1. INTRODUCTION
1.1 The problem – and three solutions
Sentences (1) and (2) exemplify a classical problem in Scandinavian grammar: How do we account for the ‘unexpected’ use of the singular neuter on the adjective?Footnote 1
(1)
(2)
These sentences may be compared with (3) and (4), which show the ‘expected’ agreement pattern, in that the predicative agrees with the subject in gender and number, the way Scandinavian predicatives usually do.
(3)
(4)
Examples such as (1) and (2), which have been referred to as ‘pancake sentences’, are common and grammatical in all three mainland Scandinavian languages, Danish, Norwegian and Swedish. In the current literature, the use of neuter form of the predicative in (1) and (2) is generally seen as agreement (and not as disagreement). Broadly, there are three different lines of account for this example of agreement:
(A) Default (e.g. Hellan Reference Hellan, Muysken and van Riemsdijk1986; Corbett Reference Corbett2006:150, 223).
(B) Semantic/‘referential’ agreement (e.g. Widmark Reference Widmark1966; Knudsen Reference Knudsen1973:40; Teleman, Hellberg & Andersson Reference Teleman, Hellberg and Andersson1999:344; Enger Reference Enger2004).
(C) Syntactic agreement. In one version of this view, the neuter form of the predicative is used because the subject in (1) and (2) is underlyingly the object of a verb that has been deleted (e.g. Faarlund Reference Faarlund1977; Faarlund, Lie & Vannebo Reference Faarlund, Lie and Vannebo1997:767).
The purpose of this paper is to defend the referential agreement analysis (B) and show its advantages over the syntactic agreement analysis (C), even in Josefsson's (Reference Josefsson2009) recent and insightful version of the latter. The details of these alternatives are further explicated in Section 1.3, but we first look at the mainland Scandinavian gender system(s).
1.2 The Scandinavian gender systems
The gender systems of the mainland Scandinavian languages are similar in many respects, but different in some (see e.g. Braunmüller Reference Braunmüller2000). In ‘standard’ Danish, Norwegian and Swedish, gender is neutralised in the plural (as in German). In the singular, Norwegian can have three ‘lexical’ genders, masculine, feminine and neuter, on determiners and on certain adjectives. Compare examples (5)–(8):
(5)
(6)
(7)
(8)
By contrast, Swedish has two genders on determiners and adjectives, usually referred to as common (or uter) and neuter, compare (9)–(12):
(9)
(10)
(11)
(12)
When it comes to personal pronouns, the situation is different. In Swedish, there are four personal pronouns in the third person singular. Reference to male humans is usually made with the pronoun han ‘he’, to female humans with hon ‘she’. Inanimate nouns that take the determiner en (i.e. those that belong to the common gender) are usually referred to with de-n ‘it-c’; inanimate nouns that take ett (i.e. those that belong to the neuter gender) are usually referred to with det ‘it.n’. Compare the noun–pronoun pairs in (13)–(16):
(13)
(14)
(15)
(16)
The situation is the same in Danish and Norwegian Bokmål as in Swedish, even if the concrete pronouns differ. By contrast, in Norwegian Nynorsk, there are in principle as many genders on pronouns as on determiners (three), and whatever gender the noun has on the determiner will also show up on the pronoun. Thus, both a boy and a bike will be referred to by using han ‘he’. In other words, there is in principle no personal pronoun den in Norwegian Nynorsk.
Agreement is never quite as simple as this, though. One complication has to do with sentences such as examples (17) and (18), which I have constructed in Norwegian Nynorsk, but analogous examples could be found in Norwegian Bokmål, Swedish and Danish.
(17)
(18)
Many scholars would refer to the use of ho in (17) and han in (18) as ‘semantic agreement’. That term may, however, be slightly unfortunate. The choice of ho in (17) and han in (18) is controlled by the sex of the referent, not by the semantics of the nouns unge ‘child’ and fjols ‘fool’, respectively. The fact that fjols is a neuter noun would presumably have to be accounted for by the observation that nouns denoting persons in a derogatory and downgrading way are often neuter in Norwegian (e.g. Trosterud Reference Trosterud2001:36); fjols is rather derogatory. The choice of the pronoun han in (18) cannot be due to this semantically-based gender assignment rule, then, but to the properties of the referent. Similarly, nouns that denote persons of unspecified sex are usually masculine in Norwegian; the use of ho in (17) cannot be due to this semantically-based gender assignment rule. For reasons like these, I shall follow Dahl (Reference Dahl2000a) in using the terms ‘referential’ rather than ‘semantic’ agreement.
1.3 Major differences between the referential agreement analysis (B) and the syntactic agreement analysis (C)
Not every difference between the referential agreement analysis (B) and the syntactic agreement analysis (C) can be commented upon here. The major differences relate to the view on agreement and gender.
To Enger (Reference Enger2004), who advocates the referential agreement analysis (B), and who follows Corbett (Reference Corbett1991, Reference Corbett2006), pronominal gender is essentially the same thing as gender on attributive and predicative controllers.Footnote 2 That is, it is basically the same phenomenon at work in all the examples we have looked at, and different agreement effects that one and the same controller can have on different targets are explained by appeal to the Agreement Hierarchy (Corbett Reference Corbett1979, Reference Corbett2006). On this account, the Animacy Hierarchy (a.k.a. Individuation Hierarchy, see Sasse Reference Sasse, Jacobs, von Stechow, Sternefeld and Vennemann1993) is also relevant. Subjects in pancake sentences are characterised by being inanimate or low in individuation, and that is their link to neuter agreement.
