2.1 f-structure

f-structure captures the abstract functional information associated with a sentence. This includes both grammatical functions (gfs), e.g. subj(ect) and obj(ect), as well as grammatical features, e.g. tense, case and def(initeness). A special type of functional feature is pred, which is a pointer into the semantics of a predicate, takes a semantic form as its value, and captures the argument(s) (if any) a predicate requires in terms of grammatical function. f-structure representations take the form of attribute-value matrices which consist of a set of attribute(/feature)-value pairs. An example f-structure for the Icelandic sentence in (1) is given in (2).

Any f-structure must satisfy certain wellformedness conditions, see (3)–(5) (Bresnan et al. 2015).

A consequence of assuming f-structure as an independent level of representation for abstract functional information is that grammatical functions such as subj and obj are viewed as primitives of the theory. As such, unlike in some other generative approaches, subj and obj need not be exclusively defined in terms of structural position. This flexibility thus neatly allows for accounts of languages where structural position plays a strong role in encoding grammatical functions, e.g. English, those where morphological marking is the dominant encoding means, e.g. Latin, or languages which use a mixture of means (see Nordlinger 1998 for a relevant typology). I discuss this further in Section 4 in relation to AC.

2.2 c-structure

Since abstract functional information is captured at f-structure, c-structure captures information purely about category and constituency, determined on the basis of constituency tests and linear word order, and is represented as a tree diagram. With respect to category, there are lexical categories, e.g. N, V, P and functional categories, e.g. C, I and D. The functional category relevant to this study is I, as I discuss in Section 3 in relation to Icelandic verb-second. Another consequence of assuming f-structure as an independent dimension for abstract functional information is that a functional category is only motivated at c-structure when there is evidence that functional information, e.g. finiteness or definiteness, is associated with a fixed structural position (Kroeger 1993; Börjars et al. 1999). Again, this distinguishes LFG from some other generative approaches in which any functional information which is present in the sentence must be represented in terms of its own functional projection (see Rizzi & Cinque 2016 for an overview of such approaches).

With respect to constituency, LFG applies a “What You See Is What You Get” approach and allows for both endocentric (headed) and exocentric (headless) phrases, and c-structures need not be exclusively binary-branching. Endocentric phrases follow a version of the X-bar schema (Bresnan et al. 2015), while exocentric phrases lack a c-structure head and are captured under a non-projective category, S, which is not headed by something of the same category as itself.² Endocentricity can be viewed as one of two competing principles for c-structure organisation, the second being predicate-argument locality (Austin & Bresnan 1996), which is in essence a preference for structures with closer proximity between a verb and its arguments including the subject, and which can give rise to exocentricity. Furthermore, languages may mix endocentric and exocentric structures and typologically this flexibility can be visualised as a spectrum with a range of possibilities, see e.g. (6) (Bresnan et al. 2015), where C stands for any lexical category and FP any functional projection.

(6)	Endocentricity/exocentricity scale

2.3 i-structure

This paper deals with information structure, a domain where terminology is notoriously problematic. I adopt an approach in which information-structural features are derived from two primitive binary properties, [± new] and [± prominent]. This provides the four-way division in (7), where focus is [+New, +Prominent] and topic [–New, +Prominent] (Butt & King 1996; 1997; based on ideas from Vallduví 1992; Choi 1999).

In this view, topic and completive.information are similar to the traditional topic-comment distinction, and focus and background.information in line with the traditional focus-background distinction. That is, topic is old information which is relevant in the current context and hence prominent, while completive.information is new information which is not prominent in the discourse. focus, by contrast, is new information which is prominent. background.information is old information which may be a necessary part of the sentence for syntactic reasons or required to further clarify the relation between what is already known and the new information introduced in the sentence.

I assume a separate level of representation for information structure, i-structure, following work by Butt & King (1996), King (1997) and Butt et al. (2016). Like f-structure, i-structure is represented as an attribute-value-matrix. An example i-structure for the second sentence in (8) is shown in (9).

The values for information-structural features in the i-structure in (9) are shown within curly brackets to indicate that in each case the value is a member of a set. This allows for sentences which have e.g. more than one topic or focus (see e.g. Dalrymple & Nikolaeva 2011). The pred.fn notation, as opposed to the standard pred feature from f-structure, indicates that the predicate value is picked up from the value of pred in the f-structure. Essentially, pred.fn links information at i-structure to the semantics of the lexical item referred to at f-structure (see Butt et al. 2016).

2.4 The projection architecture

Correspondences between the various linguistic dimensions are captured in terms of a projection architecture. Various versions and modifications of LFG’s projection architecture have been proposed (see e.g. Kaplan 1987; 1989; Falk 2001; Asudeh 2006; 2012; Bresnan et al. 2015; Dalrymple et al. 2019), but most of these nuances are not relevant to this paper. What is relevant is the correspondence between c-structure and f-structure, i.e. between phrase structure and grammatical functions (in the context of AC) and that between c-structure and i-structure, i.e. between phrase structure and information-structural features (with respect to DC).

In terms of the correspondence between c-structure and f-structure, I uncontroversially assume that this is formally handled via the correspondence function φ, whereby c-structure nodes are related to f-structures (Bresnan et al. 2015; Dalrymple et al. 2019); see the overall projection architecture proposed by Asudeh (2006) in Figure 1. The details concerning φ are not strictly relevant to this study, and so I will just explain how mapping between c-structure and f-structure is handled notationally. To capture the correspondence between structural positions and functional information, phrase-structure rules, which in turn generate c-structure trees, are annotated with functional information. An annotated c-structure showing the structure-function correspondences for modern English is provided in (10), where it is relatively uncontroversial that SpecIP is a subject position and the complement of V an object position.

Figure 1

Parallel projection architecture of LFG (Asudeh 2006: 369).

In the tree in (10), ↓ and ↑ are metavariables over f-structure variables and serve to relate every node in the c-structure to its corresponding f-structure. ↓ denotes the f-structure corresponding to that node itself, and ↑ denotes the f-structure corresponding to that node’s mother node. Multiple c-structure nodes may correspond to the same f-structure, in which case they are annotated as ↓=↑: This indicates that the functional information associated with a given node is the same as the functional information associated with that node’s mother node. The mapping between f-structures and c-structures will differ between languages, since c-structures differ considerably cross-linguistically, while f-structures are largely invariant.

