Grouping in Multiple Attribution: Advantage Albanian Frans Plank (Konstanz) 1. Ambiguous grouping In noun phrases with more than one nominal attribute there is a potential for structural ambiguity. As is illustrated in (1), with the help of brackets, the second attribute may belong either with the first attribute (1a) or with the head of that first attribute, or rather with the constituent consisting of the head plus the first attribute (1b). (1) the Academy of Sciences of Albania a. b. [the Academy [of Sciences [of Albania]]] [the Academy [of Sciences]] [of Albania] In this particular example, the reading that corresponds to the bracketing in (1a) may seem less natural than that corresponding to (1b), involving an academy devoted to such sciences as are peculiar to Albania as opposed to an academy run by Albania devoted to the sciences. However, in other instances of multiple attribution, such as (2), readings that correspond to hierarchical structures like (1a), where each successive attribute’s head is the adjacent nominal, are clearly preferred. (2) [the source [of the prosperity [of the peoples [of the Socialist countries]]]] Indeed, overall, head-recursion, as in (1a)/(2), would seem to be more common than head-sharing, as in (1b). My aim here is to survey the principal ways and means available to languages to overtly distinguish such alternative groupings in attributive constructions. Essentially, what tends to be overtly marked in constructions is what relationship a part bears to other parts or to the whole (i.e., being a head or an attribute) and/or which parts are related to each other (cf. Plank 1995). Although neither relationship-identification nor relatedness-indication is inherently designed to take care specifically of grouping, the latter type of marking is far better up to this extra job than the former. But relatedness-indication too is reluctant to be exploited for grouping beyond the first level of attribution. 2. Dependent-marking without agreement 2.1. One possible policy is not to distinguish alternative groupings in attributive constructions at all, or to rely on means other than the grammar of words and inflections, namely on intonation and pauses and perhaps other hints of prosodic phrasing. Languages such as English or Latin which encode attributes by means of grammatical forms identifying an NP as bearing this particular relationship, i.e. by attributive cases or adpositions, may not have many other choices for suggesting alternative groupings. 2.2. However, provided an attribute’s position relative to its head is variable, there may be a choice, as it were coming for free, namely to utilize constituent order to disambiguate such attributive constructions. Thus, it is possible in English to unequivocally represent the hierarchical relations of (1b) by moving one attribute in front of its head, with the two attributes thus flanking their shared head on both sides: (1') b. Albania’s Academy of Sciences What ends up following the head, marked by the preposition of, is its immediate attribute; the outer attribute, marked by the genitive enclitic ’s, precedes. In the same spirit though somewhat different in the letter, in order to make clear that NPs form a monotonous hierarchy of attributes, with each NP an immediate attribute to the next, Classical Greek would successively embed genitival attributes between article and noun, putting up with potentially complex, hard-to-process self-embedding structures (3a) with items of the same word class (article, noun) immediately juxtaposed; the alternative, suggesting a non-monotonous sequence, would be to extrapose an attribute to the right of the head, introduced by a resumptive article (3b) (Goodwin 1894: 208-210): (3) a. taŸ th`ı tw`n pollw`n yuch`ı o[mmata the.nom/acc.pl.neut the.gen.sg.fem the.gen.pl.masc manymascgen.pl soulfemgen.sg eyeneutnom/acc.pl ‘the eyes of the soul of the multitude’ b. taŸ th`ı yuch`ı o[mmata taŸ tw`n pollw`n the.nom/acc.pl.neut the.gen.sg.fem soulfemgen.sg eyeneutnom/acc.pl the.nom/acc.pl.neut the.gen.pl.masc manymascgen.pl ‘the soul-eyes of the multitude’ However, there may be independent considerations precluding such disambiguation by alternative ordering and perhaps concomitant coding distinctions. Thus, in Old English and its early Germanic relatives, (2) would most likely have been arranged as in (2'): (2') the prosperity’s source of the peoples of the