Why Incorporation? CES 6

1. Problems
 A set of interesting works have been recently published which throw light to the old discussion about Nivkh incorporation fifty years ago as well as to the theoretical problems on incorporation and polysynthesis. In this paper, I wish to find some plausible answers to the following questions:

(i)  The old discussion about Nivkh incorporation has not been yet settled conclusively. The recent argument on incorporation will help us to bring it to a conclusion in favor of Panfilov’s view  that it is not an incorporative language.  
(ii)  It is often assumed that incorporation is a typical feature of polysynthetic languages. But we have at least a counter example, namely, the languageItelmen. This fact causes us to suppose that the grammatical techniques of incorporation and polysynthesis work somehow differently, although they make use of the common device for concatenation of affixized elements around
  the verb stem.   
(iii)  In the theoretical discussion about incorporation in recent years, it has  been often claimed that a single lexical category is affixed to the verbal head.
 But we have a set of examples of Ainu
 and Chukchee which show that not only affixed nouns, but also a phrase can be incorporated.
   This fact is related to the theoretical understanding of the notion in general as well as to Panfilov’s typological supposition on that Nivkh. 
(iv) The syntactic interepretation of incorporation proposes to derive incorporative structure from a certain “underlying” structure via transformation Move-α.
  It has been recently developed into
 a new version which is characterized by the theory of parametric conditions to derive polysynthetic and incorporative structures from a universal configuration. The theory contains important

  observations, but it stands yet on the basis of problematic idea that
structure is derived transformationally from a universal

(v) Incorporation can not be often differentiated from composition. But the
 affixation to a nominal
 head can not be regarded as incorporation because of its
 less restricted concatenation of the com
ponents as well as of possible semantic
 variation of compound words. But composition is an
indispensable grammatical
  technique which any language utilize of to make a complex word
 regardless of
 word class distinction. Therefore, it is surely wrong to reduce composition into
 incorporation and vice versa. Crucial is to distinguish their functions.
2. Discussion on Nivkh Incorporation
  In the middle of the last century there was a series of discussion on the topic whether Nivkh has incorporation or it has to be regarded as an agglutinative language. The discussions were held mainly in the Leningrad branch of the Linguistic Department of the Soviet Academy of Sciences, especially between the linguists E.A.Krejnovich and V.Z.Panfilov, who are both most prominent nivkhlogists until today.

2.1. Krejnoivch for Incorporation
 Krejnovich asserts that Nivkh is an incorporative language, since incorporation is visible in the concatenations of nominal modification and the structure of direct complements of transitive verbs and, in both cases, it is morpho-phonologically marked by the consonant alternation. First, obverse  the following examples (Krejnovich 1966,48,):

(1) a. ranr     urgur       rod’
    sister-ABS well(ADV)  help-AORIST/INF
    (the sister helped well)
    b.  «t«k        urgur ph-ranr             _tod'
      father-ABS     self-sister-ABS  help-AORIST/INF
     (father helped well own-sister)
      According him, tod’ has allomorphs rod'/dod' and  =” indicates
   incorporation .       

 In the examples above, the phonological circumstance of the transitive verb rod’/tod’ is identical in both cases, namely, directly after r. When an adverb precedes it (1a), the initial consonant of the verb remains r-. But in the second case (1b), when the direct complement preceds the verb directly, the initial consonant of the verb changes into t-. He ascribed this phonological change to the incorporation of the direct complement into the transitive verb, because this progressive consonant change indicates a tight combination of morphemes just like in the case marking by a suffix (2a) in the following examples (Krejnovich 1958, 30):

(2) a. attachment of a case marker:
   i)   n'          _«t«k             _roX
        my      father-STEM  to-DATIVE  
       (to my father)
   ii)  n'        _nanX   _toX
         (to my  eder brother)
   iii)  n'         _oåla  _doX
            (to my child)
    b. nominal modification
   i)   n'         2«t«k             _r«f
          my     father-STEM  house
     (my    father’s house)
   ii)  n'    2nanX   2t«f
       (my eder brother’ house)
   iii)  n'   =oåla  =d«f
          (my child’s house)
     c. direct complement + transitive verb
   i)   n'          2«t«k             2rod'
      my     father-STEM  help-AORIST/INF  
      (to help my father)
   ii)  n'     2nanX   =tod'
         (to help my elder brother)
   iii)  n'     =oåla   =dod'
            (to help my child)

With these examples, he proposed the following hypotheses:

[K1]  The three types of concatenations in (2) are equivalently tight because the initial consonant of the last morphemes /r- ‾ t- ‾ d-/ changes in the same circumstances /-k__, ‾ -X__, ‾ -a__/ in the same way. That is, both nominal modification and transitive complement structure make a tight construct just like the suffixation of a case marker. From this fact, Kejnovich follows that the constituents in (2b) and (2c) are incorporated into a word-like unity like (2a).
 This assumption of Krejnovich was the basis for the whole discussion in Soviet Academy at that time. The crucial point was the question whether the consonant alternation indicates the incorporative combination of morphemes or it has to be regarded only as a phonological phenomenon like sandhi. Krejnovich stoods on the first view point, and the second view point was asserted by Panfilov.

[K2]  The examples in (2) include three sets of allomorphs for each morpheme.

   (3) a. sufix: {-roX: -toX: -doX}
     b. noun: {r«f: t«f: d«f}
     c. trsansitive verb: {rod': tod': dod'}

 Here, the question arises which one of the allomorphs in each set is “original”, i.e. the lexical from. The representative dictionary of this language, Sovel’va/Taksami 1976, chooses the forms {-roX, t«f, rod'} for the lexical representation, respectively. They are right in that these forms appear in a neutral position where no consonant alternation can occur, typically in the context: [-nasals__], e.g. qhan-roX (to the dog), n’ oqon t«f (the house of my financé), n’ oqon rod' (to help my financé).  
 Krejnovich showed later (Krejnovich 1960) the idea that Nivkh has fundamentaly two types of forms, free and bound forms. The formers are used as lexical representations, while the latters appears in the context of tight concatenation of words, i.e. in his incorporative structures.

[K3] The consonant alternation occurs in the same way in case of suffixes, nouns and verbs; the consonant (-)r- , e.g. changes in the different phonological context in the same way as follows:

(4) a. suffix {-roX}:        -r- -r- /-k__,  -r- -t- /-X__,  -r- -d- /-a__
    b. noun {ruv}:          r- r- /-k__,   r- t- /-X__.    ?
    c. transitve verb {rod'}:  r- r- /-k__,   r- t- /-X__,  r-d- /-a__

This phenomenon supports both Krejnovich and Panfilov. But their interepretation was different: in this same fact, Krejnovich saw incorporation, but Panfilov, on the contrary, sandhi.

