Grammaticalization and structural scope increase: possessive-classifier-based benefactive marking in Oceanic languages *.
This article discusses the development of possessive classifiers into benefactive markers in Oceanic languages. On the basis of Tabor and Traugott's (1998) diachronic string comparison, this change episode will then be demonstrated to involve structural scope increase contrary to the widely held assumption that scope decrease is a manifestation of grammaticalization. The article also identifies as an empirically testable hypothesis the strong connection between scope increase, and the formal identity between source items and their grammaticalized descendants. Moreover, the recent controversy over the status of grammaticalization (theory) is critically appraised with the conclusion that grammaticalization has an independent status of its own and the general mechanisms of language change should be invoked to account for instances of grammaticalization.
One of the major claims made in grammaticalization theory is the correlation between grammaticalization and structural scope decrease--also known as "(structural) condensation" (Lehmann 1995 : 143) or as "(structural) compacting" (Hopper and Traugott 1993: 130): the structural scope of a linguistic expression decreases in proportion to grammaticalization. This idea of structural scope decrease is embodied, for instance, in Givon's (1971, 1979) cline of structural types, across which grammatical change proceeds from left to right, as in (1): (1)
(1) discourse [right arrow] syntax [right arrow] morphology [right arrow] morphophonemics [right arrow] zero
The same idea is contained also in Lehmann's (1995 : 13-14, 143-147) "phases of grammaticalization," as in (2):
(2) syntacticization [right arrow] morphologization [right arrow] demorphemicization loss
These two clines refer essentially to the same thing, the main difference being that (1) captures different structural types undergoing processes of grammaticalization, and (2) different processes of grammaticalization bearing upon structural types (for similar views, also see Langacker 1977; Heine and Reh 1984; Heine et al. 1991; Hopper and Traugott 1993; and Bybee et al. 1994; inter alia). Thus, syntacticization in (2) covers the transition from discourse to syntax in (1), that is, "a free collocation of potentially uninflected lexical words in discourse" changing into syntactic constructions "whereby some of the lexemes assume grammatical functions" (Lehmann 1995 : 13). Morphologization in (2) refers to syntactic constructions developing into synthetic ones, that is, syntax to morphology in (1). Demorphemicization in (2) describes the transition from morphology to morphophonemics in (1). Lastly, loss in (2) represents the final stage of grammaticalization: grammatical categories may completely lose their form and content, that is, zero in (1). For example, relational nouns (e.g. ride in by the side of) may potentially develop into adpositions (e.g. beside), into agglutinative case affixes and then into fusional cases, as predicted by Givdn's and Lehmann's formulations (e.g. Lehmann 1995 : 76-87, Hopper and Traugott 1993:6-8, 106-108).
This series of developments can be seen to involve gradual reduction in structural scope of the formative in question. Thus a relational noun or an adposition for that matter--takes a noun phrase with case as its complement. To put it differently, a noun phrase with case comes under the scope of a relational noun (e.g. Hopper and Traugott 1993: 107). If and when a relational noun is grammaticalized into an agglutinative case affix, it will be attached to a caseless noun phrase, if and when an agglutinative case affix becomes a fusional case, it will be "fused" onto a noun stem. The structural scope of the grammatical categories in question is then reduced by degrees. In terms of constituent structure, then, this represents a change from (3) to (4) to (5) (taken from Lehmann 1995 : 144).
In (3), the structural scope of the relational noun is over an NP with case, whereas in (4), it is reduced to be over a caseless NP. Finally, in (5) the scope of the erstwhile relational noun is over an N stem.
Another well-recognized example of scope decrease comes from the development of bound personal pronouns. Crosslinguistically, bound personal pronouns tend to arise from independent personal pronouns (e.g. Givon 1976, 1979; Hopper and Traugott 1993: 131; and Lehmann 1995 : 39-42; inter alia). Independent personal pronouns occur at the clause level. But when they become bound personal pronouns, they tend to be attached to lexical verbs. This change episode can easily be translated into a decrease in structural scope (e.g. Lehmann 1995 : 144). Bound personal pronouns, having been reduced to bound morphemes, must be hosted or borne by words (verbs in most cases), whereas independent personal pronouns, being separate words, can stand on their own, on a par with other words in clauses. Because the more grammaticalized tends to mean the more reduced in phonetic substance (e.g. Heine and Reh 1984; Bybee et al. 1994; and Lehmann 1995 ), the correlation between grammaticalization and scope decrease makes sense.
Change episodes of scope decrease abound in the literature (e.g. Hopper and Traugott 1993; Heine et al. 1991). Thus it does not come as a surprise that structural scope decrease has been interpreted to be a manifestation of grammaticalization itself (e.g. Lehmann 1995 ). Scope decrease seems to be widely accepted also as a criterion for identifying instances of, or as a general principle of, grammaticalization (Tabor and Traugott 1998: 231, 263).
Though structural scope decrease is well attested, there do exist "hard-to-dismiss" counterexamples to the scope decrease claim or, more generally, to the direction of change in (1) or (2), that is, so-called unidirectionality (Tabor and Traugott 1998; also see Jeffers and Zwicky 1980; Janda 1980, 1995, 1996, 2001: Matsumoto 1988; Campbell 1991, 2001b; Joseph 2001; Joseph and Janda 1994; Davies 1994; Newmeyer 2001; and Kim 2001; but cf. Haspelmath 1999, 2002). Perhaps one of the best known counterexamples may be the history of the English possessive. What started as a genitive suffix in Old English functions as a clitic in Present Day English (Joseph and Janda 1994; but cf. Janda 1980; Menn and MacWhinney 1984: and Zwicky 1987). The structural scope of the Old English genitive suffix (e.g. ) was over a caseless lexical item, whereas the structural scope of the Present Day English possessive clitic (e.g. )--appearing at the end of possessor NPs even in the presence of post-head modifiers (e.g. of England in )--is over a whole phrasal constituent.
(6) ooes cyning:es sweoster Ecgfrid:es the:GEN king:GEN sister:NOM Ecgfrid:GEN 'the sister of Ecgfrid the king' (Tabor and Traugott 1998: 237) (7) the Queen of England's visit
Clearly, the Old English genitive suffix had a narrower scope than the Present Day English possessive clitic has.
Other counterexamples come from the domain of syntax (e.g. Tabor and Traugott 1998:230-231). In the history of Spanish and Portuguese, shifts "from reduced (VP) to nonreduced (IP) clauses" are reported to have occurred with the effect that a series of changes from biclausal to monoclausal to biclausal causatives can be documented (Davies 1994). In Kayardild (Australian; Bentinck Island, Queensland, Australia), subordinate clauses have been reanalyzed as independent in a change episode which Evans (1988, 1993) refers to as "insubordination." These change episodes totally unexpected under the scope decrease claim--point unequivocally to structural scope increase.
Counterexamples like these change episodes have led Tabor and Traugott (1998) to suggest that structural scope decrease as embodied in the cline in (1) or (2) should not be taken to be a criterion for identifying instances of grammaticalization, but should be regarded only as a hypothesis that needs "to be empirically tested with an explicit definition of "scope change'" (Tabor and Traugott 1998: 231). Moreover, they argue that structural scope increase also is an empirically testable hypothesis "worth looking into further," because it "turns out to be surprisingly robust" (Tabor and Traugott 1998: 225, 263). In other words, the scope decrease claim and the scope increase claim both need to be empirically tested, with systematic exploration of their formal possibilities to be carried out (Tabor and Traugott 1998:231-232).
The first part of the present article, as a direct answer to Tabor and Traugott's call for such systematic exploration, documents one change episode in Oceanic languages: the development of possessive classifiers into benefactive markers. Contrary to the scope decrease claim, this change episode involves structural scope increase. What started as part of a direct object NP has assumed the status of an adjunct NP, on a par with a larger phrase that contains that direct object NP. Moreover, it will be suggested that change episodes of scope increase, as opposed to scope decrease, involve no or little change in the formal identity between source items and their grammaticalized descendants.
The second part of the article (i.e. Section 5) attempts to help resolve a recent controversy over the status of grammaticalization and also the need for grammaticalization theory (i.e. Haspelmath 1998, 2002; Campbell 2001a, 2001b; Janda 2001; Joseph 2001; and Newmeyer 2001). These issues may not seem to be directly related to the first part of the article, but it is important to discuss them here because the development of possessive classifiers into benefactive markers in Oceanic languages is taken to be an instance of grammaticalization within grammaticalization theory. If the status of, and the need for, grammaticalization (theory) were in doubt, the status of the change episode as an instance of grammaticalization would also be likely to be in doubt. Opponents of grammaticalization (theory) should recognize the independent status of grammaticalization and the need for grammaticalization theory. The extreme pro-grammaticalization position held by Haspelmath (1998), on the other hand, suffers from one conceptual flaw: grammaticalization itself is taken to be the main mechanism of language changes with grammaticalization. General mechanisms of language change, reanalysis in particular, should instead be appealed to in accounting for instances of grammaticalization. Thus, this article is an attempt to develop the study of one change episode in a well-defined genetic group into a general contribution to research on grammaticalization.
2. Tabor and Traugott's definition of structural scope change
As a preface to the main discussion, Tabor and Traugott's (1998) explicit definition of scope change will be reviewed in brief. As a means of comparing scopes across different states of a language, Tabor and Traugott (1998) draw upon the notion of c-command, a theoretical construct which plays an important role in Government and Binding Theory (e.g. Haegeman 1994 and Ouhalla 1994, based on Reinhart 1981). The notion of c-command is based on the dominance relationship between different nodes in constituent structure: node A c-commands node B if A does not dominate B, and every branching node which dominates A also dominates B. (2) Following the standard procedure, Tabor and Traugott (1998: 233) equate c-command with scope, and assume that if node A asymmetrically c-commands node B (that is, node A c-commands node B, but not vice versa), then node A has greater scope than node B.
Then, in order to facilitate a direct comparison of scopes across earlier and later states of a given formative, Tabor and Traugott (1998: 234) propose what they call "diachronic string comparison" (DSC), in which each word in an older construction can be replaced with a descendant form from some specific later stage to produce a legitimate sentence from that later stage. Note that DSCs are made at the level of S-structure, as opposed to D-structure (Tabor and Traugott 1998: 261; Elizabeth Traugott pers. comm.). (3) For instance, the development of main verbs such as will into auxiliary verbs in English is an example of scope change, because, when the Old English construction involving wille is compared with the Present Day English construction with its descendant form will, there is an increase in structural scope, as illustrated in (8). (4)
(8) Main verb to auxiliary verb a. Min Drihten, ic wille gangan to Rome. (Old English) my lord I want go:INF to Rome 'My Lord, I will/want to go to Rome.' b. My Lord, I will go to Rome. (Present Day English) c. [CP Min Drihten [IP ic [I'O [VP wille [VP gangan [PP to [NP Rome]]] ]]]] d. [CP My Lord [IP I [I' will [VP go [PP to [NP Rome]] ]]]]
Between (8c) and (Sd), the scope of the expression in question (wille in Old English vs. will in Present Day English) has clearly enlarged. When a comparison is made between these two positions, the element will in (8d) is located at a higher level (i.e. I') in the constituent structure than is the element wille in (8c) (i.e. vp). "Given this alignment of brackets, there is a clear sense" in which the position of will in (8d) has scope over that of wille in (8c) (Tabor and Traugott 1998: 234).
By making use of c-command and DSC, Tabor and Traugott (1998) formulate their "c-command scope-increase hypothesis" as follows:
(9) The c-command scope-increase hypothesis: When an item undergoes gradual syntactic reclassification, resulting in a state in which diachronic string comparison can be applied, then its c-command scope increases.
They support this hypothesis by describing four change episodes of scope increase from English: the development of the -s possessive (as discussed above), the VP-gerund, adverbial and conjunctive instead (of), and the discourse marker anyway.
