Semantic and grammatical genders in Swedish—independent but interacting dimensions

Semantic and grammatical genders in Swedish—independent but interacting dimensions

Lingua 116 (2006) 1346–1368 www.elsevier.com/locate/lingua Semantic and grammatical genders in Swedish—independent but interacting dimensions$ Gunlo¨...

194KB Sizes 17 Downloads 29 Views

Lingua 116 (2006) 1346–1368 www.elsevier.com/locate/lingua

Semantic and grammatical genders in Swedish—independent but interacting dimensions$ Gunlo¨g Josefsson Department of Scandinavian Languages, Lund University, Helgonabacken 12, SE-221 00 Lund, Sweden Received 25 August 2003; received in revised form 16 April 2004; accepted 14 June 2004 Available online 31 March 2005

Abstract The purpose of this article is to outline the basic properties of the gender system in Swedish. I argue that there are two gender dimensions: one grammatical and one semantic. The two systems are independent of each other, but they interact. Crucially, the pronouns den (it.common) and det (it.neuter) can be used both as grammatical and semantic pronouns. The locus of semantic pronouns (as well as related, classifier-like elements) is the topmost layer of the nominal extended projection, a Semantic Phrase. Understanding how the semantic gender system is constructed, in particular the properties of the fourth gender, SUBSTANCE, – which lacks number – helps to explain otherwise strange cases of apparent disagreement on predicative adjectives. Grammatical gender is best analysed as a feature that is merged into a projection, not as an inherent marking on nouns. Cases where the gender feature is absent have traditionally been explained in terms of default agreement. The proposed analysis shows that agreement between the subject, which in such instances is a null semantic pronoun, and the adjective is true agreement. # 2005 Elsevier B.V. All rights reserved.

The purpose of this article is to outline the basic properties of the two gender dimensions in Swedish – grammatical gender (common vs. neuter) – and semantic gender. The two $

Earlier versions of this paper have been presented at different occasions: The Grammar Seminar at Lund University in November 2000 and May 2003, Grammar in Focus, Lund University in February 2002, GLAC in Bloomington, Indiana, April 2002, and Genus i Grammatikken in Oslo, November 2002. I thank the audiences at these occasions for comments and discussion. I also wish to thank Christer Platzack and two anonymous reviewers for valuable comments. E-mail address: [email protected]. 0024-3841/$ – see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.lingua.2004.06.011

G. Josefsson / Lingua 116 (2006) 1346–1368

1347

dimensions are partly independent and partly interact with each other. The discussion starts with a presentation of two puzzling agreement phenomena that will be explained in light of the proposed gender theory. Swedish is a language where predicative adjectives agree with their subject (or object) in grammatical gender (common vs. neuter) and number: (1)

a

b

c

Bil-en a¨r gro¨n-Ø. car-common.sg.def is green-common.sg ‘The car is green.’ Hus-et a¨r gro¨n-t. house-neut.sg.def is green-neut.sg ‘The house is green.’ Bilar-na/hus-en a¨r gro¨n-a. car-common.pl.def/house-neut.pl.def is green-pl ‘The cars/houses are green.’

As (1c) indicates, predicative agreement in grammatical gender is not morphologically expressed in the plural. With a clausal subject, agreement appears to be in neuter singular. Compare (2) below to (1b): [Att Carl flyttar till Frankrike] a¨r tra˚kig-t. [that Carl moves to France] is sad-neut.(sg?) ‘It is sad that Carl will move to France.’

(2)

Two observations regarding predicative agreement will serve as points of departure for the rest of this article. First of all: Although a subject consisting of conjoined DPs triggers agreement in the plural (cf. (3a) below), (3b) shows that a subject consisting of conjoined clauses does not trigger agreement in the plural. (3)

a

b

Bo och Carl a¨r tra˚kig-a. Bo and Carl are boring-pl ‘Bo and Carl are boring.’ [Att Bo sa¨ljer huset] och [att Carl flyttar till Frankrike] a¨r tra˚kig-t/*tra˚kig-a. [that Bo sells his house] and [that Carl moves to France] is sad-neut/*sad-pl ‘It is sad that Bo will sell his house and that Carl will move to France.’

The first question is why we do not get the plural ending -a, yielding tra˚kiga as the predicative adjective in (3b). Second, there are cases where the subject consists of a single common gender noun where agreement is nevertheless in neuter and what appears to be singular. Note that senap ‘mustard’, gra¨dde ‘cream’, and mjo¨lk ‘milk’ are all common gender nouns in Swedish.1 (4)

1

a

Senap a¨r gul-t. mustard is yellow-neut.(sg?) ‘Mustard is yellow.’

It should be pointed out that Senap a¨r gul (mustard is yellow.common) ‘Mustard is yellow’, according to some informants, is grammatical as well, though considered slightly less well-formed than the alternative in (4a).

1348

G. Josefsson / Lingua 116 (2006) 1346–1368

b

Men gra¨dde och mjo¨lk a¨r vit-t. but cream and milk is white-neut.(sg?) ‘But cream and milk are white.’

The goals of this paper are to provide an account for the (dis)agreement phenomena illustrated in (3b) and (4) and outline the structure of the two gender dimensions in Swedish and describe how they interact. The rest of this article is organized in the following way. In Section 1 I introduce the adopted framework. Section 2 contains an elaboration of the two gender dimensions or systems in Swedish (and probably in other cognate languages as well). In Section 3 I apply the conclusions regarding the gender systems to account for the two (dis)agreement phenomena just described. In Section 4 I discuss and reject an alternative approach according to which -t in examples like (3b) and (4) is analyzed as default inflection. Section 5 contains a summary and conclusion.