2.2. Panfilov against Incorporation Theory
 Panfilov did not agree with the incorporation theory of Krejnovich. He pointed out crucial negative evidences as follows:

[P1]  Incorporative concatenation is marked by some phonological rule, e.g. vowel harmony, and the consonant alternation can be a good candidate for the marker. However, that of Nivkh involves only plosives and fricatives of the initial consonant in contact. This is surely a remarkable phonological phenomenon, but insufficient to indicate such a important grammatical relation like incoroporation. He formulated the morphophonological rules of the alternation in (Panfilov 1954), which is more detailed than that of (Krejnovich 1958). But according to Panfilov, these rules are applied purely as sandhi rules and, therefore, are indifferent to any grammatical relation.

[P2] Panfilov’s second argument is related to the grammatical status of the modifier and complement. He thinks that they are not bound forms, but free forms. In the follwing examples, the nouns occur with comitative case marker which make up free forms:

(5) a. modifier with comitative marker
     kheq-xo hyik-xo     _zif-ku
     fox-COM rabbit-COM trace-PL
     (trace of a fox and a rabbit)
   b. transitive complement
     hoghat men khu-gho  pundi-gho _bo-t               vod’-ghu
     then  both arrow-COM bow-COM  take-in-hand-PART ging-PL
     (then both went with arrow and bow in hand)

According to him, presupposed that a combination of free forms cannot be incorporated, both nouns with comitative marker -gho‾-xo make no incorporation.

[P3] Panfilov assumes that, if a morpheme is incorporated, it has no overt syntactic relationship with a consitituent outside of the syntactic structure. That is, if pundi-gho (with bow) in (5b) above is incorporated with bo- (take into hand), it cannot make nominal conjunction with khu-go (with arrow). This is also the case with the following examples:

(6) a. modification of a conjunct nouns
     n’   urla  pila  mu
     (my  pretty big  ship)
   b. complement of a conjunct nouns with adjective modifier
     h« «ghmu-«rux  malghola zavod-gho  fabrika-gho    _djesqut-had’ (Panfilov

     this war-time-LOK many  factory-COM workshop-COM destroyed-were
     (in the war time, many facrtories as well as workshops were destroyed)
   c. adverbial modification
     if  lele  mat’ki pXov -co   _nr&«n«d'(Panfilov 1954,20)
       he very small  round fish  find-FUTUR-FIN (He interpreted nr&«n«d' in
   past tense.)

All these cases show that it is unreasonable to regard only the last noun of modifiers or complements as an incorporated element, because the modifiers are the whole phrase urla pila in (6a) and the whole complement malghola zavod-gho fabrika-gho in (6b) and lel mat’ki pXov -co in (6c). The incorporation, if ever, has to involve whole of these phrases. From this observation, Panfikov asserts that the structures like (6) cannot be regarded as incorporation.

[P4]  Panfikov remarks, too, that in an incorporative structure, the syntactic relationship among the components must be visible. During the whole discussion in Leningrad, this view was presupposed as a matter of fact. And Panfilov regarded it as the well-formedness condition for incoporative structures. But as far as the cited examples are concerneed, the condition was satistfied in any way, because they all shared the structure of nominal modification or direct complement. So is the matter, too, if they are interpreted as agglutinavie. The syntactic difference lies only in that agglutinative syntactic constituents are extracted from the verbal complex. And the structure of Nivkh is just as such. Therefore, any one can claim that Nivkh is an agglutinative language.
 In connection with the discussion, there is one more point to note: the direct complements of Nivkh transitive verbs include terms which appear as dative or directive in other languages like Russian:

(7) a. kuvy  nux       th«d’ (Panfilov 1960,55)
     thread needle-ABS pass-FIN
     (pass a thread through a needle)
   b. n’i   huxt taqu     _xrod’ (Krejnovich 1958,27)
     1SG  robe hanger-ABS hang-FIN
     (I hung a robe on a hanger)
   c. n’i   t’aqo ph-«t«nk_khind’ (ibid.)
     1SG  knife  own-brothe-ABS_gave-FIN
     (I gave a knife to own brother)

 The nouns in absolute case in (7) are the direct object of the verbs which are, therefore, transitive in this language.

 [P5] Pnafilov added a semantic condition for a well-formed incorporation: the incorporated elements have to preserve their inherent meanings. They may not change the meaning like in counpound words, e.g.:

(8) a. vut’zif (iron-way= railway) (Krejnovich 1958,31)
   b. khEq-zif (fox-way) (ibid.)
   c. pila-Na (long animal=snake) (ibid.)

Note that the initial constonant of the second noun in (8a,b) changes according to the rules which are applied to incorporative structures, too. But this supports the incorporation theory of Krejnovich as well as the sandhi interpretation of Panfilov.

2.3. Open Questions
 The discussion on Nivkh incorporation in Leningrad has brought, in fact, no definite conclusion. Instead, it has pointed out many important questions to make clear the grammatical phenomenon of incorporation. We choose here some of them:

[Q1] A well-formed incorporative strcture is subject to some syntactic conditions. In the discussions, the followings are mentioned:

(9) a. A syntactic relationship between the incorporated terms and the head must
    be visible. 
As in our case two types of structures, nominal modification and
    direct complement, are
in question, the condition is satisfied automatically.
    But when a complex noun phrase is
“incorporated”, we have no marking to
    select an plausible interpretation as the following
examples show:   

(10) a. if [NP [A lele mat’ki](&)[A pXov] co] _nr&«n«d'. (=(6c) )
     b. if [NP [A lele mat’ki][NP[A pXov] co]] _nr&«n«d'.

(9) b. The incorporated nouns are expected to be stems, namely, bound forms,
    but no free forms. They
may not attach any affix to indicate e.g. a number
     or case marker. If this is realy the case, the
 following sentence is a clear
     counter example againstincorporation:

(11) hoghat men [NP [WORD khu-gho] [WORD pundi-gho]] _bo-t vod’-ghu. (=5b)
     -gho: comitative case marker

(9) c. According to Panfilov [P3], a incorporated term may have no syntactic
    relation with any term
out of the incorporative complex. If this is true, the
    whole NP in (10a) in the interpretation
(10a) if [NP [A lele mat’ki](&)[A p
Xov] co] _nr&«n«d'
    is not
well-formed as incorporation. But, if the interpretation (10b) is
     possible, it is the case of a phrase incorporation.

   [Q2] The semantic well-formedness condition [P5] says that incorporated terms preserve their lexical meaning. On the contrary, composition makes a semanmtic units which do not preserve the original meaning and can have a figurative meaning. Does this mean that there are two kinds of lexical units of the same internal structures as follows?

(12) a. [WORD X  _Y] making a semantic unit composition
    b. [WORD X  _Y] preserving meanings of X and Y incorporation

[Q3] The exact rules for consonant alternation of Nivkh has been not yet established. There are some proposals; among them, the rules of Panfilov 1954 is the most detailed. But his proposal leaves yet some problems open. We will point out some of them.

(13) a. The consonant alternation rules have to explain possible allomorphs. They
     must indicate
which allomorph is the unmarked one, namely, the lexical

     b. Panfilov formulated the rules as sandhi of Nivkh. But they do not explain
    the case
     (1): in
(1a), the initial consonant of the following verb r- changes into
     t- when an object noun
preceds it (1b), but in (1a), it remains unchanged
     after an adverb. Is the example (1) a crucial
counter argument aginst
      sandhi interpretation?