In this article, Tabor and Traugott's c-command scope-increase hypothesis will be put to the test by examining the development of possessive classifiers into benefactive markers in Oceanic languages. But first, the change episode needs to be introduced.
3. Possessive classifiers as benefactive markers in Oceanic languages
In Oceanic linguistics, two different types of possession are generally recognized: alienable and inalienable possession (Lynch 1973, 1996, 1998; Lichtenberk 1983). In alienable possession, the possessor is seen to have control either over the possession itself or at least over the fact of possession, for example, items of disposable property or items that the possessor owns or controls in one way or another (Lynch 1998: 123). This type of possession contrasts with inalienable possession, in which the possessor is seen to have little, if any, control over the fact of possession, for example, parts of the body or some kinship terms (Lynch 1998: 122). (In Oceanic linguistics, alienable possession is referred to also as dominant, and inalienable possession as subordinate.)
In Oceanic languages (see Pawley and Ross  for a genetic classification of Oceanic languages), inalienable possession is most often associated with direct suffixation of possessive pronouns to possessed nouns (Harrison 1988; Lynch 1973, 1996, 1998: 122-123; and Lichtenberk 1983). (5) This type of construction, however, will be of no relevance to the topic of the present article. Alienable possession, on the other hand, tends to be expressed by means of auxiliary elements termed possessive classifiers (Harrison 1998; Lynch 1973, 1996, 1998: 123-124; and Lichtenberk 1983). In this type of construction, possessive pronouns are added to possessive classifiers, not to possessed nouns; when the possessor is expressed by a noun, the possessive classifier, depending on languages, may also carry a so-called construct suffix or an appropriate third person possessive pronoun. The possessive classifier construction is exemplified in (10), taken from Mokilese (Micronesian subgroup, Oceanic; Mokil Atoll and Ponape) (Harrison 1976: 129).
(10) Mokilese nime-n woall-o pill-o PCL-CONSTR man-that water-that 'that man's water'
The function of the possessive classifier system is not just to express possession but also to place possessed nouns into different categories by specifying a semantic or culturally significant relation between the possessor and the possessum (i.e. possessed nouns) (Lichtenberk 1983). For example, given possessor X and possessum Y, the "edible" possessive classifier expresses that Y is food to X; the "drinkable" possessive classifier indicates that Y is a drink to X; and the "plant" possessive classifier specifies that Y is a plant to X; and so on. In (10), thus, the possessive classifier nime- indicates that the possessive relation between the possessor and the possessum is such that the latter is something drinkable to the former--as opposed to water for washing. It is also possible that possessum Y, for example, coconut, can enter into more than one relationship, for example, food, drink, and plant, with possessor X (Lynch [1973: 76] refers to this phenomenon as "overlap."). This is why Lichtenberk (1983: 148) chooses to refer to possessive classifiers as relational classifiers as opposed to "sortal" or "mensural" classifiers in numeral classifier systems (cf. Allan 1977; Aikhenvald 2000; also see Lucy 2000 and Senft 2000b for a critical discussion of nominal classification systems and misconceptions associated with terms such as sortal and mensural classifiers).
The number of possessive classifiers varies from language to language, but most Oceanic languages are said to have at least a two-way contrast in their possessive classifier system (i.e. two possessive classifiers). The Oceanic languages in the Micronesian subgroup (with the exception of Gilbertese, which has no possessive classifiers) (6) are atypical in that they possess a very large number of possessive classifiers, ranging between fifteen and twenty (and possibly more), and indeed few outside Micronesian are known to have more than four possessive classifiers (Lichtenberk 1983:154-156; Harrison 1988: 64; Song 1997). The possessive classifier system is well-established in Oceanic languages to the extent that at least three possessive classifiers have been reconstructed for Proto-Oceanic (POc): *ka-'edible,' *na- 'general,' and *ma- 'drinkable' (Pawley 1973; Lichtenberk 1985). (7)
In Oceanic languages, possessive classifiers are enlisted in the service of expressing benefactive role, as illustrated in (11), where the possessive classifier nih- is exploited to encode benefactive role (Harrison 1976: 133) (see Song 1998 for additional documentation).
(11) Mokilese ngoah insingeh-di kijinlikkoauoaw nih-mw 1SG:SBJ write-ASP letter PCL-2SG:POSS 'I wrote a letter for you.'
(Note that in Mokilese the sequence of the possessive classifier and the possessor, for example, nih-mw in , can either precede or follow the possessum, for example, kijinlikkoauoaw, in , when the possessive classifier is used to encode benefaction, as opposed to possession [Harrison 1976: 133; also see Section 3.1]; this explains why the relevant word order in  is different from that in .)
The direction of exploitation is taken to be from possessive classifiers to benefactive markers, not the other way round. There is reason to decide on this direction (but cf. Heine 1997a, 1997b, and also Heine and Kuteva 2002:54 55 for the opposite direction attested in non-Oceanic languages). In Oceanic languages, reflexes of the POc possessive classifiers and relevant constructions express either possession or both possession and benefaction, but never benefaction alone (Song 1997: 58, 1998: 270). Margetts (2004) and Lichtenberk (2002: 447) also accept this distributional fact as a strong argument for assuming that the possession-marking function was the original one and that the benefaction-marking function was a later development.
Moreover, it is not the case that in (11) the benefactive NP has been realized on the surface as the possessor of the direct object NP, as is suggested by Croft's (1985: 41) "indirect object lowering" (i.e. some kind of synchronic "rule" designed to handle the surface realization of a recipient or beneficiary as the possessor of the direct object NP). Indeed the term "indirect object lowering," as Margetts (1999: 326, 2002: 619) correctly points out, is problematic or misleading in the context of Oceanic languages in that it takes a ditransitive construction with an indirect object NP as basic and a possessive-classifier-based benefactive expression (e.g. ) as an alternative or secondary construction. For many Oceanic languages--insofar as it can be determined from the available data--the possessive-classifier-based benefactive expression is the basic construction, there being no alternative construction in which the beneficiary could be expressed as an indirect object NP (Song 1998).
The development of possessive classifiers into benefactive markers in Oceanic languages must be regarded as an instance of grammaticalization in terms of: (i) fossilization of pragmatic or discourse strategies in the morphosyntactic structure (Givdn 1971, 1979; Lehmann 1995 ; Traugott and Heine 1991b: 2-3; also see Margetts 2002, 2004; Lichtenberk 2002); and (ii) reduced structural autonomy with reference to the verb (Haspelmath 1998:336 340; cf. Lehmann 1995 : 109 113).
The change episode can be regarded as an instance of grammaticalization in the sense that what initially emerges as a pragmatic inference grammaticalizes into a benefactive construction (see Traugott and Heine [1991b] for further support of this particular view). As will be explained in detail in Section 3.4, the change episode has its origin in pragmatic inferencing: the possessor is pragmatically interpreted or construed as a beneficiary. This pragmatically generated interpretation then "gradually [and increasingly] becomes manifested in the morpho-syntax of [the languages documented in Sections 3.1, 3.2, and 3.3]" (Margetts 2004: 447). Thus the change episode fits the description of the initial part or phase of grammaticalization in Givon's (1971, 1979) or Lehmann's (1995 ) work (i.e.  or , respectively): the fossilization of pragmatic or discourse strategies in the morphosyntactic structure.
Moreover, grammaticalization represents language changes in which linguistic expressions shift from a lexical to a grammatical status or from a less grammatical to a more grammatical status (e.g. Kurytowicz 1965; also see Campbell and Janda  for a useful survey of definitions of grammaticalization). The change episode in question (i.e. possessive [right arrow] benefactive) falls within the purview of this definition as well. Lehmann (1995 : 109-113), for instance, speaks of a grammaticalization hierarchy from locative to oblique to direct cases, for example, directional [right arrow] dative, dative [right arrow] accusative, and instrumental [right arrow], ergative. In these changes, the earlier cases are more concrete or semantically richer than the latter, which are "desemanticized" (Haspelmath 1998: 339; also see Heine et al. [1991: 148-168] for detailed arguments for a very similar grammaticalization hierarchy of cases). In other words, as one moves from locative to oblique to direct cases on Lehmann's grammaticalization hierarchy, the meanings of these cases become less concrete and more grammatical. (This is the basis of the traditional notion of concrete vs. grammatical case [Lehmann 1995 : 112].) But it is not easy to see how possession is more concrete or semantically richer than benefaction or vice versa; possession and benefaction seem to be equally concrete (or abstract) or semantically rich (but cf. Heine et al. 1991: 151).
Haspelmath (1998: 339), however, interprets Lehmann's grammaticalization hierarchy of cases in terms of reduced structural autonomy. By reduced structural autonomy is meant "the tightening of the relation with the verb" (also see Lehmann 1995 : 110 for a similar perspective). For example, if its case (e.g. ergative) is directly governed by the verb and/or intimately connected with the meaning of the verb, the relation with the verb of a given nominal expression can be said to be "tight." If, on the other hand, its case (e.g. instrumental) is not directly governed by the verb and/or not intimately connected with the meaning of the verb, the relation with the verb of a given nominal expression can be thought to be "loose." (Bear in mind that the distinction between tight and loose is not an either-or distinction, but a continuum.) Thus changes from non-direct to direct cases are understood in terms of reduction in structural autonomy. A change from the instrumental to the ergative case, for example, can be looked upon as a shift from a less grammatical (= loose) to a more grammatical (= tight) status. In other words, the degree of tightness of the relation with the verb is equated with the degree of grammaticalization.
The advantage of Haspelmath's interpretation is that the concept of reduced structural autonomy can be extended to changes from (more) nondirect to (less) nondirect cases, for example, possessive [right arrow] benefactive. The relation between the benefactive NP and the verb is much tighter (or less loose) than that between the possessor and the verb. The benefactive NP, albeit an adjunct (i.e. not directly governed by the verb), is connected with the meaning of the verb. Indeed, the very notion of benefaction depends on the action (or non-action as the case may be) denoted by the verb (e.g. if X had not carried out an action, Y would not have benefited from that action.) The possessor, on the other hand, is not at all connected with the meaning of the verb; rather, it is associated with the meaning of the possessum alone. The notion of possession exists independently of the action denoted by the verb. In fact, the relation with the verb is nonexistent insofar as the possessor is concerned. This disparity is not unexpected, because the function of the possessive classifier is to encode the relation between the possessor and the possessum (in other words, an NP-level relationship), whereas the function of the benefactive marker is to encode the relation between the benefactive NP and the verb (in other words, a VP-level relation). For this reason, the benefactive NP, not unlike other major constituents of the clause, is subject to (clause-level) grammatical rules or restrictions, whereas the possessor is not. In English, for instance, the former may be clefted or moved to different positions in the sentence (e.g. The teacher baked a cake for the student [right arrow] It was the student that the teacher baked a cake for), whereas the latter, independently of the possessum, may be unable to undergo such grammatical operations (e.g. The student liked the teacher's cake [right arrow] * It was the teacher's that the student liked cake; cf. It was the teacher's cake that the student liked). The NP-level relation points to a high degree of structural autonomy with respect to the verb, and the VP-level relation reduced structural autonomy with respect to the verb. In terms of Haspelmath's reduced structural autonomy, therefore, the change episode in question (i.e. possessive [right arrow] benefactive) is a shift from a less grammatical to a more grammatical status.
It is important to emphasize at this juncture that, even in languages with highly grammaticalized possessive-classifier-based benefactive marking, it is not necessarily the case that possessive classifiers no longer carry out their original function of expressing possession (see Section 3.4 for further discussion). They certainly do so. In point of fact, this kind of polysemous situation is commonly observed in instances of grammaticalization. Thus, when a formative undergoes grammaticalization, the new meaning(s) and function(s) that it has acquired tend(s) to coexist--at least in intermediate stages--with its original meaning and function. (This is captured in Hopper's [1991: 22] "principle of persistence".) For instance, Bybee and Pagliuca (1986) discuss a number of semantic distinctions (or meanings) of the auxiliary verb will in Present Day English: future (or prediction) meaning and other modal meanings (e.g. willingness and intention). The future meaning developed historically out of the modal use of will: the modal meanings were already found in Old English, whereas the future meaning was established only in the Middle English period (Bybee and Pagliuca 1986). These meanings all coexist in Present Day English will.