The place of i-structure within the overall projection architecture requires some detail, as here various proposals have been put forward. In this paper, I follow Butt & King (1997) in assuming a model where i-structure projects from c-structure as defined by the function ι, which can be considered as a parallel to the ϕ function which relates c-structure nodes to f-structures.³ This is in line with Asudeh (2006), see again Figure 1. In a discourse-configurational language, where certain structural positions encode certain information-structural features, specific c-structure nodes will map to information-structural features at i-structure (see e.g. Butt & King 1997 on Urdu/Hindi and Gazdik & Komlósy 2011 on Hungarian). For example, one can capture a language where the clause-initial position can exclusively host any topical grammatical function, in terms of the annotated c-structure in (11) (cf. Butt & King 1997 on Urdu/Hindi).

Notationally, the DC is captured via the first annotation line on SpecIP in the tree in (11), where the arrows annotated with ι indicate projection to i-structure. The ∈ symbol captures the fact that the SpecIP topic may be part of a set, thus allowing for sentences with more than one topic. The second line of the annotation on SpecIP is the standard mapping to f-structure (see above), which associates the position with any type of grammatical function (provided, of course, it is a topic in information-structural terms).

3.1 Evidence for endocentricity

The finite verb in Old Icelandic is positionally restricted. Word order patterns characteristic of verb-second (V2) are robustly attested, e.g. (12), and Old Icelandic is standardly assumed to be a V2 language (Eythórsson 1995; Rögnvaldsson 1995; Faarlund 2004).

Strikingly, Old Icelandic does not exhibit verb-third or verb-later (than third) word orders (Rögnvaldsson 1995; Walkden 2014; Booth 2018), thus contrasting with early West Germanic varieties where such patterns have been observed (e.g. Kiparsky 1995; Axel 2007; Walkden 2015). The only exception to V2 are verb-initial (V1) word orders which occur in constructions involving pro-drop, e.g. (13a), impersonal constructions, e.g. (13b), presentational constructions, e.g. (13c), and the so-called “narrative inversion” construction, e.g. (13d); see Booth (2018) and Sigurðsson (2018) for more on these types.

As such, Old Icelandic exhibits a fixed position associated with finiteness, with one available prefinite position (henceforth prefield) which can be optionally occupied by a wide variety of phrasal constituents; I return to the nature of prefield in Section 6.

In line with the theoretical assumptions outlined in Section 2, these facts can be captured at LFG’s c-structure in terms of a left-headed endocentric functional projection, IP, headed by the finite verb in I, with SpecIP as one available prefinite position, cf. the tree in (14).

This is in line with previous proposals for the Icelandic left periphery within LFG (Sells 2001; 2005; Booth et al. 2017; Booth & Schätzle 2019; Booth & Beck 2021). The IP-rooted account outlined here contrasts with other formal accounts of Icelandic clause structure which assume an additional CP-projection to allow the subject to remain in SpecIP in non-subject-initial V2 clauses as well as in V1 clauses (e.g. Sigurðsson 1990). However, as Booth & Beck (2021) point out, this is motivated by the assumption that SpecIP is a unique subject position, within approaches to grammar where subjects are structurally defined; the subject is assumed to be fixed in SpecIP and the finite verb must thus occur in different functional heads (typically I and C) in order to account for the observed word order possibilities. As outlined in Section 2, in LFG subjects are captured at f-structure and need not necessarily be associated with a fixed structural position at c-structure. This more abstract view of subjecthood, and the flexibility with which subjects can be associated with c-structure, means that in principle an analysis where the subject is assumed to occupy a fixed structural position – as in the more standard generative analysis described above – is equally acceptable as an alternative analysis where rather the finite verb is fixed, with two available positions for the subject. In fact, as I show in Section 6, there is empirical evidence with respect to information structure which indicates that the prefield and and the immediately postfinite position are not one and the same position. As such, I opt for the IP-rooted account which neatly captures the key intuition borne out by the Old Icelandic data, namely that the position of the finite verb is restricted.

3.2 Does Old Icelandic have a VP?

The original discussion on (argument) configurationality in Old Icelandic featured a claim concerning its structural organisation by Faarlund (1990), specifically that the language lacks a VP-constituent. Faarlund’s evidence for the absence of a VP-constituent centred on negative evidence, namely the fact that structures involving VP-fronting, VP-clefting and VP-pronominalisation did not appear to be attested. However, in historical linguistics this type of negative evidence is famously problematic; the absence of a structure in a language stages’s written attestation carries much less weight than synchronic judgements of a structure as ungrammatical. Instead, it is sensible to investigate whether also positive evidence against a VP-constituent can be found, taking into account a broader range of types of evidence. In this section, I test Faarlund’s original claims against the (M)IcePaHC data and also consider further diagnostics for constituency. In order to abstract away from V2, I only examine sentences with a finite auxiliary and a nonfinite lexical verb which is transitive, specifically examining the behaviour of the nonfinite lexical verb and its object. As I discuss below, Faarlund’s claim regarding the absence of VP-pronominalisation has since been challenged by Rögnvaldsson (1995: 13), who claims that tests for VP-constituency “appear to give the same results for Old and Modern Icelandic”. As I show in this section, if one extends the diagnostics to a broader range, this is not the case; subtle differences between Old Icelandic and the modern language can in fact be found.

3.3 Old Icelandic as an IP-S language

In this section, I have shown empirical evidence that Old Icelandic exhibits both endocentrity in terms of V2 and also a certain level of word order freedom in the postfinite domain, with evidence both for and against a VP-constituent. In this respect, Old Icelandic exhibits a clausal shape which is reminiscent of diverse languages which exhibit some second position phenomenon – whereby a functional feature is firmly associated with the second position in the clause – but where word order after this second position is relatively free. Such properties are in fact relatively common among those languages which have been classically labelled as (argument)-non-configurational and which have been analysed within LFG, e.g. Tagalog (Kroeger 1993), Warlpiri (Simpson 1991; Austin & Bresnan 1996) and Wambaya (Nordlinger 1998).

