What are the Limits of Polysynthesis?

Oxford Handbooks Online

What are the Limits of Polysynthesis?  
Michael Fortescue
The Oxford Handbook of Polysynthesis
Edited by Michael Fortescue, Marianne Mithun, and Nicholas Evans

Print Publication Date: Sep 2017 Subject: Linguistics, Morphology and Syntax
Online Publication Date: Nov 2017 DOI: 10.1093/oxfordhb/9780199683208.013.14

Abstract and Keywords

Of the various labels for morphological types currently in use by typologists

‘polysynthesis’ has proved to be the most difficult to pin down. For some it just represents
an extreme on the dimension of synthesis (one of Sapir’s two major typological axes),
while for others it is an independent category or parameter involving incorporation and
bound pronominals with far-reaching morphosyntactic ramifications. If the nub of
polysynthesis is the packing of a lot of material into single verb forms that would be
expressed as independent words in less synthetic languages, what exactly is the nature of
and limitations on this ‘material’? This chapter investigates the limits—both upwards and
downwards—of what the term is generally understood to cover and suggests a rule-of-
thumb definition. Cognitive constraints on its maximal extent are also considered.

Keywords: types of polysynthesis, bound pronominals, incorporation, cognitive constraints, morphosyntactic

ramifications

7.1 Introduction: the Diversity of Polysynthesis

THE apparently simple question posed by the title of this chapter actually covers several
distinct questions, none of them particularly easy to answer. At the most superordinate
level ‘limits’ can refer to limits on the complexity displayed by any polysynthetic
language, but it can also refer to the very definition of polysynthetic languages—what
counts as one in the first place? The question as to the most extreme manifestations of
polysynthesis can itself be broken down into several sub-questions: what are the limits on
the pure number of morphemes per word—and should one include clitics as well as roots
and affixes in that count? How many lexical roots are possible in a single word? How
many functions can incorporated nouns serve? How many distinct slots can there be in a

Page 1 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

templatic form of morphology—and how many individual lexical affixes can fill the
individual slots? How many pronominal referents can a polypersonal ‘word-sentence’
refer to? How recursive can a polysynthetic word be (with successive changes of syntactic
status)? And how many different (types or tokens of) adverbials can be drawn into the
verbal complex? Ideally, we would like to be able to say in each case that no polysynthetic
language can be found that has more than ‘X’ (where X is a number referring to any of
these ‘limits’). There is of course no guarantee whatsoever that a maximal number on any
one score means a particularly high number on any of the others. One can easily sift
through the literature and find languages that take the prize on one or another of these
counts (the five hundred or so lexical affixes of southern Wakashan languages, the great
number of templatic slots in northern Athabaskan languages, the recursivity of Eskimo,
the multiple incorporation of stems in Chukotian, and so on), but it is unlikely that any
one language would be able to maintain a maximal score on all the vectors listed.

A major problem for determining ‘X’ on any of these measures is that for many of these
senses of ‘limit’, the answer must remain fuzzy, a matter of production and
comprehension constraints and individual skill or style. Take the question of recursivity.
The Eskimo word in particular is theoretically extendable ad libitum, although it becomes
increasingly (p. 116) more difficult to find words with more than two or three (productive)
switches back and forth between verbal and nominal base before being capped off by a
suitable inflection. Lexicalization of complex bases (stems plus affixes) may extend this
number somewhat, but limitations of memory/intelligibility will soon curtail further
expansion. Despite all this, I would claim that it is useful to have a rule-of-thumb
definition of what counts as polysynthetic. I shall present such a definition in what
follows, one that bears certain consequences for the question of quantitative limitations
that have been listed.

First of all, it should be stressed that polysynthetic languages, far from representing a
single homogenous ‘type’, are arguably the most diverse and complex languages on
earth, at least as regards their verbal morphology. Nor are they just any language with
long words.1 Boas described polysynthesis thus: ‘A large number of distinct ideas are
amalgamated by grammatical processes and form a single word, without any
morphological distinction between the formal elements in the sentence and the contents
of the sentence’ (Boas 1911: 74). One might add that the criterion for deciding what
counts as a ‘single word’ is generally understood to be the obligatory ordering of
constitutive morphemes (with no interruption or cut-off possible between the first and the
last), under a unitary prosodic ‘packet’ and with at least some degree of morphophonemic
‘glue’ between constituents. The only potential problem with this is the status of clitics (I
shall return to this in connection with Wakashan languages).

Since the time of Boas there have been various attempts to characterize the exact purport
of the term more precisely. All of these remain however rather impressionistic. For some
researchers the term just refers to an extreme elaboration of the dimension of synthesis,
one of Sapir’s two orthogonal typological axes, along with morphological
‘technique’ (Sapir 1921: 20–130). These correspond more or less to Comrie’s ‘index of

Page 2 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

synthesis’ as opposed to his ‘index of fusion’ (cf. Comrie 1983: 42–3).2 For others it is an
independent category or parameter with far-reaching morphosyntactic ramifications (cf.
Baker 1996). Whether languages like Eskimo, lacking noun incorporation and
compounding, count as polysynthetic has remained a matter of theoretical controversy,
despite their displaying holophrasis and great morphological complexity.

One of the principal goals of the present handbook is to narrow down the extension of the
term while at the same time expanding our understanding of its content and possible
theoretical ramifications. The starting point for this endeavour is the following question:
If the nub of polysynthesis is the packing of a lot of material into single verb forms that
would be expressed as independent words in less synthetic languages, what exactly is the
nature of and limitations on this ‘material’? Beyond the sheer number of such constitutive
morphemes, ‘core’ or ‘prototypical’ polysynthesis involves holophrasis, which can be
defined in a narrow technical sense: verb forms that can stand alone as independent
clauses without any additional lexical arguments, with all core arguments referred to by
bound pronominals. This feature appears central in the sense that it has implications
beyond itself, particularly (p. 117) as regards the nature of syntactic structures in the
language. However, according to a widespread view as to what qualifies as a
polysynthetic language, not all languages that display holophrasis in this sense will
necessarily qualify as core polysynthetic. It would also seem necessary that more than
one lexically ‘heavy’ morpheme can appear in complex verb forms, whether this be an
additional lexical stem—nominal, verbal, or adverbial—capable of standing alone, or a
bound morpheme functionally equivalent to a stem, such as the lexically heavy suffixes of
Eskimo-Aleut and Wakashan or the Algonquian ‘medials’ and ‘finals’ with verb-like
meanings. Both these major features are symptomatic of strong head-marking tendencies
– although polysynthetic languages are not exclusively head-marking.3 Nor are all head-
marking languages polysynthetic (e.g. many Nilo-Saharan languages).