Some of the hallmarks of possessive-classifier-based benefactive marking as an instance of grammaticalization come from three different areas: (i) word order change, (ii) generalization, and (iii) specialization (Bybee and Pagliuca 1985:62 63, 67, 71 75; Hopper 1991: 25-28; Heine et al. 1991: 109, 157: Hopper and Traugott 1993:100 103, 113-116; but cf. extension in Campbell 2001b: Harris and Campbell 1995).
3.1. Word order change
In Kusaiean (Micronesian subgroup, Oceanic; Kusaie Island, the Caroline Islands and Nauru), the determiner ah must occupy the final position of NPs (Lee 1975: 237). The sequence of the possessive classifier and the possessor nominal, when expressing possession, appears in between the possessum rais and the determiner ah in (12), for it is part and parcel of the possessive NP (or the direct object NP in this case), rais la-l Sohn ah "John's rice.'
(12) Kusaiean nga mole-lah rais la-I Sohn ah 1SG:SBJ buy-ASP rice PCL-3SG:POSS John DET 'I have bought John's rice."
In (13), on the other hand, the "possessive classifier" and the "possessor" encode a nominal with benefactive role. (Note that hereafter erstwhile possessive classifiers, possessors, and possessums are enclosed within double quotation marks.) In other words, they are no longer part of the direct object NP, with the possessive classifier having been grammaticalized as a benefactive marker. This is why the sequence in question appears to the right of or after the determiner ah, which marks the right periphery of the direct object NP.
(13) Kusaiean nga mole-lab rais ah la-l Sohn 1SG:SBJ buy-ASP rice DET PCL-3SG:POSS John 'I have bought the rice for John.'
The development of the possessive classifier into the benefactive marker has thus resulted in the possessor physically "moving" out of the erstwhile possessive NP, as it were, and becoming a separate benefactive NP. The benefactive-marking function of the possessive classifier system in Kusaiean is further supported by (14) (Lee 1975: 262).
(14) Kusaiean Sohn el mole-lah ik la-1 Sepe ah John 3SG:SBJ buy-ASP fish PCL-3SG:POSS Sepe DET la-1 Sru PCL-3SG:POSS Sru 'John has bought Sepe's fish for Sru.'
In (14), the possessive classifier la- is used twice in the same clause, once for the expressing of possession, and once again for the encoding of benefaction. Note the absence of the determiner ah in the second sequence of the "possessive classifier" and the "possessor," la-l Sru; the sequence cannot be understood to mean something like "Sru's (fish)," with the "possessum" deleted or understood.
A similar word order change seems to have taken place also in Lenakel (South Vanuatu subgroup, Oceanic; West Central Tanna, Vanuatu). In (15), the sequence of the "possessive classifier" and the "possessor" is separated from the "possessum" by the locative phrase, le niki-nhamra 'in the bush.' Identifying the possessive classifier taha- as a general benefactive marker or "introducer," Lynch (1978: 72, 93) points out that, if taha-m is placed immediately after the direct object NP him ker 'a breadfruit' in (15), it is more likely to be interpreted to express possession than benefaction.
(15) Lenakel i-[??]m-[??]lh n[??]m ker le n[??]ki-nhamra 1:EXC-PST-pick breadfruit one LOC LOC-bush taha-m PCL-2SG:POSS 'I picked a breadfruit for you in the bush.'
Compare (15) with (16), in which the possessive classifier expresses possession.
(16) Lenakel kuri miin taha uus mil aan k-n-ai-ak[??]mw dog PL PCL man DL that 3NSG-PERF-PL-run.away ita already 'Those two men's dogs have run away.'
In (16), the possessum nominal is followed immediately by the sequence of the possessive classifier and the possessor.
This kind of word order change demonstrates clearly that the use of possessive classifiers for benefactive marking cannot be looked at from the perspective of Croft's (1985) indirect object lowering rule or some other rule akin to that. If such a rule were to account for this, the possessive classifier and the possessor, when encoding a beneficiary, would be expected to occupy the same sentence positions as when expressing possession. This is so, because the benefactive nominal, under indirect object lowering, is thought to be merely realized on the surface as the possessor of the direct object NP. However, they do appear in different positions in languages like Kusaiean and Lenakel, depending on whether they encode possession or benefaction. This shows that Croft's indirect object lowering is inappropriate for the change episode under discussion.
The "possessor" nominal may also be used on its own, without the "possessum" nominal being expressed in the same sentence, as long as the "possessive classifier" is there to encode the benefactive role of the "possessor" nominal. This is to say that the use of the possessive classifier for benefactive marking may also be extended or generalized to verbs which lack direct object NPs. Evidence of this generalization is observed in Kusaiean, in which the possessive classifier can be used for benefactive marking in conjunction with "derived" intransitive verbs as well, although the latter must co-occur with their "included" or incorporated object nominals, as illustrated in (17) (Lee 1975: 263, 270-277; see Sugita 1973 for a detailed discussion of object incorporation in Micronesian languages).
(17) Kusaiean nga twetwe mitmit nahtuh-1 Sepe I sharpen knife PCL-3SG:POSS Sepe 'I am knife-sharpening for Sepe.'
In (17), the "included" object nominal (or the "possessum") mitmit, amalgamated with the verb, is no longer associated with the "possessor." There is evidence for the status of the "included" object nominal (Lee 1975: 270-271). First, the "derived" intransitive verb twetwe contrasts with its transitive counterpart, twem. Second, the incorporated object nominal mitmit in (17) cannot be modified by a determiner or a numeral. Finally, directional suffixes cannot be attached to the end of the "derived" intransitive verb twetwe but to the end of mitmit or, more accurately, twetwe mitmit in (17). By contrast, directional suffixes must be attached to the end of the transitive verb twem.
In Mokilese also, the use of possessive classifiers for benefactive marking is not only found in transitive clauses, as in (18), but also in clauses without overt direct object (or "possessum") NPs or in intransitive clauses, as in (19) (Harrison 1976: 133).
(18) Mokilese ngoah insingeh-di kijinlikkoauoaw nih-mw 1SG:SBJ write-ASP letter PCL-2SGPOSS 'I wrote a letter for you.' (19) Mokilese lih-o doadoa a-h woman-that sew PCL-3SG:POSS 'That woman sews for him.'
Being subject to fewer restrictions on its occurrence, the use of possessive classifiers for benefactive marking is permitted in a wide range of grammatical contexts in Mokilese (and Kusaiean). Note that the English translations provided in (18) and (19) are not free translations for something like "I wrote your letter" or "The woman sews his (something)." These are, in fact, two of the sentences that Harrison (1976: 132-133) uses in order to discuss the "benefactive use of the possessive classifiers" in Mokilese; "[i]n such sentences [as (18) and (19)], the possessive classifier indicates the person who benefits from the object (or action), rather than its owner" (Harrison 1976: 133).
Further evidence for the grammaticalization of possessive classifiers as benefactive markers comes from the fact that one of the multiple possessive classifiers, which are all used to express possession in a given language, is chosen and pressed into the service of encoding benefaction. In other words, one possessive classifier is specialized for purposes of benefactive marking. Such is the case in Lenakel, in which out of the five possessive classifiers mainly the general one taha- is allowed to encode benefactive role.
(20) Lenakel a. uus-suaas ka r-im-am-asumw taha man-small that 3SG-PST-CONT-garden PCL r[??]m-n iuok[??]t to nimwa taha-k father-3SG:POSS near LOC house PCL-1SG:POSS 'That boy was gardening for his father near my house.' b. i-[??]m-[??]lh nim ker le niki-nhamra I:EXC-PST-pick breadfruit one LOC LOC-bush taha-m PCE-2SG:POSS 'I picked a breadfruit for you in the bush.'
In (20a), there is no "possessum" nominal which the first taha-marked nominal could be associated with, even if taha- were to be interpreted as the general possessive classifier (i.e. possession) (Lynch 1978: 72, 93). In fact, taha- is used twice here, once as a benefactive marker and then once again as a possessive classifier proper. It is also important to note that the semantics of the general possessive classifier taha-, when used to encode benefactive role, must be abstract enough--"it lacks certain specific features of meaning" (Bybee and Pagliuca 1985: 63)--to be compatible with that of the nominal which would otherwise be associated with it as the possessum, for example, the direct object NP him ker 'a breadfruit' in (20b). Lenakel thus contrasts with Kusaiean and Mokilese, in which different possessive classifiers must be selected, depending on semantic or cultural relationships between benefactive and direct object NPs. In the latter languages, the possessive classifiers can only be used to encode benefactive role in the same limited set of environments in which they express possession. This suggests strongly that the possessive classifier taha- in Lenakel is at a far more advanced stage of grammaticalization as a benefactive marker. It is found in a much wider range of semantic contexts.
3.4. Different stages of grammaticalization
The Oceanic languages discussed above--and other Oceanic languages where sufficient data are available--can be placed at different stages of grammaticalization. This has so far been only hinted at in an informal manner. For instance, Lenakel has been identified as being at a far more advanced "stage" of grammaticalization than Kusaiean and Mokilese. This kind of exercise can in fact be done in a more precise manner by drawing on Heine's (2002) scenario of grammaticalization. This particular scenario comprises four context-based stages of grammaticalization (but cf. Diewald's  three context types; cf. Margetts 2004): Stage I: initial stage; Stage II: bridging context (5. la Evans and Wilkins 2000: 550); Stage III: switch context: and Stage IV: conventionalization. At Stage I, there is an expression with a "normal"--original--meaning occurring in different contexts (in the present case, the possessive classifier encoding possession alone). At Stage II, an inferential mechanism is triggered to the effect that a new meaning emerges as a more plausible interpretation for the expression in question (in the present case, the possessive classifier expressing benefaction as well as possession). Note that the original meaning cannot be ruled out at this stage, because the new meaning is cancellable or defeasible. Switch contexts are characteristic of Stage III. Switch contexts, which represent specific contexts for the new meaning, are incompatible or in conflict with the original meaning so that the original meaning is ruled out in those specific contexts. Finally, at Stage IV the new meaning is "freed from the contextual constraints that gave rise to it; it may now be used in new contexts" that is, other than bridging and switch contexts (Heine 2002: 86). Moreover, the new meaning is conventionalized with the effect that it can co-occur with the original meaning side-by-side in the same clause (Heine 2002: 85). Thus, there is now no room for confusion or ambiguity between the original and new meanings. It must be noted that these four stages are not meant to be discrete but there is a continuum between Stage I and Stage IV (Heine 2002: 86).
Some Oceanic languages seem to be at Stage II (or "bridging context") in that the new meaning of benefaction is inferentially or pragmatically generated, and the old meaning of possession is still possible. Though due to lack of detailed data, it is not possible to define exactly what the relevant bridging context(s) may be, it can at least be gleaned from some of the examples available from published descriptive grammars that possessive NPs in direct object function, not in subject function or in prepositional phrases, can give rise to the new meaning of benefaction (Margetts 2004). The possessor in the direct object NP can be construed as a beneficiary, for example, The man built the woman's house [right arrow] The man built a house for the woman, but the possessor in the subject NP as in The woman's pig ran away or in a PP as in The man ran with the woman's pig cannot. Moreover, the activity denoted by the verb must be of such a kind as "brings the possessor into possession of the object" so that "[t]he referent of the grammatical possessor can be construed as [the] beneficiary of [the] activity expressed by the verb" (Margetts 2004, based on Croft 1985). Such verbs can be characterized as verbs of creation (e.g. bake, build, cook, sew, write), verbs of transfer (e.g. give, send, pass) or verbs of obtaining (e.g. hunt, kill, fish) (Croft 1985: 45). Thus, the possessor in the direct object NP in a sentence with the verb denoting the activity of building a house (e.g. The man built the woman's house) is much more likely to be construed as a beneficiary than the possessor in the direct object NP in a sentence with the verb denoting the activity of chasing a pig (e.g. The man chased the woman's pig).
