional
distinction (for this or some etymon) constituting the crystallization 
of the motionals as a distinct position class.  And, indeed, there is
clear evidence of exactly such a stage for {ebli}.  Dependent stems are
subject to three intensive reduplications, two productive and one of very 
limited lexically-governed distribution.  There is a Class 10 dependent 
stem {pbe:li'} 'back and forth' which looks like an intensifying
reduplication (KG:143) of a hypothetical *be:li'.  Except for the
first vowel this would match {ebli}, but the source for the vowel can be
inferred from the irregular intensive form of {oyGi} 'up, above, over',
which Barker gives as |oye:Gi| (KG:146).2  Two other pairs of dependent 
stems seem to reflect an old alternation of zero with long vowel marking 
a simplex/intensification opposition:

     {elwy} 'by the fire, along the edge', into water'
     {eli:w} 'on the very edge'

     {aLn} 'alongside'
     {oli:n} 'off the edge, side, overboard, on the edge'

A few other dependent stems show a suspicious long vowel and
semantics consistent with an intensified origin, but have no
corresponding non-intensified form in Barker's materials, e.g.
{ote:g} 'deep into', {aba:ni} 'to the limit, end of a place',
{aqye:tn} 'right beside'.
     Thus we can identify {pbe:li'} as an old
reduplication (by a formation attested only in other fossilized
forms) of *bli (Class 10) 'return', which is now directly
attested only in {ebli}.  So for two of the five members of
classes 17-19, {ebli} and {ye:g}, we can establish an origin in
dependent stems.  For the other three members of these classes
there is no such clear evidence.  However, all have meanings
which are attested in other languages as deriving from motional
predicates, and since they all share with the dependent stems the
ability to form an bipartite stem, they can be inferred to have
the same origin in this category as {ye:g} and {ebli} can be
shown to have.


Classes 20-21

The next class, 20, consists primarily of large morphemes, 3, 4,
or 5 segments long, two of them disyllabic.  Most have fairly
specific senses.  Thus overall they look like relatively
recently-grammaticalized morphological elements.  (The one
exception is {o:t}).  One has a transparent etymology in Klamath: 
{damn} 'over and over' is obviously related to the independent
verb stem {damn} ~ {damnon} 'travel, wander'.
     Class 21 has only one member, {i:} BENEFACTIVE.  The
relation of this to class 10 DMS {oy} 'giving a sg. obj.' is
transparent, both semantically and in form.  Like other DMS's,3
{oy} manifests its initial vowel only following classificatory
prefixes, so that in the position where {i:} occurs only the |y|
allomorph would be possible.  The etymology is not perfect, for
the two forms are not identical.  Both forms show irregular
allomorphy anyway, however, and if we are attempting to look
beyond synchronic position class it doesn't seem unreasonable to
join together {oy}, with unpredictable |wy| and |y| allomorphs,
with {i:}, with |i:y|, into a single etymon with unpredictable
|wy|, |y|, |i:|, and |i:y|.  (Since the essential difference
between the class 10 and class 21 forms of this etymon is simply
in weight, it may be that the relationship here is mediated by
the intensive derivation discussed above).


The Outer Classes

In classes 22-25 we begin to encounter basic tense/aspect/mood
and syntactic categories, with what is clearly very old morphology.  
At this point we need to begin to keep separate at least two, probably
three subsystems of the verb, for the choice of one or another member of
one category is compatible only with certain possible choices from later
categories.  Barker notes the functional heterogeneity of this class as
compared with previous ones.  There are three "aspects", {astg} 'tried to', 
{dgi} 'want s.o. to', and {Wi:} 'almost, nearly', whose semantics, as
Barker notes, would be consistent with inclusion in class 20.  His 
classification reflects ordering among those of these "aspectuals" which 
co-occur in his corpus; since the possibilities of cooccurrence are clearly 
at least in part semantically determined, there are no compelling grounds
for interpreting the ordering of classes 20-22 as having historical
significance.


The nominalization system

The other three members of class 22 are a polite singular
imperative {ang}, which I will discuss below, and two
nominalizers, {k'} 'instrumental nominalizer' and {w} 'past
nominalizer'.  These latter can be distinguished from the other
members of this class by the possibilities of what can follow;
{k'} and {w} must be followed by the class 23 nominalizer {y}. 
This in turn must be followed by the class 24 nominalizer {s}. 
The final class, 25, consists of two members which can occur only
after this {s}; the other members of class 24 are terminal.
     Thus we can identify, across the last four position classes,
a nominalization system:

            k'          i 
               -(y)-s-    
            w           t 

The {(y)-s} is the core of the nominalization system, with the
other two subsystems {k'}, {w} and {i}, {t} built on it.


