2.1 The one-article analysis

It is widely assumed that there is one definite article in MI, viz. (h)inn, with art and def being two surface variants (Magnússon Reference Magnússon1984; Sigurðsson Reference Sigurðsson1993; Pfaff Reference Pfaff2007, Reference Pfaff2009, Reference Pfaff, Sleeman, Van de Velde and Perridon2014, Reference Pfaff2015; Harðarson Reference Harðarson2016; Ingason Reference Ingason2016).Footnote ¹⁴ The two major (synchronic) motivations for this assumption are

1. i. complementary distribution (recall that DD-a is ungrammatical in MI, (7-a)),

2. ii. identity of stem and inflection in both elements (the central difference is the presence/absence of ‘h’: hinn vs. -inn; see fn. 15)

A movement analysis of the article was first proposed by Magnússon (Reference Magnússon1984), modified by Sigurðsson (Reference Sigurðsson1993). The gist of the analysis is that the sequence article – A – N can surface as such, in which case the article is realized as art (= pattern (II)), or a constituent containing A and N moves before the article as a result of which the article cliticizes onto the noun and is realized as def (= pattern (I)):Footnote ¹⁵

1. (15)

   

For the purpose of this article, I will adopt this idea; in the following, I will summarize an additional argument developed in Pfaff (Reference Pfaff2015, 29-48, 91-115) that relates position and interpretation of the adjective involved. One initial observation is that there seem to be semantic differences between the two articles.Footnote ¹⁶ Sometimes, pattern (I) is bad, seemingly for semantic reasons, and pattern (II) has to be used instead:

1. (16)

   

1. (17) (Thráinsson Reference Thráinsson2007:89, fn. 2)

   

Prima facie, this could mean that def – but not art – is incompatible with certain adjectives or adjectival meanings, and hence, that the two articles are not semantically equivalent (and by extension, that they are two distinct lexical items). Upon closer inspection, however, these examples do not illustrate a contrast between the two article elements per se. This becomes clear when we take into account that the corresponding pattern (III) versions are fine:

1. (18)

   

Since pattern (III) does involve def, and since it is fine precisely in those contexts where pattern (II) is fine (with the same interpretation), it is evident that the real contrast observed above is not about art vs. def, but rather about patterns (II)/(III) vs. pattern (I). In particular, it is the adjective’s position relative to the article (art or def) that is relevant; what patterns (II) and (III) have in common is that the adjective immediately follows the article – regardless of whether free or suffixed:

1. (19)

   

This can be captured in terms of movement where patterns (I) and (III) differ in which constituent is fronted before the article. Conversely, what patterns (II) and (III) have in common is that the adjective remains in situ, in the post-article position:Footnote ¹⁷

1. (20)

   

Notice that the crucial point is not about linear surface position, but about height in terms of structural position. Following the logic of this analysis, if two adjectives are present, the lower one is expected to move along with the noun to a pre-article position, while the higher one is stranded:

1. (21)

   

According to (21), the postnominal adjective in a “mixed” pattern (I+III) is merged in a higher position than the prenominal one. Thus a well-formed mixed pattern should display the reverse surface order of adjectives in the corresponding pattern (II) version, and in particular, the postnominal adjective (A₁) is expected to have scope over the prenominal one (A₂). These expectations are borne out:

1. (22)

   

It is, more generally, the case that only adjectives that are merged sufficiently high in the structure can surface in an immediate post-article position, i.e. in patterns (II)/(III). Pfaff (Reference Pfaff2015) makes the somewhat vague notion “sufficiently high” more precise and proposes a projection labeled ixP (“index phrase”) such that adjectives can be merged inside ixP or outside/above ixP. In (23) below, both αP and βP are functional projections hosting adjectives in their specifiers (cf. fn. 17), the crucial difference being their relative positions to ixP. Thus an adjective merged in Spec-αP indicates an ixP-internal position, and and adjective merged in Spec-βP indicates an ixP-external position. Now ixP is the relevant constituent that is affected by movement resulting in N-def (patterns (I)/(III)/(IV), but also unmodified definite nouns). Therefore, if movement takes place, ixP-external adjectives like “svokallaða” in (23-b) never move alongside the noun and thus always occur in the post-article position, see also (23-a). On the other hand, ixP-internal adjectives like “kalda” in (23-b) always move along (cf. Pfaff Reference Pfaff2015, 119-127):

1. (23)

   

Being merged in that high position, in turn, has semantic effects. Put informally, the adjective provides some comment or evaluation on the referent denoted by the lower noun phrase; besides so-called, we typically find evaluative adjectives (famous, popular …) in patterns (II)/(III). This is the source of the apparent semantic differences between articles, cf. (16)-(18), fn. 16. The relevant interpretation is determined by occurring in that high post-article position, not by the shape of the article.

Beyond this distinction between a higher and a lower domain for adjectives, all adjectives in patterns (I)-(III) have something in common: they occur in the weak inflection. On Pfaff’s account, this is a consequence of the initial configuration prior to (potential) movement where a definiteness morpheme, namely the article, c-commands all adjectives; weak inflection is a reflex of this configuration (see next subsection).

