Referential vs. non-referential null subjects in Middle Norwegian

Abstract

This paper investigates the relationship between referential and non-referential null subjects in Middle Norwegian. It argues that overt, non-referential subject det arose before the loss of referential null subjects, contrary to the predictions of much previous work. A diachronic analysis compatible with the empirical findings is sketched out.

1. INTRODUCTION

Like its contemporary Scandinavian sister languages, Old Norwegian allowed null (i.e. omitted) subjects. Non-referential subjects (NonRefSs) were always null, as in example (1). ¹

1. (1)

   

Referential subjects (RefSs) were mostly overt but could also be null, as in example (2).

1. (2)

   

In Modern Norwegian, as in Swedish and Danish, both RefSs and NonRefSs must generally be overt, as illustrated in example (3). In most varieties of Norwegian, overt NonRefSs have the form det ‘it’.

1. (3)

   

Null subjects in the history of Scandinavian have received an increased level of attention in recent years; see Sigurðsson (1993), Falk (1992, 1993a, 1993b), Håkansson (2008, 2013), Kinn (2010, 2011, 2014, 2016a, 2016b), Rosenkvist (2009), Faarlund (1990, 2012, 2013), Heltoft (2012), Walkden (2014), Lander & Haegeman (2014) and Kinn et al. (2016). ² However, the previous studies mostly focus on either RefSs or NonRefSs; they do not consider the relationship between the two subject types. When RefSs and NonRefSs are discussed in connection, some version of what I will refer to as the prerequisite hypothesis is mostly assumed, implicitly or explicitly. The prerequisite hypothesis can be stated as follows:

1. (4) The prerequisite hypothesis: The loss of null RefSs was a prerequisite for the rise of overt NonRefSs.

Although it is often adopted, the prerequisite hypothesis is not founded on empirical studies of data from the period in which Norwegian (as well as the other Scandinavian languages) changed into non-null-subject languages (non-NSLs). The main aim of this paper is, therefore, to test the prerequisite hypothesis against data from the Middle Norwegian period (ca. 1370–1550). I will argue that the prerequisite hypothesis, at least in a strong version (see Section 2), is not corroborated by evidence from Middle Norwegian. I will also sketch a new diachronic analysis of the loss of null subjects in Norwegian that is compatible with the empirical findings. The study is conducted within a generative framework, but in addition to theoretical issues, much attention will be devoted to the empirical results.

The paper is organised as follows: in Section 2, I discuss the prerequisite hypothesis in more detail. In Section 3, I present relevant Middle Norwegian data and show that they are not compatible with a strong version of the prerequisite hypothesis. In Section 4, I discuss a weak version of the hypothesis. In Section 5, I sketch out a diachronic analysis of the loss of null subjects in Norwegian. Section 6 concludes the paper.

2. THE PREREQUISITE HYPOTHESIS

The prerequisite hypothesis can be found in various works and is arrived upon on different grounds. In this section, I will provide some general remarks and clarifications about the hypothesis, before I discuss how it is instantiated in Rizzi (1982), Håkansson (2008) and Faarlund (2012, 2013).

2.1 Some general remarks and clarifications

The prerequisite hypothesis, as stated in (4), is phrased in descriptive terms, but is consistent with central ideas of Principles and Parameters theory (P&P). More precisely, it can be derived from one of the implicational relationships stated in the original formulation of the null subject parameter (Rizzi 1982), ³ namely that the presence of null RefSs implies the presence of null NonRefSs (more on this in Section 2.2). The idea of an implicational relationship between null RefSs and null NonRefSs has been widely accepted, also in recent works (see e.g. Roberts & Holmberg 2010:12), and it is corroborated by cross-linguistic empirical evidence. There is a very strong tendency for languages allowing null RefSs to also have null NonRefSs; Gilligan (1987) concludes that the implication seems to hold in a sample of more than 100 languages. In a more recent study, Nicolis (2008:291), who considers data from a range of Creole languages, states that the implication appears to be of an “absolute” nature.

It should be noted, though, that some (potential) counter-examples have been observed in the literature. One case is Övdalian, which allows null RefSs in the first and second person plural, but requires NonRefSs to be overtly realised (Rosenkvist 2010:236). Other cases are e.g. Old High German, Finnish and Dominican Spanish. ⁴ In the latter languages, overt and null NonRefSs co-exist (Axel 2007:295, 303, Hinzelin & Kaiser 2007, Nicolis 2008:276–277). Whether null subject systems of this type are actually counter-examples to the implicational relationship between null RefSs and null NonRefSs depends on whether one adopts a strong or a weak version of the implication. In what I will refer to as the strong version, a language that allows null RefSs will require null NonRefs; there is, in other words, no room for overt NonRefs. In what I will call the weak version, languages with null RefSs will allow null NonRefSs, but not necessarily require them; there is, in other words, room for some overt NonRefSs. Nicolis (2008) adopts the weak version; he is thus able to reconcile Finnish and Dominican Spanish data with the implication. ⁵ Camacho (2013:42), on the other hand, operates with a stronger version and takes data of the Old High German/Finnish/Dominican Spanish type to indicate that the implication between null RefSs and null NonRefSs does not hold universally.