There is, I repeat, great variation among polysynthetic languages: they are not all
complex in the same way. Thus in Fortescue (1994: 2602) the following sub-types were
distinguished: pure incorporating (as in Chukchi); recursive suffixing (as in West
Greenlandic); lexical (or field-) affixing (as in Atsugewi); and various hybrid combinations
of these—also a fourth ‘serial verb compounding’ type (as in Yimas) was mentioned.
Mattissen (2003: 281–8) has a somewhat different (but compatible) breakdown into sub-
types. She classifies them according to two cross-cutting dichotomies which can be
combined in various ways. The first dimension indicates the principal word-formational
type (i and ii below), and the second the principal method of organizing the morphology
internally (iii and iv).4 She defines them as follows:

i) The affixal type allows only one lexical root per verb complex (and has non-root
bound affixes for all other morphemes within the word).
ii) The compositional type combines more than one lexical root (whether noun plus
verb or verb plus verb) in complex verb forms in an ad hoc manner.

Page 3 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

iii) Templatic organization is characterized by a fixed number of slots in a fixed

order.
iv) By scope ordering, successive affixes are chained freely, as conditioned by their
semantics, scope relations, and compatibility restrictions.

Compositional languages may display both noun incorporation and heavy verbalizing
affixes, the latter a species of what Mattissen calls non-root bound morphemes more
typical of the affixal type of language. Hence her ‘mixed’ types (op. cit.: 284–5). Within
the compositional type there is a distinction between those that (primarily) incorporate
nominal stems and those that incorporate (or serialize) verb stems—though many allow
both. There is also diversity of a functional nature, along a continuum ranging from
simple word-internal compounding to ‘syntactic’ incorporation (when the ‘incorporate’
may display some degree of referentiality and/or adjectival modifiers be stranded when
the NP head is incorporated—see Sadock 1991: 88–91).5 The degree of productivity of
verb-internal (p. 118) morphemes and incorporative constructions can vary considerably,
as can the number of templatic ‘slots’, the number of items filling them, and the
transparency of incorporates. The kinds of morphemes representing fully
grammaticalized categories can also vary—thus animacy distinctions in Algonquian
languages, aspectual ones in Southern Wakashan, deictic ones in Northern Wakashan,
and distinctions of subject/object ‘figure’ in Athabaskan are all obligatorily marked and
pervasive, contributing significantly to the distinctive profiles of these families.

Another way of looking at Mattissen’s four-way distinction is to say that polysynthetic

languages split into those that allow canonical incorporation and those that don’t, and
that cross-cutting this is the degree to which a given language exploits the natural
rigidity of morphology, a continuum from maximal (purely templatic, compensated for by
syntactic flexibility) to minimal (purely scopal, with syntax-like flexibility). Languages
displaying canonical incorporation tend to have various restrictions on what can be
incorporated, while polysynthetic languages of the affixing type have fewer restrictions
on the single stems that can be quasi-incorporated. Much of the diversity attested may be
due to different historical pathways into polysynthesis, for which see Fortescue (2016).

Although this diversity makes it difficult to delineate a purely linguistic ‘macro-type’

behind polysynthesis, there is a positive side to it: a careful comparison of the traits
shared by languages both within such sub-types and across them promises, by virtue of
the very complexity of their morphological systems, to reveal many more
‘(micro-)implicational universals’ of the type: noun subject incorporation presupposes
noun object incorporation, or bound indirect object pronouns presupposes bound direct
object pronouns. Further research will hopefully show that such relationships (in so far as
they hold) tend to link up into somewhat larger functional clusters of interrelated
universals. It may eventually be possible to list a set of interrelated features common to
each of the sub-types that have been suggested, although doing so may prove difficult,
given the existence of hybrid types and varying stages of grammaticalization.

Page 4 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

However diverse in internal structure and constituent types the polysynthetic ‘word-
sentence’ can be, I shall suggest in the following section that the source of this
characteristic holophrasis is a certain way of codifying the underlying ‘macro-event’ (to
use Talmy’s term). All polysynthetic word-sentences display what Bohnemeyer et al.
(2011: 48) call the ‘macro-event property’, by which they mean that sub-parts of the
complex clause cannot be individually modified by temporal operators (tense and/or
aspect), such operators by necessity taking scope over the whole clause.6 In polysynthetic
languages tense and aspect are typically marked by affixes within the holophrastic word-
sentence (though sometimes by displaceable clitics), with scope over the whole word.
Recognizing this will actually help us narrow down the limits of ‘what counts as a
polysynthetic language’. The resulting tentative definition will hopefully harmonize with
the intuitions of both typologists and specialists in (p. 119) individual language families.
En route I shall present arguments as to why certain languages should be excluded and
others included within the core category.

7.2 A Limit for the Polysynthetic Word-

Sentence: The Macro-Event?
For Talmy, the ‘macro-event’ is the ‘conceptual integration of a complex event’ (see Talmy
2000, vol. 2, Ch. 3 on the typology of ‘event integration’). It consists of one main or
‘framing’ event defining the abstract core of the event (including its ‘upshot’) plus one or
more subordinate or ancillary event (‘co-event’) filling in the more specific and
perceptually vivid content of the macro-event, as in Figure 7.1 (from Talmy op. cit.: 221).7

A macro-event may be compacted into a single clause (as typically in polysynthetic

languages—but not limited to them) or expressed as more than one syntactic clause
(more typical of analytic languages). In polysynthetic languages this integration is
maximal, potentially compacting the macro-event into a single word—though there is
often a choice between maximally integrated and less integrated constructions,
depending for instance on the referenciality of nominal elements in the clause or on the
relative independence of the verbs in a serial chain.

Now although this appears to be a reasonable starting point for investigating

polysynthetic ‘holophrasis’, questions start to pile up when one looks at specific
manifestations, applying Talmy’s ‘macro-event’ framework to all varieties of polysynthetic
language across the board. How, for instance, can Talmy’s conception of the ‘framing-
event’ be applied to Eskimo word-sentences containing ‘double transitivizer’ affixes like -
niRaR- ‘say that’—in example (1) (from Fortescue 1984), or to multiple applicative
constructions as in Ainu sentence (2) (from Shibatani 1990), or to word-sentences
containing more than one ‘clause’, as in the Bininj Gun-wok sentence (3) (from Evans
2003)? And what of a sentence like (4) in Yimas (from Foley 1991), which contains two
serially combined lexical verbs, or the Nuuchahnulth word-sentence (5) (from Nakayama

Page 5 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

1997) containing numerous clitic elements that include an (inverse) subject pronominal,
or sentence (6) from Koyukon, a language displaying discontinuous morphemes? And
what, for that matter, of the Nivkh verb (p. 120) in (7) (from Mattissen 2003) that clearly
does not contain a whole clause (this is spread over several independent words)?

