The Bipartite Stem Belt: Disentangling Areal
                   and Genetic Correspondences

                         Scott DeLancey
                      University of Oregon

     Even where genetic relationship is clearly indicated
     ... the evidence of diffusion of traits from
     neighboring tribes, related or not, is seen on every
     hand.  This makes the task of determining the validity
     of the various alleged Hokan languages and the various
     alleged Penutian languages all the more difficult ...
     [and] point[s] up once again that diffusional studies
     are just as important for prehistory as genetic studies
     and what is even more in need of emphasis, it points up
     the desirability of pursuing diffusional studies along
     with genetic studies.  This is nowhere more necessary
     than in the case of the Hokan and Penutian languages
     wherever they may be found, but particularly in
     California where they may very well have existed side
     by side for many millennia.  (Haas 1976:359)

The term "bipartite stem" (Jacobsen 1980) refers to a pattern of
compound stem construction found in northern California and
Oregon, which crosses genetic boundaries, occurring in Hokan and
Penutian languages, but also seems to roughly correlate with
plausible genetic subunits (Northern Hokan, Plateau Penutian +
Maiduan).  While the pattern has clear functional analogies to
patterns of stem construction in Salish, Wakashan, Algic, and
other North American languages, its specific structural
manifestations clearly distinguish the languages of what I am
calling the "Bipartite Stem Belt" from those to the north and
south of it.  Different aspects of this bipartite stem pattern
appear to be different ages, so that their distribution in the
area suggests a chronology of various prehistoric contacts.  This
evidence is consistent with other linguistic and archaeological
evidence, and can potentially be used to bolster and complement
recent suggestions about Penutian dispersal.
     The pattern shows the most elaborate development in a "core
area", a set of languages including at least Washo, Klamath,
Sahaptin, Yana, and Atsugewi.  (Some or all of Molala, Achumawi,
and Shasta very likely belong to this core area as well, but the
situation is not clear from the data available to me).  Elements
of the pattern are shared in many surrounding languages,
particularly Nez Perce, Maiduan, Pomoan, and to an extent Numic;
this larger set of languages is what I mean by the Bipartite Stem
Belt.

1    Elements of an areal pattern

The areal pattern which we are discussing has been recognized
since the beginning of systematic California linguistics (Dixon
and Kroeber 1919, Sapir 1916, Taylor 1961, Jacobsen 1966, Sherzer
1976, DeLancey 1988).  Discussion of its defining features is
often couched in terms of "instrumental prefixes" and "locative(-
directive) suffixes".  While they have the advantage of
familiarity, these terms are misleading in two important ways--in
the assumption of greater (in the case of the "instrumentals") or
less (in the case of the "locatives") semantic specificity than
the categories of the languages actually manifest, and in the
implication, contained in the words "prefix" and "suffix", that
we are dealing with a clearly grammatical as opposed to lexical
phenomenon.  At the most general level, we are better to refer
simply to initial and final stem elements, or simply "initials"
and "finals".  For the more specific grammatical categories which
define the bipartite stem pattern I will refer to "lexical
prefixes" (LP's) and "locative-directive stems" (LDS's).

1.1  Lexical prefixes

The principal feature which has historically been recognized as
subject to areal spread in Northern California is a set of
initial stem elements, usually but not always analyzed as
prefixes.  In all of the languages which have this category some
or all of the members can have reference to the shape of an
instrument, and the category is traditionally referred to as
"instrumental prefixes".  But in all languages for which I have
data some members of the category can also refer to the shape of
a Theme argument, and in the more elaborated systems
characteristic of our area bound stems referring to manner of
motion also occupy this same positional slot.  In the core
languages the positional category also includes a motley set of
bound stems with no discernable semantic connection to any of
these fields.
     We find the closest approximation to a truly "instrumental" 
category in languages like Maiduan, Takelma, and Numic.  In all
of these most of the prefixes or initial stem elements index a
body part or the shape of an object, but some express more
abstract adverbial notions, while the shape classifiers may index
an instrument or some other kind of argument, typically a Theme
(DeLancey 1991a).  For example, in Konkow (Maiduan) /b˘-jolmŤ:n/
'bash w. a rock-like instrument', the prefix |bV| classifies an
instrument, but in /b˘-ki:n/ 'rock-like obj. is lying, put rock-
like obj. down' it indexes the Theme of a predicate referring
indifferently to location or motion (intransitive or transitive). 
Likewise in Takelma the 17 body-part prefixes often express the
instrument with a transitive verb, but they can also index object
arguments of various semantic functions, and several of them also
have semantically extended functions reminiscent of the
"adverbial" LP's of Sahaptian or the adverbial functions of some
LP's in Klamath.  A similar pattern is reported for the Numic
languages, where similar-sized sets of prefixes, primarily but
not exclusively referring to body parts, have a semantic range
extending beyond instrumentality to include "liquid, heat, and
cold ... degree or direction of force, location, transitivity,
and also other grammatical features" (Nichols 1974:155).  Both
the Takelma and the Numic categories are directly relatable to
noun-incorporation constructions, though the inventory
(particularly in Numic languages) will include morphemes whose
nominal origins are quite opaque.  It is worth noting here a
conclusion that I will suggest later, that the existence in a LP
system of a certain number of members whose lexical origin is
transparent is not in itself an argument for the recent origin of
the category, as languages with such a category tend to continue
to recruit new members, and to lose old, resulting in a slow
relexicalization of what may otherwise be relatively stable
categories.  A very similar pattern, as distant genetically and
geographically from either of these as they are from one another,
is found in Pomoan.  Again, the size of the stem-initial category
is quite similar, and the semantic range quite comparable, to the
languages we have just discussed (Oswalt 1976:15-7).
     In the core languages, we find much larger and more
elaborated sets of initial stem elements, of sufficient semantic
variety to make it impossible to retain the term "instrumental"
even for its sentimental associations.  In Klamath, for example,
there is a category of initial stem element which, while it
includes a number of morphemes referring to body parts or shape
categories as instruments, also includes elements which index the
shape or other semantic category of a Theme argument, morphemes
referring to manner of motion, and a miscellaneous set with no
particular semantic coherence.  Since no more specific term
accurately describes the set, I adopt Jacobsen's (1980) term
"lexical prefix", adapted from Salishan studies (where "lexical
suffix" refers to a functionally similar category of stem-final
elements), for the equivalent set of initials in Washo.  These
"prefixes" appear to have more than one origin; some are quite
clearly verbs, and the bipartite stems which they form are at
least etymologically best thought of as verb compounds.  Jacobs
specifically notes the same thing of the Sahaptin LP category:

     it cannot be maintained that anterior roots are true
     prefixes; tagging them prefixes or alternatively,
     anterior roots or quasi-roots, does not properly
     describe their usage, which varies from nearly genuine
     prefixation to true compounding and in some cases
     complete root independence.  (1931:153-4)

Even for the much smaller and more cohesive LP category in
Maiduan, the analysis of the category is subject to debate: 
Dixon (1911) and Ultan (1967) label them prefixes, but Shipley
(1964) considers the Maidu LP's "root morphemes", while noting
that this is as much a historical as a synchronic question.
     A very intriguing pattern is that the LP's in many of these
languages share similar phonological profiles and behavior.  They
are typically phonologically light, and in a geographically and
genetically diverse range of languages (Nez Perce, Klamath,
Washo, Maiduan, Pomoan) they undergo some degree of vowel harmony
with the stem to which they are attached--even when the language
does not otherwise show vowel harmony.  This fact definitively
ties the languages that share it into an areal complex; we can
imagine two languages independently, or semi-independently,
innovating an instrumental or more general lexical prefix
category--a noun incorporation construction would be a typical
source--but there is no imaginable reason why Pomoan and Maiduan,
for example, would both independently develop a vowel harmony
rule applying only to this category.

1.2  Locative-directive stems

Languages of this area are rich in lexicalizations of this
subcategory of motion verb, representing elaborate
classifications of the semantic field of direction, path, and
location (as opposed to types and manners of motion like 'go',
'walk', 'run', etc.).  Stems representing this semantic category
are not, of course, necessarily an overtly distinct grammatical
category, but one of the characteristic features of the bipartite
stem pattern is some degree of grammaticalization of verbs of
this type.
     We can illustrate the specialization of the category with
Chinookan, which is otherwise not a participant in the bipartite
stem pattern.  Dyk (1933) describes the existence in Wishram of
what are clearly in some sense bipartite stems:  "rather loose
combinations of two-stems ... easily broken up in their component
parts each of which is freely combined with others" (1933:76). 
(He also notes the existence of "several" compound stems one or
both elements of which are unique to them; a feature widespread
among languages of the Bipartite Stem Belt proper).  The order of
elements is fixed, that is, there is a set of initial and another
of final compounding elements, with no overlap.  Members of each
class can also occur as independent simple stems.  Initial stems
are apparently a large open class, but there are only something
like 40 final elements, and only half of these combine
productively.  These also occur independently; all are
"intransitive verbs of motion with specific directional
connotations" (p. 77) which as finals combine only with primary
motional stems.  In compound stems they are exactly parallel to
the locative-directive "suffixes" of Klamath or the "posterior
roots" of Sahaptin.
     Examination of Aoki's magnificent dictionary (1994) of Nez
Perce provides a very similar picture.  There are about three
dozen verb stems which he states occur only preceded by a LP. 
About half of these have meanings related to motion or direction. 
As in Wishram, they combine predictably with motional initials,
but they also occur with LP's with other (e.g. shape-classifying)
meanings:

     (l) hsa 'to (go) up', as in |co-l hsa-| 'lift a pole'
     ({cŁ:-} 'with a pointed object or a pole-like object),
     |wa:-l hsa-| 'fly, jump up; ({we:-} 'fly, move fast'), etc.