Finite systems

Since class 25 is relevant only to the nominalization system, we
now have only two position classes left to deal with.  Class 23
has 8 members, excluding the nominalizer {y}.  These can be
sorted into four imperative or hortatory morphemes {at}, {ek},
{i}, and {n'a}, three finite morphemes, {a}, {wabg}, and {at}, 
and a subordinating conjunction {ank}.  Of these, {wabg} can immediately
be banished to some preceding class, as it is attested in Gatschet's texts
as occurring with the declarative {a}.  On semantic grounds it would fit
in with the "aspects" of classes 20-22; like many other young future
constructions (e.g. English will) this 'has a connotation of will
or intent as well as future action' (KG:168).  Since it occurs
following class 21 {i:} (KD:176: |wqat'-l'Glgi-i:-wapg| 'will come
to clear forest for someone'), it would have to have been in class 22.  So
far I have not seen forms in which any of the "aspectual" members of that
class, {astg} 'tried to', {dgi} 'want s.o. to do', {Wi:} 'almost,
used to, nearly did', occur with {wabg}, although {astg} and {Wi:}
both occur with class 24 {a}.  Thus {wabg} in Gatschet's data appears to
fit perfectly with these in both morphology and function.
     I have already argued that the strict ordering of classes 20
and 22 with respect to each other and to the benefactive {i:}
which constitutes class 21 described in KG probably has
only limited historical relevance, its development probably
having a lot to do with the intrusion of the benefactive into a
formerly undifferentiated aspectual class, with cooccurrence
restrictions semantically determined.4  (Note that there are
exx. in KG and KD with {damn}, {nannwi}, and {samni} all
occurring with {wabg}).
     Now, {wabg} is the only member of class 23 which can occur
before two of the members of class 24, {dk} and {wk}, just as {y}
is the only member which can occur before the last member of 24,
{s} (cp. White 1973:5).  The other seven must all be terminal in
the verb complex.  Since {dk} and {wk} also must be final, and
since of the members of Barker's class 23 they co-occur only with
{wabg}, which historically belongs to a preceding class,5 we
can now eliminate class 24 in the finite system, and incorporate {dk}
and {wk} into the preceding class.
     The members of this can now be segregated on functional
grounds into three sets--two indicative endings, three
participial forms, and four imperatives:

     {a}  indicative
     {at}    can, able, ought to

     {ank}     having done
     {dk}  being in a state of
     {wk} by, because of, in order to, from

     {i} (|i|, |i:k|, |a|)  'imperative singular'
     {at}                'plural imperative'
     {ek}                '1st sg. hortatory'
     {n'a}                    '1st pl. hortatory'

     The "indicative" ending {a}, with its spare phonological
profile and thoroughly bleached semantic content, we can take to
be very old; from its form and meaning it is a fair guess that it
may be the oldest of the suffixes.


The imperative system

As Barker notes, the class 22 "aspect" {ang} 'polite imperative'
fits well semantically with the last of these sets.  He places it
in class 22 because it occurs before {at} 'pl. imperative'.  In
principle we could as well place {ang} in 23, and put {at} in a
separate later category.  In any case we have only shaky grounds
for postulating any differences in age among the five members of
the functional imperative system.
     Barker also notes the possibility of analyzing {n'a} as
{n'}, which would then belong to class 22, followed by indicative
{a}.  His rejection of this on grounds of semantics and symmetry
is probably correct; it seems unlikely that {a} would co-occur
with an imperative, particularly with only one.  In any case, the
|n'| of this form bears comparison with the Wintu 1st person
jussive {n} (with allomorphs |n|, |en|), which Pitkin (WG:122)
relates to the first person pronominal root.


The participial system

The three participial forms may be analyzable, if the |k| element
common to all three is not coincidental.  In fact White (1973:74-
5) suggests on phonological grounds that {wk} should be analyzed
as class 22 {w} 'past nominalizer' plus class 22 {k'}
'instrumental nominalizer'.  She notes that "it is impossible to
establish a definite semantic relationship" to support the
phonological inference, but it is not that hard to imagine
*{w}+{k'} 'because of the result of', developing into {wk} 'by,
because of, in order to, from'.  Gatschet states (1890:387-8)
that {wk} also occurs with the declarative {a}.  However, he
notes that it usually does not, and I have not yet noted clear
cases of this sequence in his tests, and since in his discussion
he confuses {wk} with the DMS {iGog} and several /ok/ sequences
which appear not to have any morpheme in common, at least
synchronically, we may provisionally take Barker's description as
historically as well as synchronically valid.
     Gatschet (1890:407-8) notes that Modoc has {an} instead of
{ank}, and that {an}} occurs also here and there in Klamath
texts, and infers that the Klamath |k| is a secondary accretion,
probably a reduction of the copula {gi}.  Thus there are some
independent grounds for segmenting out the final |k| in two of
the three participial forms.  I suspect that for both Gatschet's
etymology of origin in the copula {gi} is the most likely.  Such
an etymology would be semantically plausible for {dk}, but this
is far from a necessary conclusion.  Gatschet records /tko/ for
this form, which would not fit with an etymological connection to
either the copula or the instrumental nominalizer.  We have
suggested elsewhere (DeLancey, Genetti and Rude 1988) that {dk}
be equated with a Tsimshianic suffix {tkw} of very
similar function, which if correct implies a considerable age for
this concatenation.  (However, it is also possible that the
Tsimshianic form is etymologically bimorphemic).


