original uncut draft of the article to
appear in:
Crosslinguistic Approaches to the Psychology of Language:
Research in the Tradition of Dan Isaac Slobin.
Guo, Jian-Sheng, et. al. eds.
Lawrence Erlbaum Associates.
Main Verb Properties and Equipollent Framing
Leonard
Talmy
Department of Linguistics and Center for Cognitive Science
University at Buffalo, State University of New York
This paper 1 argues against too free a use of "equipollent framing" as proposed by Dan Slobin. Instead, it proposes an expanded set of criteria for main verb status, and finds them applying to languages that Slobin had considered to be equipollently framed.
The background is that Talmy (1991, 2000b, ch. 3) had proposed that languages fall into two main types on the basis of where the Path is represented in a sentence expressing a Motion event -- or, more generally, where the "core schema" is represented in a sentence expressing a "macro-event". In this two-category typology, if the Path is characteristically represented in the main verb or verb root of a sentence, the language is "verb framed", but if it is characteristically represented in the satellite and/or preposition, the language is "satellite framed". A satellite is a constituent in construction with the main verb (root) and syntactically subordinate to it as a dependent to a head. Another semantic component, the co-event -- usually the Manner or the Cause of the Motion-- might then characteristically show up in a particular constituent other than the one occupied by the Path. Note that this concept of framing type makes no appeal to the presence vs. absence of a co-event or its characteristic location, but only to the characteristic location of the Path, which unlike the co-event is seen as criterial to a Motion sentence.
Subsequently, several studies -- e.g., Delancey (1989), Slobin and Hoiting (1994), Schultze-Berndt (2000), and Zlatev and Yangklang (2004) -- either suggested or were noted by others as suggesting that certain languages do not neatly fit either category of the proposed typology. The main problem claimed was that the cited languages did not clearly assign either main verb status or satellite status to the constituent expressing Path. Where linguists considered together both the constituent expressing Path and the constituent expressing the co-event, they judged that the languages did not privilege either of these constituent types as being the main verb or some other kind of head or dominant category, nor mark the other constituent type as being a satellite or other kind of dependent or subordinate category. Slobin (2004) then proposed classing such languages together in a third category of "equipollently framed" languages within a now expanded typology.
To lay the ground for any challenge or constraint on this idea, we first observe that Slobin’s concept of equipollent framing can be analyzed as actually comprising two distinguishable properties, ones that seemingly are often conflated. One property builds directly on the original basis I proposed for framing: whether the Path shows up in the main verb (root) or in a satellite / preposition -- for short, main verb or satellite. In terms of this property, equipollence is claimed if it is unclear whether the constituent type that the Path characteristically appears in is either the main verb or a satellite. But the constituent expressing the Path along with the constituent expressing the co-event are together seen as forming all or most of a complex that does serve something like a main verb function. This circumstance could occur in several ways. One way is that the constituent expressing the Path has no clear lexical category in its own right, but typically cooccurs with one or more other constituents, all of them together seeming to constitute a full main verb complex. This pattern has been thought to occur in certain polysynthetic constructions, such as the bipartite verb stems as described for Klamath in Delancey (1989), -- addressed below in section 3. Another way is that the constituent expressing the Path might well have a lexical category, that of a verb, but is not the main verb, -- rather, only one of two or more verbs in the sentence that together again seem to constitute the full main verbal complex. This pattern has been thought to occur in certain serial verb constructions -- treated below in section 4. In this paper, the term equipollence will be applied only to patterns like these that the first property pertains to.
The second property often associated with the notion of equipollent framing, but distinguishable from the first property, is that the constituent expressing the co-event is judged to be a grammatical peer of the constituent expressing the Path. This circumstance could occur in the two just-cited patterns in which the two constituents together function like a main verb complex. In the Klamath bipartite verb stem, both constituents would be equally indeterminate with respect to lexical category, and in a serial verb construction, both constituents would equally be verbs. But since these patterns can already be addressed under the first property, they might have diminished relevance here. More importantly, the second property also holds in a construction where a third constituent functions as the true main verb, while the Path constituent and the co-event constituent, either singly or jointly, are outside any main verb designation. This circumstance might well occur among the coverbs of Jaminjong -- as discussed in section 5 -- and does occur in the polysynthetic constructions of Atsugewi.
In Atsugewi (as detailed in section 3), the constituent expressing Cause and the constituent expressing Path are both equally affixes, while a third constituent yet, one expressing the Figure, is the true main verb root. Note, though, that this pattern of Path and co-event coequal as subordinates -- that is, both in subordinate constituents -- was part of the original Motion typology as described in Talmy (1972, 1985). So insofar as the concept of equipollence is proposed as an addition or a corrective to prior theory, it should not be applied to the present pattern. Accordingly, the kind of pattern seen in Atsugewi will here be said to exhibit not equipollent framing, but rather "co-satellite status", or "co-subordinate status". Under this status, that is, the co-event constituent and the Path constituent can be on a par as long as neither of them is the main verb (root). While Atsugewi does have this co-satellite status for Path and co-event, it does not at all show equipollent framing of the first-property type based on the original criterion. On the face of it, in fact, it is clearly satellite-framed, since the Path characteristically appears in a suffixal satellite to the main verb root, not in the main verb root itself. It is cases like this that require the separation of the two properties observed in the equipollence concept.
In accordance with this analysis, any challenge to the existence or extent of equipollent framing hinges on whether, in the languages at issue, the Path constituent can be argued as having (predominantly) either main verb status or satellite status. Where both the Path constituent and the co-event constituent are considered together -- and where these two are not both subordinate to a third verb constituent -- any challenge to equipollence hinges on whether one of the two constituents can be argued as having (predominantly) main verb status while the other has satellite status. Accordingly, any inquiry into what constitutes main verb status can abet such challenges, and this is what is undertaken next.
It can be stated at the outset that there is nothing in principle the matter with extending the original framing typology to include a third category of indeterminate framing, that is, Slobin’s equipollently framed category. Insofar as such an indeterminate condition may occur, it would seem that the proposed form of equipollence is the right way to view it. The proposal here, though, is that the criteria used for judging main verb status have been too few, and that an expanded set of criteria might show a broader tendency among languages seen as candidates for equipollence in fact to privilege one of the constituent types in question with main verb status. If so, then true equipollent framing might be rarer than proposed, perhaps even nonoccurrent, and if occurrent, possibly an unstable stage that a language tends to transition out of with relative diachronic speed.