Socialist countries If an attribute was complex, and in particular if it had nominal attributes of its own, it tended to be split up anyhow, with one part preceding and the other following the head. Jacob Grimm’s (1843) collection of such examples included bold splits such as this from late Middle High German, where the first attribute is lacking an overt nominal head: (4) in des hant von Riuwental into that.gen.sg.masc hand.acc.sg of Riuwental ‘into the hand of the one [i.e., the Lord] of Riuwental’ In split constructions like (2') and (4) the ordering of NP, thus, gives no clue as to the grouping of heads and attributes — or indeed a wrong clue. 2.3. Yet another measure to obviate grouping ambiguity is to alternatively use morphological and syntactic constructions for attributes. (1"), those constituents that belong together most closely appear as compounds. (1") a. b. Thus, in the Academy of Albania-Sciences the Albania-Academy of Sciences, the Science-Academy of Albania Derivational morphology might also be exploited for this purpose, with immediate attributes constructed as derived adjectives: (1"') a. b. the Academy of Albanian Sciences the Albanian Academy of Sciences, the Scientific Academy of Albania 2.4. Often the semantics of the NPs being combined will of course suffice to tip the scale — in favour of head-sharing, as in (5a), or headrecursion, as in (5b): (5) a. b. the calculation of the income tax of the Inland Revenue the calculation of the income tax of millionaires Failing semantics, more specific marking of the semantic relationships of the respective attributes can also serve to effectuate disambiguations between head-recursion or head-sharing. Thus, while a standard attributive case or adposition (such as of in English) might do with single attributes, regardless of their distinct semantic relations (6a/b), more revealing markers or constructions would come handy when these same attributes are combined (6c/d): (6) a. b. c. c'. d. d'. the calculation of the income tax the calculation of civil servants [the calculation [of the income tax]] [of civil servants] the calculation of the income tax (made) by civil servants [the calculation [of the income tax [of civil servants]]] the calculation of the income tax (levied) on civil servants An agentive or patientive marker on the second attribute, perhaps in combination with an attributive form (such as a participle) of a suitable verb, would automatically force head-sharing or head-recursive readings when topmost heads are action nominalizations. 2.5. Still, although they are potentially capable of enlisting the support of intonational, sequencing, word-forming, and relational-semantic devices if really pressed hard, languages which merely encode the attributehood of nominals, by means of genitives or functionally equivalent kinds of dependent-marking, are not very good at systematically grouping constituents in NPs with more than one attribute, and their speakers presumably often neglect to do so. 3. Head-marking, without or with agreement 3.1. If headhood were encoded as well, a difference in grouping such as that between head-recursive (1a) and head-sharing (1b) could potentially be overtly recognizable. In (1a), ‘Sciences’, carrying attributive marking (attr), would in addition have to receive a marker identifying it as a head, while in (1b) such head-marking would have to be omitted, because ‘Sciences’ is here only an attribute, the head of ‘Albania’ being ‘Academy’ or ‘Academy of Sciences’. Schematically: (7) a. b. the Academy-head Sciences-attr Albania-attr the Academy-head Sciences-attr-head the Albania-attr Although there are languages that combine head and attribute marking, for instance construct state or possessed form and genitive or dative case respectively, as in Pitta-Pitta (Australian, (8)) and Arbore (Cushitic, (9)) (see further Plank 1995: 42), I am unaware of languages that use it in this manner for purposes of grouping. (8) kupakupa-Na t2it2i-wara old.man-dat brother-possessed ‘the old man’s brother’ (9) hikic-i hóggattu-t axe-construct laborer-gen ‘the laborer’s axe’ Ironically, what comes closest to the pattern of dual-function marking in (7b) is the omission of attributive marking if an attribute is itself a head of a further attribute, as attested in Middle High German and occasionally also later. Nouns could shed their genitive case and appear in their bare (nominative) form