The assumption that art and def are two manifestations of the same element merged in a position above adjectives, couched in an analysis along the lines of (20)/(21), captures, in an elegant and simple fashion, the commonalities of patterns (II) and (III), the ordering and scope relations in mixed patterns, and the weak adjectival inflection. Apparent semantic differences between the two articles, cf. (16)/(17) turn out to be epihenomal insofar as they arise as a corollary of the respective adjective being merged in a high position. If art and def were two distinct elements and occupied distinct positions, these correlations would be a mere coincidence.

2.2 Adjectival inflection

In Scandinavian, weak adjectival inflection is typically associated with definiteness, whereas strong inflection is a default in the sense that it occurs elsewhere (indefinite noun phrases, predicative contexts). MI moreover provides the interesting case of pattern (IV), the hallmark of which is that – even though it represents the same surface string as pattern (I) – the modifier involved is strongly inflected.

Pfaff (Reference Pfaff2015, Reference Pfaff2017) proposes that an adjective is weakly inflected if it is merged in a position c-commanded by a definiteness feature (contributed e.g. by the definite article). Given what was said about the common base position of patterns (I)-(III) in the previous subsection, it is therefore expected that the adjective in those patterns is weakly inflected. Obviously, this does not apply to pattern (IV). Pfaff argues that the reason for this is that pattern (IV) modifiers are merged in a higher position, outside the c-command domain of the definiteness feature as a result of which the weak inflection cannot be triggered, and the default strong inflection occurs:Footnote ¹⁸

1. (24)

   

I will adopt (a version of) the idea that weakly inflected adjectives must be c-commanded by some definiteness morpheme, see Sect. 5.4 (also Sect. 4.1).

3.1 Patterns with art and/or def

A first glance at tables 3/4 illustrate some interesting states of affairs:

• Pattern (I), albeit being the standard pattern in MI, is exceedingly rare up to the 16th century; during that period, all other patterns (except for DD-b) are more frequent. Conversely, pattern (II), albeit being a marked pattern in MI, was the dominant pattern for many centuries, notably, during the OI period.

• Patterns (III) and (IV) are marked patterns in MI and have never been dominant; they have been marked patterns all along. Nonetheless, by consistently ranging in those low percentage numbers, they actually show a certain persistency or stability, differently from pattern (I).

• Something similar can be said about DD-a: although not very frequent, it shows a certain stability from the 12th to the 19th century. It is perhaps even more interesting to note that that pattern only disappeared relatively recently.

3.1.1 Pattern (I)

The development of pattern (I) is one of the most surprising facts about the results reported in tables 3–5. The proportional differences between earlier and later centuries are staggering, and so are the numerical and proportional discrepancies between patterns (I) and (II), especially before the 17th century. For that period, there are only 10 attestations of pattern (I) in IcePaHC as opposed to 639 attestations of pattern (II). This tendency is strongly supported by the findings from MÍM: I have only been able to identify 11 unambiguous cases of pattern (I) as opposed to more than 1,700 cases of pattern (II). This means that pattern (I) is virtually non-existent during the OI period, and really only caught on in the 17th century, while pattern (II) is the dominant pattern during the OI period and still a prominent pattern up to the 19th century. It is therefore worthwhile having a closer look at the oldest attestations of pattern (I); (25) lists all occurrences from the 12th to the 16th century in IcePaHC, and (26) gives the attestations in MÍM (∼ 13th/14th centuries) found so far:

1. (25)

   

1. (26)

   

Table 5.

Patterns in Old Icelandic: raw numbers and proportions (based on MÍM)

It is possible that some more instances can be found, but this will hardly change the fact that pattern (I) is extremely rare. Conversely, it may be the case that not all examples listed above are genuine. For instance, (25-a) seems to be the oldest attested example, but possibly, it may have to be analyzed differently; compare:

1. (27)

   

According to (27-a), the sequence betri kosturinn is indeed a pattern (I) noun phrase, in predicate function.Footnote ¹⁹ But according to (27-b), the noun is part of the subject, while the adjective is part of the the predicate. This latter construal is feasible insofar as OI noun phrases may be discontinuous and extraction of the demonstrative sá is attested elsewhere, cf. (28-a), and predicative adjectives often precede the subject with the copula preceding the adjective, cf. (28-b) (both examples are from Njáls saga):

1. (28)

   

If this latter construal is correct, the oldest attestation of pattern (I) is actually from the 13th century, viz. (25-b). Note also that example (25-e) is problematic because the ending -a of the adjective is deviant; the expected (weak) ending would be -u. Alternatively, it could be the case that it is meant to indicate the strong inflection (expected form -ar). In that latter case, this example would actually have to be counted as an instance of pattern (IV).Footnote ²⁰ Finally, it is noteworthy that those few early attestations of pattern (I) mostly involve functional adjectives without genuine descriptive content: defective adjectivesFootnote ²¹ and ordinal numerals, or superlatives.