Click to view larger

Figure 7.1 The conceptual structure of the macro-
event

(Talmy 2000)

(1)

(2)

(3)

(4)

(5)

Page 6 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

(6)

(7)

Despite these apparent difficulties, it is not unreasonable to suggest that, for instance, in
(1) the suffix -niRaR-, which takes in its scope everything that precedes it and determines
the transitivity of the whole clause, represents a framing event, here of the type Talmy
calls ‘action correlation’ linking two events with different agents.9 This is not just simple
causation, which would be a matter of a co-event. In the Ainu case (sentence 2) the two
applicative (p. 121) morphemes draw different oblique elements into the body of the
clause, but cognitively a unitary macro-event is still being described and it could be
argued that the applicative constructions reflect co-events. As for the Bininj Gun-wok
sentence (3), the incorporated ‘clause’ (marked by a special affix IVF) clearly expresses a
subordinate co-event (of ‘concomitance’ or simultaneity)—the main event just happens to
be expressed by elements around it. In sentence (4) from Yimas, one of the verb roots
arguably refers to the main framing event (‘searching’, which takes an externally
expressed object), while the other refers to a subordinate event of manner (‘walking’).

Sentence (5) from Nuuchahnulth, (6) from Koyukon, and (7) from Nivkh likewise refer to
single macro-events. In the case of (5) the ‘doing together with someone else’ affix, even
though it requires an ‘empty’ stem to attach to, can be analysed as referring to the main
framing event—another example of Talmy’s ‘action correlating’ type. In (7) only part of
the referring expression (the object of the verb)—that is, half of the coordinated NP—is
incorporated into the verb complex. This is a matter of ‘stranding’, required in this kind
of incorporating structure and does not effect the status of the sentence as referring to a
unitary event.

Sentence (7) is also significant in another way: it leads us to question whether such a
structure really is ‘polysynthetic’ since the subject of the main clause is not incorporated
into the complex verb. Subjects, whether lexical or pronominal, are independent words in
this language, separable from the verb by other material. We shall need to turn to the
matter of definition in order to see if we can fit languages like Nivkh into a general
characterization of the phenomenon—certainly it would be desirable to be able to do so,
given other, strongly polysynthetic aspects of its verbal morphology (including noun
incorporation). Our rule-of-thumb definition of prototypical or ‘core’ polysynthesis should
allow for such a decision.

Page 7 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

Approaching the question of limits from the perspective of form rather than content,
another useful starting point lies in the nature of the holistic ‘word’. In discussing the
word in Central Alaskan Yupik (a quintessentially polysynthetic Eskimo language),
Miyaoka (2012: 26–30) distinguishes between free/independent words, weakly or strongly
bound phrases, and free phrases, as in Figure 7.2.

A free word in Yupik—a single prosodic ‘articulus’—may be of any degree of complexity

(by affixation), and an inflected verb may stand alone as a complete sentence/clause. A
free word can easily be distinguished from a bound phrase in this language, that is, from
the combination of a free word and a clitic (itself a word but one requiring a host). There
are also non-clitic bound phrases, a word-like phrase consisting of two words capable of
standing alone but still forming one articulus with an internal boundary between the
words (e.g. CAY aŋyaq(-)qaa ‘is it the boat?’, where qaa on its own means ‘is that right?’).
There is a continuum between strongly and weakly bound phrases, with weakly bound
phrases close to free or (p. 122) syntactic phrases. Strongly bound phrases tend with time
to become fixed compounds—of these Yupik displays few if any. Depending on the
language type concerned, there may be no clear-cut divide between morphology and
syntax. This descriptive apparatus can nevertheless be applied to all languages that one
might wish to characterize as polysynthetic.

Click to view larger

Figure 7.2 Miyaoka’s (2012) analysis of the ‘word’

Page 8 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

7.3 Towards a Definition of Polysynthesis

From the arguments presented so far it should be evident that the definition of core
polysynthesis that I have been homing in on is something like this:

• To qualify as core polysynthetic a language must display holophrasis (i.e. be able to

represent a whole clause—including all bound core pronominals—by a single word)
and must allow more than one lexically ‘heavy’ morpheme within the holophrastic
verb, whether it be lexical or affixal.

This combines an inflectional and a derivational component, note (bound pronominals on

the one hand and more or less open-ended derivational/incorporative productivity on the
other). This is not meant to claim that polysynthetic languages have to produce
holophrastic sentences of this kind, only that they can choose to do so. Most such
languages have less synthetic alternative constructions, utilizing the difference for
pragmatic or referential purposes. As regards bound pronominals, this must include
object (patientive) as well as subject (agentive) affixes—clitics may usually be included
here (as I shall discuss in connection with Nuuchahnulth). ‘Pro-drop’ or simple zero-
marking of all verbal arguments does not count.

It is important to understand what is meant by ‘heavy morpheme’ in the definition just

given. It refers of course to semantic as opposed to phonological heaviness. As a rule of
thumb, what is meant to be covered by the term—apart from actual independent stems or
their bound variants—are items that could in theory be represented by an independent
word or stem in the language concerned, that is, to which there exist semantically
comparable (but not identical) words in the language, although lexical stems will typically
be more specific than comparable word-internal morphemes. Mithun (2009: 13) has in
fact suggested that the heavy verbalizing affixes of Eskimo-Aleut must derive historically
from full verbs in incorporating constructions. Certainly there is evidence that auxiliaries
*ŋu- ‘be’ and *ŋit-/ŋil- ‘be without’, now affixes, were once independent stems.10 But this
may not be the whole story: incorporated auxiliaries could combine with various
preceding affixes or bring their own modifying affixes with them, producing new
lexicalized combinations like West Greenlandic -sinnaa- ‘can’ from nominal affix -sinnaq
‘only’ plus ŋu- ‘be’. Some synchronic verbalizers conceal earlier combinations of a more
general verbalizer plus a verbal extender, such as Inuit -niaq- in the sense ‘hunt X’, which
is actually a combination of *-ɣ- ‘catch’ (< *-nəɣ- ‘get’) plus verbal extender* -ni(C)aR- ‘try
to’.