when they governed a genitive preceding or also following them (Behaghel 1923: 162, 165-167): (10) a. der welte lon betrachtung the.gen.sg.fem worldfem.gen.sg reward.nom.sg consideration.nom.sg ‘the consideration of the reward of the world’ (instead of lon-es gen.sg) b. ein wissage Cristes tot a.nom.sg prophecy.nom.sg Christ.gen.sg death.nom.sg ‘a prophecy of Christ’s death’ (instead of tot-es gen.sg) c. das Brod des Wort Gottes the.nom.sg bread.nom.sg the.gen.sg word.nom.sg God.gen.sg ‘the bread of the word of God’ (instead of Wort-es gen.sg; is in the genitive, though) this attribute’s article The dual status of an NP as head and attribute, and the concomitant head-recursion reading, is thus encoded by the non-use of dependent-marking on attributes. More common, on the other hand, are overt markers of heads which simultaneously relate a head to its attribute by virtue of agreement, with attributes perhaps in turn identified as such by markers of their own. And such head-marking can conceivably be exploited for purposes of grouping. 3.2. Consider Colloquial German, where attributes, obligatorily preceding their heads, are in the dative and heads agree with them in person, number, and (though only in the singular) gender by means of possessive pronouns. The presence of a possessive pronoun at the same time identifies the following nominal as a head. Head-recursive (1a) is accordingly translated as (11a), with agreement indicated in subscript and attribute and head inflection in parentheses, and with Albania replaced by a (former) country with a case-inflecting definite article. (11) a. der DDR ihren Wissenschaften ihre Akademie ‘the GDR3sg.fem (dat.sg) its3sg.fem Sciences3pl (dat.pl) their3pl Academy (nom.sg)’ This encoding strategy is unviable if the head already has a possessive pronoun independent of attribute-agreement; e.g. meine Lösung dieses Rätsels ‘my solution of this riddle’ — *diesem Rätsel seine meine Lösung ‘this riddle (dat) its my solution’. Even without such added complications, (11a) is a veritable challenge, even though it could hardly be more monotonously regular, with each NP as the immediate attribute of that NP which it precedes: processing tends to get derailed at the second dative in a row. Since (1b), in a sense, differs from (1a) in that ‘Sciences’ is not a head, all that apparently needs to be done to (11a) to account for this is to omit the possessive pronoun that accompanies ‘Sciences’ owing to its headhood, along the lines of the contrast between (7b) and (7a): (11) b. der DDR den Wissenschaften ihre Akademie ‘the GDR (dat.sg) the Sciences3pl (dat.pl) their3pl Academy (nom.sg)’ However, I have yet to find a speaker of Colloquial German able to make sense of such sequences of NPs. And in fact, upon reflection, (11b) turns out to lack crucial structural information: what is unexpressed is that ‘the GDR’ is an attribute to ‘Academy’ or ‘Academy of Sciences’. The possessive pronoun ihre in (11b) actually happens to be homonymous between 3rd person plural and 3rd person singular feminine; but this does not seem to enable it simultaneously to cross-reference both den Wissenschaften and der DDR. To take care of this second attributive relation, an additional possessive pronoun is needed, agreeing in person, number, and gender with the attribute der DDR and preceding the head or rather the head and its other attribute: (11') b. der DDR ihre den Wissenschaften ihre Akademie ‘the GDR3sg.fem (dat.sg) its3sg.fem the Sciences3pl (dat.pl) their3pl Academy (nom.sg)’ In the abstract, the intended grouping is achieved in this manner. However, I have yet to hear a real speaker of Colloquial German going to such lengths in order to distinguish groupings such as (1a) and (1b). Systematic though (11'b) is, it is virtually unprocessable. 