Almost all of these begin with a morphophonemically variable /l/,
which may be the remnant of an older morphological mark of verb
concatenation.  Several of these are transparently analyzable,
showing recent expansion of the category; for example {(l) htq'i}
'out of water; up and out', analyzable into {l hsa} 'up' and the
verb stem {taq'ˇ:} 'out of water, come to shore'.
     In Klamath we find a much larger set of locative-directional
stems, with well over 100 members.  Except for their number,
these are very like the corresponding category in Nez Perce: they
are bound morphemes, occurring always as the final element of a
compound stem (though certain of them can occur after others);
they have unique morphophonemic behavior, each having a
characteristic initial vowel which surfaces only with certain
initial stem elements (see DeLancey 1991); and they combine
freely with both motional and shape-classifying initials.  As in
Nez Perce, a few members of the category are clearly recent
developments, showing that the category has continued to expand
over time, but most are synchronically opaque and not related to
any independent stem.
     The Nez Perce LDS's clearly represent a highly
grammaticalized category:  a relatively small (15-20) set of
bound morphemes, with characteristic morphophonemic behavior. 
The Chinookan category is less grammaticalized; though it is
about the same size, the fact that its members occur as free as
well as bound stems immediately makes us think more of lexical
stems than grammatical affixes.  The Klamath pattern deviates in
the other direction; as in Nez Perce, these are a closed class of
bound morphemes, with peculiar and characteristic morphophonemic
behavior--but there are ten times as many, with concomitant
semantic variety and specificity.  Nevertheless it is clear that
we are looking here at essentially the same phenomenon, and,
given the geographical proximity of the historic distribution of
the languages, it is hardly imaginable that its occurrence in all
of them is coincidental.
     Here in a nutshell we can see the problem with the
traditional "stem" vs. "affix" model, as several scholars have
recently noted (Jacobsen 1980, Langdon 1990).  In Wishram, there
can be no question of the verbal nature of the LDS's, since they
occur freely as motion stems in their own right.  Aoki has no
choice but to analyze the Nez Perce stems as verb stems, despite
their bound status, because they fall into exactly the same two
stem classes as all other verbs.  Barker calls the Klamath
category suffixes, but in fact the same argument can be made as
in Nez Perce for considering them to be ultimately verbal in
nature.  As we will see below, while Klamath lacks the luxuriant
stem alternations of Yok-Utian or Takelma, it does show two very
interesting irregular stem classes:  -i'-stems and n-stems.  Both
of these are probably quite old; the n-stems probably represent
inheritance from Proto-Plateau, and the -i'-stems very possibly
from Proto-Penutian.  Both categories are well-represented among
the locative-directive "suffixes", which thus in a fundamental
and organic way behave like other verbs in the language.

1.3  A taxonomy of bipartite stem types

The typological complex which we have to sort out here is more
complicated than simply a matter of possessing both instrumental
prefixes and locative-directive suffixes.  The core languages are
characterized by the fact that a majority of the verb stems of
the language are bipartite, consisting of a LP and a second
element.  The second element is not always a LDS, but typically
most or all of the potential second elements in a language are
bound morphemes which cannot occur without a LP (raising further
problems for a simple stem vs. affix analysis).  In Washo, Yana,
Klamath, and Sahaptin, at least, there are several different
kinds of bipartite stem, depending on the type of LP and second
element.  I will use Klamath data to illustrate; Jacobs'
description of Sahaptin, Jacobsen's of Washo, and Sapir's of Yana
mention all of these, and each language has one or two additional
minor types.  Talmy (1974), in his description of Atsugewi, works
harder to reduce the semantic range to a unified, rather abstract
principle, but examination of his examples suggests that the
overall system is similar.
     Klamath has about 500 simplex stems, but the majority of
stems consist of two (or occasionally more) bound elements. 
These bipartite stems are of four types (DeLancey 1995ms).  LP's
referring to instruments or actions combine with stems referring
to a change of state.  Most of these change-of-state stems can
occur unprefixed, in which case they are intransitive; with an
instrumental LP they are transitive, e.g. /ntew'-/ 'break,
shatter a surface with a round instrument' ({n-} 'round
instrument', {tew_i'} 'thin surface to break, shatter').  LP's
referring to the shape or other semantic category of a Theme
argument combine with bound LDS's to create stems, which can be
used both intransitively and transitively: e.g. /lew-/ 'puts a
round obj. into water, flat place (intr. also)' ({lV} 'round
object', {ew} 'in(to) water, a flat place').  LP's referring to
manner of motion also combine with LDS's, creating complex motion
verbs. e.g. /howw-/ 'run, jump into water, flat place' ({ho_d}
'sg. run, jump', {ew} (as above)).  We must also recognize a
miscellaneous category of LP's or combinations of LP and final
which do not fit any of the above categories: /dalmni/ 'looks up'
({dEl} 'look' and {amni} 'up, upstream, uphill'), /lGacw-/ 'has a
stripe on the hair, head' ({lGE} 'striped', {acw} 'on the head,
hair').  There is also a significant number of bipartite stems
one element of which occurs only in that stem and no other, and
other bound stem elements which do not easily fit into any of the
general classifying, instrumental, motional, or change-of-state
or LDS categories.