(p. 123) The term ‘heavy morpheme’ refers at all events to morphemes with a stem-like
lexical content, whether nominal, verbal, or adjunct. This is not particularly difficult to
determine with ‘verbalizing’ affixes of the Eskimo kind or ‘lexical affixes’ of the Wakashan
kind, which may be compared with verbal and nominal stems respectively, but abstract or
general verbalizers depending for lexical content on the nominal stem they attach to (as
in many agglutinative languages) do not count. It should also be borne in mind that in
Page 9 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

some languages meanings such as ‘want to’ and ‘intend to’ when expressed
morphologically may retain full lexical semantics, whereas in others they form a
grammatical category (entering into paradigmatic oppositions/choices). In some
polysynthetic languages prosody may provide assistance to distinguish the two cases, as
in Chukchi, where (newer) lexical incorporates tend to maintain secondary stress as
opposed to the same morphemes when they have become fully grammaticalized—for
example pələtku- ‘finish’, which loses its independent stress when occurring as an
aspectual affix of completed action.

As regards adjuncts (adverbial or adjectival) the degree of lexical heaviness may be more
in doubt. This is in part the inevitable result of grammaticalization processes, whereby
morphemes may lie anywhere on a cline between fully lexical (and concrete) and fully
grammaticalized (more abstract). They may even occur in more than one meaning
reflecting different ‘layered’ historical stages (for instance, morphemes referring to
concrete motion on the way to becoming aspectual or tense morphemes). The difference
between incorporated adjunct and heavy adverbial or adjectival affix may also be
obscured (where the source lexeme is lost or shifted in meaning). This does not cause any
great practical difficulty in applying the definition as given, since any language having
incorporated heavy adjuncts will almost certainly have the possibility of other lexically
heavy morphemes (nominal or verbal) incorporated within the polysynthetic verb too.

In order to give a better idea of what I mean by ‘lexically heavy’ affixes, consider the
following types in one polysynthetic family, namely Eskimo (as in Fortescue 1983). Some
of these are verbalizing (those with ‘X’, indicating a suitable nominal stem) and some
verb-extending. All may be considered ‘heavy’ in having concrete behavioural, perceptual
or mental semantics beyond the expression of abstract grammatical or class-shifting
categories.

Lacking X
Feeling X
Having X
Acquiring X
Movement to/from/via X
Acting or seeming like X
Doing with or providing X
Judging or saying
Wishing or waiting for
Asking to
Striving after or intending

Affixes of (adverbial) manner could also be added—these are more concrete/perceptible

than, for example, aspectual affixes, and arguably display a degree of lexical weight.
Simple adverbial affixes (including path directionals, aspectual, modal, and valency-
changing (p. 124) affixes) I do not count as heavy, but affixes of manner I do if they

Page 10 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

display types of meaning corresponding to lexical verbs or verbal phrases (‘by swimming’,
‘by walking’, etc.).11

The definition I have given appears more categorical and less disjunctive than the loose
list of ‘typical’ polysynthetic traits in Fortescue (1994), which included: noun stem
incorporation within the verbal complex; a large inventory of bound morphemes hand in
hand with a limited stock of independent stems; derivational processes being productive
in the formation of individual words as minimal sentences; bound pronominal marking of
subjects and objects; integration of locational, instrumental, and other adverbial elements
into the verbal complex; productive morphophonemic processes resulting in allomorphy
for both stems and affixes; non-configurational syntax; and head-marking type of
inflection. But it is still the case that no one concrete feature is treated as criterial under
the present definition—neither noun incorporation, polypersonalism, head marking,
recursivity, nor the presence of applicatives, etc., all of which can be found individually in
non-polysynthetic languages. Thus, for example, colloquial Arabic is more ‘polypersonal’
than West Greenlandic (allowing three bound pronominal affixes on verbs—indirect
object, direct object, and subject—while West Greenlandic allows a maximum of two), but
surely no one would want to call Arabic polysynthetic. Only ‘entanglement of syntax and
morphology’ from the list remains a feature largely restricted to polysynthetic languages
—though also here not all polysynthetic languages share it (especially ‘younger’ ones
closer to their non-synthetic syntactic origins).

My definition is broader than that of Baker (1996: 17–18), which requires both bound
pronominals and noun incorporation for a language to be considered polysynthetic (in his
narrow technical sense). This excludes, among others, Eskimo, Navajo, Haisla, Abkhaz,
and Yimas, which specialists would all want to call polysynthetic, and Baker has received
considerable criticism from such quarters (see, for example, Evans and Sasse 2002: 4).
My definition is nevertheless narrower than Mattissen’s (2003: 277) in so far as she does
not demand polypersonalism (i.e. all core pronominals being bound) of polysynthetic
languages. In Appendix 2 I present an approximate list of those languages which can be
considered ‘core polysynthetic’ according to my definition. It should be understood that
the definition is really only a practical rule of thumb: lexicalization and
grammaticalization may gradually reduce the ‘heaviness’ of originally heavy incorporated
morphemes to more abstract or (p. 125) grammatical ones, resulting in a transitional grey
zone. Within the definition there is obviously room for a continuum from weakly to
strongly polysynthetic—something that can be roughly quantified in terms of average
number of productive morphemes per (verbal) word in running text.

Note that the definition precludes Spoken French and Bantu languages as core
polysynthetic (though both of them have sometimes been considered ‘polysynthetic’):
they both fall short as regards the second part of the definition. One can nevertheless
characterize them as ‘mildly’ or ‘peripherally’ polysynthetic, drawing the boundary
somewhat arbitrarily.12 Japanese, on the other hand, has multiple verbal (and adjectival)
stem serialization but lacks bound pronominals entirely, so falls short as regards the first
part of the definition. The definition does, however, include Yimas since this language

Page 11 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

allows multiple verbal stem serialization in one word (although it does not have
productive noun incorporation), also Eskimo languages since they display ‘heavy’
verbalizing affixes, and Na-Dene languages such as Koyukon since many of its prefix
strings are ‘heavy’ in this sense—this is even so in those southern Athabaskan languages
that do not display noun incorporation.