3.3. Other languages using a similar encoding strategy for attributive constructions, however, are far less reluctant to group than is Colloquial German. One of them is Turkish (Lewis 1967: 45-47). Compare (12a) and (12b). (12) a. Ford-un aile-si-nin araba-sı Ford-gen family-3sg-gen car-3sg ‘the car of the family of Ford’ b. Ford-un aile araba-sı Ford-gen family car-3sg ‘the family-car of Ford’ A head is identified by its person-number suffix that agrees with the attribute, which in turn is in the genitive (in definite izafet) or in the absolute form (in indefinite izafet). Lacking this person-number suffix, aile in (12b) cannot be a head of Ford’un. (Its indefiniteness accounts for the absence of a genitive suffix.) Since ‘car’ in (12b) has two attributes, this NP is actually incomplete, as is (11b) vis-à-vis (11'b) in Colloquial German; but, interestingly, the more complete version with two person-number suffixes on the head is as ungrammatical as its Colloquial German counterpart: (12') b. *Ford-un aile araba-sı-sı And this ungrammaticality is apparently not due to morphological haplology; (13) two person-number suffixes seem to be disallowed even if not identical: *Ford-lar-ın aile araba-sı-ları Ford-pl-gen family car-3sg-3pl ‘the family-car of the Fords’ What is presently unclear to me is which person-number suffix is discarded, that of the attribute sequentially closer to (-sı) or more distant from (-ları) the head shared by the two attributes. 3.4. To summarize, whenever heads are marked so as to agree with attributes it is in principle viable to group constituents in multiple attribution, owing to the possibility of marking a nominal for both attributehood and headhood (as in (12a)) or for attributehood only (as in (12b), where attribute status is actually marked by position rather than by a genitive). However, not all languages of this kind put their resources to such use, and instead unjudiciously require attributes followed by other attributes to be head-marked even if they do not have this status (as in (11)). What seems generally problematic here is to mark nominals for double headhood, as theoretically required by the presence of two non-coordinate attributes, as in (11'b), (12'b), and (13). 4. Agreeing dependents 4.1. Regardless of whether heads are overtly identified as such, the marking of nominal attributes so as to agree with their heads automatically ensures overt grouping, requiring no extra measures to distinguish between monotonously recursive attribution, as in (1a), and head-sharing deviations from this pattern, as in (1b). 4.2. Thus, in Albanian (Buchholz & Fiedler 1987: 418 passim) nominal attributes, in addition to being in the genitive or ablative case and following their heads, are obligatorily accompanied by a proclitic article (or rather, connecting particle) agreeing in case, number, gender, and definiteness with the head. (Declension classes as distinguished for nouns in the singular, indicated by Roman numerals below, although interrelated with gender, are not involved in this agreement.) Here are head-recursive (1a) and head-sharing (1b) in Albanian: (14) a. Akademi-a e Shkenca-ve të Shqipëri-së academyIII.fem-nom.sg.def ART.nom.sg.def.fem science-gen.pl.def ART.gen.pl.def AlbaniaIII.fem-gen.sg.fem.def b. Akademi-a e Shkenca-ve e Shqipëri-së academyIII.fem-nom.sg.def ART.nom.sg.def.fem science-gen.pl.def ART.nom.sg.def.fem AlbaniaIII.fem-gen.sg.fem.def In both versions, the two attributes are invariably in the genitive. The article on the second, however, takes different form, depending on the case, number, gender, and definiteness of the corresponding head: të, as in (14a), identifies the head as being genitive plural definite, thus grouping Shqipëri-së with Shkenca-ve; while e, as in (14b), identifies the head as being nominative singular definite feminine, thus grouping Shqipëri-së with Akademi-a. There is in fact massive homonymy in the article paradigm, but since article choice is determined by no less than four categories, normally heads can be uniquely identified. As heads remain unmarked in this strategy, no difficulties comparable to those encountered by Turkish (12'b, 13) arise from multiple headhood. However, there could be a problem in distinguishing multiple attributes at different hierarchical levels, as in (1b), from asyndetic coordinate or appositive ones, likewise linking up to the same head. Albanian solves this problem admirably, by dropping the article from all but the first of coordinate or appositive attributes, as seen in (15), contrasting with (14b) in this respect. (15) në shërbim-Ø të socializm-it, [të] liri-së dhe [të] pavarësi-së in serviceIV.neut-acc.sg.indef ART.acc.sg.indef.neut socialismI.masc-gen.sg.def, freedomIII.fem-gen.sg.def and independenceIII.fem-gen.sg.def ‘in the service of socialism, freedom, and independence’ 4.3. The basic principle of attributive encoding in languages practising Suffixaufnahme — as primarily found in the Ancient Near East (HurroUrartean, perhaps Anatolian Indo-European), in