[Q4] The phonologically motivated allomorphs of transitive verbs in Nivkh bring out a special problem. Krejnovich 1966 showed the following paradigm:

(14) a.             i Ä d’    (lexical form: kill/take)
     b.             i Äu d’   (transitve verb in aorist: killed/took)
     c.            co-xu-d’  (fish killed, fish took)
     d.            co-xu-nivkh (man who killed/took fish)

 The paradigm shows that

 (i) “-Ä-“ and “-xu-“ are allomorphs with the meaning kill/take,
 (ii) i- in (14a,b) is exchanged with the direct complement “co” (fish)
   in (14c,d) and

 (iii) “-d’” (ending of a lexical form as well as the marker of aorist)
   appears as the marke and can
be exchanged into a noun, e.g.
   “nivkh” (man) in case of nominal modification (14d).

The prefix i- has allomorphs j- and e-, perhaps according to the old vowel harmony. It causes the fricativazation of the initial consonant of transive verbs, like x-Ä- , as in the examples above. Question is, whether the prefix is really pronominal or is it possible to explain it in a different way. Shiraishi 2002 tends to regard it as prethetic, i.e. it has no morphosyntactic function, but it is only phonologically driven.

[Q5] In comparison with Chukchee, Panfilov 1954 notes that Nivkh is difficult to regard as an incorporative language because of its non-polysynthetic structure. It has no morphematic frame which includes incorporative elements into circumfixes, e.g. for agreement markers. Moreover, Nivkh is dominant in suffixation and it has no marking to indicate where an incorporative verb complex begins, if any. According to him, Nivkh is definitely an agglutinative language, in all grammatical aspects, it is neither incorporative nor polysynthetic. But there are persistently the opinions that the language is something more complex. Mattisson 2001 is inclined to capture the language as a polysynthetic and incorporative language. In order to make this problem clear, we will observe in the next chapter how polysynthesis and incorporation interract with one another.

3. Polysynthesis and Incorporation
 It is an empirical fact that most of polysynthetic languages utilize the morphological technique of incorporation.  If the verb stem is surrounded by affixes, especially by circumfixes indicating persons and numbers, e.g. in Chukcchee, an incorporative structure is visible explicitely,. However, polysynthesis is not necessarily coocurrent with incorporation. We have at least a clear counter example against the obligatory coocurrence of them.

3.1. Polysynthesis of Itelmen and Ket
 In the last years, there have been published two detailed grammatical descriptions of polysynthetic
languages, namely, Volodin 1999 on Itelmen and Werner 1997 on Ket grammar. We cite below the polysynthetic verb frames of these languages for comparison:

(15) polysynthetic verb frame of Itelmen


 where -5: mode indicators, t-/t’-,φ,n- for indicatives and m-,q-,q’-
   for imperative, and k-/k’-/x- for
      -4: indicator for conjunctive k’-
      -3: reciprocal prefix lu-/lo-
      -2: left part of anti-causative (anti-passive) circumfix en-/an-, ne-,na-
      -1: prefix for transitivazation-causation lin-/Èen-,«n-,«nt-,t-
      1: suffix for derivation -te,-se/-sa,-la,-o
      2: right part of circumfix of causative/anti-passive -l,-N,-w
      3: diversion-suffix -sxen
      4: suffix for an Aktionsart -ala
      5: suffix for an Aktionsart -ata
      6: suffix for an Aktionsart -zo,-t,-st
      7: indicator for desiderative -aÈ/-a
      8: detransitivization (suffix) -/-φ
      9: suffix for participlization -/-È
     10: conjunctive indicator -k/-ka/-ke
     11: infinitive indicator -s
     12: aspect suffix -qzu/-qzo/-qz/-φ
     13: tense suffix, -s/-z./-φ,-al for finite forms, -kiÈh/-kiÈh/-kia/n/
for infinitives
     14: person suffix (a) -miN,-w«nnen,-win,-en,-n,-sxen,-mi/N,-w«nne/
    N,-e/n,-/n, -ne/n,
 -xkmiN etc.
     15: person suffix (b) -kien/-kean,-,-x,-sx,-en,-wen,-we/n,-sxe/n etc.

 Of course, every element does not occur always. But in most indicative finite sentences, the mode index -5 and a person suffix 15 have to occur. In case of subject-object agreement, both person suffixes 14 and 15 occur obligatorily. The shortest sentence has a verb form with three elements, mode+root+person, e.g. xilqaq k’-le-knen (it became cold).
 Note that the complex verb form (15) includes no position in which a incorporative elements can enter. Namely, it excludes incorporation by its nature.
 Now, compare this verb frame with that of Ket. The following is the maximal model of the possible arrangement of verb constituents formulated by Werner 1997:




















where 14: prefix indicating subject (obligatory), e.g. di tA:l’
      13: root 3
      12: root 2        
      11: derivator, e.g.n-,N-
      10: prefix for causation q-/R-
       9: prefix for indicating versions, e.g. direction, comitativity,
      benefactivity ba-,bo-,...

       8: prefix indicating subject/object ba-bo,ku-ku,...
       7: prefix for definity t-,d-,...
       6: prefix indicating continuity t-
       5: prefix for tense a-,-,so/su-
       4: prefix for subject 2, for object/instrumental  b-/m-/p-
       3: prefix for aspect n-/l-
       2: prefix for subject 1 d-/t-,..., for object 1 d-/t-...., version indicator a-,...
       1: prefix for imperative d-
      0: root 1
      -1: suffix indicating derivation -N,-n
      -2: plural suffix -n
      -3: juncture -ka etc. 

 The positions 13 and 12 in the chain (16) are prescribed for roots/stems of some nominal or verbal category. They are combined with the root/stem of the head verb 0, so that the chain [R3 (13) + R2 (12) ... R1 (0)] makes the internal verb complex of the whole verb frame (16). The verbs in the position 0 are relatively few and they have wide, often abstract meanings. They belong to the socalled function verbs whose meaning become concreter when they are combined with R3 (13) + R2 (12). Krejnovich 1968 wrote in his Ket grammer that verbs in this position require incorporative counterparts R3 and R2 , so that Ket verb structure is inherently incorporative. This is especially the case with the verbs like -t in the example below:

(17)   da-nan’-bet-q-(i)n-daN-et (Werner 1997, 156)
         14 13  12 10  3  2  0
      (she let us bake bread)

 Note that in this structure of the verb form, the incorporative elements occur in the definite positions in the chain. It is a well-known rule that a polysynthetic verb form requires a rigid ordering of internal categories. In most cases, it begins with a model and an agreement affix, then it followed by causative, passive, aspectual, temporal markers and ends with an agreement affix indicating subject/object inflection again. But here we see that, at least in this type of polysynthesis, the incorporated element is not directly attached to the head, but it occurs discontinuously in the definite preverbal position. We do not know yet why the incorporated element stands discontinuously and how this feature is related to the functional meaning of R in the position 0. We suppose merely that this polysynthetic verb frame prescribes the position of incorporated elements. Otherwise, incorporation will be impossible and the possible incorporative elements must be placed outside of the verb frame. 