Under the referential agreement analysis (B), there is a reason why neuter should be possible with predicative adjectives and in personal pronouns. This has to do with the Agreement Hierarchy (Corbett Reference Corbett1979, Reference Corbett2006:206–237). Figure 1 shows my version of the hierarchy, containing three ‘pegs’ for three different kinds of agreement controllers.
Figure 1. The Agreement Hierarchy, based on Corbett (Reference Corbett2006:207), simplified.
For any controller that permits alternative agreements, the likelihood of referential (semantic) agreement will rise monotonically as we move towards the right along the Agreement Hierarchy (Corbett Reference Corbett2006:207). So if referential agreement is possible on the predicative, it will be possible on the personal pronoun too, but not necessarily the other way around. Importantly, gender agreement is subject to the Agreement Hierarchy, whether it shows itself on attributes, predicatives or pronouns.
My simplified version of the hierarchy differs from Corbett's original in that the relative pronoun is not included here. The relative pronoun is excluded because the Norwegian word som, traditionally referred to as a relative pronoun, is probably better seen as a subjunction (e.g. Faarlund et al. Reference Faarlund, Lie and Vannebo1997:25; Lie Reference Lie2008), and those words that are now seen as relative pronouns do not belong in any but the most formal (perhaps even obsolescent) versions of written Bokmål.
By contrast, for Josefsson (Reference Josefsson2009), who follows Teleman (Reference Teleman and Teleman1987), and who advocates syntactic agreement (C), two basically independent entities are involved:
[T]he neuter feature has a dual nature. First of all it is a morphosyntactic feature associated with nouns, in other words a ‘lexical gender feature’. As such the neuter gender does not carry any meaning; there is simply no element of meaning shared by all neuter nouns. Information about the lexical gender is necessary for the spell-out of the definiteness and indefiniteness articles, as well as for agreement on adjectives (at least in their singular forms). This means that the lexical gender is operative within the boundaries of the DP. In the pronominal system, however, neuter vs. common gender may have a meaning. The pronouns den (it.COMMON) ‘it’ and det (it.NEUTER) ‘it’ can be used as deictic pronouns, namely in cases when they make reference directly to entities in the world, without going by way of a noun or a noun phrase. As deictic pronouns, the difference between den and det is that den refers to a bounded entity, whereas det refers to an unbounded entity, more specifically a substance or a proposition. In Josefsson (Reference Josefsson and van Riemsdijk1999, Reference Josefsson2006) han ‘he’, hon ‘she’, den ‘itBOUNDED ENTITY’ and det ‘itUNBOUNDED ENTITY’ represent the four semantic genders in Swedish. The details of this gender system are not of direct relevance here; what is important is that neuter is a meaningful category within the pronominal domain. Josefsson (Reference Josefsson2009:40; emphasis added here, footnotes left out)
Corbett (Reference Corbett2006:21–22) makes the following statement:
The question as to whether agreement is only a local phenomenon is rarely asked. Opposing views are stated, almost as facts, with little discussion. There is a divide here, though by no means an absolute one, between those who have treated agreement as a prime focus of study as opposed to those who come to it as one of a set of syntactic phenomena to be accounted for. The former . . . typically assume that the feature values of anaphoric pronouns are determined by agreement mechanisms. . . . On the other side, those who come to agreement as just one syntactic phenomenon of many often assume that it is a local phenomenon, and so exclude examples like (Mary. . . she). This is a convenient delineation for syntax, but we shall see evidence to question it. . . . there is no one point at which agreement phenomena can be neatly divided into two in a principled way.
This view will be supported in this paper.
Enger (Reference Enger2004) presents arguments against C, in particular against the deletion analysis. Josefsson (Reference Josefsson2009) accepts much of this criticism, but argues that many of the problems may be solved by using Butt's (Reference Butt2003, Reference Butt, Amberber, Baker and Harvey2010) concept of ‘light verbs’. Josefsson's paper clearly represents real progress in a difficult area; for example, she disentangles specificity and definiteness in a way that Enger did not manage. In terms of the tripartite division of approaches to pancake sentences above, Josefsson carries on the syntactic agreement analysis (C), sharing Faarlund's (Reference Faarlund1977) view that the agreement pattern in (1) and (2) above is ‘really’ ordinary syntactic agreement:
In my view, the introduction of semantic agreement in the sense ‘non-syntactic agreement’ is an undesirable solution, a device that could be applied only if standard analyses fail. This is not the case here. (Josefsson Reference Josefsson2009:38)
In brief, Josefsson's (Reference Josefsson2009:35, 38–40) claim is that the neuter in pancake sentences such as (1) and (2) is to be accounted for by a ‘null neuter element’ which is contained in the topmost XP (or NP, in more old-fashioned terms) of the subject phrase.