But what of Nivkh (as in the example given in sentence 7), which does not display
complete holophrasis, since the pronominal subject of the clause must be expressed as a
free constituent? And what of Wakashan languages (as in sentence 5) in which core
elements of the clause are expressed by displaceable clitics rather than affixes? As
regards the former, the rather special kind of polysynthesis displayed by Nivkh—
Mattissen’s ‘dependent-head’ organization—may be said to force the subject pronominal
to precede any object phrase. This is because it requires, in conjunction with the strict
SOV syntax of the language, objects of any kind to immediately precede the head verb,
and this includes bound (pronominal) and free NPs. So if the subject were a bound
pronominal, any free NP object would have to precede it, breaking the strict dependent-
head relationship that cuts across the morphology/syntax boundary. Once pronominal
subjects have to precede NP objects, other constituents like adverbial phrases may
intervene between the subject and the verb. In other respects Nivkh qualifies as
polysynthetic. Perhaps it can be characterized as ‘polysynthetic with
qualifications’ (hence the question mark after it in Appendix 2) since it still meets the first
criterion—but only as regards bound object (not subject) pronominals. Usually
incorporation of the former presupposes that of the latter. In fact, many verb-forms in
Nivkh (including most ‘converbs’) do reflect the subject/agent, it is only the plain
indicative that does not, and, moreover, the Nivkh verb can stand alone as a complete
sentence without subject marking (Mattissen 2003: 205–6). Note that ‘with qualifications’
does not necessarily mean ‘mildly polysynthetic’, just that one crucial aspect of ‘core
polysynthesis’ is only partially realized. Thus many of the highly synthetic Amazonian
languages discussed by Aikhenvald in Chapter 15 of this volume do not mark all
pronominal object referents by bound markers on the verb and thus must be
characterized as ‘polysynthetic with qualification’. This underlines the fact that highly
complex (verbal) morphology and polysynthetic ‘type’ are not necessarily equivalent
under any definition of the latter.

Other North American languages that may best be qualified as ‘polysynthetic with
qualifications’ according to my definition include Aleut, Tsimshianic, and Haida, which
display (p. 126) pronounced polysynthetic traits (heavy verbalizing affixes in the first two,
a wealth of instrumental prefixes in the last), but in which pronouns are not always bound
to the verb (they display what Mattissen this volume, Chapter 40 calls ‘defective
polypersonalism’). Unlike Eskimo, Aleut object pronominals are only bound in the 3rd
person anaphoric object construction. In the case of Tsimshianic (Nishga), pronominal
subjects of transitive verbs are independent words (1st/2nd person) or enclitic (3rd). In
the case of Haida, object pronominals usually cliticize to verbs but weaker elements can
in turn cliticize to them. When more than one pronoun is involved, only the one nearest
the verb cliticizes to it. The chaining of verb roots and adverbials in the verb complex in
Page 12 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

Haida is moreover very loose and open phonologically, but such chains can nevertheless
be analysed as representing polysynthetic wholes, given the obligatory position of shared
inflectional material following them. As with Nivkh, we are dealing with a language
where the border between morphology and clausal syntax is not watertight. Foley (1991:
280) also characterizes Yimas in the same way: as a language with a hazy distinction
between word and phrase and between clause and sentence.

It is easier to consider Wakashan languages like Nuuchahnulth as core polysynthetic

without any ‘qualification’, despite the fact that pronominal inflections are clitics (at least
according to the analysis of Davidson 2002: 252ff.).13 It should be pointed out that clitics
in Wakashan do not quite fit Miyaoka’s characterization of a clitic as ‘a word with a
certain independence, but one requiring a host’ (which fits the case of Eskimo), since
clitics in these languages can hardly be said to have any degree of independence. Like
Eskimo clitics they are typically added to a wide variety of hosts and generally refer to
the phrasal/clausal level rather than just the verbal semantics. However, the latter point
can also be made of affixes (lexical and aspectual) in these languages, so this does not get
us very far. The only essential difference between affixes and clitics here concerns their
relative ordering (clitics follow affixes) and some rather minor phonological differences in
behaviour. Neither clitics nor affixes are obligatory: a stem (with or without an aspectual
or lexical suffix) may represent a well-formed clause in most instances. Third person
singular subject is generally marked by zero and even 1st/2nd person clitics may be left
out in certain circumstances (cf. Nakayama 1997: 84).14 Bound object pronominals of any
person are only present in imperative paradigms today (they are better preserved in
related Ditidaht and Makah). The point is that pronominal arguments that are clitic, in
the absence of any other independent words, can bind directly with the verb to form a
unitary word-sentence. There is no question of obligatory free-standing pronominals in
any construction. This is not to say that clitic core argument pronominals will always
contribute to polysynthesis. They may not in languages where such morphemes are more
loosely associated with predicates, occurring always in a fixed clausal position (usually on
other hosts) or where they can appear either as clitics or as independent words.
Neighbouring Salish may be a case in point (see Watanabe this volume). For a broader
typological coverage of clitics see Aikhenvald (2002a).

To sum up: rather than stating that a language X simply is or is not polysynthetic,
(p. 127)

one should say something like: ‘X is a highly/mildly polysynthetic language of the

predominantly compounding or affixal (and scopal or templatic) type—with or without
qualifications’. Further consequences/expectations will arise from this characterization,
and any ‘qualifications’ (deviations from my definition of ‘core’ polysynthesis) can be
specified. There will usually be some diachronic or areal reason lurking.

7.4 Conclusions

Page 13 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

How does my tentative definition relate back to Talmy’s notion of underlying ‘macro-
event’? Let us take up the case of Eskimo recursivity again: however many shifts between
verbal and nominal base may accumulate in a single word, it appears that the result will
always correspond to a single ‘macro-event’. Thus consider sentence (8) following (from
Fortescue 1984), in which a verbal stem is verbalized, then nominalized, then re-
verbalized, then nominalized again, and finally verbalized (to be inflected as a verb).

(8)

The affixes here vary in productivity. The first affix (underlying -t-) is limited to a handful
of stems indicating animals, whereas all the others are quite productive (fully so in the
case of -u- ‘be’, which can attach to any nominal base). Nevertheless, there is clearly a
single ‘macro-event’ described here (an ascription of a quality to an entity). The first four
morphemes define a single abstract quality, while the addition of the next two extends
this to a type of entity displaying this quality to a high degree. The specific entity
exemplifying this type is only referred to (following the copular affix) by the 3rd person
inflectional ending. Now Talmy states that it is the framing event (whether expressed by a
satellite or a verb) that reflects the main ‘upshot’ of the macro-event and not the
semantically richer co-event (Talmy 2000: 282ff.). The element in the clause that
determines this ‘upshot’ will determine its syntactic transitivity (as well as, for instance,
its telicity). That certainly applies to the final (intransitive) suffix -u- in (8), which requires
the following inflection to be intransitive. By contrast, in sentence (1) the final suffix -
niRaR- required a following transitive inflection, despite the base to which it attaches
being intransitive. X-u-, and X-niRaR- thus represent different types of superordinate
framing event, determining the overall ‘profile’ of any clause in which ‘X’ is spelled out in
detail. Moreover, under negation it is the assertion associated with the framing element
that is negated—thus -niRaR- followed by an affix of negation would negate the reporting
event, not that reflected by the base. This too is a characterization of morphemes
reflecting the ‘framing event’ (Talmy op. cit.: 219). It is also the same element that is
demanded in imperatives and asked about in interrogatives.