3.2. Three Types of Polysynthesis
 Volodin 1999 schematized the polysynthetic verb form of Itelmen as in (18a), namely, as a chain of morphemes in which a verbal root/stem R stands in the centre and is surrounded optionally by affixes with various grammatical functions. As well known, all languages of Chukoto-Kamchatka language group, Chukchee and Koryak as its representative examples, have also polysynthetic verb structure, but they are all incorporative; they include at least one incorporative element in a verb frame. Their structure of the languages look like as (18b). We will add here the structural scheme of Ket of Werner 1997.

(18) a. Itelmen:                 (m) + R + (m)
     b. Chukoto-Kam. languages:  (m) + (r) + R + (m)
  c. Ket:                    (m) + (R3) + (R2) + ... + R1 + (m)
    where ( ): optional, m: morpheme(s), R: root/stem, r: second root
   (Volodin 1999),
and  R3,R2,R1 (Werner1997).

 Volodin 1999 does not mention the internal structure of the second root r in (18b), e.g. in Chukchee. But Kurebito 1998 gives the intereting examples as follows:

(19) a. qora-              g«tka-mla-g/e  (=Kurebito 1998 (26a))
      reindeer-ABS.SG leg-break-3SG.SUBJ
    b. qor-en             g«tka-lg«n  mle-g/e (=Kurebito 1998 (26b))
      reindeer-POSS  leg-ABS.SG break-3SG.SUBJ
     (The reindeer’s leg was broken.)
    c. epeqej              tur-k/eli-nni-g/i (=Kurebito 1998 (16a))
      grandmother(ABS.SG) new-cap-stitch-3SG.SUBJ
     (The grandmother stitched me the new cap.)

 In (19a) the noun g«tka (leg) is incorporated and attached directly to the verb. But if it modified by a possessor noun in (19b), the complex noun qor-en g«tka-lg«n (reindeer’s leg) cannot be incorporated, but both the possessor and the head noun are “stranded out of NI”. But remarkably, the noun phrase tur-k/eli ([NP new [N cap]]) is incorporated in (19c). This implies that Volodin’s r in (18b) is not a morpheme, but it can be a syntactic constituent, except for a possessive complex noun phrase like in (19b).
 Let us examine if the roots R3 and R2 in Ket can be different constitutents. In the example (17), they are made of two words nan’-bet (bread bake), which constitute a verb phrase [VP complement - transitive verb], namely, a syntactic constituent. It has to be asked further, whether the R’s can be separate constituents. Werner 1997 shows an example:

(20)  d-  Ãl'taN-  u-   Ä- a-  vet
     14  13 12    8      5  0
     (I drag it out.)

  Here, we have alternative interpretations: the adverb Ãl' and the verb root taN are governed by the verb vet independently, or rather that they constitute an infinitive verbphrase [VPinf adverb verb]. The first interpretation says that R3 and R2 are governed by R1 separately, and the latter that R3 and R2 make a syntactic constituent which is governed by R1. We can not determine now which interpretation is correct. We suppose that both are admissible, and let Werner’s R2 and R3 remain as they stand. Anyway, the condition is obvious in that R2 and R3 have to be governed by the head verb R1.

3.3. Non-incorporative Polysynthesis
 The next question is, where the syntactic component governed by the verb does occur in an incorporative and a non-incorporative polysynthetic structure. Let us compare some sentences of Itelmen and Chukchee with the same syntactic structures. The following Chukchee examples are cited from Kurebito 1998 and were translated in Itelmen by Ono Chikako together with her native friends.

(21) a1. Chuk: «ll/a-ta       ine-t-ir/-«-N-g/i (=Kurebito 1998 (3a))
           mother-INSTR(ERG) 1SG.OBJ-make-jacket-E-make-3SG.SBJ
           (My mother made me a jacket.)
     a2. Itel: qoxc       k-ci/nNit'-in  k«mmanke
           lether-jacket 1SG.DAT  INF.III-stitch-3PERF 1DAT
                  (The lether-jacket (she) stitched to me)
     b1. Chuk: g«m-nan   t-«-m«lg-ejp-«-g/en  nuterg-«-n(=Kurebito 1998 (36a))
            1-ERG  1SG.SUBJ-E-grass-fill-E-3SG.OBJ  hole-E-ABS.SG
            (I filled the hole with grass.)
     b2. Itel: lepxe  mc’el  k-tXnu/-in
           basket rowanberry-INSTR  INF.III-fill-3PERF
           ((She) filled the basket with rowanberry.)

 These examples show that the syntactic arguments governed by the head verb  are placed outside of the polysynthetic verb frame in Itelmen (21a.b), while the corresponding arguments of Chukchee are incorporated in the preverb position.
 An incorporated noun is generally assumed to be indefinite. This principle holds true for the examples (19), too: inalienable possession g«tka (leg) is incorporated (19a), but incorporation is impossible, when a possesser noun precedes it (19b). The sentences in (21) are interesting:  qoxc (lether jacket) in (21a2) in Itelmen can be definite, while ir? (jacket) of Chukchee in (21a1) is indefinite. And the incorporated m«lg (grass) in (21b1) in Chukchee is indefinite and the nuterg (hole) in absolute case definite. The relationship is perhaps true for (21b2) in Itelmem sentence, in which the definite lepxc (basket) precedes the indefinite mc’el (rowanberry). The question is, how the quantificational difference of nouns in Itelmen is indicated, because they are all placed out of the polysynthetic frame. If the nouns exchange their position in the sentence (21b2) like (21b3) below, is the opposite interpretation really possible?

(21) b3. Itel: mc’el lepxe    k-tXnu/-in
          (With rowanberry (she) filled a basket)?

The quantificational difference of the nouns in Itelmen are represented perhaps by some syntactic and even suprasegmental devices. But we do not know yet how it is done. This belongs to open questions which must be investigated further. 

4. Notion of Incorporation
We have seen in the preceding chapter that a polysynthetic structure does not necessarily involve incorporation. Polysynthesis and incorporation seem to belong to different dimensions of morphosyntactic techniques which particular languages make use of for establishing their own grammatical architectures. In this chapter, we will discuss some theoretical issus on this matter in order to make clear what is incorporation. The key point is that both incorporation and polysynthesis are the grammatical techniques to articulate (< Lat. articulus (joint)) morphemes into various linguistic units. In this sense, they are related to another technique of articulation, namely, composition as Sapir 1916 and 1921 mentioned.