1.4 Contributions of the present paper
When a problem has become a classic, the way pancake sentences have (recall Section 1.1 above), it is usually a tricky one, and so may merit extended discussion. An additional reason for pursuing the Scandinavian pancake sentences in particular, however, is that there are different research traditions behind the different analyses. In brief, Enger (Reference Enger2004) borrows much from Corbett's (Reference Corbett1979, Reference Corbett1991, Reference Corbett2006) typologically-inspired work on gender and agreement and he borrows some ideas from functionalist approaches, while Josefsson is inspired by Teleman (Reference Teleman and Teleman1987) and minimalism. There is a difference in the view on agreement (whether it can be split neatly according to targets) and in the view on ‘semantic agreement’ (see Section 1.3 above); in my view, to analyse an agreement phenomenon as referential or semantic is fairly standard. Such differences in basic premises may make the discussion difficult, but hopefully also more interesting, from a theoretical point of view.
The present paper brings forward the following three new arguments in the debate:
• In diachrony, referential (semantic) agreement develops first with pronominal targets, and extends only later to targets higher on the Agreement Hierarchy (Section 4).
• Referential (semantic) agreement can be possible on predicative targets with nouns that have a very general meaning (Section 5), just as we would expect in a semantic account.
• Structures with an overt pronominal element in the topmost part of the NP that have been invoked as parallel to pancake sentences in Swedish do not exist in Norwegian, but pancake sentences exist in Norwegian nevertheless (Section 7).
Further, I argue that if gender on attributive and predicative targets and gender on anaphoric pronouns really were two distinct kinds of grammatical features, they would not be expected to correlate as much as they do (Section 2); splitting up the two kinds of gender is not well supported by the facts (Sections 3, 6). I suggest that the evidence for a phonetically null neuter element controlling neuter agreement on predicative adjectives in pancake sentences is exactly the neuter agreement on these predicative adjectives – in other words, that there is no independent justification for the putative explanation; the explanandum is used as an explanans (Section 7).
2. THE CONNECTION BETWEEN NEUTER AND NEUTER
To begin with, there is a connection between different uses of the neuter that seems to be lost under the syntactic agreement analysis (C). Consider first examples (19) and (20):
(19)
(20)
As already noted above, Josefsson (Reference Josefsson2009) is inspired by Teleman (Reference Teleman and Teleman1987), who argues that there are two gender systems in Swedish. One is semantic and found mainly with pronouns, and the other is grammatical, found in the NP; the two are claimed to be almost completely independent of each other (Teleman Reference Teleman and Teleman1987:108). On this view, the neuter is meaningful for pronouns, but, apparently, not for predicative adjectives, and clearly not inside the NP. That is why the neuter in (2) or (20) would be accounted for differently from the neuter in (19); they are seen as different things.
And yet, if these neuters are so different, we would not expect them to correlate, but they usually do. Consider first examples (21)–(24):
(21)
(22)
(23)
(24)
The difference in the pronouns in the English translations – it in (23), they in (24) – is meant to show that in (23), reference is made to pancakes as ‘stuff’, and in (24) to a specific set of pancakes. Consider next the ungrammaticality of (25)–(26):
(25)
(26)
The ungrammaticality of (25) should be contrasted with the grammaticality of (23), (26) with (21) and (2).
In examples (21)–(24), the neuter form of the predicative adjective correlates with neuter on the pronoun. If two things are unrelated, we do not expect them to correlate so systematically. If (25)–(26) were acceptable, they would count in favour of an analysis by which gender is ‘split’ in two, but they are not.
‘Splitting’ gender into one semantic and one non-semantic part implies, in my view, describing one phenomenon as two. Such splitting has the drawback that the Scandinavian lexical gender system emerges as completely asemantic – unlike every other gender system we know, for which the claim of a ‘semantic core’ holds (Corbett Reference Corbett1991:8). In Scandinavian, as in all other languages attested so far, nouns denoting humans are assigned their gender predominantly on semantic grounds. No language is known in which the ‘gender subsystems’ are completely independent, see Dahl (Reference Dahl2000b), yet that is what Teleman and Josefsson appear to claim for Swedish. We return to the semantics of gender in Section 6.
According to the referential (semantic) agreement analysis, by contrast, the claim is that we are dealing with what is essentially one phenomenon; the different kinds of gender are expected to correlate broadly. Already Widmark (Reference Widmark1966:111) says that if an anaphoric pronoun can be thought of as a manifestation of the inherent gender of the noun (a traditional view in Scandinavian grammar), there is no reason to explain the predicative adjective differently.Footnote 3
According to the referential agreement analysis (B) – and, for that matter, also to the default agreement analysis (A) – there is no ordinary syntactic agreement in examples (1) and (2). Also the neuter form of the predicative would be very unlikely to occur with a non-neuter form of the pronoun, recall (25) and (26).
This has to do with the Agreement Hierarchy (Section 1.3 above); a language that allows semantic/referential agreement on the predicative adjective while disallowing it on the pronoun is simply ruled out. That is, the referential agreement analysis (B) is restrictive, and in a way that fits the realities on the ground. If we assume that gender is something entirely different on pronouns than on predicatives, we shall have to do some additional ‘patching up’ in order to explain why these effects, which line so well up with the hierarchy, are found.
3. THE DISTINCTION BETWEEN AGREEMENT ON PREDICATIVE ADJECTIVES AND ON PRONOUNS IS NOT SHARP
My point is not to deny the differences between the agreement on predicative adjectives and on pronouns. The claim is only that the difference is a gradual one, so that drawing a sharp border-line will entail arbitrariness. This is in line with Corbett (Reference Corbett2006:229):
Whichever way we attempt to split agreement into two phenomena, we do not thereby solve the problem of the distribution of agreement options. . . . Most importantly, agreement choices can be found at the extreme ends of the Agreement Hierarchy.