In this paper, my point of departure will be the strong version of the implicational relationship between null RefSs and null NonRefSs, and, accordingly, a strong version of the prerequisite hypothesis. The claim to be tested is thus the following: a language will not only allow but will also require null NonRefSs as long as null RefSs are licit; consequently, no overt NonRefSs are found. There are two reasons why I use the strong version of the hypothesis as my starting point. First, the strong version is the one that most immediately follows from the works to be discussed in the following sections, in particular from Rizzi (1982). Second, from a heuristic point of view, it is preferable start by testing the strongest prediction; the weaker version can then be taken into consideration if the strong version turns out not to hold. Note, however, that even a strong version of the prerequisite hypothesis is compatible with certain “subject-like”, non-referential elements. Crucially, though, such elements must have some identifiable linguistic function apart from that of being a subject (from the viewpoint of mainstream generative grammar, they must do something more than value the uninterpretable φ-features of T). Vietnamese has a non-referential element nó that has been argued to encode specificity (Greco et al. 2015). Non-referential ele in non-standard European Portuguese has been argued to be a type of discourse marker emphasising the illocutionary force of a sentence (Carrilho 2008). Elements like nó and ele do not conflict with the implication between null RefSs and null NonRefSs if we can safely assume that they are more than, or perhaps something else than, plain NonRefSs.

The distinction between the strong and the weak version of the prerequisite hypothesis concerns the predictions made for any given language synchronically. Before we move on, one point should be clarified with regard to diachronic predictions: the prerequisite hypothesis does not state that overt NonRefSs necessarily arise in any language that loses null RefSs. However, if overt NonRefSs arise in a language, we do not expect that language to be one that (still) allows null RefSs.

We now turn to the works of Rizzi (1982), Håkansson (2008) and Faarlund (2012, 2013) to see how the prerequisite hypothesis is instantiated there.

2.2 Rizzi (1982)

Rizzi (1982) does not discuss Scandinavian in detail but is relevant by virtue of being one of the most influential works in early P&P theory. Rizzi (1982) put forth the null subject parameter, whereby the null subject properties of a language are determined by the settings of two related parametric options, rendered in (5) (see Rizzi 1982:143):

1. (5)

   1. a. INFL can be specified [+ pronoun].

   2. b. INFL can be referential.

INFL is, in Rizzi’s (1982) framework, subject-verb agreement. For any subject to be null, Ref or NonRef, INFL must be specified as [+ pronoun]. INFL being specified as [+ pronoun] is a sufficient condition for null NonRefSs; null RefSs, on the other hand, additionally require a positive setting of the parameter in (5b), stating that INFL can be referential. ⁶ It follows that a language may have null NonRefSs but require RefSs to be overt; Rizzi (1982:143–144) mentions Icelandic, a variety of Dutch, and certain northern Italian dialects, e.g. Paduan, as examples of this. ⁷ Cf. example (6), from Paduan (Rizzi 1982:143):

1. (6)

   

The opposite pattern, whereby RefSs can be null, but NonRefS are overt, is “excluded for intrinsic reasons (if an inflection cannot be pronominal, it cannot be referential either)” (Rizzi 1982:143). The typological prediction of Rizzi (1982) is thus clear, but it also entails a diachronic dimension: we would not expect a language to develop overt NonRefSs while still maintaining null RefSs. If null RefSs are licensed, the language should also be capable of licensing null NonRefSs. Overt NonRefSs would be unnecessary, and Rizzi (1982) offers no way of reconciling them with null RefSs.

2.3 Håkansson (2008)

Håkansson’s (2008) study focuses on null RefSs in Old Swedish; it also proposes a diachronic account of how null subjects were lost. In Håkansson’s analysis, the loss of null subjects followed from a change in the status of Spec-TP (Spec-IP in his terminology). ⁸ According to Håkansson, Spec-TP changed from being an A'-position in Old Swedish to an A-position reserved for the subject at the modern Swedish stage. I will not discuss the details of the analysis here; the important point in the present context is the following: subjects, according to Håkansson (2008), can only be omitted from A'-positions. Thus, the change of Spec-TP from an A' to an A-position not only implied obligatory movement of the subject from its base-generated position in the V-domain to Spec-TP, but it also implied that subjects could no longer be deleted from Spec-TP. ⁹ Håkansson (2008:225) proposes a scenario in which the transition of Spec-TP into a pure subject position happened in three stages, summarised in (7):

1. (7)

   1. 1. Subjects of transitive verbs began to move obligatorily to Spec-TP. ¹⁰

   2. 2. The rule in 1 was generalised to all RefSs, causing movement of all RefSs from their base-generated, VP-internal position.

   3. 3. The rule in 2 was further extended in the sense that overt NonRefSs were generated in Spec-TP. This is the situation in Modern Swedish.

As is evident, overt NonRefSs do not enter the scene until the last stage, after the changes regarding RefSs are completed. Obligatorily overt RefSs can thus be seen as a prerequisite for the rise of overt NonRefSs. According to Håkansson (2008:225), the model in (7) can also be applied to Modern Icelandic: Modern Icelandic has only completed the two first stages, which is why RefSs are obligatorily overt, whereas NonRefSs are not.

2.4 Faarlund (2012, 2013)

Faarlund (2012, 2013) discusses null arguments in the history of Scandinavian and other languages. He treats null arguments as inherently silent pronouns whose presence or absence in a language is independent of the functional categories of the clause; the (im)possibility of null arguments only reflects properties of the pronominal system itself. ¹¹ Faarlund adheres to common generative terminology in calling silent pronouns pro, but as opposed to Rizzi (1982, 1986) and much subsequent work, he does not assume any particular syntactic licensing conditions (like pronominal INFL).