2    A comparative problem: Plateau Penutian and Maiduan

The classic Sapirean version of Penutian recognizes three major
groupings in our area:  Californian, consisting of the four
California stocks (Miwok-Costanoan, Yokutsan, Wintuan, and
Maiduan), Oregon Penutian, consisting of Takelma, Kalapuya, and
the Coast Oregon languages, and Plateau, consisting of Sahaptian,
Molala-Cayuse, and Klamath-Modoc.  (Chinookan, Tsimshianic, and
"Mexican Penutian" were the other branches (1929/1990)).  While
there is increasing evidence for the validity of something like
Sapir's Plateau branch, the "California Penutian" unit, inherited
from Dixon and Kroeber's pioneering work, has turned out to be a
phantom.  That is not, of course, to say that there is no
evidence that these languages are related--but there is no
factual basis for a claim that they compose a unified subgroup of
Penutian to the exclusion of other putatively Penutian languages
(cf. Hymes 1964), and it is now established that Wintuan
(Whistler 1977) and Maiduan (Shipley and Smith 1977) represent
separate, independent migrations, almost certainly from different
proximate origins; very possibly they are related at no shallower
level than Proto-Penutian.

2.1  Klamath as a Penutian Language

I will not discuss lexical evidence for a Penutian affiliation
for Klamath and Sahaptian here, except to assert that there is as
much evidence for the relationship of Klamath to Yokuts, and for
Klamath to Maiduan, as for Yokuts and Maiduan, and thus no
argument on the basis of lexical evidence for a special
relationship between Maiduan and Yokuts which excludes Klamath. 
Moreover, Klamath shows some fossilized structural features which
make it look more Penutian than is sometimes assumed.  As
Silverstein points out, the evidence from Takelma and Yok-Utian
suggests that, in a language related to these, we should expect
"morphological debris found at the end of stems, irregularities
in lexical form under derivation and inflection" (1979:660). 
Although Barker's presentation of his data in terms of
"morphophonemes" obscures the similarities to other languages,
Klamath does in fact have stem classes of verb characterized by
irregularity in their derived and inflected forms.  The oldest
and most interesting from a comparative Penutian point of view
are the glottal and i-stems.  A number of Klamath verbs have a
final /i/ before consonants, but not vowels.  Most of these also
have the final glottal morphophoneme which Barker writes as |'|. 
This glottalizes a preceding sonorant when the |_i| is not
realized; thus with the stem {qew_i'} 'break in two' we get forms
like /qew'a/ 'break in two', with the |i| lost before a vocalic
suffix, and the stem sonorant therefore glottalized, vs.
/qe:witk/ 'pl. broken, destroyed', with the |i| retained before a
consonant, and the glottalization therefore lost.  While we
cannot directly equate this irregularity with similar patterns in
other languages unless we can identify specific cognate stems, it
is intriguingly reminiscent of stem alternations in Yokuts.

2.2  Klamath and Sahaptian

The genetic relationship between the Sahaptian family (Sahaptin
and Nez Perce) and Klamath is recognized even by conservative
Penutian skeptics as very plausible, and considerable lexical and
grammatical evidence has been presented for it (Aoki 1963, Rude
1987, DeLancey, Genetti and Rude 1988, DeLancey 1991b).  These
languages also share both the bipartite stem pattern and several
specific roots participating in it, both as initials and finals.
     For example, Nez Perce {wŚl‚:} 'run, move quickly', which
Aoki (1970) analyzes as a LP, and Klamath {wle} 'run (few, four-
legged animal)', which Barker analyzes as a bound verb stem which
requires a LDS, in fact have identical compositional behavior,
occurring in both languages only in compounds like:

     Nez Perce: /wila-l hsa-sa/ 'I am running uphill' ({l hsa}
     'to (go) up'), /wile-l‚hne-ce/ 'I am running down' ({l‚hne}
     'down, downward, downhill'), etc.

     Klamath: /wle-qwe:L-/  'few, four-legged animals run down a
     hill' ({eqwe:L} 'down the hill, out of a tree, downslope'),
     /wle-Yanc'-/ 'few, four-legged animals run along the edge of
     a cliff, along a bank' ({oY_n} 'along a mountainside, cliff,
     twisting riverbank;), etc.