Attributive modifiers must surely come within the domain of local agreement [unlike ‘anaphoric agreement’, that is], if such a distinction is drawn, and yet we can find semantic agreement here . . . On the other hand, the personal pronoun would be expected to fall under anaphoric agreement, and yet syntactic agreement can be found here . . . I conclude that there are no good grounds for dividing agreement domains into two.
3.1 Syntactic agreement in the pronominal domain
Corbett's view fits the Scandinavian facts. ‘Syntactic agreement’ is possible also on pronouns, as the following noun–pronoun pairs illustrate:
(27)
(28)
3.2 Referential agreement inside the NP
Conversely, ‘referential agreement’ is possible also within the noun phrase. We first turn to a construction that is found especially in the spoken language. It is difficult to give an idiomatic, English translation of examples like these, so my suggestions are tentative.
(29)
(30)
Examples (29)–(30) indicate, in my view, that the distinction between referential and syntactic agreement does not warrant drawing a sharp distinction between inside and outside the NP (or, for that matter, at any one particular point in the Agreement Hierarchy, see Section 1.3 above).
The issue is tricky, however. There are good reasons to assume that the apparently similar (31) is not one NP but two in Swedish (Josefsson Reference Josefsson2009), and likewise, that (32) is not one NP but two in Danish (Sten Vikner, p.c.).
(31)
(32)
In those languages, one may insert a definite determiner after the pronoun, compare (33) and (34).
(33)
(34)
However, as shown by Johannessen (Reference Johannessen2008:173 et passim), in Norwegian (and Icelandic), the determiner, in her terms (pronoun in my terms), is internal to the NP, while in Danish and Swedish, it is not.Footnote 4 In Norwegian (as in Icelandic), the pronoun cannot co-occur with other preposed definite determiners, and Johannessen concludes that nothing
forces us to conclude that the PDD [pronominal distal demonstrative, i.e. the pronoun] should be DP-external. On the contrary, the fact that the PDD and the preposed definite determiner never co-occur indicates that they occupy the same syntactic position. (Johannessen Reference Johannessen2008:179)
Thus, examples like (29)–(30) above do not necessarily carry over from Norwegian to Danish and Swedish, since they are one phrase in Norwegian and not in Swedish and Danish. But neither do analyses of Danish and Swedish, where two phrases may be assumed, necessarily carry over to Norwegian.
It is, admittedly, unusual to analyse (29)–(30) as referential or semantic agreement. In the literature, a pronoun such as han in (29) is often referred to as a pronominal psychological demonstrative. Still, the reason for choosing han, ho in pre-nominal position in (29)–(30) is the same as that for choosing han, hon in examples such as (35)–(36):
(35)
(36)
In both cases, the use of han, hon has to do with the sex of the referent.
We also find a second, much more marginal type of referential agreement inside the NP in Norwegian Bokmål:
(37)
The noun phrase in (37), with a feminine determiner preceding a masculine noun, is clearly a marginal example in Norwegian. When I was first made aware of it, my immediate reaction was to call it wrong; many colleagues have reacted similarly, presumably because (37) conflicts with intuitions of many speakers. But this and similar examples do occur in web texts (see Enger Reference Enger2013), if only infrequently, so presumably, they do not conflict with the intuitions of all speakers. Korsæth (Reference Korsæth2010), the first to draw attention to examples like these, says simply that such sentences smack of semantic agreement. While examples such as (37) are marginal in Norwegian and not relevant at all for Swedish, they support the conclusion to be drawn from examples (29)–(30): Referential (semantic) agreement inside the NP is possible in Norwegian.
The same is the case for Danish, as a third kind of examples show. Compare (38) and (39), where the usually common nouns mælk ‘milk’ and vodka ‘vodka’ are used with neuter agreement:
(38)
(39)
See also Hansen & Heltoft (Reference Hansen and Heltoft2011:232).
While (29)–(32) concerned sex, (38)–(39) have to do with individuation. Examples like (38) and (39) are, to the best of my knowledge, ruled out in both Swedish and Norwegian, at least in the standard languages. However, the point is that the neuter form of the article is used together with a mass reading in these Danish examples, just as in the Norwegian examples such as (2) and (4), presented at the outset and repeated here as (40) and (41), respectively, for the reader's convenience:
(40)
(41)
The neuter correlates with ‘stuff’ reading, the common with a more individuated reading – whether inside the Danish NP, as in (38)–(39), or outside the Norwegian NP, as in (40)–(41).
Under the syntactic agreement analysis (C), the opposition between Danish det mælk and mælken (and det vodka and vodkaen) should presumably not be semantic, since it is inside the NP. (There are no arguments for a two-phrase analysis of det mælk in (38) and det vodka in (39).)
Admittedly, arguments based on (38) and (39) above are based on Danish data and do not necessarily carry over to the analysis of Norwegian and Swedish, but they are good arguments against the claim that gender inside the NP is asemantic – at least in Danish. Related Norwegian examples where mass-noun status would seem relevant for neuter agreement inside the NP are found in (42a, b):Footnote 5
(42)
These are examples where there is a real choice of agreement within the NP in Norwegian (42). The choice has to do with meaning; choosing neuter indicates that the usually ‘common’ controller is conceptualised as a mass, as something lower on the Animacy/Individuation Hierarchy, just as in the Danish examples (38) and (39).This is referential agreement; it does not square with the claim that gender inside the NP does not have any meaning.