Many of these accounts (Kroeger 1993; Austin & Bresnan 1996; Nordlinger 1998) capture these observations via an IP structure where the second position phenomenon is captured under I, and make use of the exocentric non-projective category S (see Section 2) to capture the post-second freedom. The assumption for the relevant languages is that this S is the complement of I, see the tree in (33).

The intuition behind S is that it allows for languages which favour predicate-argument locality (see Section 2) by permitting “internal subjects” (Austin & Bresnan 1996).⁹ As such, S is similar to the small clause (SC) proposed for Irish by Chung & McCloskey (1987), in that it is distinct from IP and can dominate a subject and a predicate without an overt expression of tense. The internal structure of S can vary typologically (Kroeger 1993), as I return to below.

In this respect, such languages, which I term IP-S languages, differ from languages where I takes a VP as a complement, e.g. English, cf. (10) above. At the same time, the presence/absence of a VP-constituent in a particular language is to some extent independent of its status as an IP-VP or IP-S language. Since the internal structure of S is assumed to vary cross-linguistically (Kroeger 1993), this allows for a flat structure where the verb and its arguments are sisters to one another, e.g. (34a) (cf. Kroeger 1993 on Tagalog; Austin & Bresnan 1996 on Warlpiri) or equally a structure where S contains a VP-constituent e.g. (34b) (cf. Chung & McCloskey 1987 on Irish).

On the gradient view of constituency as outlined in Section 3.2, and since there is mixed evidence for and against a VP-constituent in Old Icelandic, it seems reasonable to assume that Old Icelandic allows for both (34a) and (34b); an endocentric VP has to some extent developed, cf. (34b), but is not yet fully solidified, with the result that we also get word orders which are licensed by the older flat structure within S, cf. (34a). By comparison, for modern Icelandic, where the evidence for a VP-constituent is considerably stronger, it seems reasonable to assume that the structure in (34b) is more dominant.¹⁰ As such, I propose the structure in (35) as the basic clause structure for Old Icelandic, building on the assumptions regarding the left periphery outlined in Section 3.1, see (14). As outlined already, the mixed evidence regarding the status of a VP-constituent can be captured via assuming two alternative expansions of S, a flat expansion as in (36) and an expansion with a VP as in (37).¹¹

Under the flat expansion of S in (36), any grammatical function (gf) can occur, as well as any number of adjuncts, and nonfinite verbs. The functional annotation (↑gf)=↓ reflects the assumption that no specific grammatical functions are assigned within S, but that the information with respect to the specific grammatical functions comes from other sources, i.e. morphological marking (see Nordlinger & Bresnan 2011). The freedom of order between the nonfinite verb, its object, the subject and adjuncts is captured via S being the mother node; the sisters are assumed to be, in syntactic terms at least, freely ordered (see also Kroeger 1993 on Tagalog). By comparison, in the alternative expansion in (37) the verb and its object(s) form a VP-constituent to the exclusion of the subject.

Note that the diachronic proposal here, as supported by the data in Section 3.2, is that both expansions of S in (36) and (37) are available in Old Icelandic and that the structure with a VP in (37) becomes increasingly dominant over time, and by modern Icelandic is overwhelmingly dominant. While gradient models of constituency have been for the most part restricted to non-generative approaches to syntax (as discussed in Section 3.2), non-discrete models of constituency have also been considered within generative syntax (see e.g. discussion in Carnie 2010). In LFG specifically, diachronic accounts of gradual structural change have been captured in terms of the “growth of c-structure” (Vincent 1999; Börjars et al. 2016), whereby endocentric phrasal organisation gradually emerges over time. Another approach to gradual change at clause structure has been developed by Clark (2004) in terms of stochastic optimality-theoretic LFG. Moreover, recent work by Bresnan (2021) has shown how LFG’s parallel architecture can be adapted to produce a hybrid model with combines the formal representation central to generative approaches with usage-based insights (e.g. Bybee 2006) via an exemplar-based model of the mental lexicon (Pierrehumbert 2001; 2002). Developing a formal model to capture the gradience involved in an increasingly consolidated VP-constituent in the history of Icelandic would be beyond the scope of this paper. Here, I just point out that the assumptions and approaches which could underpin such a model are not without precedent, provided one considers insights from a broad range of approaches to language.

In sum, on the account outlined here, Old Icelandic is an IP-S language exhibiting a mixture of endocentricity and exocentricity at c-structure, and a somewhat established VP-constituent, which, by comparison, has become increasingly consolidated by the modern stage. This in turn has bearings on the language’s status with respect to argument configurationality, as I discuss in Section 5.

4.1 Argument configurationality

Early work on (argument) configurationality (Hale 1982; 1983) cast it in terms of a binary parameter whose “non-configurational setting” could account for various syntactic phenomena, most notably free word order, discontinuous expressions and extensive use of unexpressed arguments (cf. Baker 2001, “narrow configurationality”). However, various authors have since argued that these three phenomena are in fact independent of one another, and that discontinuous expressions and unexpressed arguments in particular are fairly superficial typological features exhibited also in many (argument-)configurational languages (e.g. Speas 1990; Kroeger 1993; Austin & Bresnan 1996). In this paper, I adopt the now standard view within LFG that argument configurationality is a gradient property and strictly concerns how a language marks grammatical relations (e.g. Nordlinger 1998; see also Baker 2001, “broad non-configurationality”). In a radically argument-configurational language such as English, the difference between subject and object is positionally marked (subject in SpecIP, object as complement of V). By contrast, in a radically non-argument-configurational language, other means must be employed for marking grammatical relations, e.g. case-marking and/or agreement patterns (cf. Kiparsky 1987; 1988; 1997; Nordlinger 1998).