According to Faarlund, pro is a lexical item that must be robustly represented in the primary linguistic data (PLD) to be acquired. If the frequency of pro drops below a certain threshold, it will not be picked up by children during language acquisition, and a new grammar with only overt pronouns will emerge. In Mainland Scandinavian, Faarlund argues that the frequency of pro started to drop because of the weakening of subject-verb agreement morphology: although pro was not syntactically conditioned by subject-verb agreement, the agreement morphology did promote recoverability of pro on a pragmatic level. When the agreement morphology was weakened, overt pronouns became a more preferable option for communicative reasons. This made the frequency of pro drop below the critical level, which in turn led to its loss.

Faarlund (2012, 2013) does not propose a stepwise loss of null subjects; in his account, both referential and non-referential pro were lost simultaneously in the history of Scandinavian. Thus, according to Faarlund, the loss of null RefSs was not directly a condition for the loss of null NonRefSs. Still, Faarlund’s works are relevant in the context of the prerequisite hypothesis: the frequency changes that triggered the loss of all null subjects must have started with null RefSs. This follows from the fact that null NonRefSs in Old Norwegian did not have overt counterparts. When overt NonRefSs were not a part of the lexicon, speakers cannot have chosen overt NonRefSs simply as a matter of preference, for communicative reasons.

In terms of variation between overt and null subjects, the predictions made by Faarlund’s account are, in practice, the same as those made by Rizzi (1982) and Håkansson (2008). In a text produced by a single grammar, we may expect to find null NonRefSs, although there are no null RefSs. This type of output is produced when pro is still a part of the lexicon, but the speaker opts for overt pronouns when these are readily available, i.e. in the referential contexts. We do not expect overt NonRefSs to co-occur with null RefSs. Det ‘it’ is not reanalysed as an overt NonRef until pro is gone (Faarlund 2013:272), and when pro is gone, null RefSs are no longer possible.

Having shown how (versions of) the prerequisite hypothesis is instantiated in some works on null subjects, I will now discuss whether it captures the situation that can be observed in Middle Norwegian.

3. EVIDENCE FROM MIDDLE NORWEGIAN

3.1 Middle Norwegian data

Middle Norwegian (ca. 1370–1550) is, as previously mentioned, the language stage between Old and Modern Norwegian. The syntax of Middle Norwegian in many ways resembles that of Old Norwegian (Mørck 2013:669): it is V2 in the sense that the finite verb seems to move to C in main clauses. Like in Old Norwegian, verb-initial declarative main clauses are rather common. In subordinate clauses, however, there is variation between the Old Norwegian word order, which in most cases means V-to-T movement (Faarlund 2004), and the Modern Norwegian word order, in which the verb is commonly analysed as in situ (Vikner 1995:140, Vittersø 2004).

I have investigated a sample of 136 Middle Norwegian charters from the Diplomatarium Norvegicum (DN). The charters date from the period 1371–1548. The documents in my sample exhibit a low degree of influence from Swedish and Danish, and many of them have a high proportion of narrative contents in addition to the usual introductory and concluding formulas (the protocol and the eschatocol). I take the view that formulaic contexts must not necessarily be disregarded as linguistic evidence (see Grøtvedt 1961:225 and Mørck 2011:41); however, none of the examples discussed in this paper are formulas.

In most cases, we can safely assume that a charter is the work of one individual scribe (Wetås 2008:89). It can be rightfully asked, however, whether this means that the charter is a reflection of a single grammar, or I-language (Chomsky 1986). The question is important, as the aim of this study is to investigate the compatibility, or incompatibility, of null RefSs and overt NonRefSs within individual grammars.

Following Kroch (1989), many works in historical syntax have adopted the idea that a single speaker may have two or more competing grammars, which can all underlie the production of one individual text. I adhere to the view that speakers may have more than one grammar, and that this may result in intra-textual syntactic variation. As Roberts (2007:331) and Walkden (2014:24–25) point out, however, it is problematic to invoke grammar competition without identifying any functional motivation for switching between grammars. It is not random whether a bilingual speaker in a given context chooses to use, say, German or French. The competition between two grammars that are more similar, like an Old English grammar with a head-initial vs. a head-final TP (see e.g. Pintzuck 2002), is not fundamentally different; we would expect the choice of grammar in every given context to be guided by stylistic, sociolinguistic or other functional factors. The charters that I have investigated are, mostly, short documents, rarely exceeding a couple of pages, treating one specific matter. Abstracting away from the question of formulas, I find it hard to justify the idea of competing grammars being at work in such contexts, unless there is specific and independent evidence of this. ¹² My default assumption is thus that all syntactic phenomena occurring within a single charter can be derived by a single grammar.

With a few exceptions, I have only considered variation between overt and null subjects within individual charters. This leaves us with a rather limited set of relevant data, as the documents must exhibit a very specific combination of properties to qualify as evidence: not only must they contain a null RefS, which is a rather low-frequency phenomenon in Middle Norwegian (see Mørck 1990 and Kinn 2016a, 2016b); they must also exhibit a syntactic environment in which an overt NonRefS can potentially occur. However, although demanding much of the data, the method has an important advantage: it minimises the risk of attributing the properties of different varieties/idiolects to a single I-language.