Thus possibly, some apparent attestations of pattern (I) may have to be discarded, and even that small number of pattern (I) noun phrases (10 + 11) is too generous. And even for the authentic cases, there seems to be some kind of restriction: if pattern (I) at all, then preferably with some functional adjective.

The observation that pattern (I) is virtually absent in older stages is interesting in light of the fact that it has become the standard pattern in Modern Icelandic.

3.2 sá as adjectival article

In Icelandic, the adjectival article hinn (art) has survived to this day, even though its relevance has diminished due to the prominence of pattern (I). In the NIScan languages, on the other hand, it has almost entirely disappeared while the distal demonstrative sá has taken its place (det; see introduction). It is therefore interesting to note that the same development, a competition between hinn and sá, was in progress in Icelandic, but with a different outcome: sá did not replace hinn.

In this subsection, I will examine two more patterns: DD-b and Dan both of which involve the distal demonstrative sá.

3.2.3 Adjectival article?

The brief appearance of DD-b coincides with the strong increase in frequency of Dan; both patterns use sá. Given that grammaticalization of (distal) demonstratives into articles is a well-known development, one likely interpretation of the data in table 4 is that sá was in the process of grammaticalizing into a definite article from the 16th century onwards. But this statement must be relativized: the increased frequency of sá is only observed in the context of adjectival modification. Table 2 illustrates that, globally (i.e. irrespective of modification), the frequency of sá remains relatively constant throughout the centuries – unlike the frequency of the suffixed definite article (def); the relevant columns of that table are repeated below:

At least initially, it seems as though sá was specifically developing into an adjectival article and about to supercede hinn in that function (which is, in fact, what had happened in the NIScan languages some centuries earlier (see e.g. Stroh-Wollin Reference Stroh-Wollin2009, Reference Stroh-Wollin2015a). Table 9 shows the frequencies of adjectival articles, where art represents patterns (II)/DD-a, and sá represents patterns DD-b/Dan:

Table 6.

Pattern (III): details (based on IcePaHC; cf. table 3)

Table 7.

Pattern (III) in Old Icelandic: details (based on MÍM; cf. table 5)

Table 8.

def and sá, basic numbers in percentage (from table 2; IcePaHC)

Table 9.

Adjectival articles: numbers in percentage (based on the numbers in table 3; IcePaHC)

Table 9 shows that, in the 16th century, sá becomes the dominant adjectival article, and, given minimal pairs like (37) and (38), this means that it is in direct competition with hinn.Footnote ²⁸ However, hinn continues to exist (in that function), and in the 19th century, it re-gains the upper hand, in contrast to what happened in the NIScan languages.

It should be emphasized that sá has retained its semantics/function as a distal demonstrative throughout, and in some instances of DD-b/Dan, it may actually have a demonstrative interpretation.Footnote ²⁹ But most examples of the period in question do have an unambiguous article (= simple uniqueness) interpretation (see also (38)):

1. (39)

   

This observation and the striking (proportional) increase of sá specifically in the context of adjectivally modified noun phrases anyway support the assessment that sá entered a process of developing into an adjectival article as in the NIScan languages – at least for a certain period.Footnote ³⁰

Yet another aspect must not be ignored: even though hinn became the dominant adjectival article once more during the 19th century, it was pattern (I) that has eventually become the dominant pattern, not pattern (II). In this sense, the develoments discussed above do not merely illustrate one process, but an intricate interaction of several processes/competitions, minimally: sá vs. hinn vs. pattern (I).

3.3 Summary

The two most striking observations made in Sect. 3.1 are (i) the virtual absence of pattern (I) prior to the 17th century, and (ii) the fact that the double definiteness pattern DD-a existed in Icelandic until the early 20th century. The fact that art and def could coocur for a long time suggests that, during that time, those two elements occupied different positions in the noun phrase structure and had a different categorial status (compare this with what was said about MI in Sect. 2.1).

The developments sketched in Sect. 3.2 in all likelihood indicate an incipient grammaticalization process where the demonstrative sá was en route to supercede hinn as an adjectival article. Pattern Dan increased markedly in frequency in the 16th century. This acceleration is likely to have been due to Danish influence, especially in the wake of the Reformation (see fn. 27). Double definiteness pattern DD-a had been around since the 12th century and continued to exist for a long time. Notwithstanding what caused its emergence in the first place, it is conceivable that the existence of that pattern in conjunction with the increase in frequency of sá boosted the development of pattern DD-b. If the macro process really was about the element sá, it explains why both DD-b and Dan gain in frequency during the 16th/17th centuries. But it would seem as though, at some point, the Danish influence was so strong that DD-b disappeared – while DD-a continued to exist.

Approximately between the 16th and 19th century, there was a competition between two adjectival articles; hinn continued to exist, and was not replaced by sá. In the course of the 19th century, the Danish pattern largely disappeared, and hinn won the competion over sá as an adjectival article. In parallel, since the 17th century, pattern (I) had steadily gained in frequency, and eventually becomes the dominant pattern.