Given this narrower definition of (argument) configurationality, one might ask precisely what types of evidence can be taken to indicate AC or a lack thereof and, perhaps more importantly, which cannot. In this view the presence/absence of two classic “non-configurational” characteristics, discontinuous expressions and unexpressed arguments, will not directly bear on a language’s status with respect to AC. At the same time, subject-object asymmetries, which are often taken to indicate an (argument-)configurational language, are also no watertight diagnostic, since many non-(argument-)configurational languages have been observed to exhibit subject-object asymmetries, e.g. Warlpiri (Hale 1983; Baker 2001) and Tagalog (Kroeger 1993). This in turn fits with the design principles behind the parallel architecture of LFG which, as discussed in Section 2, assumes that grammatical functions (at f-structure) are independent of structural configuration (at c-structure); such an approach permits subject-object asymmetries even in a non-argument-configurational language, since it is assumed that these operations refer to abstract functional information captured at f-structure, without necessarily resorting to structural properties at c-structure; see Mohanan (1982) for discussion.

Similarly, one might ask how endocentric/exocentric organisation at c-structure intersects with AC. From typological evidence alone, it is clear that the relation between AC and endocentricity/exocentricity is complex. A standard view within LFG (e.g. Nordlinger 1998) is that non-argument-configurational languages will have free word order in terms of the ordering of arguments, but need not necessarily have an exclusively exocentric structure; some level of endocentricity may be present, provided specific structural positions are not exclusively associated with particular grammatical functions. As discussed in Section 3.3, some of the classic non-(argument)-configurational languages, e.g. Warlpiri, do not encode grammatical relations via structural position, but exhibit some second position phenomenon which motivates an endocentric IP. In turn, this view means that one is left with a rather narrow set of linguistic characteristics which can be taken as evidence for or against AC, specifically those characteristics which reveal the status of the relation between grammatical functions and (i) phrase-structure configuration and (ii) morphology, since this will situate a particular language on the scale of argument configurationality in (38).

With respect to the relation between grammatical functions and phrase-structure configuration, i.e. structural encoding, indication as to how strong this relation is in a particular language will come from (i) constituency tests and linear word order patterns which show whether the verb and its object do/do not form a phrasal constituent (VP) to the exclusion of the subject and (ii) linear word order patterns which show to what extent subject, verb and object have a fixed order relative to one another and occur in specific positions (cf. Nordlinger & Bresnan 2011). In terms of the other relation, between grammatical functions and morphology, relevant considerations are to what extent specific grammatical functions are expressed via dependent-marking (i.e. case-marking on nominals) or head-marking strategies (i.e. agreement-marking on verbs).

The gradient nature of AC is evidenced both synchronically and diachronically. From a synchronic perspective, the sheer fact that many languages encode grammatical relations via a mixture of means, structural and morphological, indicates that the gradience of AC should be taken seriously (Nordlinger 1998; Nordlinger & Bresnan 2011; Snijders 2015). For example, there are languages which combine structural means with head-marking morphology, e.g. Chicheŵa (Bresnan & Mchombo 1987), and similarly languages which combine structural means with dependent-marking, i.e. case, e.g. modern Icelandic (Zaenen et al. 1985), as I return to in Section 5. On the other hand, diachronic evidence from various languages indicates that languages change their means for marking grammatical relations over time (e.g. Kiparsky 1997; Vincent 1997; Hewson & Bubenik 2006; Ponti & Luraghi 2018). This means that individual language stages will naturally exhibit intermediate stages along a particular change’s trajectory, i.e. gradient levels of AC. In computational work, recent corpus-based approaches have made important strides towards quantifying the extent to which languages use structural and/or morphological means to encode grammatical relations (e.g. Montemurro & Zanette 2011; Futrell et al. 2015; Levshina 2019; Namboodiripad to appear). Although such studies are often not explicitly connected with the issue of (argument) configurationality, their findings in turn reinforce the need to recognise the gradient nature of AC in theoretical work.

Still, previous acknowledgement of the gradience of configurationality in the wider theoretical literature, and attempts towards a typology of configurationality, have generally relied on the broader set of properties typically associated with non-configurationality as diagnostics. Payne (1993), for instance, argues for a distinction between noun-phrase configurationality and verb-phrase configurationality, based on evidence from Panare. Baker (2001), meanwhile, allows for three types of (argument)-non-configurational language, primarily based on which specific types of subject-object asymmetries a language exhibits. In this paper, I will examine Old Icelandic with respect to the specific diagnostics for AC outlined here, namely the extent to which grammatical relations are encoded via (i) structural means and (ii) morphological strategies.

4.2 Discourse configurationality

In early work, the term discourse configurationality (DC) was motivated as a way to capture “topic-prominent” and “focus-prominent” languages (Kiss 1995). According to the original criteria in Kiss (1995), a language’s status as discourse-configurational is evidenced by the fact that topic and/or focus is expressed through a particular structural relation, i.e. associated with a particular structural position. In line with this, DC has been taken to be a rather narrow term, which can only be applied to languages where a particular structural position is systematically and exclusively associated with a discourse function e.g. topic or focus. Surányi (2015), for instance, argues on this narrow definition that discourse-configurational languages should be kept distinct from what he calls “discourse prominence”, i.e. where information structure plays a role in word order, but where the relevant discourse functions are not systematically and exclusively associated with particular structural positions.

Whatever one’s position with respect to how broadly/narrowly the label DC should be employed, it is clear from typological evidence that languages vary along a continuum with respect to how they mark topic and focus, just as they vary in their encoding of grammatical relations. As languages often make use of multiple means to mark grammatical relations, they also typically employ a range of strategies to encode topic and focus (e.g. position, intonation and/or topic-/focus-markers) (see e.g. Oshima 2005; Büring 2009; Zimmermann & Onea 2011). On this basis, a classically discourse-configurational language in which topic and focus are associated with particular structural positions, e.g. Finnish (Kiss 1987; Vilkuna 1989; Kiss 1995; Surányi 2015) can be viewed as being towards one end of a continuum, which I propose as a scale of discourse configurationality, see (39); cf. Sasse (1995), Öhl (2010) and Latrouite & Van Valin Jr. (2019) who also suggest typological spectra relating to DC.