In (1) is an expanded set of proposed factors that tend to indicate that a language treats a particular constituent type as its main verb or verb root. Quite possibly none of these factors is criterial for main verb status. Rather, different subsets of the factors apply to a specific constituent type in different languages, with no individual factor emerging as crucial. The more factors that converge on a particular constituent type in a language, the more that that constituent type is being privileged with main verb status. Some languages exhibit what can be considered a split system of main verb status in that one subset of the factors applies to one constituent type, while another subset of factors applies to another constituent type.
(1) factors that tend to mark a particular constituent type as the main verb (root)
Of two constituent
types in a language that can be considered for having main
verb status,
one of them ranks higher for that status--
a. morphology
if it can take
inflections or clitics for such semantic categories as
tense, aspect, mood, evidentiality, negation, causation,
voice, transitivity,
or the person, number, and gender of the subject (and
object).
b. syntax 2
if it functions as
a head directly or nestedly in construction with
such other sentence constituents as:
adverbs;
particles for place, time, aspect, quantity (e.g., floats),
negation, etc.;
or a subject or object nominal.
c. cooccurrence patterns
if its presence is
required across a range of construction types,
while the other constituent type need not or can not be
present
in some of those construction types.
d. class size
if it has more
morpheme members or is open-class
while the other constituent type has fewer morpheme members
or is closed-class
e. phonology 3
1. if its morpheme members have a greater average phonological length.
2. if its morpheme members vary over a greater range of phonological length or pattern.
3. if its morpheme
members include phonemes ranging over a greater portion
of the phonemic inventory of the language.
f. semantics
1. if the meanings
of its member morphemes tend to have more substantive
content
greater specificity,
and a greater number of more varied conceptual components
associated together
in more intricate relationships,
while those of the other constituent type tend to have less
of these.
2. if the meanings
of its member morphemes range over a greater variety
of concepts and types of concepts and trail off into more
outlying conceptual areas,
while those of the other constituent type
tend to fit a more stereotyped semantic category.
3. if it seems to
contribute the criterial component of "actuation"
to the proposition
that is otherwise represented by the sentence.
Before using them to help resolve less clear cases, the factors in (1) can be checked out for English. Here, all the factors except the (1e) phonological ones seem to hold. To illustrate, we can consider for main verb status the constituent type instantiated by the morpheme roll and the constituent type instantiated by the morpheme down in the sample sentence My neighbor seldom rolls down his shades. The former constituent type ranks higher for main verb status first because it exhibits factor (1a) -- e.g., here taking the inflection -s representing present tense, habitual aspect, and indicative mood, as well as third person and singular number for the subject. The constituent type here instantiated by down does not take inflections. The former constituent type also exhibits factor (1b). Here, for example, roll is the head of the construction it forms with down, not vice versa. And it further functions as the head of constructions -- involving various degrees of nesting -- that it forms with the temporal particle seldom, with the object nominal his shades, and with the subject nominal My neighbor. Down does not do any of these. The former constituent type further exhibits factor (1c) in that some representative of it must be present in a range of sentence types, whereas the constituent type here represented by down can or must be excluded from many of those sentence types. By contrast, the reverse pattern -- that is, sentence types in which the down type of constituent must be present, while the roll type of constituent is optional or blocked -- is minimal at best. The former constituent type additionally exhibits factor (1d) in that it is an open class with hundreds of morpheme members, whereas the constituent type here represented by down is a closed class with only a few dozen members. Finally, the former constituent type exhibits all three parts of factor (1f). Its member morphemes on average have greater and more specific semantic content, with more semantic elements of different types together -- as roll here does relative to down. They also range over a greater variety of meanings -- as, say, roll, burrow and gush do relative to down, out and across -- where the latter tend to fill out a more stereotyped semantic category of path. (To be sure, the greater specificity and range of the former constituent type accord with its greater class size, though, in principle, these two factors need not be correlated). And last, they provide the actuating or dynamizing feature for a proposition -- as roll, but not down, does in the example sentence.
Note that the factors in (1) are on purpose formulated generically, not in terms of Motion or any of its components such as Path or Manner. The reason is that main verb status should be independently based on properties neutral to the issue that prompted its explication. A quick look at Spanish might illustrate the need to emphasize this point. Consider a sentence like La botella entr´ flotando a la cueva, ‘The bottle entered floating to the cave’ -- that is, "The bottle floated into the cave". The constituent type here instantiated by entr´ -- let’s call it constituent type 1 -- ranks higher for main verb status than the constituent type here instantiated by flotando -- let’s call it constituent type 2 -- with regard at least to the first three factors of (1). Thus, constituent type 1 takes many of the inflections indicated in (1a), while constituent type 2 takes none of them. It has more of the syntactic "head properties of (1b) than constituent type 2. And it has the cooccurrence privileges of (1c): it must occur across a range of construction types for which constituent type 2 is only optional. Assuming as for English that any (1e) phonological differences between the two constituent types are negligible, what about the class size and semantic properties of factors (1d) and (1f)? Consider the findings if we were to allow the approach of limiting the examination to characteristic Motion-expressing sentences, and hence of limiting constituent type 1 to morphemes expressing Path and constituent type 2 to morphemes expressing Manner. Constituent type 1 would now be smaller in class size than constituent type 2, since the former would range only over those morphemes expressing basic Paths (Path verbs), while the latter would cover the rather larger group of morphemes expressing Manner (Manner verbs). And with respect to the semantic factor (1f), the Paths expressed by the morphemes of constituent type 1 would be semantically rather spare and sterotyped, while the Manner-expressing morphemes of constituent type 2 would cover a more varied and more intricate set of meanings. For these two factors, then, constituent type 2 would rank higher in main verb status than constituent type 1. However, (1) is deliberately set up to address the entire morpheme complement of each of the two constituent types under comparison, not just some subset of that complement. On that basis, one would need to consider all the morphemes that can serve as constituent type 1, not just the Path verbs, as well as all the morphemes that can serve as constituent type 2, not just the Manner verbs. It is not certain how this intended comparison would turn out for Spanish under a full analysis, but it is likely that the class size and semantic diversity of the two constituent types would at least be more comparable, and perhaps tilted in favor of constituent type 1.