4.1 Adjectival article and weak inflection

There is a particularly intimate relationship between (h)inn and weakly inflected adjectives: from the oldest attestation onwards, (h)inn only occurs with an immediately following weak adjective (except for demonstrative uses, see (3), Sect. 1.2). This dependency is not entirely mutual, but there is a strong tendency: the further we go back in time, the more a weak adjective is dependent on cooccurring (h)inn, the closer we come to the present, the more independent weak adjectives become. Perridon (Reference Perridon and Ottóson1996) identifies six occurrences of bare weak adjectives (i.e. weakly inflected adjectives not preceded by (h)inn) in the Swedish runic corpus: æningi/æninga “only-one”, bæzti “best”, fyrsta “first”, þriðia “third”, and ungu (uku) “young”. However, Stroh-Wollin (Reference Stroh-Wollin2012) argues against interpreting that last one as weak adjective, and suggests instead that it has to be read as name. If so, the only remaining bare weak adjectives are functional adjectives and superlatives.Footnote ³¹ Otherwise, weakly inflected adjectives are preceded by the adjectial article (h)inn in the oldest attestations (or sá/þæn in Old East Norse, see Sect. 6.1).

In OI, bare weak adjectives can be found, but they are still exceptional. Again, normally, a weak adjective is preceded by (h)inn. Occasionally, weak adjectives occur with other formally definite elements, such as demonstratives (see Sect. 4.3 below on possessives); as for sá + A.wk, see the discussion on pattern Dan (tables 3/5 and Sect. 3.2.2); as for þessi (“this”) + A.wk, I have found 13 instances in MÍM (so far). However, here it should be pointed out that demonstratives often do not occur with a bare weak adjective; quite often there is an additional art preceding the weak adjective (at least 100 instances in MÍM), cf. (40), and sometimes, demonstratives also occur with strong adjectives (at least 20 occurrences in MÍM), cf. (41):

1. (40)

   

1. (41)

   

This is a clear indication that demonstratives are not on par with art in OI, neither categorially nor functionally. On the other hand, in MI, we systematically find dem + A.wk instead of patterns like (40)/(41).

The specific relationship between (h)inn and weak adjectives is summed up by Stroh-Wollin (Reference Stroh-Wollin2009, 7): “(h)inn […] seems to be just a formal element preceding adjectives with so called weak inflection”; Börjars and Payne (Reference Börjars and Payne2016, 2) propose that “(h)inn […] is a functional element whose role it is to allow a weak adjective to function as an adj”. Taken together, all this suggests that bare weak adjectives, in some sense, are not “complete” adjectival constituents, and art can be understood as an adjectival complementizer, as it were. I will adopt the view that one function of art is to complement (or license) a weak adjective (see (24) in Sect 2.2) rendering it a “complete” AP (see also (64)). Conversely, on this same view, strong adjectives should be considered complete.Footnote ³² In the following subsections, I will present some observations that illustrate how strong adjectives and the sequence art + A.wk appear to behave as two varieties of the same category: AP.

4.2 Absolute superlatives and predicative APs

Pattern (II) noun phrases involving a superlative adjective often occur in predicative position yielding a so-called absolute/indefinite superlative interpretation:

1. (42)

   

Under certain cirumstances, the noun may be missing:Footnote ³³

1. (43)

   

In examples like these, the sequence art A.wk is interpreted as a predicate (of type <e,t>) – just as other APs in predicative position. Furthermore, this sequence can be coordinated with “proper” APs headed by a strongly inflected adjective:

1. (44)

   

In other words, interpretation suggests that the bare sequence art A.wk behaves like a predicate, and coordination suggests that it behaves like an AP headed by a strong adjective. But while an interpretation as a referential elliptic noun phrase is not an option for examples like (43) (∼ “he was the most angry one”), it does not yet follow that art A.wk does constitute an AP. In fact, one might argue that, strictly speaking, the above merely shows that the sequence can have a predicative semantic type on a par with strong APs. It does not rule out, as such, that that sequence is an elliptic predicative NP (also of type <e,t>). After all, predicative NP & AP coordinations are well-known (‘she is [a linguist] and [proud of it]’). In the next subsection, I will loook at another distributional criterion.

4.3 Complementary distribution

In MI, adjectives following a possessive must be weakly inflected (as in NIScan more generally), both in pre- and postnominal position, even in the absence of a definite article. Moreover, possessives cannot be followed by art:

1. (45)

   

In contrast, in OI/Old Norse, adjectives following a possessive are usually strongly inflected,Footnote ³⁴ both prenominally and postnominally, (46). Alternatively, a possessive precedes the sequence art + A.wk, (47):Footnote ³⁵

1. (46)

   

2. (47)

   

On the other hand, bare weak adjectives are highly exceptional in this context (the numbers are based on MÍM):

1. (48)

   

The generalization is that, in certain contexts, strongly inflected adjectives and the sequence art A.wk are in complementary distribution, but bare weak adjectives are excluded from the same position. Put slightly differently, there are two varieties of AP: [_AP A.str ] and [_AP art A.wk ], that occur in the context possessive + AP.