3.2 Co-occurrences of overt, non-referential subjects and referential null subjects

My data set contains 33 instances of (the precursor of) det ‘it’, often spelled thet, that I have analysed as overt NonRefSs. ¹³ Contrary to the predictions of the strong prerequisite hypothesis, I have found that NonRef det does co-occur with null RefSs. Due to the nature of my data and methodology, the instances are too few to make meaningful statistical generalisations, but as existential evidence, my findings should be clear enough. ¹⁴ Some relevant data are rendered in the example pairs in (8)–(12). The examples labelled a contain null RefSs, whereas the ones labelled b contain overt NonRefS. The examples in each pair are taken from the same individual charter. Consider, first, example (8):

1. (8)

   

The null RefS in example (8a) is found in a subordinate clause; this shows that we are not dealing with topic drop/discourse ellipsis of the type found in Modern Norwegian (see note 1). The construction with det (spelt þet) will be discussed in more detail in Section 3.3; observe, however, that þet occurs in a post-verbal position. This means that it cannot be analysed as an expletive topic of the type found in Modern Icelandic, i.e. as an element that can only fill the preverbal position of declarative main clauses (see Section 3.3.1, Faarlund 1990 and Thráinsson 2007:312).

In example (9a), the null RefS occurs in a main clause; ¹⁵ the overt NonRef pronoun thet in (9b) is found in a conditional subordinate clause:

1. (9)

   

In example (10), the null RefS occurs in a main clause. The overt NonRef thet is found in an impersonal passive construction:

1. (10)

   

In example (11a), the null RefS is again found in a main clause, as is the NonRef thet in (11b). ^{16 17}

1. (11)

   

Consider, finally, example (12): ¹⁸

1. (12)

   

The null RefS in example (12a) is found in a main clause. The preverbal position is occupied by a conditional subordinate clause; again, we see that null RefSs in Middle Norwegian do not obey the distributional restrictions that apply to the modern discourse ellipsis. ¹⁹ The overt NonRef thet in (12b) is found in the postverbal position of a particular type of existential construction (Faarlund et al. 1997:1093). The construction bears resemblance to a cleft in that it involves a subordinate clause introduced by som ‘that, who’. Unlike prototypical clefts, however, the function of the construction is not to pick out and focus a referent; rather, it states (or, in this case, hypothesises) the existence of somebody. The existential nature of the sentence is underpinned by the fact that it is most naturally translated into English with there, not it. ²⁰

Having presented evidence that null RefSs and overt NonRefSs co-existed in Middle Norwegian, contrary to the strong version of the prerequisite hypothesis, I will now discuss the properties of overt NonRefSs in Middle Norwegian in some more detail.

3.3 NonRefSs with special functions?

A way of reconciling overt NonRef det in Middle Norwegian with the strong version of the prerequisite hypothesis could be to assume that NonRef det ‘it’ was not a “pure” non-referential subject before the loss of null RefSs, but rather had some other function (see Section 2.1). In this section, I will discuss that possibility. In the literature on overt NonRefSs in Scandinavian, relevant proposals have been put forth by Faarlund (1990) and Falk (1993a, 1993b): Faarlund claims that NonRef det ‘it’ was initially an expletive topic, while Falk suggests that it was initially a quasi-argument only and later also a marker of the existential construction. I will argue that none of the proposals can fully account for the situation in Middle Norwegian and that the status of the prerequisite hypothesis thus remains dubious.

3.3.1 Expletive topics

Faarlund (1990) suggests that overt NonRefSs started out as expletive topics of the type found in Modern Icelandic. Expletive topics have the function of satisfying the V2 rule, and they can, as previously mentioned, only occur in the preverbal position (Spec-CP) of declarative main clauses. Cf. the Modern Icelandic example (13) (Faarlund 1990:190):

1. (13)

   

Faarlund’s account predicts the earliest instances of NonRef det appear in Spec-CP. However, his proposal is only backed up with data from the end of the 15th century onwards, mostly written by Danish authors; Faarlund has not investigated Middle Norwegian. We have already seen that NonRef det in Middle Norwegian is not restricted to Spec-CP in main clauses (cf. again examples (8b), and (12b)). The study of Kinn (2010) indicates that these cases are not exceptional: in Kinn (2010), all instances of NonRef det from the 15th and 16th century are found either post-verbally in main clauses, or in subordinate clauses (Kinn 2010:110). Empirical evidence thus indicates that Faarlund’s (1990) proposal is on the wrong track: overt NonRefSs did not start out as expletive topics, and co-occurences of overt NonRef det and null RefSs cannot be ascribed to the presence of this grammatical element.

3.3.2 Quasi-arguments

Falk (1993a, 1993b) distinguishes between quasi-argumental and expletive NonRefSs and proposes that NonRef det in Swedish was used for quasi-arguments before it took on the function of a general NonRefS. ²¹ Quasi-arguments are typically subjects of meteorological, or “atmospheric”, expressions (Rizzi 1986) but can also occur with predicates that denote what Faarlund et al. (1997:679) call “a situation or event that is perceived”, or “states or processes” (Falk 1993a:80). Although being NonRef, quasi-arguments are assigned an external theta-role, a property they share with RefSs (see Chomsky 1981:323ff, Rizzi 1986, Bennis 1986, Vikner 1995:224). In Falk’s analysis, their “dual” status is the reason why quasi-arguments could be expressed overtly at a particularly early stage: if a quasi-argument was interpreted as being basically NonRef, it would be null. However, if it was interpreted as being basically argumental, like referential subjects, it could be overt (Falk 1993a:236–237).