4.1. Two Techniques of Articulation
 The technique of articulation consists of two operations: one of them should be called “technique of synthesis” as proposed by Sapir 1921 and the other “technique for word formation”. The first technique is used to correlate linguistic forms with notions. It determines what type of formal units represents a semantic unit. The second works for word formation. It operates to determine how  free forms are composed.
                                        4.1.1. Affixation as Syntthetic Technique  As Sapir 1921 notes, this technique of linguistic representation brings about a variety of linguistic types. If a single morpheme corresponds to a lexical or grammatical meaning, we call the procedure an analytic representation, and if not, a synthetic. When a single notion is represented with many morphemes, especially with discontinuous ones, it is often called polysynthetic. There is, in general, no clear-cut boundary to distinguish polysynthesis from synthesis, but a possible feature of polysynthesis is the utlization of circumfixes.
 Itelmen does not include the direct complement of a verb into the polysynthetic verb frame, but represents it consistenly in an analytical way. On the contrary, Chukchee includes it into a verb frame prefixing it directly to the head verb. Comparing the polysynthetic verb frames of Ket and Itelmen, they are common in that grammatical categories are placed in definite positions as affixes. But lexical categories are dealt with differently. Ket treats them as affixes, placing them in the positions R3 and R2 in the same way as other grammatical categories like causative marker, etc. Itelmen, on the contrary, affixizes a set of grammatical categories and arranges them around the verb stem, too, but it extraposes the complement out of it, not as an affix but as a syntactic unit.
 Note that in synthetic and polysynthetic structures, lexical as well as grammatical categories are all affixized. This procedure of affixization is the necessary condition for synthesis and incorporating, namely, for the grammatical procedure to include lexical and grammatical categories into a verb frame.
 Let us make a sketch on the types of synthetic technique:

(22) Technique of Synthesis
   a. analytic: lexical and gramatical categories correspound with independent
   morphemes which
 are concatenated together to make up a verb(-complex).
   b. synthetic: lexical and grammatical categories are affixed to a verb
   c. polysynthetic:
    1. non-incorporative: no lexical categories involved in a verb frame, while
categories are affixized to the verb
    2. incorporative
    2.1. simple incorporative: one affixized lexical category is involved in a verb

    2.2. complex incorporative: more affixized lexical categories and/or
    constitutent of a verb
phrase are involved in a verb frame
  (subsidary note: At a symposium of Japan Linguistic Society in June 2005, about
  a "non-slot" polisynthesis has been talked. The talk came persumably from a
  misunderstanding of the grammatical function of "slots" and from the misinterpretation
  of the data.)

4.1.2. Technique of Word Formation
 For the technique of word formation, Sapir 1921 distinguished three different operations: isolating, agglutinative and fusional ones. But the word formation techniques do not necessarily correspond to language types. A language ulilizes them simultanously. In order to see how they work, let us take an example from agglutinative verb complex of Japanese.

(23) a. [A [V tabe]-[V sase]-[V rare]-[A ta=katta]]
         east- CAUSE- PASS-  OPT=PAST
        (wished to be made to east)
    b.                          A-complex
                                V        Aff                      
             ┌─┴─┐ TENSE
              V         A  -katta
                    ┌─┴─┐  -ta=
            V         V 
                 ┌─┴─┐  -rare-
       V          V
                tabe-    -sase-

The lexical head of the complex is the verb-stem [V tabe-], while the morphological feature of the whole complex is determined by the affix [A ta=katta]. Therefore, the complex word as a whole is an adjective because of its adjectival inflexion. The lexical head stands in the leftmost, but the morphological index in the rightmost position. The agglutinative verb complex of Japanese is subject to the recursive formation rule (24), where V stands for three morphological categories: verbs, adjective and adjectival verbs which we call a“verbal”:

(24) a. V= [V V + V]
     b. V= verbals

 An agglutinative structure has some problems to solve. For example, we do not know yet precisely how converbs (Russian “depricastie”, Japanese -te form etc.) have to be treated in a verbal complex.
Any way, in an agglutinative structure, the representation of lexical and grammatical meaning is analytical in a large extent and the units of concatenation appear in a bound forms, most of which are affixes or affixized categories. The concatenation is chematized as follows:

(25) [WORD (affix)...lexical head...(affix)]

This is, in some sense, similar to the pattern of non-incorporative polysynthetic structure of Itelmen Volodin showed as(18a).
 The third techinique of concatenation, fusion, is characterized by the alternation of pholonogical segments within a word, just as we see an internal inflection of some European languages. In “Paleoasiatic” language, the technique is widely ulilized to indicate the change of grammatical function of words. When a paradigm of segmental units (26b) appears in an (poly-)synthetic structure in (26a), it motivates to change grammatical functions like ( 26c):

(26) a. [ (m) +...X... R ...Y...+(m)]
     b. X and Y make the paradigms for grammatical units.
     c. {a free form a bound form},{N V},{transitive intransitve},...

A fusional word has morpho-phonological allomorphs with various grammatical functions, so that they constitutes a paradigm. If such a paradigm occur in a (poly-)synthetic structure, the language has the typological feature: (poly-)synthetic and fusional at the same time. Ket is characterized, therefore, as fusional, polysynthetic and incorporative.  The types of technique of articulation Sapir 1921 envisaged are indispensable to think about the typological aspects of languages, especailly that of paleoasiatic and American languages. A question is now at stake: is a grammatical procedure of incorporation one of the techniques of concatenation? If yes, does it belong to that of synthesis or of word formation? So far, one thing is clear: incorporation affixizes syntactic elements governed by the head verb and attaches to it, making a compositional word.

4.2. Syntactic Interpretation of Incorporation
4.2.1. Tranformational Derivation of Incorporation
 A syntactic interpretation of incorporation was proposed in Baker 1988. In the theoretical frame work of Government and Binding (Chomsky 1981), he defined incorporation in general as  changing of grammatical functions on the lexical level. A lexical category N0, say, a noun stem, moves to another lexical category V0 on a higher position in the D-structure and is prefixed to it, making a complex category [V N0 + V0] which functions as a verb on the lexical level. The transformation takes place as a normal Move-α. He cites an example of Onongada (an Iroqoian language) like (see p.80, Baker 1988):

(27) a. D-structure  S   b. S-structure S
      ┌─┴─┐                   ┌─┴─┐
               NP        VP                    NP       VP
                   ┌─┴─┐          ┌─┴─┐        
              Pat     V        NP             Pat    V       NP                       
                    htu-/t             ┌─┴─┐
                    ‘lose’       N            N        V   N
                                hwist        hwisti  -htu-/t ti
                              ‘money’      ‘moneyi   -lose’

  His basic idea lies on the syntactic derivation of a lexical complex V = [X0 + V0] from the configurational hierarchy of the D-structure by means of X0 -movement. If X0 is a noun, we have noun incorporation. Baker mentioned incorporation of verb, preposition and passive, namely, the cases in which X0 is a verb, etc. The verb incorporation makes e.g. a causative verb complex. He includes all cases of the socalled converb constructions into the notion of verb incorporation except for passive. For he thinks about a lot of special problems to solve about this transformation. For esample, passive can be a INFL-category and, if it is incorporative, the incorporated category is not necessarily verbal, but it can be a variety of lexical categories or clitics.    
 Bakers proposal of syntactic interpretation of incorporation is surely a remarkable theoretical contribution with many rich examples of various languages. But it shows at the same time some shortages which have to be discussed about further. The followings are the most important ones:

(28) a. sufficient generalization?
        Bakers theory of incorporartion includes all sort of grammatical
     category of the type
[V X0 +V0] in the notion of incorporation. It arises
     at least three questions:

    (i) Is an incorporation applied only to verbs? In other words, is there no
    incorporation which
creates nomial or adejctival complex such as
    [N X0 +N0] /[A X0 +A0]?