Let me now apply the factors in (1) to Atsugewi, a Hokan language of northern California and the language of my fieldwork. Atsugewi is a polysynthetic language, that is, the core of the sentence is a complex constituent in turn consisting of a number of morphosyntactically distinguishable constituents that occupy distinct position classes in a specific sequence relative to each other, all of them morphologically bound. This constituent as a whole gains some ranking as main verb in that it takes many of the kinds of inflections listed under factor (1a), and it relates syntactically to other sentence constituents much as described under factor (1b). On this basis, I call this polymorphemic constituent a "verb complex". But what about the distinct constituent types within this verb complex. Might one of them exhibit enough of the remaining factors to merit status as the main verb root of the complex? The evidence below converges on just such a conclusion.
Delancey’s (1989) analysis of Klamath -- a Penutian language geographically near Atsugewi -- stands as the main claim to equipolent framing within a polysenthetic verb. I am not familiar enough with Klamath to raise questions about its analysis directly. However, Delancey’s paper cites Atsugewi as behaving in a way similar to Klamath, and proposes an areal basis for such similarity. But the conclusion below that Atsugewi does single out and privilege a particular bound constituent type as the verb root at least removes Atsugewi from Delancey’s claim. In turn, it suggests another look at Klamath from the present perspective, with the possibility that some of the arguments advanced here for Atsugewi might apply to Klamath as well and diminish its claim as an exemplar of equipollence.
In one of its most characteristic patterns, an Atsugewi verb complex that expresses a Motion event has at its center a tripartite stem, that is, a stem consisting of three distinct constituent types, all of them bound morphemes (themselves in turn surrounded by potentially numerous derivational and inflectional affixes). The first of the three constituent types has morpheme members that prototypically refer to the kind of immediately prior event that caused the Motion event -- what I label as the "Cause" -- or to what can simply be taken as the Instrument. The central constituent type has morpheme members that prototypically refer to the kind of object or material that functions as the Figure of the Motion event. The third constituent type has morpheme members that prototypically refer to the combination of a particular Path and type of Ground object within the Motion event.
Of these three constituent types, the central one referring to the Figure ranks highest for status as main verb root under the remaining factors in (1). Thus, to start with factor (1d) concerning class size (with factor (1c) reserved for later), the Figure-specifying constituent type has hundreds of morpheme members -- and there is some evidence that new morphemes can be more easily added to this type, so that it has some claim to open-class status. By contrast, the Cause-specifying constituent type has only some two dozen members, while the Path+Ground-specifying constituent type has only some fifty members, both constituent types being clearly closed-class.
The Figure-specifying constituent type also ranks higher on all three phonological properties in factor (1e). The morphemes of this constituent type first average a greater length and, second, they vary more in pattern than those of the other two constituent types. Thus, the Figure-specifying morphemes range from having no vowel and consisting of from one to three consonants, to having one vowel with various numbers of consonants on either end, to having two vowels with varying numbers of consonants at either end and in the middle. But the Cause-specifying morphemes are mostly CV in shape, the main divergences being that two of the forms add a continuant consonant after the first C, and two add one after the V. And the Path+Ground-specifying morphemes are mostly VCC or CVC in shape. In addition, the Figure-specifying morphemes have virtually no constraints on the phonemes that can occur in them. But the Cause-specifying morphemes can include stops only from the plain series, not from the glottalized or aspirated series; of the phonemically distinct dentals "r/l/n", they can morphophonemically include only r; and they lack the phoneme "q". As for the Path+Ground-specifying morphemes, the vowel that occurs in them is preponderantly "i", and none of the three "q" stop phonemes occurs in them.
considering for now
only the first two semantic properties under factor (1f),
the Figure-specifying constituent type again ranks higher
than the other two constituent types. With regard to
property (1f1), some of the Cause-specifying morphemes do
refer to relatively ccontentful Instruments, such as the
wind or buttocks. Likewise, some of the
Path+Ground-specifying suffixes refer to relatively
contentful Ground objects, such as liquid, a container, or
someone’s face/head. But the Figure-specifying
morphemes include many with a still greater amount,
specificity, and intricacy of content. Examples are one
referring to a linear flexible object suspended from one end
(e.g., a sock on a clothesline, a killed rabbit suspended
from one’s belt, a flaccid penis) and one referring to
fabric that gets bunched up or unbunched in the process of
moving (e.g., curtains getting opened, a sock getting put
on). And, with respect to property (1f2), the
Figure-specifying morphemes appear to cover a wider range of
concepts. For instance, beyond the previous two examples,
the kinds of Figure they refer to range from charcoal lumps,
to anatomically contained fluid, to a water-borne canoe
gliding lengthwise. True, the Cause-specifying morphemes
have a certain range of their own, covering natural forces,
a linear object engaged in various actions, body parts, and
sensory stimuli. But they basically cover only these four
semantic domains and make only a few distinctions within
each of them.
And what the Path+Ground morphemes specify for the Ground is
for the most part a geometric type of schema.
Let me return now to the factor of cooccurrence patterns in (1c). The largest class of Figure-specifying morphemes must occur in the tripartite stem described at the outset -- that is, they must be directly preceded by a Cause-specifying morpheme and followed by a Path+Ground-specifying morpheme. But there is also a class of Figure-specifying morphemes that, while still requiring a Cause morpheme on the left, can occur without a path+Ground morpheme on the right. Further, there is another class of Figure-specifying morphemes that requires a Path+Ground morpheme on the right, but that refuses any Cause morpheme on the left. Thus, Figure-specifying morphemes occur across a certain range of construction types, across which the other two constituent types either do or do not occur. To round out the picture a bit, there are several additional classes of morphemes that occupy the same position class as the Figure-specifying morphemes but that do not specify the Figure. Some of these classes follow each of the three patterns of requirement or refusal just cited for different classes of Figure-specifying morphemes. In addition, one class can occur by itself -- with neither the Cause nor the Path+Ground constituent type accompanying it. By contrast, neither the Cause constituent type nor the Path+Ground constituent type can occur by itself in a verb complex. And the two of them can not occur together without a Figure-specifying constituent or one of its semantic alternatives occurring between them. The upshot of this set of cooccurrence patterns is that the constituent type that specifies the Figure (or certain semantic alternatives) is criterial to the verb complex, whereas the other two constituent types are not.
