Grammaticalising by Growing Syntactic Structure: The History of North Germanic Nominal Morphosyntax
Grammaticalization is a well-attested historical mechanism whereby originally independent lexical items lose part or all of their lexical content and come to express a grammatical function or meaning (Meillet 1912). This process typically takes place in a particular syntactic environment (Traugott 2003). For example when combined with an infinitive, the Swedish verb komma 'come' develops into a marker of futurity (Hilpert 2009: ch 3) and loses its sense of physical displacement: e.g. datoriseringen kommer att påverka arbetsinnehållet 'computerisation will [lit. comes to] influence the content of our work'. In other grammatical environments the original meaning may be preserved: e.g. han kom hem sent 'he came home late'. Given the need to refer to the constructional context, is hardly surprising that a number of scholars have sought to deploy Construction Grammar (CG) to model these changes (Hilpert 2009, Hoffmann & Trousdale 2011). An alternative, construction-free, model is proposed within the Minimalist Program (MP) by Roberts & Roussou (2003) and van Gelderen (2011), in which functional projections for tense and other grammatical categories are assumed as part of UG. In the present paper we suggest that LFG by contrast offers a way of characterising morphosyntactic change in which not every emerging pattern is interepreted in constructional terms and where a universal hierarchy of functional categories is not taken as an a priori given. To this end, we consider here another type of grammaticalisation, also affecting a unit bigger than the word, but which requires a different approach to morphosyntax from than that assumed in either CG or the MP, and which allows for the possibility of structure or configurationality to emerge over time (Kiparsky 1995, Vincent 1997). Three aspects of the architecture of LFG make it particularly suited to analysing this phenomenon: the fact that non-configurational patterns are not constrained to fit into a universally binary-branching endocentric phrase-structure (Austin & Bresnan 1996, Nordlinger 1998); the fact that overt evidence is required to establish the existence of projecting functional categories (Economy of Expression, Bresnan 2001); and the fact that different types of linguistic information can be represented in separate parallel dimensions of structure.
We illustrate our argument through an analysis of the stages by which some modern North Germanic languages have developed a system of marking definiteness in which an item, sometimes called an 'end-article', attaches to an unmodified noun but a modified noun takes a freestanding definite article, as in Danish huset 'the house' vs det gamle hus 'the old house'. By contrast in Swedish the modified item has both the end article and the freestanding one: det gamla huset. We will argue that there is evidence in the development from Old Norse to Faroese that the noun phrase developed from a flat NP structure, with order determined largely by information-structural constraints, to a DP, where definiteness is associated with a particular position and order is determined by syntactic constraints. The evolution of the North Germanic DP thus parallels the account offered by Kiparsky (1995) for the emergence of the CP in Germanic from the flatter clause-structure of Indo-European.
In Old Norse, definiteness could be expressed as a bound morpheme on the noun (hestrinn 'horse-DEF) or the adjective or as a syntactic element (hinir auðgu 'the rich') while word order inside the NP was relatively free. The syntactic definiteness marker was not obligatory (thus hestr var allvænligr 'the horse was particularly beautiful'), but was restricted to environments where the noun is modified by an adjective or where a weak adjective is substantivised. If the noun was flanked on either side by adjectives, the syntactic marker associated with the adjective is often repeated (hina beztu menn ok hina vitrustu 'DEF best men and DEF wisest'). Possessive determiners and adjectives followed the noun in information structurally unmarked noun phrases (dóttir mín hefði dreymt 'daughter my had dreamt'); however, if the possessive relation was emphasized or contrasted, the possessive preceded the noun (þessi er þin dóttir eigi mín 'this is your daughter not mine'). On the basis of this and other data, we conclude that the noun phrase had a flat structure, with one information structurally privileged position before the noun as (1). In proposing this analysis we assume that the semantics of reference is accounted for independently and that a D-projection is not required for there to be (in)definite reference (compare the distinction drawn by Chierchia 1998 between NP and DP languages).
(1) Old Norse NP / \ / \ Top/Foc N' / | \ ... N ... [ PRED '...' DEF + ]
In older forms of Faroese, the noun became the morphological locus of definiteness marking, irrespective of the presence or absence of adjectival modification. Ordering was no longer used for information structural purposes, but each constituent of the phrase was associated with a strict syntactic position. We take this as evidence that the noun phrase was developing more structure, in other words was becoming more configurational.
In modern Faroese, there is both a syntactic and a morphological definiteness marker. Only the morphological one can occur as the sole exponent of definiteness, and when it is the sole exponent, the definite noun occurs on the left edge of the noun phrase. When material precedes the noun, a syntactic marker is required, which then occurs on the left edge. The data from modern Faroese suggest that for referential noun phrases, (in)definiteness has to be marked on the left edge, but it can be marked morphologically on the noun or syntactically, hence definiteness is associated with a particular position. Furthermore, the order between other elements cannot be varied for information structural purposes. This is evidence of a functional D category having developed and an articulated DP structure having replaced the flat structure where information structure determined the order, viz:
(2) Modern Faroese DP | D / \ / \ D NP / \ AP N [ PRED '...' DEF + ]
By comparison with (1), the c-structure in (2) has become more articulated, the f-structure has remained constant and there is no longer a privileged position to encode i-structure. Meanwhile, co-occurrence restrictions on m-structure and c-structure must be invoked to account for the different distribution of the definiteness marker on modified and unmodified words.
In summary, we argue that the historical stages modelled here capture all the effects of the different early and modern North Germanic nominal syntax, but at no stage is it necessary, or enlightening, to reify the concept 'construction'. Special constructions such as the English way-construction (Asudeh, Dalrymple & Toivonen 2008) or the Swedish directed motion construction (Toivonen 2002) may still emerge but in our view they do not form the model on which all syntactic patterns are built. Neither a CG-style approach nor one in which syntactic movement must parallel functional or semantic changes can capture the way in which in the data studied here, as the role of information structure has diminished, the mapping between c-structure, m-structure and f-structure has changed. On our account there are only two basic mechanisms of change: a) the emergence of new pieces of syntax (grammaticalization in the classic sense), and b) shifts in the mapping principles between the different levels of structure (i-, s-, f-, c-, m-structure), all of which are independently required to characterise the syntax of natural languages.