Given the correspondence in form and meaning and in position
class and combinatorial behavior, the two stems are indubitably
cognate.  (This is one of the most widely attested and best known
pan-Penutian etyma; cf. Shipley 1966).  We have good reason to
suppose that their status as bound motional stems which occur
with grammaticalized LDS's is likewise common inheritance from
Proto-Plateau.
     But this hypothesis entails that the existence of a
specialized class of LDS's occurring as bound final elements of
bipartite stems is common inheritance, which in turn implies that
at least some members of this category in the two languages
should be cognate.  And examination of the bound second-position
stem elements with which this etymon combines in the two
languages does turn up a few plausible cognates; the most
promising that I have noted are:

     NP {(l)eyl‚:k} 'into (typically into a hole)'; Kl. {(o)ne:g}
     'in(to) a hole'

     NP {(l) htq'i} 'out of water; up and out', composed of
     {l hsa} 'up' and {taq'ˇ:} 'to come out of water, to come to
     the shore', cp. Kl. {(o)tq'ag} 'up out of (as water, dirt, a
     hole, etc.)'; also possibly related here are NP {(l)‚ht}
     'out'; Kl. {(o)dg} 'out of a container', {(o)dG} 'taking
     away, removing, depriving', 

     NP {(l)‚hyek}, 'upstream, upriver' and {letˇyek} 'high'; Kl.
     {(o)ye:g} 'up, raising, lifting'

     NP {lawˇ:} 'to leave (sometimes to get food)'; Kl. {(o)wi}
     'spreading out, scattering'

However, four plausible sets (and a few more speculative
similarities) among the 15 or so LDS's in Nez Perce, and the 125
or so in Klamath, is not an impressive showing.  If the sets
given here are valid, then, since they correspond in grammatical
category as well as form and meaning, they constitute sufficient
basis to attribute this bound category to the proto-language. 
Nevertheless, the small number of cognates, and the huge
disparity in size between the sets in the two languages,
demonstrates considerable independent secondary development in
the daughter languages.
     Thus though there is some reason--in the overall structural
parallelism and the close correspondence of isolated elements of
the system like *wile--to attribute parts of this stem
composition system to a common ancestor of Sahaptian and Klamath,
it is nevertheless clear that there have been extensive
independent developments in the languages since their separation,
including in particular the innovation in one or (almost
certainly) both of many new members of the various bound stem
classes.
     In fact, the same is true even within Sahaptian. It is
easier to find strong resemblant pairs of LP's or bound LDS's
between Nez Perce and Sahaptin than for either of those and
Klamath, but still a very substantial number of both initials and
finals are unique to one Sahaptian language or the other.  Some
sets establish specific independent developments in the two
languages.  For example, Aoki (1994) lists as a bound LDS in Nez
Perce {(l)‚hyek} 'upstream, upriver', and an independent verb
stem {tol y} 'go upstream'.  In Jacobs 1931 we find the Sahaptin
"posterior root", i.e. a bound stem, {-tuni-} 'upstream', cognate
to the Nez Perce independent stem.  Thus we see that the two
languages have chosen different roots to grammaticalize for this
meaning.  This is a recurrent pattern in the languages of the
Belt:  indications that the LP's and LDS's are quite old as
grammaticalized categories, together with data showing continual
expansion of the categories and replacement of forms over time.