India (commonly in Indo-Aryan, at least once in Dravidian), in the Caucasus (Kartvel, Tsezic), in Siberia (Chukchi, Evenki), in Ethiopia (Cushitic), and in Aboriginal Australia (Plank 1990, (ed.) 1995) — is identical to that of Albanian (or also Classical Greek, where the repetition of the definite article with nominal attributes, as seen in (3b), is not obligatory, though): in basic attributive constructions, nominal attributes, in addition to receiving some attributive case, agree with their heads in case and perhaps number and further inflections, actually copying these categories from their heads. Thus, grouping is conveniently achieved by simply copying case etc. from the appropriate head onto the respective attribute. For instance, in an Old Georgian example such as (16), the final nominative suffix of the first attribute, following upon a plural oblique suffix identifying this noun as a plural attribute, links up this attribute with a nominative head. (Definite articles, abbreviated as ART, are postnominal.) (16) sasxdomel-eb-i igi ms9idel-ta-j ma-t 7red-isa-ta-j stall-pl-nom ART.nom seller-pl.obl-nom ART-pl.obl pigeon-gen-pl.obl-nom ‘the stalls of the sellers of pigeons’ The second attribute from top, being in the genitive case and not number-marked, is linked up doubly: first with its immediate head, shown to be plural oblique by the first suffix after genitive, then, somewhat redundantly, with the head of its own head, shown to be nominative by the final suffix. However, what seems to be the rule in recursive attributions in languages of this kind are reduced patterns of marking (Plank 1990, 1995: 64-69). One such pattern is to omit the Suffixaufnahme linkage from the first attribute from top, as was a common option in Old Georgian ((17); see further Boeder 1995: 174-179, 181-182); others are to link up second and further attributes only with their immediate heads, as optionally in Hurrian ((18), where Suffixaufnahme of the head’s directional case is also omitted from the first attribute from top; Wilhelm 1995: 128-129, Wegner 1995: 142-143), or to link them only to topmost heads, as in Dyirbal, where double genitives are impermissible ((19); Dixon 1972: 106). (17) 6li7e-n-i sasupevel-isa ca-ta-jsa-n-i key-pl-nom kingdom-gen heaven-pl.obl-gen-pl-nom ‘(the) keys of the kingdom of (the) heavens’ (18) sén(a)-iffu-u8e-né-vá-d-an ast(i)-î-ve ni0ár(i)-ída brother-1sg.poss-gen-def.sg-gen-1sg.abs.pro-connective wife-3sg.poss-gen dowry-3sg.poss-dir ‘to the dowry of the wife of my brother’ (19) Nay-gu-d3in-du yabu-Nu-ˆ d3in-gu baNun guda-Ngu 1sg-gen-ligature-erg mother-gen-lig-erg classII (erg) dog-erg ‘my mother’s dog’ Still, in reduced Suffixaufnahme patterns like (17), the genitive suffix on the second attribute from top (‘heavens’), following upon its own plural oblique suffix, appropriately links this attribute to its immediate genitival head (‘kingdom’), while its final plural and nominative suffixes in turn relate this genitival head plus its attribute to the plural nominative head (‘keys’) in accordance with the head-recursive structure of this NP. All that would need to be altered in order to link up the second attribute with the topmost nominal rather than with its adjacent attribute as head, is to omit the appropriate case(-number) copies, and perhaps to add case(-number) copies connecting the first attribute, now without an attribute of its own that would provide the link-up, with its head. Although the intended head-sharing reading may be somewhat unnatural, (17') differs from (17) in this manner. (17') 6li7e-n-i sasupevel-isa(-n-i) ca-ta-n-i key-pl-nom kingdom-gen(-pl-nom) heaven-pl.obl-pl-nom ‘heaven’s keys of the kingdom’ This is a constructed example, but (20) is real, and it ostensibly makes the same point: (20) sul-i sulneleb-isa-j sa6umevel-ta-j whiff-nom fragrance-gen-nom incense-pl.obl-nom ‘a whiff of fragrance of incense’ The second attribute, ‘incense’, is not linked up with the first, ‘fragrance’, lacking a further genitive between its own attributive marking (plural oblique) and the nominative linking up with the topmost head. However, although a head-sharing reading would make sense (‘a fragrant whiff of incense’), what is going on here is not the judicious exploitation of Suffixaufnahme to distinguish head-sharing and head-recursive groupings, but a reduction of morphological complexity irrespective of syntax (Boeder 1995: 