This suggests that there is indeed one—and only one—element in the morphology of the
polysynthetic word-sentence that directly reflects the main framing event, the core
schema plus activating process. In Eskimo at least, this is the final valency-determining,
lexically (p. 128) heavy affix—only if there is none present is it expressed by the
underlying semantics of the base itself (and the restrictions this imposes on possible
inflections). The polysynthetic word-sentence may, on the other hand, contain several ‘co-
event’ elements, some reflected in the choice of base, others—for example, of manner—in
additional affixes. (This is a matter that Talmy, focusing on Atsugewi, does not go into in

Page 14 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

detail.) Which co-event elements are chosen for expression in this manner is very much a
language-specific matter.

Much of the multiple category shifting in Eskimo languages is embedded in complex

nouns, as in (9) from West Greenlandic, with a stranded modifier of ‘quasi-incorporated’
kavaajaq (here in allomorphic form kavass-) ‘coat’.15 Here the nominalizing affix (-ɣaR-)
has of course nothing to do with the framing event of the whole clause—that is
determined solely by the verbalizing affix -isaaR- ‘wear’, which reflects a stative framing
event.

(9)

This latter construction is not typical of polysynthesis generally, and is probably limited to
the ‘recursive’ type which West Greenlandic represents. As I have emphasized
throughout, when we talk of polysynthetic languages in general we are really conflating a
number of different sub-types. The templatic northern Na-Dene languages and the scopal
(‘onion skin’) Chukotian ones, for instance, display quite different properties on a number
of dimensions despite noun incorporation being common to both. But in all of these sub-
types there appears to be a limit of ‘one word = one macro-event’. Thus the Chukchi
sentence in (10) (from Skorik 1961) also refers to a unitary macro-event—the complexity
stems from the complex nature of the incorporated nominal phrase.16

(10)

The lexical core of the word lies in the (inherently transitive) verb pəla- ‘leave’. In Talmy’s
scheme of things it directly reflects the ‘framing event’ (one of caused motion/position).
This is an abstract schema involving a figure and a ground expressed by the verbal base
and an associated relationship of agency. The actual concrete contents of the schema is
indicated by the choice of verb plus its incorporated object. In this case, then, the main
framing event could again be said to be concentrated in one morphological element—here
the base rather than an affix.

Even Yimas, which is rather extreme as regards the sheer number of morphemes that can
be packed into a single word, is limited by the ‘macro-event’ principle. According to Foley
(1991: 331) up to three verbal stems can be compounded into a complex verb, although
two is (p. 129) far more common. This suggests a challenge for the applicability of the
notion of the unitary ‘macro-event’ to the individual polysynthetic word-sentence. It can
nevertheless be argued that in this language there is indeed only one verb stem
corresponding to the main ‘framing’ event, namely that upon which the transitivity of the
whole clause (and its general ‘upshot’) depends, the one with the highest inherent
Page 15 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

transitivity—or if both share the same transitivity, then the final one. The whole word-
sentence refers to ‘a single complex event’ (Foley 1991: 326), but the bound pronominal
affixes refer to the arguments of the verb of highest transitivity (the subject must be the
same for all the serialized verb stems regardless of transitivity). The others can be
regarded as subordinate, either simultaneous or sequential. Note that this refers to
semantic subordination, not to linguistically marked subordination—the serialized verb
construction in Yimas is specifically referred to by Foley and Van Valin (1984: 242) as ‘co-
subordination’ (at the level of ‘nuclear juncture’). Note that this is a somewhat different
conclusion from Foley’s dependent-head analysis of Yimas (this volume), whereby the
rightmost serialized verb (just before the tense markers) is regarded as the
morphosyntactic ‘head’ of these constructions.

Sentence (4) was an example of the ‘simultaneous’ case, with the verb of highest
transitivity being the ‘main’ one determining the pronominal marking. More complex is
the case of sequential serialization, where the individual verb stems are linked by a
sequential marker -mpi- as in (11). The sequence is purely temporal, the final one being
the last in the series of actions conceptualized as a unitary whole.

(11)

Here, where both lexical verbs are intransitive, there is no difficulty seeing the last
sequence (‘start to walk’) as the core of the event (tal, literally ‘hold’, has a purely
inchoative sense here). Foley (op. cit.: 331) also gives examples with three transitive
verbs of the type ‘split-break-tie’ referring to splitting things, breaking them, and tying
them in bundles in a purely temporal sequence, also of cause/effect relationships between
serialized verbs of the type ‘hit-break’. In both cases it is the last verbal stem that bears
the main focus of the shared ‘upshot’ (the whole clause conceptualized as a single
sequential action). But in (12) there is a mixture of transitive and intransitive verbs in
sequence:

(12)

Here it is the first and second serialized verbs (not the third) that determine the
transitivity of the whole clause, and the second can be regarded as the core bearing the
‘upshot’ since it follows the first temporally. The temporal morpheme ‘at night’ refers to
the whole complex event. Under negation it is the whole verb complex in such holistic
word-sentences that is negated (as also with the marking of mood and tense), not just one
element of it (Foley op. cit.: 328). These highly compacted serial constructions contrast
with a dependent verb construction in which several separable independent verbs can be

Page 16 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

serially chained, each (p. 130) potentially taking their own affixal material (as in most of
the Papuan languages of New Guinea).

In fact, all the complex sentences (1) to (7) given in Section 7.2 can be analysed as
containing just one (albeit complex) element directly reflecting the core ‘framing event’.
To do so it is necessary to be a little less abstract about the type of framing event
concerned than in Talmy’s cognitive typology (though I do not deny that there may be a
deeper relationship of analogy between them all).17 West Greenlandic sentence (1) I have
already characterized as expressing a ‘reportative’ event with an affix representing the
framing event. In the case of (2) (Ainu) the verbal stem (‘sway’) plus its own incorporated
object (‘heart’) is the main framing element in a ‘mental event’ (it forms a unitary
idiomatic expression for ‘wonder about s.th.’). The two applicative additions refer to co-
events associated with it—the first one taking an external argument referring to the
object of the verb and the second, internal one, referring to the cognitive location of the
event (within the speaker). The applicative morphemes do not in themselves define the
‘upshot’ of the event. In sentence (3) (Bininj Gun-wok) the stem (‘go’) has the main
framing function (determining the transitivity of the whole clause), as does the stem
(‘see’) in (7) (Nivkh), in respectively an event of motion and one of perception. In (5)
(Nuuchahnulth) it is, on the other hand, an affix (‘together’) that is the main framing
element—expressing an ‘action (or here: physical state) correlation’ event. It determines
the transitivity of the whole (the stem is just a peg element hosting it).