Like gradience with respect to AC, the gradient nature of DC is well-attested cross-linguistically. For instance, we find languages which are discourse-configurational with respect to one discourse function only, such as Aghem, which has been shown to exhibit DC specifically with respect to focus (Horvath 1995), and Japanese which is discourse-configurational in relation to topics (Kiss 1995). Moreover, since topic and focus are relatively broad terms, there is also the possibility that a language is discourse-configurational specifically with respect to certain types of topic or focus. Vallduví (1992), for instance, shows that Catalan is discourse-configurational with respect to certain types of topic. At the same time, there is also evidence that a language’s place on the scale in (39) can change (typically subtly and gradually) over time, just as AC (see Section 4.1). For instance, it is well-known that word order is information-structurally driven in early Germanic (e.g. Trips & Fuß 2009; Hinterhölzl & Petrova 2009; 2010; Petrova & Hinterhölzl 2010; Bech & Eide 2014) and that, by comparison, position is generally more important at this early stage for marking information structure than in the corresponding modern languages. I return to this issue in Section 6 in relation to Old Icelandic.

5.1 Grammatical relations in modern Icelandic

As outlined in Section 3, the evidence supporting a VP-constituent in modern Icelandic is relatively strong, which led me to propose that by the modern stage, the c-structure with the VP in (37) is dominant. In other words, the object is now (more) firmly associated with a structural position within the endocentric VP (sister of V) to the exclusion of the subject. At the same time, it is clear that modern Icelandic employs both structural configuration and case-marking to encode grammatical relations, as has been widely observed (e.g. Zaenen et al. 1985; Nordlinger 1998). Accounts vary, however, as to how much weight is given to the structural vs. morphological encoding strategies in the language. Sells (1998), for instance, claims that rich case marking on modern Icelandic noun phrases allows their grammatical function to be determined without reference to a specific structural position, referencing the constructive case approach of Nordlinger (1998) as a formal mechanism within LFG for this. More common, however, are accounts which emphasise the relation between structural position and grammatical functions and thus argue, either implicitly or explicitly, for at least some degree of (argument) configurationality. Jónsson (2018), for instance, points out for modern Icelandic that subjects precede objects in main clauses in neutral word order, and that in e.g. yes-no questions only subjects, not objects, can invert with the verb, in line with what is expected of a Germanic V2 language, see the contrast in (40) (see also Sigurðsson 1989; Jónsson 1996; Thráinsson 2007).

Jónsson (2018) takes contrasts like (40) to indicate that the immediately postfinite position in modern Icelandic is a designated subject position, despite citing counterexamples where the immediately postfinite position can host an object followed by a subject in a “lower subject position”, provided the object is pronominal e.g. (41) (=(32)).

Data like (41) are well-known in the literature and are generally taken as evidence that there is a “lower subject position” available to indefinite or quantified subjects (Jónsson 1996; Bobaljik & Jonas 1996; Vangsnes 2002). An object may also precede a subject in the postfinite domain in modern Icelandic in another context, specifically when the subject is indefinite and relatively heavy, e.g. (42).

In many accounts, including Jónsson (2018), data like (41) and (42) can still be reconciled with the assumption that the immediately postfinite position remains, underlyingly, a designated subject position; “deviant” word orders where the object precedes subject in the postfinite domain are attributed to surface-level syntactic processes such as “object shift”, cf. (41) and “heavy/indefinite subject shift”, cf. (42), which disrupt the underlying positions. I return to below this point below in connection with Old Icelandic.

In the context of how grammatical relations are marked, various authors also cite the existence of non-nominative subjects, that is non-nominative-marked arguments which qualify as subjects due to their syntactic behaviour (Andrews 1976; Zaenen et al. 1985; Sigurðsson 2004). An example is provided in (43), where the dative argument henni behaves like a subject in triggering the obligatory reflexive sín; see also Maling (2001) for discussion of mismatches between case and grammatical functions in modern Icelandic.

The possibility for non-nominative subjects is taken to suggest that structural subject-encoding (i.e. position) has primacy over morphological subject-encoding (i.e. case) in modern Icelandic. On the other hand, some have raised doubts over the primacy of structural encoding. Zaenen et al. (2017) – who extend the investigation to late subjects in presentational constructions (pivots) – point out that their observations indicate that position does not uniquely identify subjects. As they state, the facts lead to the “somewhat paradoxical conclusion that in Icelandic, neither case marking nor position uniquely identify subjects” (Zaenen et al. 2017: 269). Of course, there is no paradox here if one allows for gradience in AC, as outlined in Section 4, and thus for languages which employ some combination of structural and morphological encoding.

Moreover, Zaenen et al. (2017) cast doubt over the status of what they refer to as the “canonical subject positions”, i.e. the prefield and the immediately postfinite position. They point out that while they are, in statistical terms, the most prevalent positions in which subjects are found in modern Icelandic, this is “hardly a syntactic distribution” (Zaenen et al. 2017: 270) and that, furthermore, they cannot host any type of subject. Citing Thráinsson (2007: 323), they note that the grammaticallity of a bare non-specific indefinite subject in the prefield is questionable, e.g. (44). Thráinsson (2007: 323) specifically states that such subjects “seem more natural inside the VP than outside”.¹²

Furthermore, Zaenen et al. (2017) show that the positional coding properties typical of prefield and immediately postfinite subjects do not hold for the pivots; in other words, only canonical subjects (i.e. those which are not pivots) can occur in those positions which have been taken to be “canonical” or even “designated” subject positions. In fact, as Zaenen et al. suggest, the positional coding properties may in fact turn out to be properties of topics, rather than of subjects per se, and have been conflated with subjects, since subjects are often topics (e.g. Lambrecht 1994: 131–145). Although they do not pursue this possibility any further, it would seem to be a nod in the direction of discourse configurationality, specifically the notion that the “canonical subject positions” are in fact canonical topic positions. I return to this point in Section 6 in the context of Old Icelandic.

With so much focus on subject positions and case in modern Icelandic, the other morphological marking strategy, agreement, has a tendency to be glossed over. Nevertheless, there are interesting and highly complex patterns on display here and the relation between agreement and grammatical relations is far from straightforward. Sigurðsson (1996), for instance, points out that while nominative subjects control person and number agreement on the finite verb, e.g. (45a), non-nominative subjects do not, e.g. (45b); in the latter it is standardly assumed that the finite verb shows default (3sg) agreement, in the absence of a nominative noun phrase.