Interestingly, many of the early occurrences of overt NonRefSs in Middle Norwegian are quasi-arguments. This is the case in examples (8b) and (9b), which instantiate what I will refer to as the det så construction. For convenience, I repeat the relevant parts of example (8b) in (14):

1. (14)

   

The det så construction contains a non-referential det (þet in 14), the pro-adjective så ‘so/like this’ (spelt sua in (14) and a subordinate clause, often introduced by the complementiser at ‘that’ (ad in (14). The pro-adjective så is appositionally related to the subordinate clause; Falk (1993a:211), describing the same construction in Swedish, writes that “. . . så receives its referential content from the following att-clause.” så and the subordinate clause constitute a predicate that denotes “a situation or event that is perceived” (Faarlund et al. 1997:679), and I take it that this type of predicate can assign a quasi-theta role. In the period up until 1450, I have found six instances of NonRef det, all of which occur in the det så construction. ²² This suggests that not just any quasi-argument could be overt at the earliest stage; I return to this issue in Section 3.3.3.

Det (thet) is also clearly a quasi-argument in example (11b), repeated below in (15), which is a meteorological expression.

1. (15)

   

It is worth noting that the charter containing the overt quasi-argument in (15) also exhibits a potential instance of a null expletive. ²³

Cf. example (16):

1. (16)

   

The analysis of example (16) depends on the status of the oblique argument tik ‘you’. According to e.g. Rögnvaldsson (1995), Haugan (1998), Barðdal & Eythórsson (2003) and Eythórsson & Barðdal (2005), Old Norse had oblique subjects; if that assumption is correct and can be extended to Middle Norwegian, tik is a subject, and there is no null expletive in (16). However, Mørck (1994) and Faarlund (2001, 2004:194–195, n. 1) argue that Old Norse and Middle Norwegian did not have oblique subjects. If Mørck and Faarlund are correct, it may be argued that an expletive pro is present in (16), as illustrated in (17) (see e.g. Faarlund 2013).

1. (17)

   

A less ambiguous case of an overt quasi-argument co-occurring with a null expletive is found in example (18). ²⁴ (This charter does not additionally exhibit any null RefSs.)

1. (18)

   

Co-occurrences of overt quasi-arguments and null expletives within the same, individual charter can be taken to suggest that quasi-arguments were overtly realised earlier than expletives. The argument is corroborated by the fact that I have not found any instances of the opposite pattern, whereby overt expletives would co-occur with null quasi-arguments. However, in the context of the prerequisite hypothesis, the important question is whether quasi-arguments are the only overt NonRefSs that co-occur with null RefSs. This is not the case; recall from Section 3.2 that overt NonRefSs are also found in impersonal passives and existential constructions, while RefSs within the same individual documents are still null. For convenience, I repeat example (10b), which is an impersonal passive, and example (12), which is a type of existential, in (19):

1. (19)

   

To sum up this section, we have seen that quasi-arguments are expressed overtly at a particularly early stage in the history of Norwegian. However, in the context of the prerequisite hypothesis, the crucial question is whether quasi-arguments are the only overt NonRefSs that co-occur with null RefSs. This is not the case; we have seen that overt expletives too are found in charters exhibiting null RefSs. Thus, although the distinction between quasi-arguments and expletives is relevant in the history of overt NonRefSs in Norwegian, it is not enough to save the prerequisite hypothesis.

3.3.3 Construction markers

In addition to expressions with quasi-arguments, Falk (1993b, 1993a) observes another context in which NonRef det could occur in Swedish before it became a general NonRefS, namely existential constructions. NonRef det was, according to Falk, a construction marker, i.e. pragmatic signal of existentials (see in particular Falk 1993b:165).

The idea of det as a construction marker in Falk’s sense in Middle Norwegian is only meaningful if the presence of an overt det distinguishes the existential construction from other constructions. As I have shown, by the second half of the 15th century, NonRef det appears as a quasi-argument and in existentials, as well as in impersonal passives. This range of contexts seems too wide to justify a construction marker analysis. ²⁵ I repeat the impersonal passive in example (19) in example (20):

1. (20)

   

It is worth noting that as a more general concept, the idea of det as a construction marker is actually relevant in early Middle Norwegian. The relevant construction is, however, not the existential construction, but the det så construction, discussed in Section 3.3.2. As mentioned, the det så construction involves a quasi-argument; I repeat the relevant parts of example (14) in (21):

1. (21)

   

Up until 1450, I have only found overt, NonRef det in the det så construction; other NonRefSs, including other quasi-arguments, are always null. Thus, we can discern a stage at which NonRef det is a marker of the det så construction before it became generally licit as a quasi-argument. This, however, only accounts for a subset of the quasi-argumental occurrences of det, not for any of the overt expletives that co-occur with null RefSs.

To sum up this section, we have seen that Falk’s (1993b, 1993a) idea of det as a marker of the existential construction cannot fully reconcile the distribution of NonRef det in Middle Norwegian with the prerequisite hypothesis. I have not been able to spot any alternative functions that NonRef det in Middle Norwegian might have had, apart from that of being a subject. ²⁶ My conclusion is, therefore, that the prerequisite hypothesis, at least in its strong version, cannot be reconciled with the behaviour of NonRef det in the history of Norwegian.