There is a conceptual motivation for this in that beneficiaries can indeed be thought of as new or prospective possessors: "[t]he new possessor is in the benefactive role by virtue of his coming into possession of the possessed item" (Croft 1991: 295). In other words, by virtue of her coming into possession of the house, the woman benefited from the man's activity of building it. This kind of inferencing is not plausible in the case of the man chasing the woman's pig. More likely is the scenario in which the woman owned the pig prior to the man's chase. Thus, it is not easy to imagine that the man's chase somehow contributed to the woman's coming into possession of the pig; the possessor nominal (i.e. the woman) expressed in conjunction with the possessum (i.e. pig) is there to indicate the identity of the latter more than anything else.
But then exactly how does the new meaning of benefaction arise from this kind of bridging context? Given the basic sentence The man built the woman's house, the hearer could run through the following inferencing (cf. Diewald 2002: 109-114):8
The man built a house, and it now belongs to the woman. But, if 1 (= the hearer) focus on the time during which the house was being built, it is also plausible to think that, because one cannot possess something before it comes into existence, the speaker may want to imply that, while the man was building the house, he was doing so FOR THE BENEFIT OF the woman. Therefore, it makes sense to conclude that the speaker can also mean by uttering The man built the woman's house: "The man built a house for the woman."
Oceanic languages such as Tolai (Western Oceanic subgroup, Oceanic; Eastern New Britian Province, Papua New Guinea) may be regarded as being at Heine's Stage II (i.e. bridging context). For instance, Mosel (1984: 172-173) points out that in Tolai, the possessive classifier system is exploited to express benefactive nominals IF it is understood that the possessor owns or will own what is affected or produced by the action or that it is done for the benefit of the possessor. This is the kind of bridging context that has been identified in connection with the types of verb that can give rise to the new meaning of benefaction.
Other Oceanic languages may be at Heine's third stage of grammaticalization (i.e. switch context), because the new meaning is the only possible interpretation and also because that meaning has to "be supported by a specific context (or cluster of contexts)" (Heine 2002: 85). In Sakao (North/Central Vanuatu subgroup, Oceanic; Northeast Santo Island, Vanuatu), for instance, the possessive classifier appears without the possessum nominal in the context of intransitive verbs (Guy 1974: 42-44). The interpretation of the possessive classifier as a marker of possession is no longer possible here, simply because there is no possessum nominal to go with it. This specific context, which is now incompatible with the original meaning, allows only the benefactive interpretation. This is an example of Heine's switch context.
Heine's Stage IV (i.e. conventionalization) is exemplified by Kusaiean and Lenakel. In both Kusaiean and Lenakel, the new meaning of benefaction has been conventionalized to the extent that it can occur side-by-side with the original meaning in the same clause--without causing confusion or ambiguity. Thus in Kusaiean, as in (14), the possessive classifier occurs twice in the same clause, once encoding possession and once again encoding benefaction. The same situation has been reported for Saliba (Western Oceanic subgroup, Oceanic; Saliba Island, Milne Bay Province, Papua New Guinea) (Margetts 2002). A similar comment can be made of Lenakel (Lynch 1978: 72), although, as already been noted, this language may be located in a further advanced part of Stage IV than Kusaiean.
Finally, it needs to be borne in mind that the foregoing discussion is based on the available data. With more (detailed) data becoming available (e.g. Margetts 2002 on Saliba), some of the languages discussed above may turn out to be at an earlier or later stage of Heine's (2002) grammaticalization continuum.
4. Structural scope increase and grammaticalization
It is important to see what kind of constituent structure can be motivated for expressions such as (10) and (11) so that Tabor and Traugott's c-command scope-increase hypothesis can be tested over the grammaticalization of possessive classifiers as benefactive markers. (Bear in mind that  and  are being used below as representative examples; what is to be said about them will thus apply to other Oceanic languages, subtle differences between them notwithstanding.)
4.1. Constituent structure before and after grammaticalization
To the best of the author's knowledge, there is no generative--government and binding (GB) theory in particular--analysis of possessive classifier constructions in Oceanic languages (also Diane Massam and Liz Pearce, pers. comm.), although a small amount of GB research has been done on the possessive phrase in Maori (Waite 1994; Pearce 1997). (9) Since the possessive phrase is different from the possessive classifier construction--Maori has no possessive classifiers--the possessive D(eterminer) P(hrase) structure proposed by Waite (1994) or by Pearce (1997) is of little use for purposes of the DSC to be carried out in this article. (10) Fortunately, Harrison (1976: 129-130) provides a (relatively theory-neutral) constituent structure for the possessive classifier construction in Mokilese; he assumes the constituent structure in (21) for the possessive classifier construction in (10).
However, because Tabor and Traugott's c-command scope-increase hypothesis makes a specific reference to the structural concept embedded in GB theory and other similar syntactic theories, readers may insist that (21) be modified into more "compatible" form. For the sake of consistency or compatibility, therefore, it will be assumed here that the possessive classifier (PCL) heads the possessive classifier phrase (PCLP), and that the PCLP in turn is taken to be the specifier (Spec) of the head (i.e. the possessum) of the whole phrase (see Note 11 for justification). The "revised" constituent structure, which will be adopted in preference to (21), is represented in (22). (11)
Note, however, that, regardless of which of the two constituent structures is chosen, what really matters for purposes of the DSC will remain the same: the position of the PCLP inside the NP nime-n woall-o pill-o. As a matter of fact, the "revised" constituent structure in (22) does not differ much from Harrison's original constituent structure in (21)--the PCLP in particular. (12)
The D-structure of (10) may turn out to be very different from the S-structure in (22). For instance, the latter contains a mix of right- and left-headed phrases; at the level of D-structure, phrases may be consistently either right- or left-headed. This, beyond the scope of the present article, needs to be worked out elsewhere. Nonetheless, the S-structure in (22) will serve the present purpose, for the DSC method operates on S-structure, not D-structure (see Note 3).
The sentence in (11) is mapped onto the tree diagram in (23), with the tree diagram in (24) representing the "original" (that is, prior to the grammaticalization) constituent structure of (11). In the tree diagram in (23), the erstwhile possessor NP functions as a benefactive NP, with the possessive classifier marking its benefactive role. This "new" benefactive NP should be regarded as an adjunct, because in (11) nih-mw can be omitted without causing any ungrammaticality (Sheldon Harrison, pers. comm.). The erstwhile possessum NP kijinlikkoauoaw, on the other hand, continues to serve as the direct object NP (or complement) of the sentence in (11) (as it did prior to the grammaticalization).
In standard GB theory, a layered constituent structure is assumed for VP (Haegeman 1994: 90): the adjunct benefactive NP thus forms VP with V', which in turn consists of V and the direct object NP or complement. This layered VP structure happens to be almost identical to Harrison's (1976: 144) "predicate" structure, which is made up of two immediate constituents: "VP" (=V and the direct object NP) and an adjunct. Thus, the constituent structure provided here for (11)--somewhat abbreviated for the sake of simplicity, for example, the absence of I'--seems to be entirely appropriate for purposes of the DSC to be carried out.
4.2. The diachronic string comparison (DSC)
The sentence in (11) (repeated below), in conjunction with (23) and (24), can now be mapped onto the DSC in (25).
(11) Mokilese ngoah insingeh-di kijinlikkoauoaw nih-mw 1SG:SBJ write-ASP letter PCL-2SG:POSS 'I wrote a letter for you.' (25) Diachronic string comparison: possession to benefaction a. [IP ngoah [I'[VP[V'[V insingeh-di] [NP[NP kijinlikkoauoaw][PCLP nih-mw]] ]]]] b. [IP ngoah [I'[VP[V'[V insingeh-di][NP kijinlikkoauoaw]] [NP nih-mw] ]]]
The scope of the possessive classifier and the possessor together (or the PCLP) is examined (or underlined in [25a]), because the possessor also has been affected by the process of grammaticalization. In (25a), the possessive classifier and the possessor, nih-mw, are contained in the same NP as is the head kijinlikkoauoaw, because the former are the specifier of the latter. In (25b), on the other hand, the "possessive classifier" and "possessor" are related to the verb, because they together constitute the benefactive NP. They have "'moved up" in the constituent structure, as it were, as can be seen by comparing (25a) [=(24)] and (25b) [=(23)]. There is a clear sense that nih-mw in (25b) has scope over nih-mw in (25a): the position of nih-mw in (25b) asymmetrically c-commands the position of nih-mw in (25a) (cf. Tabor and Traugott 1998: 253). The change episode in question thus clearly involves scope increase, not decrease, thereby supporting Tabor and Traugott's (1998) c-command scope-increase hypothesis. This scope increase has resulted directly from the grammatico-functional (or semantic role) change of the possessor -mw and from the concomitant change of the possessive classifier nih-. To put it differently, what started as a specifier within the direct object NP has now assumed the status of an adjunct NP, on a par with V', which contains that direct object NP and the V. In view of these changes, structural scope decrease would in fact make no sense.
4.3. Scope increase and formal identity between source items and descendants
Most previous research on grammaticalization has taken the view that grammaticalization goes hand in hand with scope decrease. That is, the more grammaticalized a given formative becomes, the narrower its structural scope becomes. Indeed there is much supporting evidence in the literature. Thus Lehmann (1995 : 143-147), for instance, recognizes structural scope decrease as one of the six "parameters" manifested typically in instances of grammaticalization. This view--under the general assumption of "isomorphism between historical development and synchronic relations among polysemous items"--has also led to the tendency to accept structural scope decrease as "a criterion for deciding what change episodes come within the purview of 'grammaticalization studies'" (Tabor and Traugott 1998: 231, 263). Change episodes such as the ones discussed in Tabor and Traugott (1998) and in this article, however, call this status of scope decrease into question. Once the status of scope decrease is questioned like this, it is not unreasonable to ask whether or not scope change per se can be regarded as a reliable criterion for identifying instances of grammaticalization. Unfortunately, there is an insufficient amount of data to help answer this question. In point of fact, it is not possible to ascertain what the change episode highlighted in this article has in common with other known change episodes of scope increase, let alone to distinguish on independent grounds change episodes of scope increase from those of scope decrease (Tabor and Traugott 1998: 262).
Nonetheless, it can be said, albeit with caution, that in change episodes of scope increase including the one discussed in this article, the formal identity between source items and their grammaticalized descendants seems to be well maintained. In other words, what Lehmann (1995 : 126-127) refers to as "phonological integrity" seems to be kept more or less intact in change episodes of scope increase. In Oceanic languages, thus, grammaticalized benefactive markers have retained the same forms as their source possessive classifiers. None of the Oceanic languages discussed here and elsewhere (Song 1997, 1998) shows any signs of reduction in phonological integrity between possessive classifiers and their descendant benefactive markers--regardless of degrees of grammaticalization. Similarly, the change episodes of scope increase that Tabor and Traugott (1998) discuss--the development of the s-possessive, the VP-gerund, adverbial and conjunctive instead (of), and the discourse marker anyway--involve no or little change in the formal identity between the sources and their grammaticalized descendants. For instance, regardless of whether anyway appears as a manner (or clause-internal) adverb or as a topic-resuming discourse marker, exactly the same form anyway is used. The Present Day English possessive clitic -s does not much deviate formally from its Old English source, the genitive suffix -(e)s. Campbell (1991: 290-291) also describes a change episode of scope increase in Old Estonian. In Old Estonian, the Proto-Balto-Finnic interrogative suffix *-ko lost out to the question particle es. This question particle had been a clitic/suffix in earlier times. (The clitic/suffix was originally only -s but, subsequently, the final vowel of the word to which it was attached was reanalyzed as part of the clitic/suffix, e.g. kelte-s [right arrow] kelt-es "from whom?") With vowel harmony lost, the clitic/suffix was subsequently reinterpreted as an independent question particle. The phonological integrity of the segments in question remained intact throughout the development from the clitic/suffix to the independent particle, namely, -es [right arrow] es. So-called "insubordinated" clauses reanalyzed as independent in the Australian Aboriginal language Kayardild are "morphologically [or phonologically] identical to [...] subordinate clauses" with main clauses ellipsed (Evans 1993: 270). In other words, no change in phonological integrity is involved between subordinate and "insubordinated" clauses.