Consider now (43) and (44):
(43)
(44)
If gender on pronouns is due to a mechanism different from the mechanism assigning gender to predicative adjectives, then lita and ho in (43) have to be accounted for by two entirely different mechanisms, as must liten and han in (44). That seems unappealing. These Nynorsk examples do not carry over to Swedish, admittedly, but the problem remains essentially the same in at least one fairly standard variety of Swedish, described in the Swedish Academy Grammar (Teleman et al. Reference Teleman, Hellberg and Andersson1999:227):
(45)
(46)
The point is simple: The choice of han depends on the same factor as the choice of nye – animacy and sex of the referent, and nye is an example of referential agreement inside the NP.Footnote 6
We have now reviewed evidence from constructions restricted to Norwegian Nynorsk and Bokmål (in (29)–(30)), Bokmål (in (37)), Danish (in (38)–(39)), Nynorsk (in (42)) and Swedish (in (45)–(46)), which indicates that there can be referential agreement inside the NP. My conclusion is therefore that the distinction between referential and syntactic agreement does not warrant drawing a sharp distinction between inside and outside the NP (or, for that matter, at any one particular point in the Agreement Hierarchy).
As observed above, Corbett (Reference Corbett2006:21–22, 228–230) argues against ‘splitting’ agreement, one of the main arguments being that agreement by means of anaphoric pronouns is of the same kind as agreement with more ‘canonical’ targets. That is what I have argued in this section. A second argument for Corbett's view, relating to the Agreement Hierarchy and diachrony, is the topic of Section 4, after a brief excursus.
3.3 Excursus: Number behaves like gender
The possibility of referential agreement is not restricted to gender. It is also found on number:
(47)
(48)
The noun ektepar will, as a rule, take plural agreement on the predicative adjective, according to Vinje (Reference Vinje2002: 217), from whom the example in (47) is taken. I have adapted the example somewhat. It is important to note that plural agreement on the demonstrative is, by contrast, clearly out, as shown in (48). It is the same factors behind the plural verb seem in the English translation of (47) as behind the plural adjective lykkelige in Norwegian. What (47)–(48) indicate is that referential (or semantic) agreement for number is subject to the same kind of possibilities and restrictions as for gender. Referential agreement is possible on predicatives, but not on attributive determiners, in a way that is compatible with the Agreement Hierarchy (Section 1.3 above). Thus the possibility of ‘unexpected’ agreement behaviour is not restricted to gender, and should therefore be subject to a more general analysis.Footnote 7
4. THE AGREEMENT HIERARCHY AND DIACHRONY
Under the referential agreement analysis (B), there is a reason why neuter should be possible with predicative adjectives and in personal pronouns. The diagram representing the Agreement Hierarchy (introduced in Section 1.3 above) is repeated here for the reader's convenience:
The possibility for referential/semantic agreement will rise monotonically towards the right (Corbett Reference Corbett2006:207). The possibility of referential agreement on attributive determiners and pronouns was noted in Section 3. Such a possibility is, by the Agreement Hierarchy, less likely than referential agreement on the predicative, which again is less likely than referential agreement on the pronoun. That is indeed what we find in Scandinavian.
There is a diachronic point to be made here. By the Agreement Hierarchy combined with the referential agreement analysis, it is to be expected that referential agreement should be possible with pronouns earlier than with predicative adjectives. As Semplicini (Reference Semplicini2012:142) puts it, ‘any change in a gender system towards a higher semanticity should have its starting point in the pronominal domain’. Again, that is what we find in Scandinavian. Already in Old Norse, in which pancake sentences appear to be categorically excluded, neuter pronouns can be used to make reference to non-neuter noun phrases. Nygaard (Reference Nygaard1905:81) presents a number of examples, only two are given here:
(49)
(50)
In short, the diachronic development is as predicted by the default and the referential (semantic) agreement analyses (A, B), while the alternative, syntactic agreement analysis (C) makes no claim as to the diachronic development.
It is perhaps worth pointing out that there are parallels in other Germanic languages. Thus, De Vogelaer & De Sutter (Reference De Vogelaer and De Sutter2011:204) observe that in both English and Dutch,
the dominant tendency in resemanticization processes appears to be the increasing usage of neuter pronouns to refer to non-neuter nouns, especially when these nouns are mass nouns.
This fits nicely with the referential agreement analysis (B).