There is one more pattern involving cooccurrence that privileges the Figure-specifying constituent type. In a special construction, a Figure-specifying morpheme of the class that otherwise requires both a Cause morpheme and a Path+Ground morpheme can be removed from the verb complex entirely, placed in front as a frozen form, and set in construction with a new generic (or light) verb that now takes all the inflections. For example, the morpheme -qput- that refers to ‘dirt’ as a Figure, and that usually occurs at the center of a tripartite stem within a verb complex referring to dirt as moving or located, can also occur before a ‘be’ verb in a construction that means ‘for there to be dirt present / in occurrence’. Neither of the other two constituent types can take part in such a construction. Thus, both within the verb complex and outside it, the Figure-specifying constituent type is singled out as the survivor across a range of construction types and so, by factor (1c) is once more accorded higher ranking for verb status.
Because of its high
ranking on factors (1c) through (1f), the Figure-specifying
constituent type (and its semantic alternatives) functions
most as the main verb. Since it is a bound morpheme within a
polymorphemic word, my practice has been to term it the main
verb root. Accordingly, the Cause-specifying constituent
type can now be definitively treated as a prefix and the
Path+Ground-specifying constituent type as a suffix. With
appeal to the semantic property in (1f3), the
Figure-specifying constituent type can now, as main verb
root, be considered to actuate the multi-affixal verb
complex it is in. (This in turn -- as the whole-word
constituent on the sentence level that functions as the verb
on the basis of factors (1a) and (1b) -- actuates the
sentence as a whole). It is because the Figure-specifying
morphemes in Atsugewi behave like the main verb root that I
originally cited Atsugewi as an example of a third major
type within my Motion-actuating typology.
4
Namely, this is the type where, of the various semantic
components
within a Motion event, it is the Figure that
characteristically appears in the main verb root along with
‘fact of Motion’. (Presumably similar arguments
could be made for Navajo as another example of this
type).
for these reasons, the Figure-specifying morphemes have been consistently glossed in my work as verbs, not, say, simply as nominals that refer to the Figure. For example, -qput- is glossed as ‘for dirt to move / be located’ -- not, say, simply as ‘dirt’. That is, the dynamizing semantic component of "fact of Motion" is incorporated directly within the meanings of these morphemes.
By the same token,
the other two constituent types have consistently not
been glossed as verbs. For example, the Cause-specifying
prefix ca- has been glossed either as an adverbial
clause, ‘as the result of the wind blowing on the
Figure’ -- or simply as a prepositional phrase,
‘from the wind’ -- but not as a verb form like
‘for the wind to blow’. Likewise, the Cause
prefix ma- has been glossed basically as the
adverbial clause ‘as the result of one’s feet
acting on the Figure’ -- or, in an agentive sentence,
as ‘by acting on the Figure with one’s
feet’ -- or simply as the instrumental phrase
‘with one’s feet’. But it has not been
glossed as a verb form such as
‘to do with the feet’ or ‘to act with the
feet’ -- a kind of gloss that appears in other works
that seem to be describing something comparable to a Cause
morpheme.
Comparably, the Path+Ground-specifying suffixes have been glossed as prepositional phrases. For example, -ic’t has been glossed as ‘into liquid’ -- not as a verb form like ‘move into liquid’. One place that this becomes an issue is for the Atsugewi morphemes for possession and change of possession. These morphemes belong to the exact same constituent type as the unproblematic Path+Ground suffixes and can occur in its position class with roughly the same sets of surrounding morphemes. But in some other treatments, apparently comparable morphemes are glossed like verbs as ‘have’ or ‘give’ -- inconsistently with the glossing of other morphemes in the same constituent type. In my work, however, the relevant Atsugewi suffixes -ahn and -ay are glossed respectively as ‘in one’s possession’ and ‘into someone’s possession’ -- that is, in the same prepositional phrase mold as the other members of the same constituent type (see Talmy 2000b, ch. 4 for a more elaborate discussion of such issues). The point here is that once a particular constituent type has been identified as a verb root and other constituent types complementarily fall into place for their respective semantic-syntactic roles, then it is best to give a consistent form of glossing to the morphemes of each constituent type -- a form that corresponds to the semantics of that type.
The conclusion from
all the preceding, then, is that Atsugewi does not have
equipollent framing. First, it has a definite main verb root
-- there is nothing indeterminate here -- a root that in a
Motion event happens to express the Figure. Next, Path is
expressed in a satellite, subordinate to the main verb root,
specifically, in the suffix immediately following the main
verb root. Thus, by the original principle determining
framing type, Atsugewi is a satellite-framed language.
Lastly, both the Path and the co-event, in particular its
Cause type, are equally expressed in satellites subordinate
to the main verb root. Thus, together, they do not comprise
all or even part of some complex that functions like a main
verb. Rather, since they are both wholly outside the main
verb root and its dependants as head, they exhibit
co-satellite or co-subordinate status -- a status stipulated
earlier as distinct from equipollence. Perhaps further
research should examine whether, in accordance with some
additional set of factors, two such co-satellites in turn
exhibit some kind of
asymmetry in syntax, semantics, phonology, etc. between
themselves,
with one more dominant and the other more subordinate, or
instead exhibit a kind of coequality. But even any uncovered
coequality at this level should not be identified with the
equipollent framing that was proposed as a new framing
type.
finally, then, Atsugewi can be considered to have a split system in its conferal of main verb status. The multi-affixal verb complex as a higher-level constituent type, exhibiting the first two factors of (1), acts as the main verb relative to the other major constituent types in the sentence. At the same time, the simplex constituent type within the verb complex that specifies the Figure (or its alternatives) exhibits the remaining four factors of (1), and so can be considered to function as the verb root within the main verb complex, -- what I have dubbed the "main verb root". It is because of polysynthetic languages like Atsugewi that my work on the Motion-actuating typology from the outset stressed the need -- insofar as verbal constituents were being considered -- to use the verb root for crosslinguistic comparisons. It is the Figure-expressing main verb root within the polysynthetic verb complex of Atsugewi that is to be compared with the Path-expressing verb root within the inflected verb form of Spanish, and with the Manner- or Cause-expressing monomorphemic V1 in an isolating language like Mandarin.