Change episodes of scope decrease, on the other hand, do seem to involve phonological deformations, that is, loss of the formal identity between source items and their grammaticalized descendants (e.g. Heine and Reh 1984; Lehmann 1995 : 126-132). For instance, take bound personal pronouns, which undergo formal reduction from independent personal pronouns as the former emerge from the latter (Lehmann 1995 : 39-42; e.g. Abkhaz sarh [right arrow] s- 'first person singular'). Also, as a relational noun changes into an adposition, an agglutinative affix and then a fusional case, the phonological shape of the relational noun undergoes reduction or attrition to the effect that the phonological integrity of the source item is undermined--possibly beyond recognition (Lehmann 1995 : 76-87, and also Campbell 2001b: 115; e.g. Estonian kaes [a (dependent) noun in the inessive] 'in the company' [right arrow] -ga [a case suffix]).
This indeed is an interesting observation worth investigating further, because it has also been observed that change episodes that move from left to right along the grammaticalization pathway in (1) or (2) do not only involve semantic bleaching (i.e. desemanticization) but also display the tendency for grammaticalized descendants to "show increasingly different phonological shapes from [...] the (functional or contentful) items from which they arose" (Janda 1995: 120). Janda (1995: 121) goes on to suggest that change episodes which maintain (or increase) the formal similarity between grammaticalized items and their sources, real or apparent, are occasionally able to resist or even reverse the direction of change embodied in the grammaticalization pathway. To put it differently, insofar as the formal similarity or identity between grammaticalized items and their sources remains intact, it is possible or even likely for change episodes to run counter to the predicted direction of the grammaticalization pathway. If scope change is a manifestation of grammaticalization (cf. Lehmann 1995 ), what this could mean is scope increase, rather than scope decrease. Intriguing as this may be, it remains to be answered why certain change episodes maintain the formal similarity between source items and their grammaticalized descendants in the first place and also why such change episodes are able to reverse the direction of change (cf. Joseph and Janda 1994).
5. The status of grammaticalization (theory) revisited
Grammaticalization and grammaticalization theory have recently come under a great deal of criticism. The severest criticism that has so far appeared in print comes from the papers collected in Campbell (2001a): Campbell (2001b), Janda (2001), Joseph (2001), and Newmeyer (2001) in particular. The main thrust of this criticism is that grammaticalization cannot be more than an epiphenomenon of well-understood mechanisms of language change such as reanalysis, extension, and borrowing--the most dominant one being reanalysis--and that these mechanisms are commonly attested also in language changes that have nothing to do with grammaticalization (Campbell 2001b: 141; also Harris and Campbell 1995). Being reduced to the mechanisms of language change, grammaticalization is not an independent phenomenon but merely a derivative concept (Campbell 2001b; Janda 2001; Joseph 2001; and Newmeyer 2001) or even a cover term for the mechanisms of language change (Newmeyer 2001). If, as claimed by these critics, grammaticalization does not have an independent status of its own, there cannot does not need to--be grammaticalization theory because there will be nothing left to explain within such a theory; a general theory of language change which subsumes the mechanisms of language change will suffice. Furthermore, grammaticalization theory, even if it is "allowed" to exist, will be unable to explain instances of grammaticalization without appeal to these mechanisms of language change. This is in stark contrast with the programmaticalization position held by Haspelmath (1998), who goes so far as to argue that "the large majority of syntactic changes are instances of grammaticalization and should be explained within the framework of a theory of grammaticalization, WITHOUT reference to reanalysis [emphasis added]" (Haspelmath 1998: 315) (but see Heine et al. 1991: 219; and Hopper and Traugott 1993: 48-56 on the role of reanalysis in grammaticalization).
In view of this controversy, it behooves the author to defend the independent status of grammaticalization and also the need for grammaticalization theory, since the development of possessive classifiers into benefactive markers has been described here as an instance of grammaticalization in the context of grammaticalization theory. At the same time, the role of the general mechanisms of language change--reanalysis in particular--in the domain of grammaticalization should be duly recognized, because the change episode highlighted in the present paper can and should be accounted for WITH reference to reanalysis. The change in hierarchical structure between (25a) and (25b) (that is, the constituent structure in  reanalyzed as that in ) is transparent enough, and concomitant with this change in hierarchical structure is a change in grammatical categories, that is, the reanalysis of the possessive classifier prefix as a benefactive prefix and consequently of PCLP as an adjunct NP. It will thus be argued here that the two opposing sides need to take a step back from their positions and then (re-)consider what their opponents have said and perhaps also what they have themselves claimed--about grammaticalization and grammaticalization theory. There are other issues concerning grammaticalization (theory) discussed by the opposing sides, but this section will concentrate on the three main issues of disagreement referred to above. (In what follows, references to the anti-grammaticalization view come mainly from Campbell [2001b], because his critical comments are more or less reiterated albeit independently--in Janda , Joseph , and Newmeyer .)
5.1. The anti-grammaticalization view: grammaticalization as an epiphenomenon
As has been noted, the opponents of grammaticalization (theory) argue that grammaticalization is not an independent phenomenon, simply because instances of grammaticalization can ultimately be explained in terms of the familiar mechanisms of language change. Therefore, grammaticalization is merely an epiphenomenon, a derivative concept, or a cover term at best. At the same time, however, Campbell (2001b: 152153) admits that there are constraints on what can and cannot grammaticalize, and that "we do need to recognize and explain these 'constraints' on what can grammaticalize and what the result of grammaticalization can be." This means that, in spite of Campbell and the other critics' claim to the contrary, grammaticalization does have an independent status of its own, because one cannot have constraints on something that does not exist as an independent phenomenon. Why would one bother to discover constraints on an epiphenomenon when that epiphenomenon is--by definition--"derivative" of an independent phenomenon? To draw an analogy, in order to understand the occurrence of fever (that is, why and when people have a fever), one does not study fever itself but its cause. There is no such a thing as "constraints" on the occurrence of fever, because fever is merely an epiphenomenon of some specific disease. One instead aims to study this disease and discover "constraints" on its occurrence. Thus, if there are constraints on what can and cannot grammaticalize, then grammaticalization cannot possibly be an epiphenomenon. There should instead be such a thing as grammaticalization as an independent phenomenon.
Moreover, if grammaticalization is an epiphenomenon to be explained with reference to the mechanisms of language change, the constraints on grammaticalization should fall out directly from the constraints on these mechanisms of language change. But clearly they do not. It is not possible to explain with reference to the mechanisms of language change what can and cannot grammaticalize. Campbell and the other opponents of grammaticalization have not demonstrated that it is possible to do so. This is because language changes that have to do with grammaticalization are indeed different from language changes that have nothing to do with grammaticalization. Thus, instances of grammaticalization constitute a distinct category of language changes that needs to be studied in its own right. In other words, grammaticalization has an independent status of its own. (13)
Although he accepts the need to recognize and explain the constraints on grammaticalization, Campbell (2001b) hastens to add that: (14)
... it is not at all clear that the explanation of these patterns and constrains [on grammaticalization] ... would most naturally reside in 'grammaticalization theory.' It seems more likely that the explanation of these facts will more naturally be found in the domain of semantic change, reanalysis, and grammatical structure in general. (Campbell 2001 b: 153)
Clearly, the implication here is that there is no need for grammaticalization theory. But this seems to be an extraordinary claim in view of Campbell's recognition of the existence of the constraints on grammaticalization. If the question as to why some, and not other, language changes give rise to grammaticalization (i.e. the constraints on grammaticalization) could indeed be answered with reference to semantic change, reanalysis, grammatical structure in general, or whatever, there would indeed be no need for grammaticalization theory, as Campbell claims. But the fact of the matter is that that is not possible. This alone calls for some kind of theory, even if that theory draws upon semantic change, reanalysis, grammatical structure, and what not. This is because grammaticalization is a distinct category of language changes, and the constraints on grammaticalization in turn should be discovered and explained within a theory that recognizes this, not within a theory that denies it.
It needs to be made clear at this juncture what is meant by grammaticalization theory. Needless to say, it is not clear what the opponents mean by grammaticalization theory, because they do not believe in it. Thus it is necessary to turn to proponents of grammaticalization in order to find that out. By grammaticalization theory, Hopper and Traugott (1993: 1) mean a framework of grammaticalization "that focuses on how grammatical forms and constructions arise, how they are used, and how they shape the language." Similarly, Heine et al. (1991: 23) make it clear that they are concerned with providing "a new framework for understanding grammaticalization." (Indeed the title of their book reads: Grammaticalization: A Conceptual Framework.) In other words, grammaticalization theory is a conceptual framework in which language changes with grammaticalization and their constraints can be identified, described and (hopefully) explained. As has been argued here, grammaticalization is an independent phenomenon, not least because there are constraints on grammaticalization. There is nothing wrong with having a framework in which such an independent phenomenon can be recognized, studied, and taken account of.
The basis for rejecting grammaticalization theory is slightly shifted, however, when Campbell (2001b: 152) further claims that, although it "is true [that there are constraints on what can and cannot grammaticalize,] ... much hinges on the form in which the claims of predictability are stated." Thus he argues that grammaticalization theory even if it exists--cannot explain why a single lexical source can develop in multiple directions and why a single grammatical "target" can come from multiple lexical sources. Because of this inadequacy, Campbell (2001b: 153) continues, at least "strong claims for the predictive power of grammaticalization are clearly exaggerated; lacking predictability, grammaticalization theory thus lacks explanatory power." Regardless of whether it is an accurate description of grammaticalization theory, this comment should not be interpreted to detract from the independent status of grammaticalization, to be explained in the proper domain of grammaticalization theory. Campbell (2001b) repeatedly attempts to discredit the need for grammaticalization theory in terms of predictability, lack of explanatory power, etc., not on its own terms. Nevertheless, he never denies the existence of constraints on grammaticalization and also the need to explain them. This is a peculiar position to hold.
Campbell (2001b: 151-152) also wonders whether his claim that grammaticalization can be explained in terms of the mechanisms of language change is merely an example of reductionism, and whether grammaticalization is an explanation at a less abstract level. He draws an analogy like this. If someone asks a question "why did the window break?," a number of answers are possible. One reasonable answer could be: "because Mary slammed it." But it is also possible to give a very "technical" answer--no matter how unreasonable it may sound to the man in the street:
"Shock waves [i.e. caused by Mary's slamming it] together with the molecular characteristics of the glass caused the window to break." Campbell likens the first answer to the explanation based on grammaticalization and the second to the explanation based on the mechanisms of language change. But he rejects this analogy only to conclude--somewhat expectedly--that grammaticalization cannot lay claim to explanatory power even at its own level, because "almost unanimously proponents of grammaticalization themselves appeal to reanalysis and other changes as the explanatory mechanisms upon which cases of grammaticalization depend" (Campbell 2001b: 152). But again, the fact that some, not other, changes give rise to grammaticalization and constraints on what can and cannot grammaticalize exist does call for the need for grammaticalization theory. To extend Campbell's analogy, if the window breaks when Mary slams it ([congruent to] language change with grammaticalization), but if it never breaks when Linda slams it ([congruent to] language change without grammaticalization), some theory of Mary's and Linda's window-slamming "styles" ([congruent to] grammaticalization theory) must explain this, and that theory is different from the theory of shock waves and the molecular characteristics of windows ([congruent to] reanalysis, extension, etc.). They are not merely more or less abstract levels of explanation.