In a weaker sense, the argument from complementary distribution could also be made for demonstratives, cf. (40)/(41) (and possibly other contexts). Complementary distribution can be seen as a direct reflection of the idea suggested in Sect. 4.1, that bare weak adjectives – differently from strong adjectives – are not complete APs in OI and require art as an adjectival complementizer.

By the same token, of course, considering (45), this reasoning entails that bare weak adjectives are actually complete APs in MI, since they now occur in contexts where they were not found previously. It also follows that art is not a component of the AP any longer in MI (see Sect. 4.5 for another difference between OI and MI).

4.4 Occurrence per adjective

In some cases of adjectival coordination both adjectives are preceded by art:Footnote ³⁶

1. (49)

   

Examples like these are a strong indication that art is more closely associated with the adjective than the noun phrase as a whole. This view is also supported by the fact that these noun phrases denote exactly one referent; both adjectives modifiy the same noun and do not belong to different noun phrases. The same point can be made more clearly. As was mentioned above, in OI, art and the distal demonstrative sá frequently cooccur in the same noun phrase with art preceding the adjective:

1. (50)

   

Examples like these show that sá and art belong to distinct categories and have different distributions. This is particularly noticeable in adjectival coordination structures:

1. (51)

   

Here we find one sá per noun phrase, but one art per adjective, which iterates the point that art belongs more closely with the adjective rather than with the noun phrase. This in conjunction with the above observations strongly corroborates the idea that art forms a constituent with a weak adjective (= AP) to the exclusion of the noun and does not associate with the xNP directly.Footnote ³⁷

4.5 Numerals

When it comes to numerals, the present analysis has two options: either (a) OI treated numerals as adjectives, in which case we should expect the adjectival article in case of numerals as well, or (b) OI did not treat numerals as adjectives, in which case numerals should never intervene between art and the adjective.Footnote ³⁸ On the other hand, if art were in fact a determiner, we would expect numerals to intervene between the two (cf. Cinque Reference Cinque2005). Therefore, if (b) were to be confirmed, there would be a strong argument for the proposal that the art is in fact associated with AP rather than xNP.

The only genuine example of art + numeral that I have found is the following:

1. (52)

   

In IcePaHC and MÍM, there are some further instances of hinn + numeral, but in all those cases, hinn has to be interpreted as as a demonstrative “the-other” (hinir tveir “the other two”), not as an adjectival article. Thus, normally, numerals are not treated as adjectives in that they occur without art. More to the point, there are no indications that numerals (can) intervene between art and adjectives in OI.

In MI, in contrast, numerals do intervene between art and adjectives (see also Pfaff Reference Pfaff2015, 31,95):Footnote ³⁹

1. (53)

   

In other words, these observations further corroborate the present claim that art is narrowly associated with the adjective (= is a part of AP) to the exclusion of the noun in OI. Likewise, based on that criterion, by comparison, it follows that art has a different status in OI and MI in that it is no longer directly associated with the adjective in MI, but with the larger noun phrase (= xNP).

4.6 Excursus: Nominal APs? A tentative sketch of the pre-history

The picture that emerges for OI is that art is not a determiner in the nominal projection, but a component of AP, and, at least in the cases considered so far, the bare sequence art A.wk can be argued to be an AP – rather than a noun phrase. However, recall from Sect. 1.2 that the oldest attestations mostly involve proper names where the (postnominal) sequence (h)inn + A acts as an epithet:Footnote ⁴⁰

1. (54)

   

Epithets of this kind, occurring with a proper name can be argued to be nominal in character in that they have a name-like function/distribution themselves (like a surname). So for instance Rießler (Reference Rießler2016, 52/53), who discusses this type of epithets (“Frederick the Great”) involving an adjectival (in his terminology: attributive) article under the heading “attributive nominalization”, treats them as NPs.

Tentatively, I propose a double construal of the sequence art + A.wk as either AP or xNP,Footnote ⁴¹ and briefly sketch a preliminary diachronic analysis. Since this proposal is (by its very nature) speculative, I will not go into much detail. The logical steps are summarized in the following table and commented upon below:

1. (55)

   

1. 1. It has been proposed that IE did not have a seperate morpho-syntactic category “adjective”, but only one category “nominal” comprising what we might otherwise consider adjectives and nouns. Instead of attribution, IE employed close apposition where two nominals constitute one macro noun phrase. A distinct category “adjective” emerges as a result of the diachronic development from IE towards her daughter languages (e.g. Werner Reference Werner, Untermann and Brogyanyi1984; Viti Reference Viti2015; Adrados et al. Reference Adrados, Bernabé and Mendoza2016; see Rehn Reference Rehn2018 for discussion and further references).