Another case to which Slobin applied his concept of equipollent framing was serial verb constructions. 5 Although I am not as familiar with serial verbs as with polysynthesis, I can present some evidence counter to equipollent framing among them.
First, Matisoff, in his (1973, 1991) treatment of the Tibeto-Burman language Lahu, describes a characteristic construction -- one that includes the representation of Motion events -- in which up to five verbs can be concatenated within distinct position slots. he is clear, though, that the verbs occurring in one of those position slots is the main verb, the "head", while the others -- what he terms "versatile verbs" -- are semantically subordinate to the head verb and occupy pre-head and post-head position slots. (All these versatile verbs can also occur as main verbs.) Most of the factors in (1) appear to correlate with Matisoff’s analysis and might be the basis for it. Without going through them all here, we can note that factor (1d) pertaining to class size seems to apply. Thus, the head position can be occupied by any of the hundreds of verbs in the language, including those referring to Manner or Cause in an expression of Motion. But the pre-head class of versatile verbs has only some dozen members; the "juxtacapital" class of versatile verbs that immediately follows the head, and that represents the Path in a Motion expression, again has only some dozen members; the "medial" class of versatile verbs that comes next has some fifteen to twenty members; the "caudal" class of versatile verbs that comes last has some eight members; and the "variable" class of versatile verbs, which can occur in several positions relative to the preceding classes, has eight members.
Or, further, we can note how factor (1c) pertaining to cooccurrence patterns might come into play. First, we need to observe that some of the versatile verbs have very similar meanings as a head and as a subordinate to a head -- like the form with the meanings ‘begin’ and ‘begin to’, respectively. But other versatile verbs have quite divergent meanings in the two roles -- like the form that means ‘send on an errand’ as a head verb and ‘cause to’ as a subordinate. Now, a sentence can have just a single verb. But if this verb is one of the versatile verbs, the meaning that emerges is always that of its head role, never that of its subordinate role. Thus, the "head" constituent type within a serial verb construction is the one that survives across a range of construction types, and, by factor (1c), thus gains additional main verb status. Thus, Matisoff’s description stands as one counterexample to equipollent framing in a serial verb language.
I turn now to
Mandarin, another language with a serial verb construction
that can represent a Motion event as well as its semantic
generalizations, using what some have proposed as
equipollent framing. When representing such events,
typically, the verb in the first position of the series,
what can be designated as V1, represents the Co-event --
either Manner or Cause. The verb in the second position, or
V2, represents the Conformation component of Path.
And a third verb or V3 can be present representing the
Deixis
component of Path. The verbs that can occur in each of the
three positions of the series generally belong to different
sets. The procedure proposed here is to compare the verbs in
each such set -- when in fact used in one position of the
serial construction -- with the same verbs when used as the
sole verb in a sentence without a serial construction. This
latter will here be designated as V0 -- that is, V followed
by a zero. To help in this comparison, I propose the
principle in (2), which can be used in conjunction with the
factors in (1) to suggest different degrees of main verb
status.
(2) principles for the degree of overlap of two otherwise
distinguishable constituent types
If a language has
two syntactically distinguishable constituent types
that share some but not all of their morpheme members,
then:
a. the degree of their divergence as distinct constituent types correlates with:
1. the proportion
of non-overlap of their respective morpheme memberships
and -- for morphemes within the overlap --
2. the proportion of morphemes whose meanings differ in the
two constituent types
and
3. the degree of such differences in meaning.
b. A morpheme
within the overlap that has basically the same meaning in
both constituent types
can seem to belong to a meta-category that spans both
constituent types
or to belong to the dominant category type even when
functioning syntactically in the other type
These characteristics tend not to hold for:
a morpheme outside the overlap
or a morpheme within the overlap that has distinct meanings
in the two constituent types
These principles can be initially checked out in English. As a backdrop, first note that there is virtually complete overlap between some pairs of syntactically distinguishable constituent types, such as the nouns that can occur in subject NPs and the nouns that can occur in object NPs. And there is a complete disjunction between other pairs, such as between determiners and auxiliaries. But now consider two other syntactically distinguishable constituent types: prepositions, which are in construction with a nominal, and satellites, which are in construction with the verb. With respect to property (2a1), there is much overlap in the morphemic memberships of these two constituent types, but at the same time each type has morphemes not occurring in the other. Thus in, on, off, up, down, across, along, through, and around can all function either as satellites or as prepositions. But away, back, ahead, forth, apart, and together function only as satellites. On the other hand of, from, at, towards, beside, among, and (in standard English) with function only as prepositions. With respect to property (2a2), among the morphemes serving for both constituent types, many have similar senses in both usages. An example is in which has a comparable meaning when functioning as a preposition, as in She is in the room, and when functioning as a satellite, as in She hurried in. But other morphemes have distinct senses. Thus, over as a satellite includes the sense ‘rotationally about a horizontal axis’, as in The pole fell over, but this sense is absent in the prepositional usage of the form. With respect to property (2a3), the semantic divergence between prepositional and satellite usages in such cases seem in general not to be very great. For example, the satellite senses and the prepositional senses of over can be fairly readily linked (see Brugmann, 1981). The two constituent types, therefore, can be judged to be neither identical nor unrelated, but rather partially overlapping and hence moderately distinct. Finally, with respect to property (2b), to a speaker with some syntactic sensitivity, a form like in with its comparable meaning in both usages might seem to belong to some meta-preposition/satellite category, or might seem, for example, to be a preposition even when functioning as a satellite as in She hurried in. But forms like apart and of would unambiguously be taken to be either a satellite or a preposition, respectively. And a form like over with its diverging senses might be starting to seem like having a foot in two different categories.
Other types of partial overlap can be found across languages. For example, one other type seems to hold in at least some noun-incorporating languages, such as Caddo, between their independent nouns, as one constituent type, and their incorporated nouns, as a second constituent type.