5.2. The pro-gramrnaticalization view. grammaticalization as the mechanism of change
The strongest argument in support of grammaticalization and grammaticalization theory comes from Haspelmath (1998, also 2002). Thus it comes as no surprise that the opponents level their strongest criticism at his position. Haspelmath (1998), as has already been noted, claims that there is no room for reanalysis in grammaticalization. Haspelmath's position departs from that held by most proponents of grammaticalization. Hopper and Traugott (1993: 32, 50), for instance, state that "[u]nquestionably, reanalysis is the most important mechanism of grammaticalization" and go so far as to claim that "[i]t is best, then, to regard grammaticalization as a subset of changes involved in reanalysis, rather than to identify the two." Heine et al. (1991: 217) point out that "typically, reanalysis accompanies grammaticalization ... or, conversely, when reanalysis takes place, this is likely to involve the grammaticalization of at least one morpheme within the structure undergoing reanalysis"--despite their (Heine et al. 1991: 217) view that the two must be strictly kept apart. Thus, broadly speaking, the position held by these proponents of grammaticalization is not very different from that advanced by the opponents of grammaticalization: reanalysis is the primary mechanism of grammaticalization. The controversy over the status of grammaticalization has arisen partly because Haspelmath's (1998) position differs from this commonly held position. In Haspelmath's view, grammaticalization itself is THE mechanism of instances of grammaticalization, as he (Haspelmath 1998: 315, 344) declares the "[t]he main mechanism of syntactic change is grammaticalization, i.e. the gradual unidirectional change that turns lexical items into grammatical items and loose structures into tight structures, subjecting frequent linguistic units to more and more grammatical restrictions and reducing their autonomy." Instances of grammaticalization should thus be explained WITHOUT reference to reanalysis (Haspelmath 1998: 315). (15)
Independently of Campbell's (2001b: 148) reservations about Haspelmath's "redefinition" of reanalysis, the latter's claim that grammaticalization can be both a phenomenon to be explained and the main mechanism with which to explain that phenomenon is far from convincing. (16) In fact, this seems to be a conceptual flaw in Haspelmath's position. Grammaticalization represents a coherent category of language changes in which linguistic expressions shift from a lexical to a grammatical status or from a less grammatical to a more grammatical status. There is no disagreement with Haspelmath over this point. But just as language changes without grammaticalization are explained with reference to such well-understood mechanisms of language change as reanalysis, extension, etc., instances of grammaticalization should also be explained by means of the general mechanisms of language change, wherever and whenever possible. This is indeed the view also held by most proponents of grammaticalization. Haspelmath's attempt to regard grammaticalization as the main explanatory mechanism of grammaticalization itself is precisely the point that the opponents of grammaticalization take issue with. Other proponents of grammaticalization also will agree with the opponents of grammaticalization on this. Hopper and Traugott (1993: 48-50), for instance, do not regard grammaticalization as a mechanism of language changes with grammaticalization. In point of fact, they (Hopper and Traugott 1993:1-2) restrict the use of the term "grammaticalization" to the framework of grammaticalization and to "the actual phenomena of language that the framework of grammaticalization seeks to address [i.e. instances of grammaticalization]." Grammaticalization is truly an independent phenomenon of language change, distinct from other instances of language change that have nothing to do with grammaticalization. That does not mean, however, that it can also be used to explain itself. To explain instances of grammaticalization, one needs to step out of grammaticalization, as it were, and appeal to the general mechanisms of language change. Grammaticalization IS a type of language change, after all. What still remains to be explained after that--or in Camp-bell's (2001b: 152) words, "what can and cannot grammaticalize"--should then be dealt with in the proper domain of grammaticalization theory.
Moreover, Haspelmath's (1998) position on the nonexistence of reanalysis in grammaticalization is untenable, because, as Campbell (2001b) also points out, his definition of reanalysis is too narrow or restrictive. This can be demonstrated by looking at one and the same change episode that is analyzed totally differently by Haspelmath and by Campbell: the development of a verb into a preposition or a complementizer in Ewe, based on Heine and Reh (1984:37-38). Haspelmath (1998: 328) believes that this change does not involve reanalysis, because "the hierarchical structure of the sentences [in (26), for example] does not change at all--it is only the category labels that change: a verb turns into a preposition or a complementizer (and of course the corresponding category VP automatically changes to PP or CP [-S'], respectively)."
The development in question is clearly an instance of grammaticalization in that a lexical element (i.e. a verb) has changed into a grammatical element (i.e. a preposition). In Haspelmath's view, a change in grammatical categories alone is not reanalysis but merely grammaticalization. Campbell (2001b: 146), on the other hand, argues that a change in grammatical categories does involve reanalysis. While it is true that the hierarchical structures in Heine and Reh's original work--regardless of their accuracy (but see below)--have not at all changed in the development, a "reanalysis" of a verb as a preposition (or as a complementizer for that matter) has indeed taken place.
Since there is a striking difference of opinion on what constitutes reanalysis, it is worth going back to Langacker's (1977: 59) definition of reanalysis, which both Campbell and Haspelmath endorse: "change in the structure of an expression or class of expressions that does not involve any immediate or intrinsic modification of its surface manifestation." The crucial word here is "structure." Does "structure" include grammatical categories in addition to hierarchical structure as in the case of Campbell's interpretation, or not, as in the case of Haspelmath's? This is a the-oretical as well as empirical question. It is a theoretical question because some syntactic theories may take hierarchical structure and grammatical categories to be conceptually related to each other. In fact, so much so that if hierarchical structure is to be regarded as "structure," then so are grammatical categories (and vice versa). In X-bar theory, for instance, phrases are projected directly from lexical heads (i.e. N, V, P, and A). Complements combine with the lexical head to form X'-projections, and adjuncts combine with X' to form further X'-projections. The specifier then combines with the topmost X' to form the maximal projection XP, be it NP, VP, PP, or AP. In other words, phrases are HIERARCHICALLY STRUCTURED projections of their lexical heads (Haegeman 1994: 146). In syntactic theories like this, grammatical categories (or lexical heads in X-bar theory) and hierarchical (constituent) structure are so inextricably intertwined that it is not possible to regard one, and not the other, as structure. (This, of course, does not mean that grammatical categories and consitutent structure are not distinct!)
Even if this point is to be disregarded, Haspelmath's argument based on the hierarchical structure remaining intact even after the change in grammatical categories is, to say the least, a very shaky one. The hierarchical structures in Heine and Reh's original analysis indeed remain unchanged after the verb changed into a preposition; since there is no change in the hierarchical structures, there is no reanalysis (Haspelmath 1998: 328). However, depending on how hierarchical structure is motivated and drawn, a change in grammatical categories may co-occur with a change in hierarchical structure. In other words, the "reanalysis of category labels" can also be seen to give rise to a change in hierarchical structure. Before the change, the phrase na Kofi in (26) is one of the two conjuncts of the coordinate structure (i.e. I poured palmwine and gave [it to] Kofi)--although this is not transparent from Heine and Reh's hierarchical structure in (26a). As the verb na changed into a preposition, the same phrase na Kofi changed into an adjunct PP, something that is omissible (i.e. I poured palmwine for Kofi vs. I poured palmwine). In X-bar theory (or other syntactic theories for that matter), this difference will be reflected in terms of where in the hierarchical structure the phrase in question will be positioned. Thus na Kofi, as a conjunct of the coordinate structure, will be placed under the lowest V', as it contains the verb na and the object NP or complement Kofi. The same phrase as an adjunct PP, on the other hand, will combine with a V' to form a higher V' projection (or the VP as the case may be). In other words, there will be a difference of at least one V'-projection between the conjunct V' na kofi and the adjunct PP na Kofi. If this is not regarded as a change in hierarchical structure the difference of one V' projection is hierarchical indeed--then one does not know what is. To base one's argument on the hierarchical structures provided by Heine and Reh (1984)--presumably in a most theory-neutral manner or for the sake of simple exposition--is injudicious and, in fact, untenable. Whether "structure" includes grammatical categories in addition to hierarchical structure is also an empirical question because whether the phrase na Kofi is analyzed as a conjunct V' or as an adjunct PP in the first place depends not only on one's syntactic theory, but also inevitably on relevant syntactic tests (i.e. data).
In fact, Haspelmath is well aware of his definition of reanalysis being too narrow or restrictive, as he (Haspelmath 1998: 330) himself admits: "I have no disagreement with these authors [e.g. Hopper and Traugott 1993], but I would prefer to reserve the term [i.e. reanalysis] for the narrower concept." Contrary to this narrow interpretation, changes in grammatical categories do come within the purview of reanalysis, because they cannot be dissociated from changes in hierarchical structure. Much of the controversy surrounding grammaticalization, as Campbell (2001b: 148) rightly points out, arises partly from Haspelmath's narrow interpretation of reanalysis. But, more importantly, Haspelmath also makes an attempt to refer to grammaticalization as THE explanatory mechanism of instances of grammaticalization. This is bound to cause problems, because even on the most intuitive level it seems to lead nowhere. It is analogous to claiming that people's mood changes (e.g. from good [[congruent to] less grammatical] to bad [[congruent to] more grammatical]) are the main mechanism of people's mood changes: The doctor's mood has changed from good to bad as his mood has changed from good to bad. No one will seriously accept this as the explanatory mechanism of the doctor's mood change. Similarly, it is difficult to accept grammaticalization as the explanatory mechanism of instances of grammaticalization: X has shifted from a less to a more grammatical status as it has shifted from a less to a more grammatical status.
The opponents of grammaticalization (theory) should recognize the independent status of grammaticalization because they, like the proponents of grammaticalization, recognize constraints on what can and cannot grammaticalize. They should also accept the need for grammaticalization theory as a conceptual framework that does not deny but recognizes the independent status of grammaticalization so that instances of grammaticalization and their constraints can be properly identified, described, and explained. The extreme programmaticalization position maintained by Haspelmath (1998), on the other hand, suffers from one conceptual flaw: grammaticalization itself is taken to be the main mechanism of instances of grammaticalization. Grammaticalization is a type of language change. Thus it comes as no surprise that such language changes also can be explained with reference to the general mechanisms of language change, for example, reanalysis, as the development of possessive classifiers into benefactive markers in Oceanic languages was described at the beginning of this section, and as, in fact, most proponents of grammaticalization believe that they should. Moreover, Haspelmath's definition of reanalysis is unacceptably narrow and is not well-founded. His unequivocal refusal to accept a change of category labels as an example of reanalysis, for instance, is found to be unjustifiable. Indeed his hard line on the nonexistence of reanalysis in grammaticalization will first need to be reconciled with that held by other proponents of grammaticalization.
In this article, possessive classifiers have been characterized as being exploited for benefactive marking in Oceanic languages, and arguments have been brought to bear in demonstrating that this is an instance of grammaticalization. Then, by drawing on Tabor and Traugott's (1998) DSC method and also their c-command-based definition of scope change, the change episode in question has been demonstrated to involve structural scope increase, not decrease, thereby adding to the increasing amount of evidence in support of the c-command scope-increase hypothesis. However, there is much evidence also in support of the opposite of this hypothesis, namely, scope decrease. Moreover, there are no independent grounds on which to distinguish change episodes of scope increase from change episodes of scope decrease, although special note has been taken of the strong connection between scope increase, and the formal identity between source items and their grammaticalized descendants (and, conversely, the connection between scope decrease and the lack of the formal identity). In view of this state of affairs, it cannot but be concluded, in common with Tabor and Traugott (1998: 231), that the jury is still out on the status of scope change as a criterion for identifying instances of grammaticalization. In the meantime, what needs to--and can--be done is to document a sufficiently wide range of change episodes of scope increase or decrease, and to investigate them carefully and systematically. This is what the present article has done. Moreover, in view of the current controversy over the status of grammaticalization and the need for grammaticalization theory, an evaluation of the two opposing positions has been carried out with the conclusion that, contrary to the opponents of grammaticalization, grammaticalization has an independent status of its own, justifying the need for grammaticalization theory, and also that, contrary to the extreme programmaticalization position, the general mechanisms of language change (e.g. reanalysis), not grammaticalization itself, should be invoked in order to account for instances of grammaticalization.