4. A WEAK VERSION OF THE HYPOTHESIS

In the previous section, I argued that co-occurrences of overt NonRefSs and null RefSs cannot be reconciled with the strong version of the prerequisite hypothesis. This introduces the question of whether a weaker version of the hypothesis would be compatible with the data. By a weaker version, I mean a version according to which a language with null RefSs may allow some overt NonRefSs, as long as it also permits null NonRefSs (see Section 2.1).

In Sections 3.3.2 and 3.3.3, I briefly touched upon a fact of relevance to the weak version of the prerequisite hypothesis, which I will discuss in some more detail here: within individual charters, we may encounter co-occurrences not only of overt NonRefSs and null RefSs but also of null and overt NonRefSs. Cf. examples (22) and (23) for evidence of variation between null and overt NonRefSs:

1. (22)

   

2. (23)

   

Example (22a) instantiates a null quasi-argument, while (22b) shows an overt quasi-argument within the same, individual document. In example (23), the NonRefSs are expletives; (23a) contains a null expletive in an impersonal passive, while (23b) contains an overt expletive in an existential construction. The data in (22)–(23) indicate that Middle Norwegian did allow null NonRefSs in some cases even after the introduction of overt NonRefSs. Thus, the Middle Norwegian data are compatible with the weak version of the prerequisite hypothesis. ²⁷

However, even though the weak version of the hypothesis is descriptively adequate, it is important to be aware that it leaves some important questions unanswered. The most fundamental one is perhaps the following: when both null and overt NonRefSs were possible, which factors influenced the distribution of the two options? As mentioned, with the exception of the det så construction and perhaps also other quasi-arguments in early Middle Norwegian, I have not been able to spot any special functions associated with overt NonRefSs. There are two ways of interpreting this: either overt and null NonRefSs were in free variation, i.e. there was no interpretive difference between the two options (“true optionality” in the sense of Biberauer & Richards 2006). Alternatively, we may assume that there was in fact a difference, but that it is not easily detectable in the preserved Middle Norwegian data. In what follows, I will briefly discuss the two options.

According to Biberauer & Richards (2006), null and overt NonRefSs are in free variation in Faroese and Afrikaans. Free variation in other Germanic varieties may be taken as corroborating evidence of free variation in Middle Norwegian. However, conceptually, the idea of free variation is not uncontroversial, at least if “free” is taken to mean that there are not even subtle, contextual (e.g. scope-related, pragmatic, or stylistic) differences between the two options. In linguistic theory, it is often assumed that different lexical items with the exact same meaning do not occur in natural languages; this is due to blocking effects (see e.g. Aronoff 1976 and Kroch 1994). ²⁸ Many syntacticians take the view that blocking effects prevent not only different words with the exact same meaning but also different syntactic derivations. This follows directly from the basic assumption that derivations are driven by features, which are items of the lexicon (e.g. Baker 2008; more on this in Section 5.1). The restriction on semantically vacuous variation is also stated on slightly different grounds by e.g. Reinhart (1995), Fox (2000), and Chomsky (2001) (see Walkden 2014:23ff for discussion). If we assume free variation between null and overt NonRefSs, we would have to admit an exception to a theoretical principle which seems otherwise well founded.

Another issue that arises if we assume free variation between overt and null NonRefSs concerns the Avoid Pronoun principle (Chomsky 1981). The Avoid Pronoun principle can be stated as follows:

1. (24) Avoid overt pronoun, whenever possible. (Frascarelli 2007:694)

Chomsky (1981:65) originally proposed the Avoid Pronoun principle in a discussion of PRO vs. overt pronouns but takes the view that it might be a more general “principle of grammar”. It has indeed been treated as such in the subsequent literature; the Avoid Pronoun principle is often alluded to both in the context of RefSs and NonRefSs. The fact that overt, referential subject pronouns are pragmatically marked, while null subjects are unmarked in e.g. Italian falls out of Avoid Pronoun, as does the non-existence of overt NonRefSs in that langauge (see, among many others, Hyams 1987, Lebeaux 1987 and Frascarelli 2007). However, if the variation between overt and null NonRefSs in Middle Norwegian is free, the Avoid Pronoun principle cannot apply: if there is free variation, it is possible to avoid overt pronouns; still, speakers use overt NonRefSs. ^{29 30}

The alternative to assuming free variation between null and overt NonRefSs in Middle Norwegian would be to postulate differences that are not observable, or at least hard to spot, due to the nature of the available data. It is difficult to exclude this possibility entirely; as Walkden (2014:27) states, “our inability to perceive the conditions on variation does not mean that those conditions do not exist.” If we commit to the view that variation between null and overt NonRefSs is in fact not free, even though we cannot identify the conditions, we can, in principle, rescue the strong version of the prerequisite hypothesis. As mentioned, even the strong version of the prerequisite hypothesis allows overt NonRef pronouns in Middle Norwegian if the overt, NonRef pronouns are somehow “more” than just subjects. By rejecting free variation, we implicitly say that one of the options must encode something more than just subjecthood. The most obvious assumption would be that the innovative, overt forms have some kind of additional function; if this is correct, they would be compatible with null RefSs in the same way that construction markers are in the analysis of Falk (1993a, b).

In practice, however, taking this approach would force us to dismiss the empirical evidence presented in Section 3.2 on purely theoretical grounds. That would, in my view, not be methodologically sound. I therefore maintain that as it stands, the Middle Norwegian data qualify as counter-evidence to the strong version of the prerequisite hypothesis. ³¹

Summing up this section, then, I have argued that a weak version of the prerequisite hypothesis is compatible with the Middle Norwegian data. However, the weak version also raises certain theoretical and methodological issues of a rather fundamental nature. It is beyond the scope of this paper to discuss these issues in any more detail; in the following, I will instead return to the diachronic development in Middle Norwegian and sketch a diachronic analysis.