Returning to Mandarin, first of all, there might be some evidence that V1 ranks higher than V2 for main verb status on the basis of factors (1b, c, d, and f) -- that is, on the basis of certain forms of syntactic, coocurrence, class size, and semantic behavior. 6 For example, in terms of factor (1d) pertaining to class size, the set of forms that can occur as V1 might be significantly larger in size than the set able to occur as V2. But the following discussion appeals only to the principles in (2) so that their function can be seen. In broad strokes, the basic observation is that V1 is always taken as a main verb, and that a V2 is equally taken as amain verb if it can also appear elsewhere by itself as a V0 with the same meaning, but is otherwise taken as a satellite subordinate to V1. What follows is the finer-grained analysis.
With respect to property (2a1), it looks like there might be a greater overlap in morphemic membership between the V1 and V0 constituent types than between the V2 and V0 constituent types. If so, the class of first-position verbs may be more of a piece with the class of solo verbs, while the class of second-position verbs would show more divergence from the class of solo verbs. Moreover, in terms of property (2a2) across the overlapping portions of morpheme memberships, the morphemes that can function both as V1 and as V0 seem to have basically the same meanings across both usages, whereas a number of the morphemes that can function both as V2 and as V0 have divergent meanings across the two usages.
The V2/V0 overlap
can be illustrated with morphemes first that do and then
that do not have the same semantic content across the two
usages. The form jin4 refers to ‘motion
into’ both as a V2, as in (3a), and as a V0, as in
(3b).
(3) a. Ta1 zou3 jin4 le gong1-yuan2. she/he walk enter PERF
park
"She/He walked into the park."
b. Ta1 jin4 le gong1-yuan2. she/he enter PERF park
"She/He entered the park."
If now jin4
is replaced by guo4, the sentence corresponding to
(3a), shown in (4a), can translate as "She/He walked
past/across the park". Here, the new form in its V2
usage represents a fairly common path concept. But when
guo4 appears as the V0 in a sentence corresponding to
(3b), shown in (4b), the sentence does not correlatively
mean "She/He passed/crossed the park". Rather, it
indicates that the subject’s movement was one within a
succession of movements being observed from some distance by
someone else. Further, the subject’s path now tends to
be that of passing to one side, rather than crossing --
though the latter is in principle possible -- largely
because the Ground object, the park, is being conceptualized
as a point due to observation from a distance. Thus,
guo4 as a V2 functions semantically as one of a
familiar series of Path specifiers, whereas it has certain
semantic idiosyncracies as a V0: a case of semantic
divergence.
(4) a. Ta1 zou3 guo4 le gong1-yuan2. she/he walk pass PERF
park
"She/He walked past/across the park."
b. Ta1 guo4 le gong1-yuan2. she/he pass PERF park
"She/He was observed to pass the park as part of a
longer route."
Talmy (2000b, ch. 3) argues that where a language characteristically represents Path, it usually also represents certain other semantic categories, including aspect. And, indeed, the V2 slot in Mandarin is the characteristic constituent type not only for the representation of Path, but also for that of aspect. But, in terms of property (2a3), those morphemes expressing aspect in their V2 usage generally seem to express meanings there that are more divergent from those in their V0 usage than in the case of the Path morphemes. Thus, in their V2 usage, both hao3 and wan2 mainly express the aspectual concept ‘to completion’. But in their V0 usage, although these forms can express comparable meanings, they tend instead to express quite distinct meanings. As a V0, hao3 usually means ‘be good’, while wan2 is usually used to refer to something like ‘be all for nothing / be done for’. Moreover, the same guo4 already seen above can also appear as a V2 to express the so-called "experiencial" aspect ‘to have already / ever Ved’ -- a meaning quite divergent from that in its V0 usage.
And now we come to the main point of the discussion based on (2). The semantic and syntactic properties of a morpheme covered under principle (2b) seem largely to determine a native speaker’s sense of the lexical category of that morpheme in its V2 usage -- specifically, whether the morpheme is functioning as a verb or as a satellite there. And this assignment in turn determines whether the V1-V2 construction exhibits equipollent framing or satellite framing. In particular, if the meaning of a morpheme in its V2 usage is basically the same as its meaning in its V0 usage, a speaker tends to regard the morpheme in its V2 usage as a verb having some prominence, roughly coequal with that of the V1 verb. The basis for this sense would seem to be that even in its V2 usage, the morpheme is still associated with its capacity to stand alone as a full main verb in its V0 usage. But if a morpheme has divergent meanings in its V2 and V0 usages, the speaker tends to regard the V2 form as more subordinate than the V1 verb, hence, as a satellite to it. The basis for this impression might be that the morpheme in its V2 usage, semantically decoupled from the V0 usage, is not associated with any full main verb function, and instead is ascribed membership in some non-verb-like lexical category, now subordinate to the V1 which retains its verb status.
I most recently consulted with Lian-Cheng Chief, a native speaker of Mandarin from Taiwan (and a student of mine) on these issues, and his judgments generally accord with the preceding summary. Thus, his judgment about the V1 and V2 in sentence (3a) is that he cannot tell which of them, if either, is functioning as a sole main verb. In fact, they both seem equally verb-like to him, each introducing a separate verbal concept -- first walking, then entering. This, then, is the first instance of what might be genuinely equipollent framing in the entire discussion so far.
But in the counterpart sentence in (4a) that contains guo4 instead of jin4, Chief’s judgment is that the V1, zou3 is definitely the main verb, while the V2 guo4 is subordinate to it. The verbal concept of walking introduced by the V1 is not succeeded by another verbal concept introduced by the V2 but rather continues, with the V2 simply indicating where the walking takes place.
Comparably, when the V2 is one of the aspect-specifying morphemes, Chief’s judgement is clearly that the V1 is the main verb and that the V2 is a satellite.