More problematic at first glance is the case of example (6) from Koyukon, a language
displaying discontinuous morphemes. Both the satellite (‘in(to) the palm’) and the stem
(‘handle a compact object’) would seem to contribute to the expression of the main
framing event (one of motion). The satellite (tl’o-) can have a stative or a motion sense,
depending on the morphological context in which it is found. The stem in turn can be
inserted into several distinct frames—here it functions as a verb ‘theme’ of classificatory
motion, ‘carry or handle a compact object’, and combines with a preceding object prefix
(also a gender prefix and a particular momentaneous aspect prefix). The satellite
introduces its own ‘postpositional’ object here. Nevertheless, the whole ‘derivational
chain’ P + tl’o ‘giving or selling to P’ can be regarded as representing a more specific co-
event that stands in a relationship of Path/Ground to the abstract classificatory motion
stem. Since it does not define the core schema of the event on its own (a subject acting on
a certain type of object) it cannot be regarded as the framer of the whole clause as
opposed to supplying an ancillary co-event. Thus there is no real problem with seeing a
single framing event here either—one anchored in the stem.

I conclude that if something like the macro-event is indeed the universal basis for
constructing individual utterances in any language, it is only the distribution of ‘event
integration’ across the syntax/morphology divide that will vary significantly from
language to language. Polysynthetic languages are those that allow the maximal
compacting of a whole macro-event into a single independent word, while by contrast an
ideal isolating language allows only one element of the macro-event at a time to be
expressed as a single word (whether Figure, Ground, activating process, or any co-event

Page 17 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

element). However, one further constraint needs to be added here as regards the kind of
co-event that can be compacted into a single polysynthetic word-sentence. As we have
seen from the examples given, it would appear that co-events can generally only be
integrated that share the same subject or (p. 131) agent with the main framing event: the
pronominal arguments may be cumulative but they refer to the whole complex event, with
shared subject.

An apparent exception is Talmy’s ‘action correlation’ type of framing event, as illustrated

by West Greenlandic sentence (1), with two different underlying subjects. However, this is
precisely a type of framing event: the inflection of the word-sentence (when one removes
the three independent words) is limited to one subject (the agent of the reporting event)
and one object (= the subject of the ‘underlying’ co-event). This is easier to see in
Nuuchahnulth sentence (5) since here it is a matter of two ‘subjects’ of the same
concerted state—the pronominal marking is again determined by the superordinate
framing element (the affix). In both cases a special kind of affix is required to perform
this kind of integration.

The limit on the type and number of morphemes that may appear in a single polysynthetic
word-sentence is then ultimately a limit on what a sentence in general may express in
terms of underlying cognitive content. This limit is equally relevant to more analytic
languages—as a limit to what may be packaged into the single multi-word clause. After
all, Talmy emphasizes the universality of his ‘macro-event’. Also sign languages are
relevant here—the ‘polymorphemic’ structure typical of such languages can be directly
compared with polysynthesis in spoken languages, classifier verbs on their own
representing holistic events. Engberg-Pedersen (1993: 253f.) specifically points out the
striking parallelism between Danish Sign Language and the sub-type of polysynthetic
language exemplified by Atsugewi, with which it shares similar lexicalization patterns.
The limits on what can be compacted into a single ‘utterance’ or ‘prosodic phrase’ in sign
languages (definable in terms of successive ‘movements’ and ‘holds’) would appear to be
quite similar to what can be compacted into a single complex verb form in polysynthetic
languages. Thus several predicates can be linked by a single maintained classifier hand
shape indicating a chain of motion and location (sub-) events so long as the subject of the
overall (macro-)event remains the same, much as in the compound serialization
constructions of Yimas (Aronoff et al. 2005: 327). Just as polysynthetic spoken languages
often combine templatic and concatinative sub-parts of their morphology, so do sign
languages combine simultaneous (classificatory) and sequential (affixal) aspects. As
Aronoff et al. (op. cit.: 335) point out, the ‘simultaneous’ aspect of sign language is
grounded in visuo-spatial cognition.

Perhaps all of this does not take us very far beyond Sapir’s and Comrie’s treatment of
polysynthesis as an extreme degree of (head-marking) synthesis, but it does, I believe,
make more explicit the elements that may be integrated into the polysynthetic word-
sentence and the constraints that cognition sets upon them. What is common to these

Page 18 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

holophrastic languages is relatively open-ended verbal morphology that takes over many
of the functions of syntax in analytic languages.

7.5 Appendix 1 A passage in West Greenlandic

(p. 132)

‘It is told that big Igimarasussuk frequently lost his wives. Each time the poor sod
married a new wife he would lose her. Before a year had passed, he would lose his wife.
Having lost his wife he would go lamenting to his wife’s relatives. One time, having lost
another wife, he began to woo a woman by the name of Masaannaaq, the only one
amongst the many men living to the north.’

7.6 Appendix 2 Core polysynthetic languages

North America:
Eskimo (all)
Aleut?
Na-Dene (all)

Page 19 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

Algonquian (all)
Siouan (all)
Kutenai?
Tsimshianic?
Haida?
Wakashan (all)
Salishan (all)
Chimakuan
Numic Uto-Aztecan?
Sahaptian?
Chinook
Klamath?
Northern Hokan (Palaihnihan, Shastan, etc., but not Pomoan)
Washo
Kiowa-Tanoan (all) (p. 133)
Iroquoian (all)
Caddoan (all)
Muskogean?
Yuchi?
Meso-America:
Purepecha (Tarascan)?
Totonac-Tapehua (all)
Mixe-Zoque (all)
Nahuatl (Uto-Aztecan)
Yucatan Mayan18
South America:
Mapudungan
Arawak (most)19
Aymara?
Nambikwaran
Mosetén
Harakmbet
Cahuapena
Yanomami
Guahibo
Tupi-Guaraní
Nadëb (Makú)?
Matses (Panoan)?
Cariban (some)?
Eurasia:
Chukotian (but not Itelmen)
Ket
Nivkh?