5. SKETCH OF A DIACHRONIC ANALYSIS

5.1 On the notion of parameter

The behaviour of NonRef det in Middle Norwegian is unexpected in the context of the null subject parameter of Rizzi (1982). It can thus be seen as counter-evidence against P&P theory more generally – at least P&P in its original formulation (Chomsky 1981). As a consequence of this, the diachronic account that I would like to sketch here is not based on P&P in the “classical” sense. Instead, it is set in the framework of what I will refer to as modern parametric theory (see e.g. Roberts 2012), where the notion of parameter is substantially different from that developed in the 1980s.

A basic assumption of modern parametric theory is the so-called Borer-Chomsky Conjecture (BCC), according to which the locus of parametric variation is the lexicon: parametric differences are a result of the presence or absence of features on lexical items, including functional categories like C and T (Baker 2008:353). Importantly, parametric options and implicational relationships are not given by UG; rather they follow from interaction of all the three factors in language design proposed by Chomsky (2005): the genetic endowment (i.e. UG), experience (i.e. the PLD) and so-called third factor principles, which are innate capacities not specific to the faculty of language. ³²

The more recent notion of parameter allows for more flexibility than Rizzi’s (1982) proposal. In modern parametric theory, the null subject properties of a language (or absence thereof) are not determined by the features of one particular lexical item, rather, they result from the featural make-up of functional categories in combination with the featural make-up of pronouns (Biberauer 2008:50). In the following, I will argue that the featural make-up of pronouns has played a particularly important role in the history of null subjects in Norwegian, in a way that is compatible with the data presented in Section 3.

5.2 DP vs. φPs pronouns and pronoun deletion

To account for how a language develops diachronically, it is necessary to make explicit assumptions about its synchronic nature. I will therefore clarify some points regarding the analysis of null subjects in Old Norwegian before I turn to the question of diachronic change. I will focus on issues that are immediately relevant to the relationship between RefSs and NonRefSs; for extensive discussions, cf. Kinn (a, b).

As briefly mentioned in footnote 6, there are weighty reasons to assume that null subjects in Old Norwegian were not directly licensed by subject-verb agreement. First, even though both subject-verb agreement and null subjects were lost in the transition into Modern Norwegian, the relative chronology between the changes does not indicate any direct, causal relationship: the loss of null subjects basically started while subject-verb agreement was still retained (Kinn 2011; cf. Håkansson 2008, 2013 for similar conclusions regarding the loss of null subjects in the history of Swedish). Second and relatedly, Old Icelandic, which is very closely related to Old Norwegian, eventually lost null subjects without any concomitant weakening of subject-verb agreement (Sigurðsson 1993). Third, Old Norwegian, like other early Germanic languages, allowed null objects without having any object-verb agreement (Sigurðsson 1993, Walkden 2014). Fourth, in Old Norwegian, null subjects are virtually always third person and almost never first and second person (more on this below), but third-person forms of finite verbs do not have any more distinct agreement morphology than the first and second persons, as we would perhaps expect if subject-verb agreement acted as a syntactic licenser. Cf. Table 1, based on Faarlund (2004:49–53) and Spurkland (1989:124).

Table 1. Old Norwegian verbal inflections, present indicative.

The asymmetry between the first and second persons on the one hand versus the third person on the other is the point of departure of the syntactic analysis that I will assume here. Note that the division between the grammatical persons is very clear in Old Norwegian: while a preference for the third person has been observed in discussions of null subjects in other early Germanic languages (e.g. Old High German; see Axel 2007:315), the Old Norwegian data seem to suggest something more than a quantitative tendency. As compared to third-person null subjects, across genres, first- and second-person null subjects are so rare that both the philological literature and more recent, generative works assume the existence of a grammatical rule that prevents them (see Nygaard 1894: 1905 and Faarlund 2012, 2013); Kinn (2016a:126ff) provides quantitative and paleographic evidence in favour of this view.

The question is, then, what it is that prevents deletion of pronouns in the first and second persons. I will argue that the answer can be found in the featural make-up of these pronouns. ³³

It is well known that pronouns differ in terms of syntactic category and internal structure, both cross-linguistically and within individual languages (Cardinaletti & Starke 1996, 1999, Déchaine & Wiltschko 2002, Höhn 2015). One of the points that varies is whether a pronoun can function as a determiner, i.e. take a lexical noun as (a part of) its complement. Déchaine & Wiltschko (2002), whose framework I am adopting, distinguish between DP and φP pronouns, of which only the first type may function as determiners.

In terms of internal structure, the crucial difference between DPs and φPs is that DPs have a D-feature that heads a D-projection, or D-layer, as illustrated in (25a). The D-feature is not present in φPs; cf. (25b):

1. (25)

   

The D-layer makes DP pronouns “demonstrably definite” (Déchaine & Wiltschko 2002:410); their ability to function as determiners is related to this semantic property. An example of a DP pronoun is we in English; when we functions as a determiner, in examples such as we linguists, we have the following structure (Déchaine & Wiltschko 2002:21):

1. (26)

   

φPs, as opposed to DPs, “lack inherent semantics”; they “simply spell out φ-features” and cannot be determiners (Déchaine & Wiltschko 2002:410–411).