    (ii) Is the incorporated element X0 only a stem? In other words, has e.g.
    a converb construction
like [V V-stem=affix  +V0] to be excluded? If yes,
    then all sort of participle constructions
have to be excluded from this
    grammatical notion.

   (iii) Can all sort of compound verbs have to be taken as incorporative? In other
    words, is it
plausible to differenciate compound from incorporated verbs only
     by means of the visibility
of any syntactic relationship?     
    b. head movement only? The transformed element X0 in [V X0 +V0] is,
    according to him, only a lexical head.
 However, as we see in Chukchee
     (see (19c)), there is a clear case of incorporation of a
phrase. The following
     is the a famous example from Ainu:

(29) a. wakka korachi notaku een poro mukar a-e-shi-sempir-ani. (CHIRI Mashiho
     1937, p.619)

     b. wakka korachi notaku   een  poro mukar shi sempir ta   a-ani.
      ‘water like    the blade sharp big  ax   self behind LOC 1/4P-hold’
     c. e: wakka korachi notaku een poro mukar

(Here, the incorporative construction (29a) corresponds to the non-incorporative sentence (29b), where “e” refers to the nominal phrase (29c). In this example the object complement      and the locative are incorporated between person affix “a-” and verb “ani” (have).
  Baker’s restriction to X0 is apparently too strict. It must be enlarged to a phrasal construct. Of course, the visibility condition or rather its semantic variant has to be guaranteed in such a case, too.

(28) c. unversal D-structure?
   Baker 1988 remained in the general frame work of GB, so that it must be taken
   for granted that
  he bases on the assumption that the derivation of incorporation
   works on a unified configuration
of “D-structure”. It is made up of the general
   formation rules of X-bar-syntax with
functional notions such as INFL, CP, VP
   and a set of grammatical procedures and restrictions
which were generally
   available at that time. The transformation Move-
α operates on the hierarchical
  configurationof grammatical functions which he supposes to be applicable to any

    language regardless of the typological differences, something very near to a
configuration which the S-structure of particular languages is derived

4.2.2. Polysynthesis Parameter
 Eight years after Baker 1988, namely, in 1996, Baker integrated his incorporation theory into the more comprehensive analysis of polysynthesis. He stands now on the new stage of generative theory proposed by Chomsky 1995, i.e. Minimalist Program. Therefore, the names like “D-structure”, ECP and other old theoretical notions dissappeared in this new book. However, the idea of the transformational derivation of incorporation remains in essense, but with a new disguise that it works as a typological parameter applied on Universal Grammer to actualize particular types of languages. He proposed a macroparameter Morphological Visibility Condition (MVC) which is formulated as below (from p. 496):

(30) A phrase X is visible forθ-role assignment from a head Y only if it is coindexed
   with a
morpheme in the word containing Y via
  (i) an agreement relationsship, or
  (ii) a movement relationship.

This goes back to his basic idea which can be characterized by the following features:

(31) a. For the derivation of polysynthetic languages from UG, they must be filtered
    out by the
morpho-syntactic condition MVC.
    b. A polythynthetic word contains an element which has to satisfy the syntactic
    condition (i) or
(ii) of MVC.
    c. The conditions are formulated with configurational syntactic structure
    traditionally utilized
for the representation of D-structure, which, he seems
    to think, is yet valid for the
interpretation of the phenomena.
    d. The incorporated element in a polysynthetic word must be coindexed with an
    argument in the
“underlying” configurational pattern. The idea about the
    transformational derivation remains
after that a universal configuration had
   “died out”.

 There are a lot of problems to be discussed further about his hypothesis. But we mention here only the followings:

(32) a. If we take it for granted that the underlying structure is formulated
    something like he shows,
 there must be yet another intermediate macro-
    parameter specifically for the representation of
non-configurational languages.
    For, in order to operate the syntactic procedure of movement
and coindexing,
    some different conditions are necessary than for the configurational
    It is, therefore, plausible to assume that there is a macroparameter for configu-
tionality which is applied before the polysynthetic.
   b. Baker writes in many places that noun incorporation is obligatory in polysynthe-
   tic languages,
but it is forbidden in all others (e.g. p.330). His MVC presupposes
    this idea, too. If so, he has 
to explain the existence of the counter example like
    Itelmen which rejects noun incorporation.

   c. His analysis of verb incorporation is insufficient, because it does not explain many
of converb constructions. Also his formulation “morphological causative is
   limited in the
polysynthetic lanugages”(p.372) is not correct. Any way, this part
    of grammar requires
 especially further investigation (see Haspelmath 1995).

 Summing up so far, the syntactic derivation of incorporation is surely an attractive interpretation, especially because it can clarify the hidden syntactic relationship of the incorporated elements in its polysynthetic circumstance. But Baker’s analysis presupposes some “underlying” structure the incorporated element is derived from. It is in essense a universal configuration from which even a word forming morpheme is taken out by means of Move-α.
 The simple cases of incorporation is that of an indefinite unmarked lexical category as Baker assumes. But we have more complex cases in which a phrasal constuction is incorporated as in the examples of Chukchee (19c) and Ainu (29a). As Sato Tomomi in the ELPR-Conference in (2002. July 13.) mentioned, Ainu, as a rule, incorporates the long/possessive form of a noun, namely, not only a lexical category, but also an affixed category and a phrasal element.
 A polysynthetic language Itelmen shows that polysynthesis does not necessarily prescribe incorporation. Of course, in a polysynthetic verb frame, we can easily recognize an incorporated element because of its occurrence within circumfixes. Bur because of this counter example we must suppose that incorporation is a different grammatical technique from polysynthesis. 

5. Typologically Relevant Oppositions
 In the discussion above, we have found some keywords crucial for the further investigation of our problems about incorporation and polysynthesis. In this chapter, we thematize them in form of the following three sets of oppositions.

5.1. Nouns versus Verbs
 In the discussion on Nivkh incorporation, Krejnovich showed the assumption that nominal modifiation as well as direct complement structures are so tight a concatenation of morphemes as case marking, so that there exist not only noun-to-verb incorporation [V N+V0], but also noun-to-noun incorporation [N N+N0].
 In the discussion of our internal seminars, Nakagawa Hiroshi mentioned once that deverbal nouns have wider possibility for nominal composition than the orginal verbs. Let us observe the following nominal compounds of Japanese:

(33) a. (o-)te-ara_i (=toilet)
      (POLITE-) hand-wash_NOMINALIZION
    b. te-ara_i (=hand-washing
    c. te-ara_i (=manual washing)
    d. kikai-ara_i (= washing with a machine or washing of machine-OBJECT )
      “_” indicates here compositinal concatenation of morphemes.