2. 2. The IE daughter language Germanic does have a separate class “adjectives”. But simultaneously, another innovation takes place, viz. the establishment of a second adjectival paradigm, traditionally referred to as weak inflection. The precise origin and original function of the weak inflection are still a matter of debate. One approach suggests that it primarily had a nominalizing function (Osthoff Reference Osthoff1876; Kovari Reference Kovari1984). Although not uncontested (see Curme Reference Curme1910; Ringe Reference Ringe2006; Ratkus Reference Ratkus2011; Rehn Reference Rehn2018 for discussion), I will adopt this idea and interpret it in such a way as to mean that a weakly inflected adjective at the early Germanic stage essentially heads a nominal projection.

3. 3. If weak adjectives are indeed at the center of a nominal projection, it is not surprising that such a projection may also host a (presumably semantically weakened) demonstrative as an immediate component. Considering that the oldest attestations of the sequence (h)inn + A.wk are, in fact, appostions (epithets), we may, in a sense, view weak adjectives as a continuation of the IE mode of modification: apposition involving nominals. In other words, in Proto Norse, demonstrative (h)inn + A.wk constitute an appositive noun phrase.

4. 4. Now assume that, at some point, when occurring with common nouns rather than proper names, the appositive nominal projection was categorially re-analysed as AP. This reanalysis at the phrasal level may have become an option because the sequence art + A.wk and strong adjectives increasingly occurred in identical contexts. It may also have been enforced by the facts that both used the same (= adjectival) stems. At the same time, (h)inn had largely lost its demonstrative force and its function as a determiner element in that context – but adopted a new function, see Sect. 4.1.

If the development of weak adjectives and (h)inn did roughly proceed along these lines, it would explain how an element that started out as an element of a nominal projection (demonstrative) ended up as an adjectival article “trapped” in an adjectival projection. It also suggests that the nominal construal of (h)inn + A.wk is the older one, and the construal as AP is an innovation; still, both construals coexisted at least throughout the OI period.

5.1 Prosodic unit reanalyzed as morphosyntactic unit

Taking the schematic (12) as a point of departure, (h)inn, is a demonstrative in stage A, but in stage B, it represents the adjectival article art (see previous section):

1. (56)

   

At some point, constellation B gives rise to cliticization to the effect that the noun and the reduced version of art form a prosodic unit, cf. examples (9)/(10):

1. (57)

   

Since this step is understood here as a purely phonological process, the article element, albeit now manifesting itself as def, is still construed as an adjectival article in AP, not an element of the nominal projection. If we follow the idea that the suffixed article developed from a demonstrative, see Sect. 1.2, we have to assume a similar intermediate stage where a prosodically (and presumably semantically) weakened demonstrative forms a merely phonological unit with the preceding noun:

1. (58)

   

Glossing over the question whether C or C’ is the relevant input constellation, the decisive step occurs now: at some point, a prosodic unit gives rise to a reanalysis as a morpho-syntactic unit:

1. (59)

   

Only this step indicates the emergence of a new element: a “nominal” article. This nominal article forms with the noun – and nothing else – a syntactic constituent, and thus has scope only over the noun. Following Lohndal (Reference Lohndal, Abraham, Leiss and Stark2007) (see Sect. 1.3), I will refer to that constituent as nP.

Moreover, I will follow Lohndal in assuming that this projection did not exist in the language prior to reanalysis. Recall, however, that, according to Lohndal, this process took place in the latter part of the Old Norse period (late 13th/early 14th century). I will not attempt to give a precise date, but I will pre-date the terminus ante quem. Given the oldest attestation of a bare definite noun, cf. (11), I suggest that reanalysis and the emergence of the nominal article/nP must have taken place, at the latest, during the 11th century, at any rate before the Old Icelandic period. This is still compatible with Stroh-Wollin (2016) who suggests that it took place around the year 1000 (but not necessarily with Skrzypek (2009, 2010) who argues that the formation of the suffixed article must have happened before 800).

5.2 Modified and “bare” definite noun phrases

Recall the minimal pairs (32) and (33) repeated below:

1. (60)

   

2. (61)

   

Based on alternations like these I argued that the article element involved in both is the same, viz. the adjectival article. However, given reanalysis and the emergence of a nominal article as suggested in the previous subsection, we are now faced with an ambiguity: the item -inn, glossed as def, can be (i) a phonological variant of the adjectival article in the context of pattern (III-a), or (ii) the nominal article; that is it can be either an element of AP or of nP. I suggest that reanalysis did not simply replace the construal of the suffixed article of stage C as an adjectival article and that the construal as a prosodic unit of stage C must have been available for a certain period even after reanalysis. What kept the construal of def as an adjectival article in pattern (III-a) alive presumably had to do with the existence of pattern (III-b), and (III-a)/(III-b) alternations like (60)/(61). Notice that this is independent of whether the nominal article derived from art or from the demonstrative, i.e. independent of whether it was C or C’ that gave rise to reanalysis D. In the latter case, it would mean that it is due to an etymological coincidence (art itself also derives from the same demonstrative) that the cliticized form of art and the nominal article appear to be identical regarding stem (-in-) and inflection. In short, according to my analysis, two distinct items, the adjectival and the nominal article, can have the same morpho-phonological shape and both be glossed as def, while (h)inn, glossed as art, is unambiguously the adjectival article. Conversely, according to the analyses by Roehrs and Sapp Reference Roehrs, Sapp, Martínez, Alcázar and Hernández2004; Faarlund Reference Faarlund2004, Reference Faarlund2007, Reference Faarlund2009; Lohndal Reference Lohndal, Abraham, Leiss and Stark2007; Laake Reference Laake2007 (see also Sect. 1.3), there is one lexical item (the definite article) that is ambiguous between two morpho-phonological shapes (free and suffixed). This seemingly rather subtle difference between analyses has concrete ramifications when it comes to the relationship between article and adjective.