Perhaps as a
development in progress, the aspect-specifying forms that
occupy the V2 position might together be coming to form a
distinct lexical class, a class that is intrinsically more
satellite-like in character. In fact, if a morpheme in the
V2 position expresses aspect even moderately, the morpheme
tends to become classed together with fully aspectual V2
morphemes and, accordingly, to seem to function as a
satellite rather than as a verb. For example, the morpheme
zhong4 as a main verb in V0 position, as in (5b),
means for a propelled object to hit rather than miss a
target at which it has been aimed. In the V2 position, as in
(5a), its meaning is quite comparable. Accordingly, it might
have retained its character as a verb in the V2 position
much as jin4 does. However, its V2 use as in
sentences like (5a) suggests something aspectual -- a
fulfillment or completion of the goal of aiming. This then
seems to trigger its assimilation to the class of fully
aspectual V2 forms and, hence, to a more satellite-like
function. Thus, Chief judges that the V1 in (5a) is the main
verb and the V2 is a subordinate form.
(5) a. Ta1 she4 zhong4 le mu4-biao1. she/he shoot hit PERF
target.
"She/He shot (an arrow) and hit the target."
b. Jian4 zhong4 le mu4-biao1. arrow hit PERF target.
"The arrow hit the target."
Thus, of all the V1V2 serial constructions considered in this discussion of Mandarin, only the one with jin4 in (3a) is a candidate for showing equipollent framing. In all the other constructions, the V2 emerges as subordinate to the V1 in what can be construed as the relation of a satellite to a main verb as head. Since the V2 forms in these constructions have expressed Path or its extension to aspect, these constructions show satellite framing.
Slobin has also applied the notion of equipollent framing to Jaminjong, a language in which both the constituent expressing Path and the constituent expressing the co-event are outside the constituent generally regarded as the main verb (see Schultze-Berndt 2000). As argued in the Introduction, this fact alone should exempt the first two constituents from equipollent status and at best accord them co-satellite or co-subordinate status. But a closer analysis of Jaminjong should come first, with these conclusions about framing returned to later.
We can apply the factors of (1) for main verb status to the language. Though my knowledge of it is still quite limited, Jaminjung seems to represent another kind of split system. A certain constituent type in the language takes the kinds of inflections outlined in factor (1a). It may exhibit some of the syntactic privileges of factor (1b) (though this needs clarification). And it is apparently the criterial constituent type in a sentence, having to be present while other constituent types need not be, in accord with the cooccurrence properties of factor (1c). On these bases, this constituent type is generally seen as the main verb, and it will here be referred to as such.
However, with respect to factor (1d), this constituent type is closed-class, with rather few morphemes as members. And with respect to the first two semantic properties of factor (1f), the meanings of the morphemes in this constituent type seem to be rather generic and to remain within rather stereotyped semantic limits.
On the other hand, there is another constituent type -- or perhaps a family of related constituent types -- often occurring in construction with the first type, that is open-class with many member morphemes, morphemes that have a wide range of rather specific meanings. This other constituent type thus exhibits at least two of the factors for main verb status, (1d) and (1f). (Whether it also exhibits the greater phonological freedom of factor (1e) still needs assessment). Perhaps for these reasons, this constituent type has been termed the "coverb".
Now, with regard to framing, Slobin has focused on two groups of coverbs, one expressing Path and another Manner, which can both occur together in a sentence along with the main verb. But it should be noted that the path coverbs apparently tend to express the geometrically more intricate Conformation part of the Path component. On the other hand, it is the main verb that expresses the Deictic part of the path component or, more generally, to express unbounded extended paths, the type that are covered under the "ALONG" case of the "Motion-aspect formulas" that are proposed as universal in Talmy (2000a, ch. 3). Thus, the main verb constituent type includes morphemes with such meanings as: ‘go’, ‘come’, ‘take’, ‘bring’, ‘procede away from’, ‘procede toward’, and ‘follow along after’. On the face of it, within my Motion-framing typology, jaminjung would apear to belong to the verb-framed type. True, not many Path distinctions are marked within the main verb. But such a pattern was already proposed and exemplified under my Motion-actuating typology as "Motion plus a minimally differentiated semantic component", and can as readily be applied here. Any accompanying Path coverb would then provide additional Path specifications as part of a fuller distributed Path representation.
The whole pattern seems rather comparable to that seen in Japanese or Korean. In those languages, Motion sentences often have a deictic ‘come’/‘go’ verb root as main verb, accompanied by verb roots in a gerundive or bound form that express Manner and/or the Conformation part of Path. The main difference is that in Japanese and Korean the Conformation-specifying roots (and, for that matter, the Manner-specifying roots) can also occur as main verbs, whereas in Jaminjong they cannot.
If this interpretation holds, then the possibility of equipollent framing for jaminjong simply disappears: this is a verb-framed language. As in Atsugewi, the Path constituent and the co-event constituent are both syntactically subordinate to the main verb as head, and so together exhibit co-satellite status or co-subordinate status.
A question raised for Atsugewi above could be pursued here as well. One could look further into the co-satellite status of the coverb expressing Path Conformation and the coverb expressing Manner to see whether a coequal pattern or a dominant-versus-subordinate pattern holds across these two types of non-main-verb constituents. And this question could be pursued equally for Jaminjung, Japanese, and Korean. But that would involve a further layer of phenomena, one not included under the original framing typology.
To sum up, this paper has first argued that the concept of equipollent framing should only be applied to cases where a constituent expressing Path and a constituent expressing the co-event together serve most or all of a main verb-like function in a sentence, not where they are both outside a third constituent that does function as a main verb. In the latter case, the two constituents show co-satellite status or co-subordinate status, but not equipollent framing. Second, even in the applicable cases, actual equipollence in framing emerges as a seemingly much rarer phenomenon than previously claimed. The arguments against the claimed cases of equipollence are based on an expanded set of criteria for main verb status, and on principles for the assignment of lexical category.
It has been a pleasure adding my paper to the others in this volume to acknowledge and thank Dan Slobin for his immense contribution. Dan’s work has closely interacted with my own over the decades, from the typology of Motion, to the structure of spatial conceptions, to the semantics of grammar. In this Festschrift for him, I would like to thank him for his pivotal ideas in these and more areas. We have agreed on much and, where we have disagreed, each has often used the other’s objections as a springboard for further developments, sometimes through several cycles, as in the case of the present paper. Best of all, our linguistic interactions have taken place within a friendship that has also spanned the decades.