  Not that the coresponding verbal expressions are impossible for both the ambiguous cases in which the prefixed elements te- and kikai- appear as object or insturmental of the verb araw-:

(34) a. *te-ara_u (=manual washing, or hand-washing)
     b. *kikai-ara_u (= washing with machine, or washing of machine)

 It follows from this comparison that, as far as this type of expressions is concerned, the verbal compounds are not possible, while nominal compound of deverbal nouns has a wide possibilities of compounding like (33) and bubun-arai (=partial washing), kinubukin-arai (=washing with silk cloth), etc. It follows, therefore, that the nominal compounds of the type (33) are constucted after the nominalization araw- to ara_i was once established. Moreover, nominal composition is neither subject to the syntactic conditions like MVC, nor necessary to preserve the inherent meaning of the components.
 Krejnovich supposed that incorporation is found in both nominal and verbal structures in Nivkh. But note that the expressions (14c) and (14d) share no grammatical parallelism: in (14c), the head is -xu- (kill) to which the object co- (fish), say, incorporated, while the head of (14d) is the noun nivkh (man) which follows the modifier co-xu-, makig a nominal phrase. In this noun phrase, we see no matter of incorporation, but only an adnominal verb phrase with a complex stem. From this, we can follow merely that the incorporated verb stem is the adnominal modifier of the noun. Therefore, the nominal expression (14d) has nothing to do with incorporation. We claim that there exist no incorporative nouns, but all cases he supposed to be incorporation are merely nominal composition.

5.2. Polysynthesis versus Agglutination
 The agglutinative concatenation of Japanese verbs like (23) is made up with the word formation rule:

(35) Y = [WORD Y - X]
    where Y is a lexical head and X an affixized category including affixes for

In case of (23), the initial Y is a verb stem tabe- and the last X is past tense affix of an adjective -katta, so that the whole Y is a complex adjective. The category X is an affix, inherent or derived one. It can be a relatively independent lexical categories which Hattori Shiro 1950 called huzoku-go, a bound word.
 The rule (35) is applied recursively, so that an agglutinavie word can make often a long chain of morphemes. Agglutinative languages have a general tendency to prefer suffixation, therefore, right- ward concatenations appear in the majority of this type of languages. But prefixation is not excluded. A reverse notation of (35) is useful as follows:

(36)  Y = [WORDX - Y]

 An agglutinative language makes use of both the rules (35) and (36) often simultanously. But it uses no circumfix, i.e. no obligatory combination of pre- and suffix representing a single meaning analytically. The analyticity of X is a salient and necessay condition for making an agglutinative word:

(36) a. Y = [WORD Y - X]/ [WORD X - Y]
     b. X: an analytic affix

 Polysynthesis is, on the contrary, free from the condition (36b). Polysynthetic affixes is analytic, or often discontinuous. The “fusion” of Sapir 1921 means, in a normal sense, a phonological alternation within a word. But fusion and circumfix share a common grammatical featurx in that they are not analytic, but synthetic and possibly discontinuous. We can, therefore, combine the notions as follows:

(37) a. Y = [WORD Y - X]/ [WORD X - Y]/[WORD x1...Y...x2],
     b. X/x1...x2: analytic, fusional and/or discontinuous,
    where x1...x2 is a circumfix.

 Both the grammartical techniques agglutination and polysynthesis make a long complex word by means of affixation. But the difference lies in that agglutination utilizes merely analytic affixes but polysynthesis not only analytic, but also synthetic and discontinuos ones. In Chapter 3, we observed that a polysynthetic verb frame consists of a set of ordered positions for lexical and grammatical categories which are represented by affixes including cirmumfixes. Obviously, this is the salient feature for polysynthetic strucrures in general.

5.3. Incorporation versus Composition
 Incorporation and composiiton are not mutually exclusive, rather partially inclusive. Recall the discussion of Nivkh incorporation. Krejnovich’s supposition that noun-to-noun-incorporation is false, and Panfilov was right in criticizing him. However, we do not know yet exactly under what grammatical conditions nominal composition is possible. Maybe, a semantic compatibility between the components is the least condition for susscessful composition in a general sense, but it is yet an open question how it is grammatically forumulated. But one thing is clear that a derivational interpretation of nominal composition runs in a wrong way because it bring us invetiably to the presupposition of an “underlying” configuration just like Baker’s.
 Verbal categories work in a different way. First, no doubt there are various types of compound verb. The affixed element X0 in [V X0 + V0] varies on lexical categories from noun to preposition. But questions arise if X0 is a pronoun or a clitic, because there must be a certain “underlying” syntactic relationship between them and the head V0. As the compositional frame [V X0 + V0] is recursively applicable, the affixed element X0 can be multiplied. So, a long compound verb can be made just as nominal composition.
 But consider the case of Ainu (29): the affixed element is no more a single lexical category, but explicitely a part of a verb phrase with a clitic. The compositional frame is not applicable because the affixed element is no moren a lexical category X0. Moreover, there must work a certain semantic guranttee between the affixed element and the head to reflect the syntactic relationship between them, something like Baker’s MVC. But there remains yet doubt if we can draw a border-line between incorporation and composition by way of the feature whether it is a single lexical unit or a phrase, namely, X0 or XP. Because, first, XP includes X0 and, second, these must be related in some semantic relationship which reflects the syntactic one. We can not yet bring about any well-motivated counter-argument against the lexicalist view of Sapir 1916 (p. 42):

(38) “Hence noun incorporation is but a particular case of verb composition,
   using that term in this
widest sense, and the objective noun incorporation
   but a particular syntactic use of a larger

 This lexicalist view is correct in any way, because nominal as well as verbal composition do exist in any language. But we find rich examples in Ainu and Chukchee which involve affixed phrases XP. Moreover, the structure is productive, in other words, it varies according to the syntactic context in these languages. The lexicalist view has to be revised in order to take this case into account. We sum up the crucial features for incorporative structures above mentioned:

(39) a. The structure concerned reflects the syntactic relationship between
     the affixed elements and
the verbal head.
     b. The affixed elements can be a phrase which is governed by the verbal head.
     c. The affixation to the verbal head is productive.

 Let us regard the morphosyntactic structure satisfying these features as incorporative and distinguish it from simple verbal composition. Namely, we suppose the binary existence of the verbal categories, incorporation as well as composition. On the other hand, nominal categories, we recognize merely composition. We call our view the weak lexicalist one, in opposition to the fully lexicalist view of Sapir 1916 (38) above.  Recall again Baker’s derivational interpretation of incorporation. It suffers at least from two shortages: first, his MVC filters neither verbal composition nor agglutinative concatenation. It can not distinguish incorporation from composition, too. Second, his incorporation is confined merely to affixes of X0. This veils over the distinction between composition and incorporation. Baker’s view stands on the  opposite position to the lexicalist one of Sapir 1916 in many points.
 Summing up so far, we have the following scheme:











 semantic compatibility

no synt. re-










or syntactic relationship 














  Typologically observed, the use of nominal composition is wide spread and they are productive. And the verbal composition is also popular. But the distinction of incorporation from composition lies in that the verbal affixes of incorporative structures must reflect their syntactic relationship to the head.
 An affixized phrase XP is not found everywhere. As we see the contrast (29a) to (29b) in Ainu examples, the speaker decides whether he represents the notion (29c) in the verb frame or extraposes it outside of it. The decision is made perhaps from his aesthetic taste, and surely, it does not come from a certain grammatical compulsion of this language. On the contrary, he has an option to choose one of the grammatical devices, namely, incorporation or extraposition which this language provides with to ulilize. This grammatical feature characterizes the architecture of the language Ainu; in other words, it has such a mathematical (=formal-aesthetic) inguenity to represent the same matter with a different nuance.