It has been pointed out that the freestanding adjectival article (art) only occurs with a weak adjective. On the present analysis, this is captured by the assumption that it is in fact an element of AP. On the other hand, Faarlund (Reference Faarlund2009, 629) proposes that there is a selectional restriction in the freestanding article art requiring it to merge with an αP (a functional projection hosting an AP in its specifier), which is another way of capturing the fact that it only occurs with adjectives. But if art has this kind of selectional restriction and if art and def are identical (which is the case on Faarlund’s account), then the suffixed article def would be expected to also select an adjective, and thus to only occur as pattern (III-a). Alternations like (60)/(61) would indeed be captured by this approach, if there were nothing more. However, the fact that “bare” definite nouns (N-def) are already widely attested in the oldest texts shows that the suffixed article is not as such dependent on an adjective; in other words, every attested bare definite noun is in violation of this selectional restriction.

On the other hand, a distinction between adjectival and nominal article along the lines sketched in the previous subsection allows us to preserve the conclusions drawn in Sect. 3.1.2 about the identity of (60) and (61) (both involve an adjectival article), while simultaneously accounting for the fact that not every suffixed article requires the presence of an adjective: the adjectival article manifesting itself as art or def categorically requires the presence of an adjective, the nominal article only occurring as def merely requires the presence of a noun.

On a related note: in Sect. 1.2, I pointed out that the suffixed article usually does not occur on proper names, but also showed that there appear to be exceptions, cf. (10). With a categorial distinction between adjectival and nominal article, rather than a mere morpho-phonological one between art and def, I can reduce the “exceptionality”: only the adjectival article can occur cliticized to a name, not the nominal article.

5.3 Double Definiteness

Consider the following examples repeated from (34):

1. (62)

   

Compare (62) with (60)/(61): while art is obviously the same element in both (62) and (61), this is not obviously the case for def in (62) and (60). On the present approach, the adjective occurs with an adjectival article in all three cases, which can be art: (62) and (61), or def: (60). In addition, independently of the adjective, the noun may carry its own nominal article, which only occurs as def: (62). This reiterates the point from the previous subsection that def allows for two construals, but it also highlights another aspect: double definiteness.

One obvious consequence of reanalysis is that, now, there are two distinct elements occupying two distinct positions: nominal article ≠ adjectival article, and their cooccurrence is not ruled out on structural grounds. This fits with the actual chronology. As was shown in Sect. 3.1.3, the oldest attestations of pattern DD-a are from the 12th century, and no other cases of double definiteness are attested before that. Double definiteness became an option only after reanalysis had taken place (given the time frame that I suggested in Sect. 5.1: 11th century).

5.4 Adjectival inflection and the lack of pattern (I)

In Sect. 2.2, I briefly discussed the idea by Pfaff (Reference Pfaff2017) that the weak inflection is licensed iff the adjective is c-commanded by a definiteness feature in MI:

1. (63)

   

By assumption, the OI nominal article argued for above also carries some definiteness feature simply by virtue of “making the noun definite” (pre-theoretically speaking). On the current conception, it has scope only over the noun and nothing else. This entails, in particular, that it lacks the capacity to “make an adjective definite”. If we equate making an adjective definite with licensing the weak inflection, the assumed badness of pattern (I), as evidenced by virtual non-attestation, can be captured by the same mechanism that governs the distribution of weak adjectives in MI: it follows from the fact that the OI nominal article is merged in such a low position that it does not c-command adnominal adjectives.Footnote ⁴² This idea converges with the one suggested in Sects. 4.1/4.3, viz. that bare weak adjectives are not complete APs (notated below as “weakP”), but require a complementizer (= art):

1. (64)

   

In (64-a), the weak inflection is licensed by the adjectival article/complemen-tizer (and weakP is thus rendered a complete AP) – irrespective of whether the nominal article is present: DD-a, or not: pattern (II). On the other hand, (64-b), which is a pattern (I) constellation, is an illicit structure. It involves a bare weak adjective, i.e. an “incomplete” AP without its complementizer. Recall that other definite determiners like demonstratives are, at least potentially, alternative licensors of the weak inflection (Sect. 4.1). Now the nominal article does have a definiteness feature, so at least in principle, it could also be an alternative licensor of the weak inflection. However, by assumption, the nominal article is merged too low in the structure and its definiteness feature does not c-command the adjective – thus the the crucial structural pre-condition is violated.Footnote ⁴³

Hence, not only does the assumption of a distinct nominal article capture the occurrence of bare definite nouns and the possibility of double definiteness marking, a specific construal as a low n⁰ can account for the absence of pattern (I) in OI.