The Pre-Klamath verb

Now, it is not clear that all four systems are reconstructible. 
The indicative, imperative, and nominalization systems seem
irreducible; while it may be possible to find secondary sources
for some of the imperative morphemes, we have no grounds for
considering the category itself to be of recent origin.  The
participial system, on the other hand, may repay further
analysis.  If we reconstruct *an, *w, and *dkw, all of
these would normally require the zero allomorph of the
indicative, so that the absence of attestation of {a} following
these is not strong evidence that all four belong to the same
category.  White's suggestion that {wk} may be built on a
nominalized form is plausible, although as I have stated I prefer
to identify the second element with the copula.  Otherwise, I
have at present little evidence for identifying any of the
hypothetical etyma involved in this system with other specific
elements of the complex.


A Reconstructed Model

Omitting reduplication as a position class, we reconstruct the
following, we can reconstruct the following scheme, where PC (for
Prefix Complex) represents the valence-changing prefixes, and the
Stem consists of various combinations of classificatory prefixes,
DMS's, and Class 7 stems:


                       NOMINALIZATION 
     PC-STEM-DEICTIC - INDICATIVE  
                       IMPERATIVE    
                       SUBORDINATION 


While the efflorescent verb described in KG seems unusual
for the area, we have noted elsewhere that the elements of the
reconstructed scheme are directly comparable to constructions in
other Penutian languages.  Thus the prefix complex has strong
similarities to a Tsimshianic category (DeLancey, Genetti and
Rude 1988), and the inflectional complex both structural and
etymological similarities to that of Wintuan (DeLancey 1988),
while the stem complex is comparable to that of Maiduan (DeLancey
1989).


Diachronic Implications

Some may find something counterintuitive about the conclusions
suggested here, in that I am claiming that the outermost layers
of verb morphology are considerably more archaic than several
inner layers.  My reconstructed history for the development of
the Klamath verb complex would begin the history of each morpheme
or true series with a serial verb or complement construction. 
The process of transition from such a construction to a
morphological one normally includes a stage at which only one of
the verbs in the construction bears finite morphology.  This is
always the more grammaticalized verb.  Both of the basic sources
which we have identified for Klamath suffixes--incorporated verbs
(as at least one member, and probably others, of class 20) and
dependent stems (classes 17-19, 21)--would thus have come between
the modern stem and the finite morphology, exactly where we find
them.


Notes


1)   Rude (1987) suggests that the stem-final -n of many
Sahaptian and Klamath verbs many trace back to a motional
element.

2)  Note that this is phonologically distinct from the
{ye:g}/{oye:g} morphemes which we previously discussed, although
it is certainly tempting to imagine an ultimate relation between
them.

3)  In the phonological literature on Klamath the rule is often
stated as applying to suffixes in general.  In fact, however, the
only suffixes which can occur directly after a classificatory
prefix are DMS's and recently-grammaticalized DMS's (i.e. classes
17-19), so it is only these than can manifest this behavior.

4)  The careful reader will note the many tenuous links in the
chain of inference which now leads from the fact that {damn}
'doing over and over' is evidently deverbal to the implied
suggestion that {wabg}, by its membership in this same broad
category, is a recent addition to the suffixal complex.  Note,
however, that it shares with the other members of this category
the stigmata of recent grammaticalization, viz. phonological bulk
and complexity and semantic specificity.

5)  In the next stage of internal reconstruction the question of
which position classes reduplicate becomes one of considerable
interest.


References

Barker, M.A.R.  1963. Klamath dictionary.  UCPIL 31.
     . 1964. Klamath grammar. UCPIL 32.
DeLancey, Scott. 1988. Morphological parallels between Klamath
     and Wintu.  J. Redden, ed., Papers from the 1987 Hokan-
     Penutian Languages Workshop and Friends of Uto-Aztecan
     Workshop, pp. 50-60. (So. Ill. U. Occasional Papers in
     Linguistics 14). Carbondale: Dept. of Linguistics, U. of
     Southern Illinois.
     . 1989. Klamath stem structure in genetic and areal
     perspective. Papers from the 1988 Hokan-Penutian Languages
     Workshop, pp. 31-39.  Eugene: Dept. of Linguistics, U. of
     Oregon.
     , C. Genetti, and N. Rude. 1988. Some Sahaptian-Klamath-
     Tsimshianic lexical sets.  W. Shipley, ed., In honor of
     Mary Haas, pp. 195-224.
Gatschet, Albert. 1890. The Klamath Indians of Southwestern
     Oregon, Part I.  Contributions to North American
     Ethnology, vol. II. Washington: Govt. Printing Office.
Rude, Noel. 1987. Some Klamath-Sahaptian grammatical
     correspondences.  Kansas Working Papers in Linguistics
     12:67-83.
White, Robin. 1973. Klamath phonology.  Ph.D.
     dissertation, University of Washington.