5. NOUNS THAT HAVE A VERY GENERAL MEANING TRIGGER ‘PANCAKE AGREEMENT’
An additional argument for the referential agreement analysis (B) concerns some facts that have so far not been addressed in the debate. The agreement pattern for a couple of nouns with a very ‘schematic’ or general sense is shown in (51)–(53):
(51)
(52)
(53)
(51)–(53) may be compared to (5) and (10), repeated here, respectively, as (54) and (55) for the reader's convenience:
(54)
(55)
The point is that (51)–(53) illustrate the normal agreement pattern for the noun ting ‘thing’ and two near-synonyms in Norwegian and Swedish, (54) for normal masculines in Norwegian, and (55) for normal common-gender nouns in Swedish. The sentences in (51)–(53) show pancake agreement; given the gender of the controller, the problem is how to account for the ‘unexpected’ neuter on the predicative (recall Section 1.1 above), which at least in Norwegian seems quite standard. Faarlund et al. (Reference Faarlund, Lie and Vannebo1997:768) point out that the noun del, which normally means ‘share, part’, will induce the neuter form of the predicative only when used in a general or abstract sense, as illustrated in (51). Consider (51)–(53) above and (56)–(57):
(56)
(57)
When used genuinely partitively, as in (57), the noun del ‘part’ takes ordinary syntactic agreement in line with its lexical gender (masculine) on the target, and when used with a more general (schematic) meaning, del takes referential agreement with neuter on the target. These facts follow naturally from an analysis in terms of referential (or semantic) agreement, as the gist of that analysis is that neuter agreement in pancake sentences should be related to a low degree of individuation (Enger Reference Enger2004; see Section 1.2 above). It is hard to see how a syntactic analysis would account for the observation that the very general noun ting ‘thing’ – and its near-synonym sak – and del, but only when used in a sense nearly synonymous to ting and sak – should trigger pancake agreement in a way different from other nouns. Compare also (58)–(59):
(58)
(59)
Sceptics may ask how we can determine that a noun has a very general meaning. My answer would be that if nouns B, C and D normally imply A, but not necessarily the other way around, then A is more general than the others. In a taxonomic hierarchy, hyperonyms are more general than hyponyms. For example, animal is more general than cairn terrier, since the latter would normally imply the former, but not the other way around – a cairn terrier is by definition an animal, but an animal can be many other things than a cairn terrier. Furthermore, if B, C and D belong to different ‘semantic fields’, then A probably has a very general meaning. Thus, ting could be implied by seng ‘bed’, inntektsskatt ‘income tax’, transformasjonsgrammatikk ‘transformational grammar’, pils ‘lager beer’ and hveteøl ‘wheat beer’. A hyperonym like øl ‘beer’ could be implied by pils and hveteøl, but not by nouns from other semantic fields, such as seng, inntektsskatt, transformasjonsgrammatikk. Ting can be implied by (or replace) most nouns that are low on the Animacy Hierarchy. That is why ting triggers neuter agreement, even though it is a masculine, lexically.
6. THE RELATION TO LEXICAL GENDER
Following Teleman (Reference Teleman and Teleman1987), Josefsson (Reference Josefsson2006, Reference Josefsson2009) rejects the idea that lexical genders have any meaning. However, they both concede that Swedish nouns denoting human beings are as a rule not neuters, and that inanimates and masses tend to be neuter. In other words, nouns at the top of the Animacy Hierarchy tend to belong to the common gender, nouns at the bottom tend to belong to the neuter; this observation is valuable and not restricted to Swedish (see Faarlund et al. (Reference Faarlund, Lie and Vannebo1997:153–156) and Trosterud Reference Trosterud2001 on Norwegian), and it does not support the conclusion that lexical gender has no meaning at all. Rather, it indicates that in parts of the lexicon, gender assignment is ‘predominantly semantic’, in Corbett's (Reference Corbett1991) terminology. On the whole, lexical gender clearly does have ‘less meaning’ than pronominal gender, but that is a difference in degree, and not in kind (see Dahl Reference Dahl2000a, Reference Dahlb).
In fact, we can even find the relevance of the Individuation/Animacy Hierarchy reflected in some ‘minimal pairs’ of gender assignment. In Norwegian Bokmål, such pairs are well known and include en gap ‘a jester, joker’ (masculine) versus et gap ‘fun’ (neuter), en fant ‘a vagabond/crook’ versus et fant ‘a trick’, en øl ‘a (glass of) beer’ versus et øl ‘a (kind of) beer’ (see Faarlund et al. Reference Faarlund, Lie and Vannebo1997:158). Conzett (Reference Conzett, Onysko and Michel2010) shows that such pairs are not as few as previously assumed; they show that exactly the same semantics can be at play in lexical and referential gender, thus undermining the ‘splitting’ approach.
Somewhat surprisingly, Josefsson (Reference Josefsson2009:66, footnote 16) finds Dahl's distinction between lexical and referential gender ‘similar’ to her own, whereas Dahl (Reference Dahl2000a:106) is at pains to emphasise that
[his distinction does not coincide with] that between semantic and nonsemantic gender: Whereas referential gender is in principle always semantic, lexical gender may be motivated by semantic and/or fomal [sic] factors or be assigned on an arbitrary, lexeme-specific basis.
Thus, Dahl is not saying that lexical gender has no meaning. In fact, Dahl (Reference Dahl2000b:586f.) is explicitly critical towards an important aspect of Teleman's (Reference Teleman and Teleman1987) – and thereby Josefsson's – analysis:
The empirical claim . . . is that the uter:neuter distinction is, in the words of Teleman (Reference Teleman and Teleman1987), ‘basically semantics-free’. This is of course in direct contradiction to the generalization made by Aksenov . . . and quoted by Corbett (Reference Corbett1991: 8) that all gender systems have a semantic core. As noted by Källström (Reference Källström1996), the claim about the non-semantic character of the uter:neuter distinction is in fact false.