6.1 From adjectival complementizer to determiner

Development (iii) follows from the construals of art for OI and MI; if they are correct, this categorial and structural change must have taken place. It is more difficult to ascertain how it proceeded, but one factor that is likely to have influenced it considerably is the competition between art and sá (see Sect. 3.2).

Now consider the following Old Danish example (early 11th century):

1. (65)

   

This example illustrates, on the one hand, that sá occurs as an article element without demonstrative function, and on the other hand, that it can be coordinated with art. Notably, coordination suggests that the two have the same categorial status, which, in turn, indicates that, in Old East Norse (= Old Swedish & Old Danish), the distal demonstrative sá was originally developing into a genuine adjectival article, i.e. an adjectival complementizer in AP, on a par with (h)inn. Evidently, there was direct competion between sá and art early on. Indeed, in Old East Norse, sá very soon became the dominant adjectival article while art disappeared. As for the Modern NIScan languages, on the other hand, there is broad consensus that the adjectival article det is an actual determiner (e.g. Vangsnes Reference Vangsnes1999; Julien Reference Julien2005; Roehrs Reference Roehrs2009). In other words, we must assume that sá has undergone further developments on its way to modern det in the NIScan languages (see Stroh-Wollin Reference Stroh-Wollin2009, Reference Stroh-Wollin2015a on (Old) Swedish) thus mirroring development (iii) for Icelandic art.

Neither IcePaHC nor MÍM contain instances of sá A & art A coordination, and considering their categorial distinctness noted throughout Sect. 4, there do not seem to be any hints at a competition between sá and art during the OI period. First indications of competition between the two can be found in the 16th century. In Sect. 3.2.3, I suggested that the competition was an indication that sá was developing into an adjectival article, a replay of what had happened some centuries earlier in Old East Norse, cf. (65). That conclusion is not, however, inevitable; the development Icelandic sá underwent during the period under consideration probably was not towards a genuine adjectival article (= complementizer) like Viking period (h)inn and Old East Norse sá. None of the criteria discussed in Sect. 4 – complementary distribution with strong adjectives, coordination, occurrence per adjective etc. – apply here. Rather it seems that sá was developing towards a determiner element more akin to modern det.Footnote ⁴⁵ On this view, the competition between sá and art also allows another interpretation: it was not was not exclusively about the development of sá, but also indicates that art was developing from an adjectival complementizer into a determiner element.

6.2 Icelandic and NIScan

One difference between Icelandic and NIScan already pointed out is the survival vs. disappearance of hinn (art). But there is another aspect that concerns the status of def. Many analyses of the Modern Scandinavian noun phrase (e.g. Vangsnes Reference Vangsnes1999; Julien Reference Julien2005; Roehrs Reference Roehrs2009) assume that the suffixed article occupies an X⁰ position below adjectives. If my analysis of the nominal article (Sect. 5) is on the right track, this would mean that def has remained in a low position in NIScan, but moved up the tree in Icelandic (= development (ii), see above).

But this actually raises an interesting question. Recall that Faarlund and Lohndal propose grammaticalization down the tree in order to capture the development of def from clitic to affix in Norwegian, see Sect. 1.3, while the present analysis suggests movement up the tree in Icelandic. Is this merely a different implementation that essentially captures the same development, viz. the well-known grammaticalization cline clitic > affix, or does it indicate that the Icelandic development is qualitatively different? Faarlund (2009) makes a strong case that def is an inflectional suffix in Modern Norwegian and Swedish, but the arguments do not easily apply to Modern Icelandic.Footnote ⁴⁶ While def as such is a bound element, it is not simply a suffix. Unlike proper (inflectional or derivational) suffixes, for instance, it does not trigger so-called U-umlaut (i.e. [a] → [ö] / [σ__].[σ [ y]]; see Indriðason Reference Indriðason1986, Reference Indriðason1994; Pfaff Reference Pfaff2007, Reference Pfaff2009 on that point). Moreover, in contrast to a simple inflectional affix, Icelandic def can itself be segmented into stem + inflection:

1. (66)

   

1. (67)

   

Also the relationship between patterns (II) and (III) and adjective ordering in mixed (I+III) patterns (see Sect. 2.1) show that def participates in word order phenomena. Taken together, these points suggest that MI def, in spite of being a bound element, behaves more like an element of syntax (∼clitic), not of morphology (inflectional affix). If this is correct, and given that the points illustrated by (66) and (67) have applied all along, it could simply be the case that, differently from Norwegian and Swedish, Icelandic def has never become an affix in the first place (which is also why I have refrained from talking about “grammaticalization up the tree”).