University of Otago
Received 3 July 2003
Revised version received
18 November 2003
Abney, Steven (1987). The English noun phrase in its sentential aspect. Unpublished doctoral dissertation, MIT.
Aikhenvald, Alexandra (2000). Classifiers: A Typology of Noun Categorization Devices. Oxford: Oxford University Press.
Allan, Keith (19773. Classifiers. Language 53, 285-311.
Bybee, Joan L.; and Pagliuca, William (1985). Cross-linguistic comparison and the development of grammatical meaning. In Historical Semantics: Historical Word Formation, Jacek Fisiak (ed.), 59-83. Berlin and New York: Mouton de Gruyter.
--; and Pagliuca, William (1986). The evolution of future meaning. In Papers from the Seventh International Conference on Historical Linguistics, Anna Giacalone Ramat et al. (eds.), 108-122. Amsterdam: John Benjamins.
--; Perkins, Revere; and Pagliuca, William (1994). The Evolution of Grammar. Tense, Aspect and Modality in the Languages of the World. Chicago: University of Chicago Press.
Campbell, Lyle (1991). Some grammaticalization changes in Estonian and their implications. In Approaches to Grammaticalization, Elizabeth Traugott and Bernd Heine (eds.), 285-299. Amsterdam: John Benjamins.
--(ed.) (2001a). Grammaticalization: A Critical Assessment. Language Sciences 23 (Special Issue).
--(2001 b). What's wrong with grammaticalization? Language Sciences 23, 113-161.
--; and Janda, Richard (2001). Introduction: conceptions of grammaticalization and their problems. Language Sciences 23, 93-112.
Croft, William (1985). Indirect object "lowering". Berkeley Linguistics Society 11, 39-51.
--(1991). Syntactic Categories and Grammatical Relations. The Cognitive Organization of Information. Chicago: University of Chicago Press.
Davies, Mark (1994). Parameters, passives, and parsing: motivating diachronic and synchronic variation in Spanish and Portuguese. Chicago Linguistics Society 30, 46-60.
Diewald, Gabriele (2002). A model for relevant types of contexts in grammaticalization. In New Reflections on Grammaticalization, Ilse Wischer and Gabriele Diewald (eds.), 103-120. Amsterdam: John Benjamins.
Evans, Nicholas (19883. Odd topic marking in Kayardild. In Complex Sentences in Australian Languages, Peter Austin (ed.), 219-266. Amsterdam: John Benjamins.
--(1993). Code, inference, placedness and ellipsis. In The Role of Theory in Language Description, William A. Foley (ed.), 243-280. Berlin and New York: Mouton de Gruyter.
--; and Wilkins, David (2000). In the mind's ear: the semantic extensions of perception verbs in Australian languages. Language 76, 546-592.
Givon, Talmy (1971). Historical syntax and historical morphology: an archaeologist's field trip. Chicago Linguistic Society 7, 394-415.
--(1976). Topic, pronoun and grammatical agreement. In Subject and Topic, Charles Li (ed.), 149-189. New York: Academic Press.
--(1979). On Understanding Grammar. New York: Academic Press.
Guy, Jaques B. M. (1974). A Grammar of the Northern Dialect of Sakao. Canberra: Australian National University.
Haegeman, Liliane (1994). Introduction to Government and Binding Theory. Oxford: Blackwell.
Harris, Alice; and Campbell, Lyle (1995). Historical Syntax in Cross-linguistic Perspective. Cambridge: Cambridge University Press.
Harrison, Sheldon P. (1976). Mokilese Reference Grammar. Honolulu: University Press of Hawaii.
--(1988). A plausible history of Micronesian possessive classifiers. Oceanic Linguistics 27, 63-78.
Haspelmath, Martin (1998). Does grammaticalization need reanalysis? Studies in Language 22, 315-351.
--(1999). Why is grammaticalization irreversible? Linguistics 37(6), 1043-1068.
--(2002). On directionality in language change: with particular reference to unidirectionality in grammaticalization. Paper presented at New Reflections on Grammaticalization 2 University of Amsterdam, 4-6 April 2002.
Heine, Bernd (1997a). Cognitive Foundations of Grammar. New York and Oxford: Oxford University Press.
--(1997b). Possesion: Cognitive Sources, Forces, and Grammaticalization. Cambridge: Cambridge University Press.
--(2002). On the role of context in grammaticalization. In New Reflections on Grammaticalization, Ilse Wischer and Gabriele Diewald (eds.), 83-101. Amsterdam: John Benjamins.
--; Claudi, Ulrike: and Hunnemeyer, Friederike (1991). Grammaticalization: A Conceptual Framework. Chicago: University of Chicago Press.
--; and Kuteva, Tania (2002). World Lexicon of Grammaticalization. Cambridge: Cambridge University Press.
--; and Reh, Mechthild (1984). Grammaticalization and Reanalysis in African Languages. Hamburg: Helmut Buske.
Hopper, Paul (1991). On some principles of grammaticization. In Approaches to Grammaticalization, Vol. 1: Focus on Theoretical and Methodological Issues, Elizabeth C. Traugott and Bernd Heine (eds.), 17-35. Amsterdam: John Benjamins.
--; and Traugott, Elizabeth C. (1993). Grammaticalization. Cambridge: Cambridge University Press.
Janda, Richard D. (1980). On the decline of declensional systems: the overall loss of OE nominal case inflections and the ME reanalysis of -es as his. In Papers from the 4th International Conference on Historical Linguistics, Elizabeth C. Traugott, Rebecca Labrum, and Susan Shepherd (eds.), 243-252. Amsterdam: John Benjamins.
--(1995). From agreement affix to subject "clitic"--and bound root: -mos > -nos vs. (-)nos(-) and nos-otros in New Mexican and other regional Spanish dialects. Chicago Linguistic Society 31, 118-139.
--(1996). The alleged unidirectionality of grammaticalization vs. the discontinuity of language transmission. Paper presented at the LSA, San Diego.
--(2001). Beyond "pathways" and "unidirectionality": on the discontinuity of language transmission and the counterability of grammaticalization. Language Sciences 23, 265-340.
Jeffers, Robert J.; and Zwicky, Arnold (1980). The evolution of clitics. In Papers from the 4th International Conference on Historical Linguistics, Elizabeth C. Traugott, Rebecca Labrum, and Susan Shepherd (eds.), 221-231. Amsterdam: John Benjamins.
Joseph, Brian D. (2001). Is there such a thing as "grammaticalization"? Language Sciences 23, 163-186.
--; and Janda, Richard D. (1994). The how and why of diachronic morphologization and demorphologization. In Theoretical Morphology: Approaches in Modern Linguistics, Michael Hammond and Michael Noonan (eds.), 193-210. San Diego: Academic Press.
Kim, Hyeree (2001). Remarks on the unidirectionality principle in grammaticalization. Folia Linguistica Historica XXII, 49-65.
Kurylowciz, Jerzy (1965). The evolution of grammatical categories. Diogenes 51, 55-71.
Langacker, Ronald W. (1977). Syntactic reanalysis. In Mechanisms of Syntactic Change, Charles Li (ed.), 57-139. Austin: University of Texas Press.
Lee, Kee-Dong (1975). Kusaiean Reference Grammar. Honolulu: University Press of Hawaii. Lehmann, Christian (1995 ). Thoughts on Grammaticalization. Munich: Lincom Europa.
Lichtenberk, Frantisek (1983). Relational classifiers. Lingua 60, 147-176.
--(1985). Possessive constructions in Oceanic languages and in Proto-Oceanic. In Austronesian Linguistics at the 15th Pacific Science Congress, Andrew Pawley and Lois Carrington (eds.), 93-140. Canberra: Australian National University.
--(2002). The possessive-benefactive connection. Oceanic Linguistics 41, 439-474.
Lucy, John A. (2000). Systems of nominal classification: a concluding discussion. In Systems of Nominal Classification, Gunter Senft (ed.), 326-341. Cambridge: Cambridge University Press.
Lynch, John (1973). Verbal aspects of possession in Melanesian languages. Oceanic Linguistics 12, 69-102.
--(1978). A Grammar of Lenakel. Canberra: Australian National University.
--(1996). Proto Oceanic possessive marking. In Oceanic Studies: Proceedings of the First International Conference on Oceanic Linguistics, John Lynch and Fa'afo Pat (eds.), 93-110. Canberra: Australian National University.
--(1998). Pacific Languages. An Introduction. Honolulu: University of Hawaii Press.
Margetts, Anna (1999). Valence and Transitivity in Saliha: An Oceanic Language of Papua New Guinea. Nijmegen: Max Planck Institute for Psycholinguistics.
--(2002). The linguistic encoding of three-participant events in Saliba. Studies in Language 26, 613-636.
--(2004). From implicature to construction: emergence of a benefactive construction in Oceanic. Oceanic Linguistics 43, 445-468.
Matsumoto, Yo (1988). From bound grammatical markers to free discourse markers: history of some Japanese connectives. Berkeley Linguistics Society 14, 340-351.
Menn, Lise; and MacWhinney, Brian (1984). The repeated morph constraint: toward an explanation. Language 60, 519-541.
Mosel, Ulrike (1984). Tolai Syntax and Its Historical Development. Canberra: Australian National University.
Newmeyer, Frederick J. (2001). Deconstructing grammaticalization. Language Sciences 23, 187-229.
Ouhalla, Jamal (1994). Transformational Grammar. Formal Rules to Principles and Parameters. London: Edward Arnold.
Pawley, Andrew (1973). Some problems in Proto-Oceanic grammar. Oceanic Linguistics 12, 103-188.
--; and Ross, Malcolm (1995). The prehistory of the Oceanic languages: a current view. In The Austronesians, Peter Bellwood, James J. Fox, and Darrell Tryon (eds.), 39-74. Canberra: Department of Anthropology, Australian National University.
Pearce, Elizabeth (1997). Genitive case in the Maori DP. Wellington Working Papers in Linguistics 9, 31-55.
Reinhart, Tanya (1981). Definite NP anaphora and c-command domains. Linguistic Inquiry 12, 605-635.
Roberts, Ian G. (1993). A formal account of grammaticalization in the history of Romance futures. Folia Linguistica Historica XIII, 219-258.
Ross, Malcolm D. (1988). Proto Oceanic and the Austronesian Languages of Western Melanesia. Canberra: Australian National University.
Senft, Gunter (2000a). Systems of Nominal Classification. Cambridge: Cambridge University Press.
--(2000b). What do we really know about nominal classification systems? In Systems of Nominal Classification, Gunter Senft (ed.), 11-49. Cambridge: Cambridge University Press.
Song, Jae Jung (1994). The verb-object bonding principle and the pronominal system: with special reference to Nuclear Micronesian languages. Oceanic Linguistics 33, 517-565.
--(1997). The history of Micronesian possessive classifiers and benefactive marking in Oceanic languages. Oceanic Linguistics 36, 29-64.
--(1998). Benefactive marking in Oceanic languages: from possessive classifiers to benefactive markers. In Case, Typology and Grammar: In Honor of Barry J. Blake, Atom Siewierska and Jae Jung Song (eds.), 247-275. Amsterdam: John Benjamins.
Sugita, Hiroshi (1973). Semitransitive verbs and object incorporation in Micronesian languages. Oceanic Linguistics 12, 393-406.