Clearly this complicates the situation with respect to AC, adding to the mixed picture with respect to the morphological encoding of grammatical relations; just as there is no one-to-one correspondence between case marking and grammatical function, there is no clear-cut pattern with respect to how agreement marks grammatical relations. Moreover, while verbal agreement in modern Icelandic is most commonly controlled by nominative subjects (Sigurðsson 1996), we also find nominative-marked objects which control verb agreement, although this is not obligatory and is limited by number to 3rd person objects (Taraldsen 1995; Sigurðsson 1996), e.g. (46).

In this respect, agreement appears to be closely connected with case, as evidenced by the fact that nominative arguments, whether subject or object, can control agreement.

In sum, it is clear that modern Icelandic employs a complex combination of structural and morphological means for the encoding of grammatical relations; neither structural position, nor case or agreement alone can be said to straightforwardly mark grammatical relations in a clear-cut way.

5.2 Grammatical relations in Old Icelandic

I now consider the encoding of grammatical relations in Old Icelandic, with respect to both structural and morphological strategies, and specifically whether there is evidence to suggest a different balance between these means in Old Icelandic compared to the modern language. Like modern Icelandic, Old Icelandic employs a complex combination of structural means (i.e. position) and morphological strategies (i.e. case and agreement) to mark grammatical relations. However, there is evidence to suggest that the balance between these two means in Old Icelandic is more strongly weighted towards the morphological compared to modern Icelandic. Firstly, there is evidence that the relationship between structural configuration and grammatical functions is somewhat weaker in Old Icelandic; as shown in Section 3, there is good reason to assume that the VP-constituent, though to some extent available in Old Icelandic, was not yet consolidated and that the flat expansion of S was more dominant. Crucially, in the flat structure, the subject and other verbal arguments are all immediate daughters of S, i.e. in a sisterhood relation to one another, cf. (36). As such, there is already evidence towards Old Icelandic being at least somewhat less argument-configurational than modern Icelandic where, as discussed, the structure with the VP in (37) is more dominant.

There are also further indications that the role of structural configuration is weaker in Old Icelandic from linear word order patterns showing which positions are available to subjects and objects. In previous work, the question of whether Old Icelandic has a designated subject position immediately after the finite verb has been debated (Haugan 2000; Faarlund 2001; Jónsson 2018). All three authors have observed examples in Old Icelandic where an object can occur in the immediately postfinite position, before an indefinite or a quantified subject, e.g. (47) (cf. (41) for modern Icelandic above). Note that this is possible when the object is pronominal, e.g. (47a) or nominal, e.g. (47b).

Faarlund (2001) and Haugan (2000) cite examples such as (47) as evidence that objects can occur in the “canonical subject position” (= the immediately postfinite position) in Old Icelandic, and thus that this is not a unique subject position.¹³ Faarlund (2001), in fact, goes one step further and shows that all of the positions available to nominative subjects in Old Icelandic are also available to objects.

An investigation of the Old Icelandic texts in IcePaHC (1150–1350) yields similar data to that observed by Haugan (2000) and Faarlund (2001). Indeed, some similar clauses where the object precedes the subject in the postfinite domain were already provided in Section 3.2 as (31), repeated here in (48), where a quantified subject can intervene between the object and a nonfinite verb.

In total, once erroneous examples are removed, there are 54 matrix clauses in the data where, as in (48), a direct object with non-nominative marking precedes a nominative-marked subject in the postfinite domain. The majority (51/54) of these 55 examples are distributed over the properties in (49).

The three remaining examples are provided in (50).

In Section 6.1, I show how the various relative orderings of subject and object exhibited in this section can be accounted for in terms of information structure, with bearings on the status of Old Icelandic with respect to discourse configurationality.

Since structural configuration seems to play a weaker role in marking grammatical relations in Old Icelandic than in modern Icelandic, we would also expect to find evidence that morphological means are stronger in this early stage of the language, in order to compensate for this. With respect to agreement, Old Icelandic appears to exhibit similar behaviour to modern Icelandic. Faarlund (2001), for instance, observes that, as in modern Icelandic, in Old Icelandic the verb never agrees with a non-nominative subject (Faarlund 2001). In other words, agreement is exclusively controlled by nominative-marked arguments in Old Icelandic, as is the case in the modern language.

More instructive than the agreement data is data with respect to case and grammatical functions. The role of case in marking grammatical relations in Old Icelandic is disputed, in line with the debate as to whether non-nominative subjects are already a part of the Old Icelandic grammar or not. On the one hand, various authors claim that Old Icelandic does have non-nominative subjects (e.g. Rögnvaldsson 1991; Rögnvaldsson 1995; Rögnvaldsson 1996; Barðdal 1998; 2000; Haugan 2000; Barðdal & Eythórsson 2003; 2009; Barðdal et al. 2012; Jónsson 2018). On this account, there would be no radical difference in the relation between case and grammatical relations in Old Icelandic compared to modern Icelandic. However, others have also provided data which points against the existence of non-nominative subjects in Old Icelandic, i.e. that the relevant arguments are not subjects at this early stage (Faarlund 1990; Juntune 1992; Mørck 1992; Faarlund 2001). Faarlund (2001), for instance, discusses what qualifies as subject tests in Old Icelandic and shows that on these tests, non-nominative arguments of the quirky type in modern Icelandic do not qualify as subjects in Old Icelandic. Such authors thus assume that certain oblique arguments have acquired subject properties over time (cf. the Object-to-Subject Hypothesis, Cole et al. 1980; Haspelmath 2001).

Many of these authors are guided by their specific theoretical approach and its corresponding view of subjecthood. Barðdal & Eythórsson (2003; 2009), for instance, assume that a subject is the leftmost argument in the argument structure of a predicate, while Jónsson (2018) uses positional criteria as evidence for subjecthood. Both of these definitions of subjecthood are ruled out here; in the LFG-approach adopted in this paper, subjects are considered independently of argument-structure and position which, as shown in this section, cannot be taken as a subject criterion to the exclusion of objects.