2.3  Plateau Penutian and Maiduan

From the earliest days of Penutian research Maiduan has been
recognized as distinctly unlike the other (California) Penutian
languages in its use of instrumental prefixes and locative-
directive suffixes, both unknown in any of the other four
California groups.  Both Dixon and Kroeber (1919) and Sapir
(1916) ascribe this to contact with the northern Hokan languages
Washo, Achumawi, Atsugewi, and Yana.  Once the California
Penutian hypothesis is abandoned, however, precisely these facts
about Maiduan stand out as a strong structural correspondence
with the Plateau languages.  Both Klamath and Sahaptian have
extensive series of initial instrumental or adverbial and final
LDS elements.  There is a small body of promising lexical and
grammatical comparisons suggesting a special connection within
Penutian between Maiduan and Klamath and Sahaptian, a possibility
which would fit well with Shipley and Smith's (1979) evidence for
relatively late migration of Maiduan into its historic area.
     While Maiduan shares LP's, grammaticalized LDS's and
bipartite stems with the Plateau languages, the fact that these
are also shared with its Hokan neighbors should cause us to
proceed with caution in attributing the similarities to cognacy,
since it raises the possibility that the development of the
pattern in all of the languages is an areal phenomenon which
postdates the breakup of their nearest common ancestor.  There
are structural differences: Maiduan languages have only 15-20
members of each of LP's and LDS's.  Thus its LDS inventory is
comparable to that of Nez Perce, but much smaller than Klamath or
Sahaptin.  From available descriptions Maiduan does not seem to
have a large store of bound initials of any type.  Shipley
(1964:38-9) notes that Maidu does have compound stems, including
stems with incorporated nouns and compounds of two verb stems as
well as stems involving a LP.  Still, from the available data,
Maiduan seems to have a significantly less extensive bipartite
stem system than the Plateau languages.
     Closer examination, however, does turn up some parallels. 
For example, Maidu {wel‚} 'run' occurs, like its evident
Sahaptian and Klamath cognates, as the first member of bipartite
stems with a second directional element: /wel‚-doj/  'run up (a
hill, a staircase)', /wel‚-no/ 'run, run along', /wel‚-sito/ 'run
across', etc.  Thus it shares combinatorial behavior as well as
form and meaning with the Plateau forms.  Unlike these, however,
it apparently also occurs unsuffixed: Shipley (1963:185) lists
the form /h n-welŠ/ 'run holding something in the arms' ({h n}
'transport by carrying in the arms'.  No corresponding
construction with this root would be possible in Klamath or
Sahaptian, where it occurs only as an initial with a LDS final.
     We have, then, no grounds to posit *wile as a bound initial
for the common ancestor of Plateau and Maiduan, as we tentatively
can for Proto-Plateau.  Still, the similar behavior and
corresponding order are significant of something, as we can see
by comparing it with the functionally equivalent construction in
Takelma, also putatively Penutian and spoken at no great distance
from the other languages.  Takelma has a set of grammaticalized
LDS's, similar in size and semantic range to the corresponding
categories in Nez Perce and Maiduan (Sapir 1912/1990).  But in
Takelma these are initial rather than final elements of the stem. 
Thus built on hi-wiliw-, the Takelma cognate to Plateau
*wile, we have compound stems like dal-hiwilii-gw- 'run
off into brush with', with the locative-directive prefix
dal- 'away into brush, among, between' and
xam-hiwˇliu? 'ran to the water' (xam- 'in river'). 
Thus, regardless of to what extent it might be possible to equate
particular Takelma morphemes with Sahaptian, Klamath, or Maiduan
forms, the compound constructions of these languages on the one
hand and Takelma on the other must represent independent
developments.
     There are a few forms among the LP and LDS series in Maiduan
with plausible Plateau connections, most notably the instrumental
prefix {w} 'sticklike instrument', cp. Kl. {w} 'with a long
instrument', Nez Perce {we} 'with chopping instrument', {wet}
'stick-like object'.  This even appears in a couple of possibly
cognate bipartite stems.  Both parts of Maidu /wyk'y't-daw/ 'cut
off' and Kl. /w-gatt'-/ 'chop in two, chop down' correspond well,
and only slightly weaker is Konkow /wˇ-c'it'-in/ 'split a small
obj. with an instrument', Kl. /w-cic'-/ 'split something thin
with a long instrument'.
     This list of comparisons pertaining to the bipartite stem
system could be extended a bit with a bit more space, and
probably slightly further with a bit more research, but it is
clear that by far the greater part of the modern bound categories
in Maiduan, Sahaptian, and Klamath are secondary independent
developments.  But once we have even a few substantive
comparisons of this sort, we have sufficient basis to reconstruct
some form of the bipartite stem pattern for a common ancestor of
Plateau and Maiduan.  However, we cannot reconstruct anything
like the luxuriant development of that pattern which we find in
Klamath and Sahaptin--indeed, it is doubtful whether that can be
reconstructed even for Proto-Plateau.  At most we can reconstruct
a more modest system like that found in the Maiduan languages,
with a few LP's and a few grammaticalized LDS's.  Very possibly
even the latter is a later development; we could explain the
parallels that we observe if we reconstructed only a somewhat
productive construction involving concatenation of motion stems,
with manner stems preceding locative-directives.