Page 20 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

Ainu
So:ra (and other South Munda)
Chintang (and other Kiranti)
Northwest Caucasian (all)
Oceanic/Australian:
Tukang Besi (Sulawesi)20
Yimas
Alamblak
Barupu
Warembori (p. 134)
Gunwinyguan (all)
South Daly River
Tiwi

Abbreviations
A or AG
agent
ALL
allative
APPL
applicative
ASS
associative
CAUS
causative (mood)
CONTEMP
contemporative (mood)
CONTING
contingent (mood)
DUR
durative
FR
far (past)
FUT
future
HAB
habitual
INCH
inchoative

Page 21 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

IND
indicative
INTR.PART
intransitive participle
INS
instrumental
INV
inverse
IPFV
imperfective
IRR
irrealis
ITER
iterative
IVF
incorporating verb form
KIN
kinaesthetic
NEG
negative
NP
non-past
O or OBJ
object
PART
participial (mood)
PASS.PART
passive participle
PFV
perfective
PC
paucal
PL
plural
PRO
proverb
REFL
reflexive
REL
relative
REP
repetitive
RM

Page 22 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

remote (past)
S or SUBJ
subject
SEQ
sequential

Notes:

(1) See Appendix 1 for a typical passage in West Greenlandic, a language widely regarded
as polysynthetic and high on the ‘index of synthesis’. It is taken from the traditional tale
‘Igimarasussussuaq’, as told by Jaakuaraq of Nuuk to Knud Rasmussen and reproduced in
Fortescue (1990b: 103). It can be seen that it consists of twenty words, corresponding to
the seventy-three words of the English translation. The language is simple, with no
particularly long words by Greenlandic standards.

(2) See also Bickel and Nichols (2007: 188ff.) who rename the synthesis dimension
‘semantic density’ (at the word level).

(3) Chukchi, for example, is dependent-marking as far as nominals are concerned (it
displays a robust case system), and Eskimo is double-marking. What counts as regards
degree of polysynthesis, however, is arguably the verb.

(4) But note that languages may combine general scope-ordering with templatic
organization of a sub-section of the morphology (as with West Greenlandic ‘sentential’
affixes). Nivkh combines free Eskimo-like ordering to the left and rigid templatic ordering
to the right of the main stem (Mattissen op. cit.: 212).

(5) Syntactic incorporation is typically displayed by scopally organized languages with

‘heavy’ verbalizing affixes affecting valency and interacting with the external syntax in a
productive manner, for example Eskimo-Aleut with its verbalizing affixes or Wakashan
with its ‘governing’ lexical affixes. Noun incorporation does not usually attain this degree
of freedom (cf. Mithun 1984: 889), though it would appear to approach it, at least in
purely incorporating languages like Sora (Anderson, this volume). Purely templatic
morphology on the other hand would seem not to favour this type of morphosyntactic
behaviour.

(6) Bohnemeyer et al. deal mainly with causal constructions, but within this context they
also discuss serial verb constructions, which are maximally compacted in polysynthetic
languages like Yimas.

(7) ‘Constitutiveness’ here refers to the support relation of filling in the conceptual ‘body’
of the temporal contour provided by the framing event. This could, for instance, be an
agent’s action in bringing about the macro-event.

(8) VII is a nominal agreement class.

Page 23 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

(9) More specifically, in which ‘an intentional Agent effects or maintains a particular
correlation between an action performed by herself and an action performed by another
Agency’ (Talmy 2000, vol. 2: 253–61). This covers the cases he terms ‘concert’ (i.e.
together with someone), ‘accompaniment’, ‘imitation’, ‘surpassment’, and
‘demonstration’ (when the two actions are complementary rather than equivalent).

(10) *ŋit- is also an ingredient in the verbal negation extension *-nRit- ‘not’ from
nominalizer *-nəR plus *ŋit-.

(11) Applying the same criteria to Cree, a polysynthetic language of the Algonquian
family displaying a rather different morphological type, at least the following non-stem
morpheme meanings (from Wolfart 1996) can be regarded as lexically heavy:

Medials: classificatory (wood, stone, fabric, etc.)

body parts
Finals: instrumental (do by hand, by cutting edge, by vision, by mouth or speech, by
tool, by vehicle, etc.)
(do) at or to a place
move
run
have X
arrange X
make X

There are also ‘pre-verbs’ with meanings like ‘go to’, ‘want to’, ‘intend’, ‘audibly’, which
probably belong here too. One kind of lexically heavy morpheme not found in either
Eskimo-Aleut or Algonquian is the spatial lexical suffixes of the Wakashan family. These
cover meanings of both Path (directional) and Ground (location)—and combinations of the
two—of considerable topographical specificity. A listing of these for Nuuchahnulth can be
found in Sapir and Swadesh (1939).

(12) One could say the same of Nama (Khoisan), which also displays a surprising degree
of synthesis but falls short on obligatory bound pronominals and productivity of
incorporation.

(13) Pronominal clitics typically appear on the first element of the clause in this strictly
VSO language, even when that element is not the main verb but some other highlighted
element, as in ʔa:ni-sa ʕac’ik-šiƛ (really-1SG know.how.to-MOM) ‘I really learnt how to do
it’.

(14) A somewhat similar situation holds in Ute (also as regards the variety of hosts
pronominal clitics can adhere to), though also 3rd person clitics are involved, in object as
well as subject function. The conditions in which they may all be replaced by zero in Ute
is primarily a matter of reference continuity (T. Givón, pers. comm.).

Page 24 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017

What are the Limits of Polysynthesis?

(15) 4SG = ‘fourth’ (reflexive 3rd) person, here indicating the same subject as that of the
superordinate predicate.

(16) Incorporating adjuncts in this manner is relatively rare in polysynthetic languages,

as are inflectional circumfixes.

(17) I shall not attempt to relate the sentences to Talmy’s distinction between satellite-
framed and verb-framed languages (problematic for polysynthetic languages).

(18) And perhaps other lowland Mayan languages with noun incorporation. All of these
are only mildly polysynthetic.

(19) Tariana, under the influence of neighbouring Tucanoan, does not mark pronominal
objects of transitive verbs on the verb (unlike most Arawak)—though subject-marked
transitive verbs with Ø-marked 3rd person object can apparently constitute a complete
word-clause (cf. Aikhenvald 2002b: 61–5). So Tariana may qualify as ‘polysynthetic with
qualifications’.

(20) And possibly some other Austronesian languages of southwest Sulawesi, the
northern Philippines and Taiwan.

Michael Fortescue

Michael Fortescue is Emeritus Professor of Linguistics at the University of

Copenhagen, now associated with St Hugh’s College, Oxford. His special area of
interest is Arctic and Sub-Arctic languages, principally Eskimo-Aleut, but also
Chukotko-Kamchatkan and Wakashan languages, for all of which he has published
(or co-published) comparative dictionaries. He has carried out field work in
Greenland, Canada, Alaska, and Siberia. Besides a comprehensive grammar of West
Greenlandic (1984, Croom Helm) his work covers many aspects of Eskimo-Aleut
languages, including the acquisition of West Greenlandic. He has extensively
published in the more general fields of comparative, typological, cognitive, and
functional linguistics. He received his MA at the University of California, Berkeley,
and his PhD at the University of Edinburgh.

Page 25 of 25

PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in
Oxford Handbooks Online for personal use (for details see Privacy Policy).

Subscriber: OUP-Reference Gratis Access; date: 16 November 2017