6. Conclusion
 We have begun the discussion with a question whether the languagen Nivkh is incorporative or agglutinative. The question arose already in the discussion of the Soviet Academy, but ended with no theoretical conclusion. There remained, however, a lot of open questions which cast light, especially to the structure of the socalled Paleoasiatic languages. We have seen above, first, that, despite the aparent criteron of consonant alternation, the incorporative characteristics have not been proved in Nivkh. But there is also no pursuasive conclusion that the indefinite object complement is not incorporative as in the examples (2c) and (7). Therefore, a restricted use of incorporation is not excluded in this language. This is in relation to the problem of the prefixed i-, e-, j- and fricative initials of intransitve verbs. For this matter and further, we will have to talk about the vestiges of the old stages of the language.
 The question whether Nivkh belongs to polysynthetic languages has been often talked about, too. The incorporative interpretation of the language seems to help this view. But we have seen in Chapter 3 that polysynthesis and incorporation are not obligatorily coexistent; they rather belong to different grammatical techniques languages utilize for the concatenation of affixized elements. The polysynthetic frame of Itelmen shows a clear counter-example against the assertion e.g. of Baker’s 1996.
 The grammatical techniques composition serves for word formation to make a string Y= [WORD X+Y]. This rule attaches X to Y making up a word Y. Crucial is that it utilizes a grammatical technique of affixation. The element which comes into the position X must be changed into an affix and attached to Y. If Y is a noun, the string works to make a nominal composition. But if Y is verbal, the difference arises: first, if X is limited to a lexical category, namely X=X0, a compound verb is made. In this case, it is ambiguous whether Y is a verbal composition or it is interpreeted as incorporation. The latter case requires the condition that it reflects a syntactic relationship between X and Y. But if X is a phrase governed by Y, the word Y is clearly incorporative. X is often an objective complement or a phrase directly governed by the verbal head. It is often included in a polysynthetic verb frame. However, recall Sapir 1916 and capture the composition in the widest sense. We must conclude that incorporation is a grammatical technique which utilizes another technique composition so as to construct a compound verb to reflect a certain syntactic relationship.
 The polysynthetic frame consists of verbal categories like agreement, causative, passive, aspect, tense etc., which are all external categories of a VP. The incorporation involves grammatical categories of a VP only, but polysynthesis the categories of IP (I=INFL) into a word frame. Therefore, incorporation fertiles a verb with affixes, and polysynthesis expands it into a IP-frame, both making up a large verb.
 Now, let us try to answer the question of the title “why incorporation?” If a language has a mathematical taste for the grammatical architecture to decorate a verb with categories of VP, she makes use of the compositional technique for incorporation. And if she wants to dress up a verb further with categories of IP, she utilizes the synthetic technique to construct a brilliant polysynthetic word. 

Baker 1988: Baker, M. C., Incorporation, Chicago
Baker 1996: Baker, M. C., The Polysynthesis Parameter, Oxford
Chiri 1973, 知里真志保1937:「アイヌ語法概説」『知里真志保著作集 4』平凡社
Hattori 1950, 服部四郎:「付属語と付属形式」『言語研究』50
Haspelmath 1995: The converb as a cross-linguistically valid category. in: Haspelmath,
 M.& E.
Koenig(eds.), Converbs in Cross-Linguistic Perspective. Mouton de Gr./Berlin,

Krejnovih 1958% Krejnovih, E. A., Ob inkorporipovanii v nivxskom qzyke, VQ 1958-6
Krejnovih 1966% Krejnovih, E. A., Ob inkorporipovanii i primykanii v nivxskom
   qzyke, VQ 1966-3
Krejnovih1968% Krejnnovih,E.A., Glagol ketskogo qzyka. Nauka Leningrad
Kurebito 1998: Kurebito Tokusu, A Report on Noun Incorporation in Chukchi, in:
Languages of the 
North Pacific Rim.Vol.4. Kyoto University
Mattissen 2001: Mattissen, J., Dependent-Head Synthesis in Nivh and its Constribution
to a
Typology of Polysynthesis, Dept. of Ling. Uni Koeln
Panfilov 1954% Panfilov, V. Z., K voprosu ob inkorporirovanii - iz materialox
   nivxskogo qzyka, Voprosy Qzykaznaniq (d.VQ) 1954-6
Panfilov 1960% Panfilov, V. Z., Problema slova ^inkorpopipovanie& v nivxskom
   qzyke, VQ 1960-6
Panfilov1966% Panfilov, V. Z., K tipoloiheskoj xarakteristike nivxskogo qzyke,
    VQ 1966-5
Sapir 1916: Sapir, E., The Problem of Noun Incorporation in American Languages.
Sel. Writings
of Edward Sapir. Vol.5 pp.250-282
Sapir 1921: Sapir, E., Language, an introduction to the study of speech, Harcourt
Shiraishi 2001: Phonologically driven allomorphy of Nivkh transitive verbs - with
i mplications
for the nature of prefix -i -
Sovl’eva/Taksami 1970: Sovel;eva, V.N. &H. M. Taksami, Nivxsko-Russkij Slovar;.
   Izd. Sov. ?nc. Moskva
Tamaura 2002, 田村 雅史2002: 動詞の語形成と場所目的語−動詞の語形成

Vall/Kanakin 1990% Vall, M. I. & I. A. Kanakin, Oherk fonologii i grammatiki
   ketskogo qzyka. Nauka Sib.
Volodin 1976% Volodin, A. P., Itel;menskij qzyk. Izd. Nauka, Leningrad
Volodin 1995: Volodin, A. P. &. H-R. Kaempfe, Abriss der Tschuktschischen
 Grammatik auf der
Basis der Schriftsprache.Tunguso-Sibirica 1, Harrassowitz
Volodin 1997% Volodin, A. P., Paleoaziatskie qzyki, Itel;menskij qzyk, i.d.     
     Qzyki Mira - Paleoaziatskie Qzyki - , Nauka Moskva
Volodin 1999: Volodin, A. P. & S.Georg, Die itelmenische Sprache. Tunguso-
 Sibirica 5.
Werner 1997a: Die ketische Sprache, Tunguso-Sibirica 3, Harrassowitz
Werner 1997b: Abriss der kotischen Sprache, Tunguso-Sibirica 4, Harrassowitz
Zhukova 1984% "ukova,A. N., Inkorporativnyj kompleks kak slovosohetanie
 qzykax hukoto-kamhatskoj gruppy VQ 1984-6