3    Toward a prehistory of the bipartite stem pattern

We have seen that those aspects of the bipartite stem pattern
which Klamath and Sahaptin, languages of the core area, share
with Maiduan can plausibly be attributed to a common ancestor
which they share at a (considerably) lower level than Proto-
Penutian.  But, of course, those aspects of the pattern which are
shared with the other core languages, all Hokan, cannot be so
interpreted.  From this we can argue that the attested system of
Klamath and Sahaptin must have developed in at least two stages. 
The initial innovation or adoption of the basic pattern--a LP
category, specialized LDS's, and compounding of motion verbs into
bipartite stems--must precede the breakup of Plateau-Maiduan,
while the dramatic expansion of the grammaticalized categories
and the dominance of the bipartite pattern in the verbal lexicon
must postdate it.  Moreover, we must suppose a geographical
movement of Maiduan in the interim, to remove it from the core
area where the areal efflorescence of the bipartite stem pattern
took place.
     As we have noted, something like the core pattern is found
in Washo, Atsugewi, and Yana, probably Achumawi, and perhaps
Shasta and Chimariko.  Except for Washo, all of these belong to
Sapir's postulated Northern Hokan branch.  The pattern is not
found throughout the branch, however; Karok definitely does not
belong, though Haas (1980) discovered fossilized evidence for an
earlier instrumental prefix category.  There is no evidence in
Haas' data of an LDS category: the second elements which she is
able to reconstruct are almost all change-of-state or other
manipulative predicates, and none appear to be locative-directive
elements.  Pomoan has a well-developed but relatively small (~20)
LP category, and a significant number of bipartite stems, but I
don't know to what, if any, extent there is a clearly
grammaticalized LDS category.
     So we have some elements of the pattern throughout the Hokan
languages of Northern California, though it appears that Pomoan,
and even more Karok, behave as outliers to the area in this
respect.  The most significant, geographically non-marginal
exception to this picture of northern California and eastern
Oregon is Wintuan, as Sapir noted long ago (1916).  Wintuan has
no trace of anything like our LP category.  It does have a highly
grammaticalized LDS set, but, as in Takelma, these are initial
rather than final elements of the stem.  This structural
correspondence to Takelma is striking in light of the evidence
given by Whistler (1977) and Golla (1993) for a relatively recent
northern origin for the Wintuan languages.
     Though it is still rough, this assemblage of facts suggests
a chronology of areal developments and population movements.  We
can posit a time when the Northern Hokan languages (regardless of
whether or not this turns out to be a valid genetic grouping)
were all spoken in a compact and contiguous area, with Proto-
Plateau-Maiduan and ancestral Washo nearby, to the northeast and
east, respectively.  At an early date we can link Takelma with
this areal complex as well.  This is the period when the LP
construction, whatever its origin, spread through all of the
languages.  The absence of any evidence of LP's in Wintuan
implies that at this date it was spoken still farther north than
the point of origin from which it moved into California.  This
hypothetical areal pattern must be quite old, and it is much less
well-defined by synchronic data than the subsequent stages.
     The next areal wave, involving grammaticalization of LDS's
and the development of bipartite stems not involving
instrumental/classifying prefixes, apparently excluded Karok and
Takelma; where Pomoan stands in this development is not entirely
clear, but my current interpretation is that it participated to
some extent in this second stage.  This again would appear to
have been early.  The comparative facts outlined above suggest
that this stage was still prior to the breakup of the common
ancestor of Plateau Penutian and Maiduan, which implies that, if
Northern Hokan including Karok is a genetic unit, it is of
somewhat greater time depth than Plateau-Maiduan; it specifically
implies that the split of Hokan, and perhaps Pomoan, from
Northern Hokan preceded that of Maiduan from Plateau.
     The third wave of areal influence is the hypertrophy of the
bipartite stem system in the core languages.  As we have seen,
this affected Washo but apparently missed Maiduan; it also seems
to have affected Sahaptin somewhat more than Nez Perce.  This
obviously places this wave subsequent to the breakup of Plateau,
and probably even of Sahaptian.  It also has some implications
for the geographical relationship of these languages at this
time: Sahaptin and Klamath must be in close contact, with Nez
Perce somewhat peripheral.  This is more-or-less the historic
situation.  It also places Maiduan at some geographical remove
from Klamath and Sahaptin, which again is consistent with
historic locations.  But Washo, presumably, must have been
farther north, in closer contact with Atsugewi and Klamath, than
it is at present.  Pomoan clearly was not a participant in this
development; if it should turn out that Achumawi and/or Shasta
are, like Atsugewi and Yana, core Bipartite Belt languages, that
implies that this last wave postdated the incursion of Wintuan
into the Central Valley--an event which would have insulated
Pomoan from areal developments to the east of the Valley.
     One problem with this outline involves Numic, all the
languages of which have a modest lexical prefix system comparable
to those of Maiduan or Pomoan.  If I want to attribute the rest
of the distribution of this category to an areal spread, then
presumably Numic must have been a participant in this complex. 
But I have just argued that this stage of the areal development
occurred quite early, while Uto-Aztecanists in general consider
the Numic expansion northward through the Great Basin to be quite
recent.  If the Numic facts are not to constitute a threat to my
hypothesis, I need to have the Numic languages--or, more likely,
a relatively undifferentiated late Proto-Numic dialect complex--
residing far enough north to participate in the areal pattern at
least 2,000 years ago, and probably a good bit earlier.  There is
controversy here already; I will only point out here that many
archeologists find the posited recent spread and divergence of
Numic quite problematic.  For example, Aikens and Witherspoon
(1986) place ancestral Numic in the desert areas of the Great
Basin (which were not constant over time) from shortly after the
initial breakup of Proto-Uto-Aztecan.  This model fits much
better with my data, and would lend itself to the interpretation
that Numic might well have been the original source for the
lexical prefix pattern.  Jacobsen (1966) points out that a Numic
dialect complex remaining relatively stable over a long period of
time, while an unorthodox idea among linguists, is not
necessarily implausible.  (In this connection he also suggests
Numic as a possible source for the lexical prefix pattern in
Northern California, which in light of the present research
remains an intriguing possibility).