Endnotes
1. This paper is a revised and expanded version of one portion within Talmy (forthcoming). My thanks to Jian-Sheng guo for comments on the paper, to James Matisoff for information on Lahu, and to these two as well as to Lian-Cheng Chief for information on Mandarin. None of them is responsible for my errors and oversights.
2. Note that the syntactic property of a constituent here associated with main verb status is that of functioning as the head of a construction. However, as argued in Talmy (2000b, ch. 3), a constituent can have other syntactic properties that are not associated with main verb status but rather with the status of being a "framing constituent". In a verb-framed language such as Spanish, these additional properties are in fact exhibited by the main verb. But in a satellite-framed language like English, these further properties are exhibited by the satellite. The syntactic properties associated with a framing constituent are that, over the scope of an entire sentence, this constituent generally is the target of negation or interrogation, determines the overall aspect, and can determine much of the argument structure. For example, in the English sentence I didn’t eat up the popcorn, the negative particle negates the satellite up in its sense of ‘completion’, but does not negate the notion that I ate.
Note that although head status is taken for granted in the formulation of factor (1b), perhaps the notion of head status itself should be rethought, much as has been done for main verb status here. Such a reanalysis might conclude that a set of properties pertain to head status, that more or fewer of such properties can hold for any given constituent type in some language, and that a kind of split head status might apply to two competing constituents that divide up some of the properties.
3. Perhaps a fourth subfactor should be added under the phonology factor. It would apply where one of the constituent types gets stronger stress than the other. This is not included here, though, because there is not enough evidence whether stronger stress shows a recurrent crosslinguistic association either with the main verb constituent type or the subordinate constituent type as this might be determined by other factors. It might at first be thought that stronger stress would correlate with a main verb constituent type. But English is an immediate counterexample since, in a verb complex with a main verb followed by one or more satellites, it is the last satellite that tends to get the strongest stress (as in: Come right back down OUT from up in there.).
4. The terms "Motion-actuating typology" and "Motion-framing typology" are introduced in Talmy (forthcoming) to help sort out a distinction at times missed in the literature. The former term refers to the procedure in which one morphosyntactic category, such as the verb root, is held constant across languages to see which semantic category, such as the Path, Manner, or Figure, is characteristically represented in it. The second term refers to the procedure in which one semantic component, such as the Path, is held constant across languages to see which morphosyntactic category, such as the verb root or the satellite and/or preposition, it is characteristically represented in. There is only one set of typological relations, but the two terms pertain to different perspectives that can be taken on them.
5. The type of construction generally considered under this aegis consists of a succession of two or more free monomorphemic verb roots: VV, VVV, etc. this kind of construction is known in the Chinese linguistic tradition as compund verbs, and Matisoff calls it concatenation. Both sources reserve the term "serial verbs" for a construction of successive verb phrases, each verb taking a direct object nominal or the like. For Slobin, however, the term "serial" covers the present type of construction as well, and this usage is followed here.
6. The distinctions among the syntactic properties of constituents that were discussed in endnote 2 may well apply to Mandarin. Like the English main verb, the Mandarin V1 seems generally to exhibit the properties of the head of a construction. And like the English satellite, the Mandarin V2 seems generally to exhibit the properties of a framing constituent. As endnote 2 indicated, the framing constituent generally is the target of negation or interrogation, determines the overall aspect, and can determine much of the argument structure.
References
DeLancey, S. 1989. Klamath stem structure in genetic and areal perspective. Papers from the 1988 Hokan-Penutian languages workshop, 31-39. Eugene, OR: University of Oregon.
Matisoff, James A. 1973. The Grammar of Lahu. UCPL no.75. Berkeley and Los Angeles: University of California Press.
Matisoff, James A. 1991. "Areal and universal dimensions of grammatization in Lahu". In Elizabeth C. Traugott and Bernd Heine, eds., Approaches to Grammaticalization, Vol. II, pp. 383-453. Amsterdam: Benjamins.
Schultze-Berndt, Eva. 2000. Simple and complex verbs in Jaminjung: A study of event categorisation in an Australian language. MPI Series in Psycholinguistics, no. 14. Wageningen, Netherlands: Ponsen and Looijen.
Slobin, Dan I. (2004). The many ways to search for a frog. In Sven Strömqvist and Ludo Verhoeven (eds.), Relating Events in Narrative. Typological and Contextual Perspectives, 219-257. Hillsdale, NJ: Lawrence Erlbaum.
Slobin, Dan I. & Nini Hoiting (1994). Reference to movement in spoken and signed languages: Typological considerations. Proceedings of the Twentieth Annual Meeting of the Berkeley Linguistic Society. Berkeley: Berkeley Linguistics Society, 487-505.
Talmy, Leonard. 1972. Semantic Structures in English and Atsugewi. Berkeley: University of California Ph.D. dissertation.
Talmy, Leonard. 1985. Lexicalization patterns: semantic structure in lexical forms. Language Typology and Syntactic Description, vol. 3: Grammatical Categories and the Lexicon, ed. by Timothy Shopen, 57-149. Cambridge: Cambridge University Press.
Talmy, Leonard. 1991. Path to realization: a typology of event conflation. Papers of the Seventeenth Annual Meeting of the Berkeley Linguistics Society, ed. by Laurel A. Sutton, Christopher Johnson and Ruth Shields, 480-520. Berkeley: Berkeley Linguistics Society.
Talmy, Leonard. 2000a. Toward a Cognitive Semantics, volume I: Concept structuring systems. i-viii, 1-565. Cambridge: MIT Press.
Talmy, Leonard. 2000b. Toward a Cognitive Semantics, volume II: Typology and process in concept structuring. i-viii, 1-495. Cambridge: MIT Press.
Talmy, Leonard. 2005. Interview: A windowing onto conceptual structure and language. Part 1: Lexicalisation and typology. [Written interview by Talmy on his work conducted by Iraide Ibarretxe] Annual Review of Cognitive Linguistics, vol. 3, 325-347. John Benjamins.
Zlatev, Jordan and Yangkland Peerapat (2004). A third way to travel. The place of Thai in motion-event typology. In Sven Strömqvist and Ludo Verhoeven (eds.), Relating Events in Narrative. Typological and Contextual Perspectives, 159-190. Hillsdale, NJ: Lawrence Erlbaum.