Whatever one assumes about whether non-nominative subjects were already established in Old Icelandic or not, it is reasonable to assume that case is becoming less important compared to structural position for the encoding of grammatical relations throughout the Icelandic diachrony. For instance, the change known as “Dative Substitution” or “Dative Sickness” is well-documented (e.g. Svavarsdóttir 1982; Smith 1996; Jónsson 2003; Barðdal 2011), whereby dative case is increasingly associated with experiencer subjects; a similar change as been observed on the basis of IcePaHC data by Schätzle et al. (2015) and more fully in Schätzle (2018). An increase in dative marking on theme arguments in the history of Icelandic has also been observed (Barðdal 2001; 2003; Maling 2002).

Moreover, in line with the linking theory developed by Kiparsky (1987; 1988; 1997; 2001), in which case, agreement and position are viewed as interacting licensing strategies for grammatical relations, Kiparsky (1997) suggests that morphological marking (i.e. case and agreement) are still relevant in modern Icelandic, but have become recessive strategies in comparison to position, which is now the dominant strategy. Booth et al. (2017) and Schätzle (2018) develop the rise of positional licensing account for the history of Icelandic further, based on diachronic investigations using IcePaHC. Specifically, they show that subjects are increasingly realised in the prefield (SpecIP on the account outlined in Section 3); see also Booth & Beck (2021). Crucially, it is observed that, compared to nominative subjects, dative subjects lag behind in this development, showing the increasing tendency for the prefield only later in the diachrony; in other words, dative subjects become overall more subject-like over time, in line with the strengthening of positional encoding of grammatical relations.

In sum, the evidence provided here with respect to the structural positions available to (nominative) subjects and (non-nominative) objects in Old Icelandic, together with the account of an increasingly dominant VP in Section 3 and previous accounts of the diachronic status of non-nominative subjects, indicate a change in status, albeit a rather subtle one, with respect to argument configurationality. Structural position is less dominant as a strategy for encoding grammatical relations in Old Icelandic compared to the modern language, and there is good reason to assume that the increasing dominance of position over time coincides with a concomitant weakening of case-marking as a strategy. Moreover, this subtle change can only be captured if one takes the view of argument configurationality outlined in Section 4.1, whereby it is a gradient property allowing for languages to shift moderately along the AC-continuum over time.

6.1 The information-structural characteristics of the prefield

In Germanic V2 languages, it is typically assumed that there is a close relationship between the prefield and topicality (see e.g. Holmberg 2015). The observation that the prefield typically hosts topical constituents in Old Icelandic (e.g. Kossuth 1978: 455; Faarlund 2004: 231) is in line with the information-structural account of early Germanic clause structure proposed by Hinterhölzl & Petrova (2010), whereby the verb is assumed to serve as an information-structural boundary, separating topic (prefinite) and comment (postfinite); see also Booth et al. (2017) who adopt this account for early Icelandic. Prefield-topics are indeed robustly attested in Old Icelandic, e.g. (12a) and (12b) above. However, there is evidence that the prefield is not exclusively associated with topics. For instance, there are some examples where the subject of a presentational sentence – which is discourse-new and thus cannot be a topic – can occur in the prefield, e.g. (51); see also discussion in Faarlund (2004) and Booth (2018).

Note that, as outlined in Section 5, in modern Icelandic a bare non-specific indefinite subject in the prefield is by contrast deemed questionable, or at least not very natural (Thráinsson 2007; Zaenen et al. 2017), cf. (44) above.

Furthermore, there is also evidence that elements in new information focus can occupy the prefield, e.g. (52) and (53), which both introduce a new individual to the discourse.

Additionally, examples can be found where a constituent which is clearly in contrastive focus occupies the prefield, e.g. (54), where the nine host angels are contrasted with God in the previous sentence.

Since topics and focus – which are both +prominent (see Section 2) – can occur in the clause-initial position in Icelandic, one might be tempted to characterise SpecIP as a position where any +prominent constituent can occur. However, this would fail to capture the fact that sentences with a clause-initial adjunct, typically a scene-setting adverb phrase, are very common in Old Icelandic, e.g. (55).

Examples like (55) are in line with observations elsewhere in early Germanic that the prefield frequently hosts constituents which are ‘discourse-linking’; more specifically, it has been shown that clause-initial discourse adverbs typically mark a sequence of foregrounded successive actions or events which do not overlap temporally, see e.g. Foster (1975); Enkvist & Wårvik (1987); Wårvik (2011); Trips & Fuß (2009); Los (2012) on Old English, Betten (1987) on Old High German and Klein (1994) on Gothic.

Moreover, the fact that stylistically fronted elements occupy the prefield (see Section 3) may indicate a further diversification of what can go in this position in information-structural terms. Although stylistic fronting in modern Icelandic has been traditionally assumed to be pragmatically neutral (Maling 1990; Holmberg 2000), more recent research has led to claims that it serves an information-structural function, e.g. contrastive focus (Hrafnbjargarson 2004; Molnár 2010) or as a backgrounding strategy (Egerland 2013). Research on stylistic fronting in Old Icelandic is notably lacking, and a precise characterisation would be beyond the scope of this paper. Here, I just point out that the nature of stylistic fronting is a relevant concern for future work on syntax and information structure in Old Icelandic.

In sum, it is clear from the data outlined here that the prefield in Old Icelandic has a dual nature in information-structural terms. On the one hand, it can host a nominal argument if that argument is information-structurally prominent; on the other, it can host scene-setting adjuncts which cannot be considered prominent, but which serve a discourse-organisational function, linking narrative events in time and space. These two options for the prefield can now be added to the c-structure tree for Old Icelandic originally provided in Section 3.3, see (56). The prefield (SpecIP) now bears annotations which assume the features [±prominent] and [±d(iscourse)-linking] at i-structure.¹⁴ I do not fully provide all the daughters of S in order to generalise over the two alternative expansions with and without the VP argued for in Section 3.3, cf. (36) and (37).