Tabor, Whitney; and Traugott, Elizabeth C. (1998). Structural scope expansion and grammaticalization. In The Limits of Grammaticalization, Anna Giacalone Ramat and Paul J. Hopper (eds.), 229-272. Amsterdam: John Benjamins.
Traugott, Elizabeth C.; and Heine, Bernd (eds.) (1991a). Approaches to Grammaticalization. Amsterdam: John Benjamins.
--; and Heine, Bernd (1991b). Introduction. In Approaches to Grammaticalization, Elizabeth C. Traugott and Bernd Heine (eds.), 1-14. Amsterdam: John Benjamins.
Waite, Jeffrey (1994). Determiner phrases in Maori. Te Reo 37, 55-70.
Zwicky, Arnold M. (1987). Supressing the Zs. Journal of Linguistics 23, 122-148.
* I am grateful to Shelly Harrison, John Lynch, Diane Massam, Liz Pearce, and Elizabeth Traugott for answering questions about the languages or theories that they specialize in, and to Ali Knott for procuring me obscure references. I benefited greatly from discussions with Anna Margens, Frank Lichtenberk, and participants in the Workshop on Benefactive Marking in Oceanic Languages and Languages of Eastern Nusantara at the 9th International Conference on Austronesian Linguistics (held at the Australian National University, 8-11 January 2002). I wish to record here my indebtedness to the Department of Linguistics, the Max Planck Institute for Evolutionary Anthropology (Leipzig, Germany), where I developed this article during the summer of 2002, and also to the Department of Linguistics, University of Canterbury (Christ-church, New Zealand), where I subsequently revised it. I am grateful to Bernard Comrie, Lyle Campbell, and Martin Haspelmath not only for their hospitality, but also for their provocative comments on earlier drafts of this article and for taking the time to discuss with me some of the issues dealt with in this article. This should not be taken to imply that they necessarily agree with everything that I have said in this article, and that I have addressed all their questions. Last but not least, l am grateful also to two anonymous Linguistics referees for their useful comments and suggestions. Needless to say, I alone bear full responsibilities for claims and conclusions made here. Abbreviations used in glosses: ASP: Aspect: CONSTR: Constructive suffix: CONT: Continuative: CP: Complementizer Phrase: D(ET): Determiner; DL: Dual; DP: Determiner Phrase; EXC: Exclusive; GEN: Genitive: INF: Infinitive; IP: Inflection Phrase; LOC: Locative; NOM: Nominative: NSG: Nonsingular; PCL: Possessive Classifier; PCLP: Possessive Classifier Phrase: PERF: Perfective; PL: Plural; POc: Proto-Oceanic: POSS: Possessive; PST: Past; SBJ: Subject: SG: Singular: Spec: Specifier. Correspondence address: Linguistics Programme, School of Language, Literature and Performing Arts, University of Otago, P.O. Box 56, Dunedin, New Zealand. E-mail: jaejung. email@example.com.
(1.) In addition to structural scope, Lehmann (1995 : 121 160) identifies five other "parameters" of grammaticalization: integrity, paradigmaticity, paradigmatic variability, bondedness, and syntagmatic variability. For instance, as a linguistic expression or formative changes over time along the cline in (1) or (2), it gradually loses its integrity, that is, phonological substance (i.e. phonological attrition) and semantic content (i.e. desemanticization).
(2.) The notion of c-command is a standard structural notion in Government and Binding Theory, but, for the benefit of readers who are unfamiliar with it, this notion can be illustrated with reference to the constituent structure in (i) below.
In (i), node A c-commands node B. The application of the definition of c-command works like this (based on Haegeman 1994:132 134): starting from node A, one moves upwards till one reaches the first branching node (i.e. node X in [i]); then one moves downwards following every branch of the tree, and every node that one finds on the way down is c-commanded by the starting node (i.e. node A), regardless of whether one moves to the right or to the left Node B, on the other hand, does not c-command node A, because the first branching node that dominates node B (i.e. node Y in [i]) does not dominate node A.
(3.) Elizabeth Traugott (pers. comm.) informs the author that reanalysis at the level of D-structure is probably not possible, because language change is change in use, not in grammars. This, however, is an empirical issue that awaits further investigation (cf. Roberts  for grammaticalization as involving the eliminating of the D-structure to S-structure movement in favor of base-generation in the S-structure position).
(4.) While Tabor and Traugott (1998: 234) acknowledge that there is still a question as to how the history of the English modal system should be analyzed, they also point out that all historical researchers agree on one thing: the modern modals occupy an auxiliary position which dominates the verb phrase, whereas the Old English ancestors of modal verbs were like main verbs, which were presumably generated in the verb phrase.
(5.) Direct suffixation of possessive pronouns to possessed nouns is exemplified in (i), taken from Mokilese (Micronesian subgroup, Oceanic: Mokil Atoll and Ponape) (Harrison 1976: 113).
(i) Mokilese mijoa-ioa face- 1SG:POSS 'my face'
Note that, when the possessor is a noun instead of a pronoun, a so-called construct suffix must also be used in (i), as in the case of possessive classifiers (see below).
(6) Gilbertese does have what Harrison (1988: 67) calls a "possessive article." This article is used to encode alienable possession, but this is qualitatively different from possessive classifiers that are attested in other Micronesian languages (Harrison 1988: 68).
(7) Pawley (1973: 163-164) only tentatively reconstructs *ma- for Proto-Oceanic (POc), whereas Lichtenberk (1985:118-119) is confident enough to do so (also see Lynch 1996, who shares Lichtenberk's confidence). Ross (1988: 185), however, believes "that POc in fact had a somewhat larger collection of ... [possessive] classifiers [than has so far been reconstructed], and that it is the most frequently used which have survived."
(8) This cannot be an example of entailment, as can be seen in (i).
(i) The man built the woman's house; he didn't build it for her, but for his own brother, who later sold it to her.
Entailment is not cancellable or defeasible like this. Thus, the meaning of benefaction can be only pragmatically inferenced or generated.
(9.) GB theory has been chosen here as representative of formal syntactic theories that make use of constituent structure and the notion of c-command. Thus, what follows can be applied to any of such theories. Moreover, it is also the theory that Tabor and Traugott (1998) assume in formulating their c-command scope-increase hypothesis.
(10.) Maori makes a distinction between so-called O-possessive and A-possessive marking (i.e. o- vs. a- genitives). The former is used to encode subordinate possession, while the latter is used to relate to dominant possession. This system, however, is qualitatively different from the possessive classifier system under discussion.
(11.) The constituent structure in (22) can be justified as follows. The PCLP is used to "specify" or "fix" the reference of the possessum, as it were; the former is taken to be the specifier of the latter. The possessum is the head of the whole phrase, because it can potentially appear on its own without the PCLP--in which case it will no longer encode the possessum--but the PCLP cannot occur alone without the possessum unless the latter is contextually understood (Harrison 1976: 131). The motivation for grouping the PCL and the possessor NP together under the PCLP in (22) is that an agreement relationship exists between them, as indicated by the use of the construct suffix -n. Moreover, it is not possible to have one without the other; the PCL and the possessor NP must co-occur with each other. Equally importantly, when the possessor is in pronominal form (e.g. first or second person possessors), it must be added to the end of the PCL. In other words, the PCL and the possessor can potentially turn up on the surface as a single word. The PCL and the possessor NP are taken to be the head and the complement of the PCLP, respectively. The motivation for this is that the relationship between the PCL and the possessor NP seems to parallel that found between the V (or head) and the direct object NP (or complement) in sentences. The pronominal possessor, as has already been noted, must appear as a suffix added to the PCL. This is reminiscent of direct object pronominal affixes being added to the verb in many languages (see Song 1994 for further discussion of this phenomenon in Micronesian and crosslinguistic perspectives). Assuming that case is assigned at S-structure (Haegeman 1994: 340), the advantage here is that the possessor NP can be seen to receive its case (genitive) from the PCL, just as the direct object NP receives its case (accusative) from the V. Evidence for the possessor NP receiving its case from the PCL, not from the possessum, is that the possessum does not have to be expressed when its identity is contextually understood or known. In cases like this, it will be difficult to see from where the possessor NP gets its case if the "absentee" possessum is taken to be its case-assigner. This problem is obviated if the case-assigner for the possessor NP is the PCL; the former must always co-occur with the latter (and vice versa). Thus, regardless of whether the possessum is expressed on the surface or not, it is ensured that the possessor NP receive its case from the PCL.
There is also an alternative generative proposal which reinterprets NP as DP, in which the head of NP is not N but rather D (i.e. the so-called DP hypothesis). As has already been noted. Waite (1994) and Pearce (1997) propose such a DP analysis for the possessive phrase in Maori on the basis of Abney (1987). If such a DP analysis is extended to it, the constituent structure of the possessive classifier construction in Oceanic languages will no doubt look different from the one in (22). However, since Tabor and Traugott (1998) assume NP, rather than DP in their DSC, this article will do likewise, if only for the sake of consistency. More importantly, regardless of which of these two analyses is taken to be the basis for the DSC in question, the fact does remain that both the PCE and the possessor NP, when expressing possession, are structurally contained within the NP nime-n woall-o pill-o. In point of fact, this is what really matters for purposes of the DSC in question.
(12.) In the revised tree diagram in (22), the possessum noun pill- and the determiner -o are placed under one and the same NP node. In Harrison's original tree diagram, on the other hand, they, together with the POSS NP, constitute the whole phrasal constituent. The motivation for this modification comes from the fact that the determiner is suffixed to the possessum noun (i.e. a single word).
(13.) Bernard Comrie (pers. comm.) points out that the claim that grammaticalization is an important notion is potentially independent of whether grammaticalization is an independent phenomenon (i.e. a primitive notion) or an epiphenomenon (i.e. a derived notion). The author agrees with him that independence and importance are indeed "logically completely distinct notions in scientific discourse." Bernard Comrie (pers. comm.) further argues that there can potentially be constraints on a derived notion also. If this is the case, the existence of constraints on grammaticalization in itself will not necessarily prove the independent status of grammaticalization. While acknowledging that this is an important philosophical question (which is, needless to say, beyond the scope of this article), the author disputes it in the present context for the very reasons given in the main text. Moreover, if it is accepted that there can be constraints on a derived notion, it is not clear what will rule out the logical possibility that constraints on X (a derived notion) completely contradict constraints on Y (a primitive notion). Under such circumstances, it will be impossible to maintain that X is "derivative" of Y.
(14.) Lyle Campbell (pers. comm.) informs the author that it is possible to explain the constraints on grammaticalization by appealing to semantic change. For semantic reasons, X can grammaticalize into Y, but A cannot grammaticalize into B. Surprisingly, though, Campbell (2001b: 119) himself points out that "[t]here are several kinds of phenomena considered to be instances of grammaticalization which do not necessarily require any shift in meaning [i.e. semantic change] at all [...]."
(15.) Haspelmath (1998:315) also acknowledges that a minority of syntactic changes are due to reanalysis, but he argues that, with grammaticalization and reanalysis constituting disjoint classes of phenomena, these changes must be explained in different terms.
(16.) Campbell has a valid point here, because Haspelmath's (1998: 330) definition of reanalysis differs from that adopted by Hopper and Traugott (1993), for instance (also see below). So Campbell (2001: 148) continues: "[...] Haspelmath confesses that it is he who is out of step with nearly all others working in grammaticalization and in syntactic change generally, and they do not accept his definition of reanalysis."
|Printer friendly Cite/link Email Feedback|
|Author:||Song, Jae Jung|
|Publication:||Linguistics: an interdisciplinary journal of the language sciences|
|Date:||Jul 1, 2005|
|Previous Article:||Adjectival modification in Mandarin Chinese and related issues *.|
|Next Article:||Aspects of the prosody of Kuot, a language where intonation ignores stress (1).|