I think it is fair to conclude that Dahl's view is more similar to, and compatible with, the referential agreement analysis (B) than with the syntactic agreement analysis (C).
Let us return for a moment to the claim mentioned in Section 2, that neuter is ‘a morphosyntactic feature associated with nouns . . . As such the neuter gender does not carry any meaning; there is simply no element of meaning shared by all neuter nouns’ (Josefsson Reference Josefsson2009:40; emphasis added). The claim that there is no element of meaning shared by all neuter nouns is probably correct, but it does not follow that the neuter gender does not carry any meaning. Consider, for illustration, a highly polysemous noun such as Norwegian krone ‘crown’, which can be used to denote a coin, the ornamental head dress worn by a monarch, and the top of a tree, among other things. Unless we posit a highly abstract Gesamtbedeutung (which may be both unfalsifiable and unlearnable), there is, presumably, no element of meaning shared by the various uses. Yet it does not follow that the noun crown does not have any meaning.Footnote 8 An element of meaning shared by many neuter nouns, which recurs in the pancake sentences, is that of being low on the Animacy/Individuation Hierarchy.
7. THE USE OF LIGHT VERBS AND INVISIBLE ELEMENTS
As noted in Section 1.3 above, two central concepts used in Josefsson's (Reference Josefsson2009) revision of the syntactic agreement analysis (C) are ‘light verbs’ and an invisible feature. This section will explore these ideas.
The status of ‘invisible’ elements may trigger an almost everlasting debate, and there are many possible positions here. My claim is not that one cannot posit invisible elements; only that they must be established on independent grounds. One may talk of ‘light verbs’ where they can be observed, as does Butt (Reference Butt2003, Reference Butt, Amberber, Baker and Harvey2010), mainly with South Asian examples. It is an entirely different matter to posit them where they cannot be observed, which is what Josefsson does in the context of Swedish and related languages.
Like Enger, Josefsson (Reference Josefsson2009:46) is skeptical about Faarlund's (Reference Faarlund1977) deletion analysis, according to which the use of the neuter form in (1) above was analysed as due to the use of the neuter in the underlying å ete pannekaker er god-t ‘to eat pancakes is good-n’, but, she says that ‘we have no reason to assume that the verb is deleted; instead the verb is null, i.e. devoid of phonological features’. Now, whether the use of the neuter form is due to a verb that first was there and then was deleted (as Faarlund argued), or to an invisible verb that has been there all along (as Josefsson argues), is not necessarily a very significant difference, at least not to my mind. A more interesting similarity, pointed out by Wechsler (Reference Wechsler2011), is that both Josefsson and Faarlund assume a ‘Silent Head Hypothesis’, by which the head of the controlling element is not there. One of Enger's (Reference Enger2004:8) arguments against Faarlund's analysis was that it seemed ‘somewhat suspicious that there are so many possible different sources of the derivation’. This point is accepted by Josefsson, but she thinks the criticism can be evaded by the concept of ‘light verbs’:
I will leave the question open as to the precise membership of the list of Swedish null light verbs. The proposed analysis directly addresses Enger's (Reference Enger2004) objection to a deletion analysis of the apparent disagreement construction. The ‘missing’ verb is a light verb, which means that the number of verbs that could fit in is restricted. According to a strong version of a general theory of light verbs, this class of verbs would be the same cross-linguistically, probably because they encode basic human activities, acts, and experiences. (Josefsson Reference Josefsson2009:52)
It is also claimed that ‘the identity of the null light verb may remain underdetermined, thus “oscillating” between different readings’ (Josefsson Reference Josefsson2009:50).
To sum up, the list of light verbs has been left open, even language-specifically for Swedish, and the identity of the light verb may remain underdetermined or even ‘oscillate’. Taken together, these moves deprive of empirical content the claim that the number of verbs that could fit in is restricted. What remains is not a very strong alternative to referential or semantic agreement. In fact, Josefsson
[does] not reject the idea that agreement in neuter is semantic in nature per se. What is rejected is the idea that it is the semantic interpretation of the subject that triggers agreement. With the solution that I propose the semantics of the subject is a function of the feature content of the subject, and this content is mirrored by the feature makeup of the predicative adjective – in the way agreement generally works in the grammar. Agreement is thus ‘semantic’ in the proposed analysis too, but in a trivial sense, and it works in the same way as number, for example: a subject in the plural, such as bilarna ‘the cars’ and husen ‘the houses’ in (1c) [bilarna är gröna – husen är gröna], has the semantic interpretation ‘more than one’, and this interpretation is reflected in the plural agreement of the predicative adjective, which, consequently, also carries the meaning ‘more than one’. (Josefsson Reference Josefsson2009:38)
On this account, pancake sentences are just like other sentences involving straightforward syntactic agreement, apparently. The problem with this claim is that in pancake sentences, the agreement is with an invisible element, in sentences involving straightforward agreement (like bilarna är gröna ‘the cars are green’), it is not.