174-184). Though in principle permitting full Suffixaufnahme, Old Georgian was prone to simplify sequences of case(-number) markers by measures such as the omission of inner cases or (indeed obligatory) haplology of two plural obliques in a row (-ta-ta- > -ta-). The even further reduced pattern illustrated from Hurrian in (18) is unambiguous as to its head-recursive grouping, insofar as the second attribute is linked to the first via its genitive. However, in the reduced pattern illustrated from Dyirbal in (19) overt marking as such is potentially misleading, because on the face of it it suggests head-sharing: the second attribute, a possessive pronoun referring to the speaker, lacks a genitive that would link it the first attribute, ‘mother’, but it has an ergative, ultimately in agreement with the topmost head. But since the first attribute itself carries an ergative suffix too, indicating that it is to be related to a head in that case (‘dog’), the ergative on the second attribute could conceivably also be construed as directly relating to the ergative copy on the attribute rather than to its source on the head. At any rate, semantics will here take care of the head-recursive sense even without elaborate marking. Where it does not, unfortunately nothing could be done about it, since Australian morphology, more resolutely than that of Old Georgian, only caters to the more moderate everyday demands of syntax. Beyond the first level of recursion, ‘morphological sequence constraints’ (Dench & Evans 1988: 35-43) strike and reduce, replace, or adjust too complex inflectional marking, careless of its possible usefulness for syntactic grouping. 5. Moral The upshot of this brief survey is that the ease with which constituents are grouped in multiple attribution first and foremost depends on the means utilized in a language to encode attributive constructions. What is on universal offer is relationship-identifying or relatedness-indicating marking, on heads and/or dependents; what is best for grouping, morphology permitting, is relatedness-indication on dependents. Considering the rationales of the different encoding strategies this is not in itself very surprising. What is intriguing, though, is that we have here another instance of the expressive power of languages, or at any rate of the effortlessness with which it is wielded, being determined by their morphosyntactic machinery. It may afford those languages some consolation which have not gone for relatedness-indicating dependent-marking that grouping in multiple attribution is not the most pressing need for grammars. (Nor, in all honesty, is it known to bear on worldviews or temperaments.) In itself, such grouping expediency is hardly a decisive influence on the choice among the ways and means available for relational coding. But then, when choices have been made, slighting supplementary benefit would be foolish. Correspondence address: Fachbereich Sprachwissenschaft, Universität Konstanz, 78467 Konstanz, Germany; e-mail: frans.plank@uni-konstanz.de References Behaghel, Otto 1923. Deutsche Syntax: Eine geschichtliche Darstellung. Band 1: Die Wortklassen und Wortformen. A. Nomen. Pronomen. Heidelberg: Winter. Boeder, Winfried 1995. “Suffixaufnahme in Kartvelian”, in Plank (ed.), 151-215. Buchholz, Oda and Wilfried Fiedler 1987. Albanische Grammatik. Leipzig: VEB Verlag Enzyklopädie. Dench, Alan and Nicholas Evans 1988. “Multiple case-marking in Australian languages’, Australian Journal of Linguistics 8: 1-47. Dixon, R. M. W. 1972. The Dyirbal Language of North Queensland. Cambridge: Cambridge University Press. Goodwin, William W. 1894. A Greek Grammar. (2nd edn.) London: Macmillan. Grimm, Jacob 1843. “Zur syntax der eigennamen”, Zeitschrift für deutsches Alterthum 3: 134-139. (Reprinted in Grimm’s Kleinere Schriften, vol. 7: Recensionen und vermischte Aufsätze, 4. Teil, 130-138. Berlin: Dümmler, 1884.) Lewis, G. L. 1967. Turkish Grammar. Oxford: Clarendon Press. Plank, Frans 1990. “Suffix copying as a mirror-image phenomenon”, Linguistics 28: 1039-1045. Plank, Frans 1995. “(Re-)Introducing Suffixaufnahme”, in Plank (ed.), 3-110. Plank, Frans (ed.) 1995. Double Case: Agreement by Suffixaufnahme. New York: Oxford University Press. Wegner, Ilse 1995. “Suffixaufnahme in Hurrian: Normal cases and special cases”, in Plank (ed.), 136-147. Wilhelm, Gernot 1995. “Suffixaufnahme in Hurrian and Urartian”, in Plank (ed.), 113-135.