Strikingly, the observations made with respect to the prefield in Old Icelandic share similarities with those made for the clause-initial position in Warlpiri and related Australian languages by Simpson & Mushin (2008). They observe that the clause-initial position can host prominent nominal forms as well as scene-setting adjuncts, which they say are “drawn” to the clause-initial position in order to be situated at the junction of two utterances. More exploration of this dual nature in Old Icelandic and beyond I leave for further research. For now, one can conclude that, being broadly associated with prominence (as well as discourse-linking), there is no evidence that the prefield in Old Icelandic is associated with an exclusive discourse function, either topic or focus and that, on the gradient model outlined in Section 4.2, Old Icelandic does not appear to be particularly discourse-configurational in this respect.

6.2 Discourse configurationality in the postfinite domain

With so much emphasis on the information-structural characteristics of the prefield in Germanic V2 languages, the postfinite domain is often neglected in this context. In this section, I take a closer look at the correspondence between word order and information structure in this part of the clause in Old Icelandic. This involves revisiting some of the word orders discussed in Section 5 in relation to argument configurationality, which showed that the relative position of the subject and object in the postfinite domain varies. Taking information structure into account, I show that these patterns can be accounted for if one assumes that Old Icelandic exhibits some level of discourse configurationality in the postfinite domain.

As shown in Section 5, the relative order of subject and object in the postfinite domain in Old Icelandic is not fixed; the object can precede the subject under certain conditions, cf. the examples in (49) and (50) above. This was taken as evidence that subject and object are not exclusively positionally encoded in the postfinite domain, in line with the flat structure within S proposed in Section 3. In fact, if one reexamines clauses where the object precedes the subject in the postfinite domain with respect to information structure, a striking pattern emerges; in each of the examples, the object exhibits properties which generally correlate with topichood (e.g. definiteness, pronominality and thus discourse-givenness) and the subject properties which tend to correlate with non-topics (quantification, indefiniteness and discourse-newness). This indicates that word order in the postfinite domain is information-structurally governed, specifically that topical material, whether subject or object, must precede non-topical material. Furthermore, this constraint can account for the examples in (31) above, whereby a subject intervenes between an immediately postfinite object and its verb; in each case, the object can be construed as a topic, versus the subject which is quantified in some way.

Further evidence with respect to information-structural structural properties in the postfinite domain is hard to come by in the corpus data, since (M)IcePaHC is not annotated for information structure. As such, one can only use correlates such as definiteness, pronominality and quantification, which are explicitly encoded in the corpus. However, more evidence can be gained by considering the behaviour of a small class of discourse adverbs (e.g. nú ‘now’, síðan ‘then’, svo ‘so’, þar ‘there’, þá ‘then’) which have been previously observed to show interesting structural properties with respect to information structure in the postfinite domain (Booth & Schätzle 2019). These discourse adverbs regularly appear in the postfinite domain directly following a topical constituent and preceding non-topical material, e.g. (57).

Strikingly, in clauses where there is a quantified argument (i.e. nontopical subject or object) in the postfinite domain, but no topic, there is also a frequent word order where a discourse adverb intervenes between the finite verb and the quantified argument. Examples of this in the IcePaHC data are plentiful (n=146), and some examples are given in (58).

By contrast, there are only 12 instances where a quantified argument precedes one of these discourse adverbs in the postfinite domain, but these involve either misannotations of the adverb or the quantified argument, or cases with two discourse adverbs flanking the quantified argument, e.g. (59).

Informatively, similar behaviour is observed in clauses which have an indefinite subject in the postfinite domain, i.e. presentational sentences, which are generally taken to be thetic, i.e. topicless constructions (e.g. Sasse 1987; Lambrecht 2000). The corpus data reveals that, in presentationals with a postfinite non-topical subject and a discourse adverb, the subject always follows the adverb, e.g. (60).

By contrast, there are no attested examples of a configuration like (61), where the subject of a presentational sentence occurs in the midfield before an adverb.

While this is not definitive evidence that such a configuration is not grammatical, given the commonness of configurations as in (60) and the absence of the configuration in (61) in the corpus data, one can at least interpret a clear preference for the order in (60).

In sum, the data indicates that there is a position in the postfinite domain which is exclusively associated with topics – in linear terms a position directly after the finite verb and before the canonical position for discourse adverbs. In light of this, the annotated tree provided in (56) above can now be furnished with more detail within S, see the updated tree in (62) with annotations capturing the projection to i-structure for the designated topic position. Assuming that the evidence provided above is indeed indicative of a c-structure position being exclusively associated with topical material, it therefore follows that Old Icelandic is at least somewhat discourse-configurational as per the definition adopted in this paper; while the prefield is associated with a variety of information-structural functions, in the postfinite domain topics are structurally expressed.

On this view, the ordering of subject and object in the postfinite domain in the modern language, including the “object shift” patterns and various subject positions outlined in Section 5, can be seen as remnants of this older stage, with DC as a dominant organising principle in the postfinite domain. In fact, it has been observed for various modern Scandinavian varieties that the “object shift position” (=the immediately postfinite position) is only available to objects which are topical (Andréasson 2008; 2009; 2010; Anderssen & Bentzen 2012), and specifically for Icelandic that shifted objects must be definite or specific, while nonshifted objects must be indefinite or nonspecific (Diesing 1992; 1996; Collins & Thráinsson 1996), suggesting such an account is on track.

Furthermore, the findings in this section with respect to the information-structural nature of the prefield and the immediately postfinite position in turn lend further support for the IP-S analysis of Old Icelandic clause structure proposed in Section 3. From the data presented here, it is clear that the prefield and the immediately postfinite position in fact have nuanced differences in their information-structural characteristics; the prefield can host both topical and focal material, whereas focal material is barred from the immediately postfinite position, this being a unique topic position. These differences speak against capturing the prefield and the immediately postfinite position under a single position (SpecIP), as in some generative accounts of Icelandic clause structure (e.g. Sigurðsson 1990), and fits better with the structure assumed here, where the prefield (SpecIP) is assumed to be structurally distinct from the immediately postfinite position (within S).