Now, in Old Norwegian, first- and second-person pronouns exhibit DP behaviour; they function as determiners in examples like (27):

1. (27)

   

3rd person pronouns, on the other hand, do not appear in contexts like those cited in (27). When a third-person pronoun co-occurs with a noun, the noun is always definite, and I follow Faarlund (2004:89–90) in interpreting it as an apposition rather than a pronoun. Cf. (28):

1. (28)

   

As Old Norwegian third-person pronouns generally cannot function as determiners, I take it that they are φPs rather than DPs. ³⁴ In Old Norwegian, the distribution of null subjects thus seems to correlate with the DP vs. φP distinction: null subjects are almost always third person, and third-person pronouns, as opposed to first- and second-person pronouns, are φPs. This suggests a syntactic derivation that allows φPs, but not DPs, to be null. The generalisation can be formally accounted for if we assume that null subjects are derived by a version of Roberts’ (2010) mechanism of pronoun deletion, whereby pronouns are deleted if their features are a proper subset of the features of a Probe with which they agree. I assume that the D-feature, which is only present in first- and second-person pronouns, interferes with the required subset – superset relation and thus prevents deletion; I refer to Kinn (a, b) for details and further discussion.

Keep in mind that both the featural make-up of pronouns and that of the functional categories with which pronouns agree are language-specific. The fact that other languages allow first- and second-person null subjects does therefore not in itself undermine my account of Old Norwegian. In languages that allow first- and second-person null subjects, the internal structure of pronouns may be different, or the relevant functional categories, or both.

5.3 Spreading of the D-feature and the loss of null subjects

We now turn to the question of diachronic change. If it is correct, as I am assuming, that there is a D-feature on first- and second-person pronouns in Old Norwegian that prevents deletion, the loss of null subjects can be analysed in terms of an extension of the D-feature to new contexts. In other words, if it can be shown that the D-feature “spread” to the third person, we potentially have a formal account of how null subjects were lost. There is in fact some evidence of such a development.

In Modern Norwegian, first- and second-person pronouns may function as determiners just like in Old Norwegian. Additionally, however, there are contexts in which third-person pronouns too exhibit DP behaviour, contrary to what was argued for Old Norwegian in Section 5.2. For example, han ‘he’ and hun ‘she’ can be used as psychologically distal determiners (PDDs) (Johannessen 2006, 2008a, 2008b). PDDs express a certain type of deixis, namely psychological distance to persons. Typically, they occur in contexts where either the speaker or the addressee does not personally know the person referred to, or when the speaker wants to signal a negative attitude to this person. Cf. the examples in (29), from Johannessen (2008b:164–166):

1. (29)

   

PDDs cannot co-occur with the definite determiner den ‘that’, and they are therefore best analysed as DPs (Johannessen 2008b). ³⁵ PDDs are not known to occur in Old Norwegian; I take them to be symptomatic of a reanalysis of the pronoun system, whereby not only first- and second-person pronouns but also third-person pronouns are DPs. ³⁶

The spreading of the D-feature to a new class of pronouns can be seen as an analogical change. In the framework of modern parametric theory, this type of change is predicted by the third factor principle Input Generalisation (IG) (Roberts 2012, Biberauer et al. 2014). IG states that once a child has identified a formal feature, like the D-feature, in the PLD, he or she will have an inclination to (over-)generalise this feature to new, but similar contexts. ³⁷

The D-feature, as we have seen, is a formal feature that enables a pronoun to have determiner properties. A reanalysis from φP pronoun to DP pronoun by analogy, along the lines suggested above, need not necessarily mean that the new determiner functions arose immediately; a pronoun may be a DP without taking a noun as a lexical complement. Relatedly, nothing in principle prevents the reanalysis from φP to DP to affect NonRefSs, or a subset of NonRefSs, before RefSs. Thus, the diachronic scenario that I have sketched is compatible with the Middle Norwegian data, in which overt NonRefSs and null RefSs co-occur. It is also worth noting, however, that the Middle Norwegian situation did not last for very long; Norwegian did, eventually, end up as a non-NSL, consistently with common cross-linguistic patterns and the typological predictions of Rizzi (1982). ³⁸ Although modern parametric theory allows for the Middle Norwegian situation to occur, it also provides a motivation for the subsequent diachronic development: in a situation where the D-feature preventing deletion is present on first- and second-person pronouns, as well as on some third-person pronouns, a further generalisation of the D-feature to all third-person pronouns seems close at hand. At that stage, null subjects are lost altogether.

Returning now briefly to the prerequisite hypothesis, the following point can be noted: we have seen that the loss of null RefSs was not a prerequisite for the emergence of overt NonRefSs. The diachronic analysis that I have sketched allows us to take another step away from the prerequisite hypothesis as it was presented in the introduction of this paper: rather than being a consequence of the loss of null RefSs, the emergence of overt NonRefSs in Middle Norwegian may have been one of the language-internal factors that contributed to this change. ³⁹

6. CONCLUSION

In this paper, I have discussed the relationship between null RefSs and null NonRefSs in Middle Norwegian. My point of departure was the idea that I have referred to as the prerequisite hypothesis, which states that the loss of null RefSs was a prerequisite for the rise of overt, NonRefSs. I have argued that Middle Norwegian data do not corroborate a strong version of the hypothesis but may be reconciled with a weaker version. I have also sketched a diachronic analysis that is compatible with the situation in Middle Norwegian.