Aikens, C. Melvin, and Y. T. Witherspoon. 1986. Great Basin Numic
     prehistory:  Linguistics, archeology, and environment.  C.
     Condie and D. Fowler, Anthropology of the desert
     West. 8-20.  Salt Lake City: University of Utah Press.
Aoki, Haruo. 1963. On Sahaptian-Klamath linguistic affiliations. 
     IJAL 29:107-12.
_____. 1970. Nez Perce texts.  UCPIL 62.
_____. 1994. Nez Perce dictionary.  UCPIL 122.
Barker, M.A.R. 1963. Klamath dictionary.  UCPIL 31.
_____. 1964. Klamath grammar. UCPIL 32.
DeLancey, Scott. 1988. Klamath stem structure in genetic and
     areal perspective.  S. DeLancey, Papers from the 1988
     Hokan-Penutian Languages Workshop, 31-39.
_____. 1991a. Event construal and case role assignment.
     BLS 17:338-53.
_____. 1991b. Chronological strata of suffix classes in the
     Klamath verb.  IJAL 57:426-45.
_____. 1992.  Klamath and Sahaptian numerals. IJAL 58:235-
     39.
_____. 1995ms. Lexical prefixes and the bipartite stem
     construction in Klamath.
_____. 1996. Argument structure of Klamath bipartite stems. 
     presented at the SSILA meeting, San Diego.
_____, C. Genetti, and N. Rude. 1988. Some Sahaptian-Klamath-
     Tsimshianic lexical sets.  W. Shipley, In Honor of Mary
     Haas:  From the Haas Festival Conference on Native American
     linguistics, 195-224. New York: de Gruyter.
Dixon, Roland B. 1911.  Maidu.  F. Boas, Handbook of American
     Indian languages, part I, 679-734.
_____, and A.L Kroeber. 1919. Linguistic families of California. 
     UCPAAE 16(3):47-118.
Dyk, Walter. 1933. A grammar of Wishram.  Ph.D.
     dissertation, Yale University.
Golla, Victor. 1993ms. Alsea (& Siuslaw, Coosan) - Wintu (&
     Patwin) comparisons.
Haas, Mary. 1980.  Notes on Karok internal reconstruction.  Klar
     1980:67-76.
Hymes, Dell. 1964. Evidence for Penutian in lexical sets with
     initial *c- and *s-.  IJAL 30:213-42.
Jacobs, Melville. 1931. A sketch of Northern Sahaptin
     Grammar.  U. Washington Pub.in Anthropology 4.2:85-292.
Jacobsen, William. 1966. Comments on linguistics.  W. d'Azevedo
     et. al., The current status of anthropological research
     in the Great Basin: 1964, 259-64. Reno: Desert Research
     Institute.
_____. 1980. Washo bipartite verb stems. Klar 1980:85-99.
Klar, Kathryn, et. al. 1980. American Indian and Indoeuropean
     Studies:  Papers in honor of Madison S. Beeler.  The
     Hague: Mouton.
Langdon, Margaret. 1990. Morphosyntax and problems of
     reconstruction in Yuman and Hokan.  P. Baldi, Linguistic
     change and reconstruction methodology,  57-72. NY:
     Mouton de Gruyter.
Nichols, Michael. 1974. Northern Paiute historical
     grammar.  Ph.D. dissertation, UC Berkeley.
Oswalt, Robert L. 1976.  Comparative verb morphology of Pomo.  M.
     Langdon and S. Silver, eds., Hokan Studies, pp. 13-
     28.  The Hague: Mouton.
Pitkin, Harvey. 1984.  Wintu grammar.  UCPIL 94.
Rude, Noel. 1987. Some Klamath-Sahaptian grammatical
     correspondences.  Kansas Working Papers in
     Linguistics 12:67-83.
Sapir, Edward. 1912. The Takelma language of Southwestern Oregon.
     F. Boas, Handbook of American Indian languages, vol.
     2, 1-296. (repr. V. Golla, The collected works of
     Edward Sapir VIII: Takelma texts and grammar. NY: Mouton
     de Gruyter).
_____. 1916.  Time perspective in aboriginal American culture:  A
     Study in Method.  Canada Department of Mines, Geological
     Survey, Memoir 90, Anthropological Series No. 13. (repr. D.
     Mandelbaum, Selected writings of Edward Sapir. Los
     Angeles: U. of California Press, 1949).
_____. 1922. The fundamental elements of Northern Yana. 
     UCPAAE 13:215-34.
_____. 1929. Central and North American Languages. 
     Encyclopaedia Britannica, 14th ed., vol. 5, 138-141.
     (repr. W. Bright, Collected works of Edward Sapir, vol.
     5, 95-104. NY: Mouton de Gruyter).
Sherzer, Joel. 1976. An areal-typological study of American
     Indian languages north of Mexico.  Amsterdam: North-
     Holland.
Shipley, William. 1963.  Maidu texts and dictionary. 
     UCPIL 33.
_____. Maidu grammar. UCPIL 41.
_____, and Richard A. Smith. 1979.  The roles of cognation and
     diffusion in a theory of Maidun prehistory.  J. of
     California and Great Basin Anthropology: Papers in
     Linguistics 1:65-73.
Silverstein, Michael. 1979. Penutian: An assessment. L. Campbell
     and M. Mithun, eds., The languages of Native America: 
     Historical and comparative assessment, pp. 650-91.
     Austin: U. of Texas Press.
Taylor, Walter. 1961. Archaeology and language in Western North
     America. Am. Antiquity 27:71-81.
Ultan, Russell. 1967. Konkow grammar.  Ph.D. dissertation,
     UC Berkeley.