There are further problems here. We first turn to examples (60) and (61), taken from Knudsen (Reference Knudsen1973:40):
(60)
(61)
Josefsson's claim is that an invisible feature determines the agreement pattern on the adjective, but one and the same noun phrase, apparently, is compatible with both plural in (60) and neuter singular in (61) on the adjective. The claim is that
the topmost XP of the subject phrase contains a null neuter element. This accounts for the neuter predicative agreement; hence the idea of default agreement or semantic agreement can be dismissed. (Josefsson Reference Josefsson2009:35; emphasis added)
The problem with this claim is simply that it is not a theory-independent empirical observation that the topmost XP contains a null neuter element. The argument for postulating the null element is exactly and only that which the null element is to account for. In other words, this is not an explanation, but a re-phrasing of an observation, formulated as an explanation of the very same observation. Given Occam's razor, the onus of argument for the null neuter element rests on those who wish to posit it, but convincing evidence remains to be produced.
A more technical objection is that the slot within which the null element is to reside is justified (by Josefsson Reference Josefsson2009:49) with reference to the Swedish construction in (62):
(62)
While such constructions may be perfectly acceptable in Swedish, they are impossible in Norwegian (Section 3.1 above, Johannessen Reference Johannessen2008:178–179). Still, Norwegian can have pancake sentences. That indicates that (62) is probably not the key to the analysis of pancake sentences.
Two final points of objection are the following: Firstly, if pancake sentences are ordinary agreement, as claimed in the quotations above, then why are the subjects unusual? Subjects in Scandinavian pancake sentences tend to be indefinite, and they usually denote masses, propositions and inanimates. By contrast, ordinary subjects in sentences with ordinary agreement tend to be definite, and they very often denote humans (and countable things). The intuition that pancake sentences have unusual subjects was elegantly captured by Faarlund's (Reference Faarlund1977) analysis, in which subjects in pancake sentences were underlyingly objects, because objects are, as a rule, non-animate and indefinite. The intuition is also captured by linking pancake sentences to the lower part of the animacy/individuation hierarchy, as done in the referential agreement analysis (B) (Section 1.2 above, Enger Reference Enger2004), and, for that matter, in the default analysis (see Corbett Reference Corbett2006:150, 223–224). If pancake sentences basically have the same agreement as others, this intuition is lost.
Secondly, while it is the case in canonical agreement that the feature content of the controller is matched by that of the target, there are several examples where such matching does not occur (Corbett Reference Corbett2006:143 et passim). If we insist that ‘the feature content of the subject . . . is mirrored by the feature makeup of the predicative adjective’ (Josefsson Reference Josefsson2009:38), we end up in some fairly difficult waters. A familiar argument going back to Pollard & Sag (Reference Pollard and Sag1994: 62) concerns examples such as (63)–(64):
(63)
(64)
By a referential account (analysis B), the difference in agreement patterns between (63) and (64) is not terribly difficult; in (63), the adjective is in the masculine because the speaker is a man, in (64) it is in the feminine because the speaker is a woman. Yet it is hard to see how we are to make further progress along the syntactic agreement analysis (C) if we insist that the difference in the predicative adjectives must ‘mirror’ or reflect the feature content of the respect subjects. One would have to assume that the pronoun eg ‘I’ in (63) carries the gender feature [+ masculine] and that in (64) the gender feature is [+ feminine]. Also for the pronoun du ‘you’ it would be necessary to assume different gender specifications depending on the agreement. Unfortunately, the consequence is a proliferation of lexical entries, and it is unclear how to link them. (It seems unappealing to treat the two instances of eg as homonyms, for example.) The idea that all agreement must (or even can) be accounted for by means of ‘mirroring’, i.e., derivation or feature copying, is open to objection (see Pollard & Sag Reference Pollard and Sag1994:62–67). In his recent survey, Corbett (Reference Corbett2012:37) notes that copying is not used in GPSG/HPSG and LFG, and that so far the notion of ‘checking’, employed within some Minimalist studies lacks a full formalisation (see also Asudeh & Toivonen Reference Asudeh and Toivonen2006). In that perspective, the claim that mirroring is the way agreement generally works seems debatable, in my view.
8. SUMMING UP
While the analysis advocated by Josefsson (Reference Josefsson2009) clearly has its merits, such as highlighting what counts as a pancake sentence and clarifying the role of specificity, it also has a number of drawbacks when set against the referential agreement analysis (B):
• The connection between different neuters is lost (Section 2).
• The distinction between different genders is not well motivated (Sections 3 and 6).
• Certain diachronic predictions are lost (Section 4).
• The agreement patterns for very ‘general’ nouns (Section 5) follows more naturally on analysis B (referential agreement).
• The use of the notion of light verbs and invisible elements is debatable (Section 7).
Finally, despite the term pancake sentences, the sentences in (1) and (2) are not so peculiar. They fit a well-established typological pattern. Siemund (Reference Siemund2002, Reference Siemund2008), Audring (Reference Audring2009) and De Vogelaer & De Sutter (Reference De Vogelaer and De Sutter2011) show parallels in other Germanic languages, applying the agreement and animacy (individuation) hierarchies (see also Corbett Reference Corbett2006).
ACKNOWLEDGEMENTS
For helpful comments on different oral versions, thanks to several different audiences in Oslo on several different occasions, to audiences in Bergen (June 2010), Tromsø (November 2011), and Vienna (February 2012). Thanks to Sten-Olof Ullström for help with Swedish data. Special thanks to Piotr Garbacz, who read and commented on a draft, and to many students in Oslo. Finally, sincere thanks to three anonymous NJL referees and the Editor Sten Vikner for helpful comments and constructive criticism.