1. Background The main framework used in this paper is Distributed Morphology (see e.g. Halle and Marantz, 1993). Concentrating on inflectional morphology, one of the core ideas in this framework is that the syntax does not handle phonological material. Phonological items are inserted post-syntactically, in a morphological component. Furthermore, there is not a oneto-one matching between a terminal syntactic node and the phonological material inserted into this node. Instead vocabulary items compete for insertion. In short, the most highly specified item wins, provided it is not marked for features not present in the terminal node. Consider the Subset Principle of Halle (1997) below: Subset Principle The phonological exponent of a Vocabulary item is inserted into a morpheme in the terminal string if the item matches all or a subset of the grammatical features specified in the terminal morpheme. Insertion does not take place if the Vocabulary item contains features not present in the morpheme. Where several Vocabulary items meet the conditions for insertion, the item matching the greatest number of features specified in the terminal morpheme must be chosen (Halle, 1997: 428). An important consequence of the Subset Principle, in my view one of the major insights into the nature of language in general, is that not all instances of a particular inflectional element or a grammatical word correspond to the same syntactic feature setup. That is, just because a vocabulary item, say a particular inflectional element, appears in a certain syntactic context, it is not by definition the case that this vocabulary item will correspond to the same set-up of morphosyntactic features in a different syntactic context. Another virtue of the DM approach is that it is strictly feature-based. Importantly, all features present in a derivation must be accounted for. We cannot simply stipulate that something is default without specifying in detail in what sense and in terms of which features. The two points in question – that there is no one-to-one

G. Josefsson / Lingua 116 (2006) 1346–1368

1349

correspondence between form and feature content and that all features must be accounted for – are natural parts of a DM framework, though they could without doubt be assumed in other frameworks as well. When it comes to the description of the gender systems of Swedish I will elaborate on Teleman (1965, 1969, 1987).2 The major point of Teleman’s gender system is that there are different but interacting gender systems or dimensions, specifically a grammatical gender system and a semantic gender system. Within the semantic gender system there is a distinction between animates and inanimates. The animate gender is further divided into males and females. A male is referred to by the pronoun han ‘he’, and a female by hon ‘she’. Within the inanimate gender the grammatical genders, common vs. neuter, are marked on the personal pronouns: den (it.common) vs. det (it.neut).3 There is no clear-cut borderline between the categories animates and inanimates. First of all, animals may fall in either category. Furthermore, the choice of pronoun, in particular when it comes to reference to pets, but also (sometimes by virtue of metaphor) to cities, countries, organizations, astronomical objects (the moon, the sun. . .) is partly a linguistic and partly an extralinguistic matter. Sometimes there is a choice for the speaker as to what point of view he/she wants to assume with regard to the referent in question. Dahl (2000) uses the term ‘referential gender’, instead of ‘semantic gender’. In my view the term ‘semantic gender’ is more appropriate however; it emphasizes the fact that it is not always ‘‘objective’’ properties of the referent that are of importance for the choice of pronoun, but rather subjective properties, such as the choice of point-of-view of the speaker.

2. The gender systems of Swedish In this section I outline the basic structure of the two gender systems of Swedish: the semantic gender system and the grammatical gender system. The purpose is to show that Swedish has four semantic genders, each one corresponding to a particular personal pronoun. The properties and feature set-up of the 4th semantic gender will help us to understand the alleged disagreement constructions exemplified in the beginning of this article (cf. (3b) and (4)). The first claim I want to make is that clauses (including clause-like constituents, infinitival phrases, etc.), mass nouns, and substance nouns lack a number feature.4 Let us first consider event-denoting categories. The idea that clauses lack a number feature follows rather straightforwardly from Grimshaw (1990). Grimshaw discusses 2

An analysis quite similar to Teleman’s has recently been launched by Dahl (2000). An alternative to the terms ‘animate – inanimate’ would be ‘humans – non-human’. However, by choosing the term ‘animate’ I want to emphasize the idea that humans make a distinction between on the one hand those who ‘‘are like ourselves’’ in a certain sense, i.e. beings that we think of as endowed with a capacity to think, feel and/or experience the world in the same way as we do, i.e. the human way — and on the other hand those who do not. As pointed out above, this does not imply that animates really have a capacity to think, feel and experience the world like us, only that we categorize them in these terms. 4 Consider Kamiya (2001) for a similar idea (about nouns). Kamiya (whose work is based on Muromatsu, 1998) discusses primarily Japanese data, but the findings are easily carried over to Swedish. Consider also Vangsnes (2001), who argues that mass nouns lack a Number head in their extended projection. 3

1350

G. Josefsson / Lingua 116 (2006) 1346–1368

so-called ‘‘complex event nouns’’, which are nominalizations that have an event structure and consequently are provided with the same argument structure as the corresponding verbs. She shows convincingly that complex event nominals do not pluralize (Grimshaw, 1990: 54). So, if there are event nominals that cannot be pluralized it is quite reasonable to believe that this is due to the fact that they lack a grammatical number feature. And if the type of nominalization that is most similar to verbal constructions lacks a number feature, it follows without much ado that clauses lack a number feature too. Grimshaw mainly discusses English, but a Swedish correlate to a complex event nominal would be the nominalization ma˚lning (painting), which relates to the verb ma˚la ‘paint’. The word can either mean ‘an act/event of painting’ (in which case it is a complex event nominal) or ‘picture’. If a plural affix is added the meaning is disambiguated; the word ma˚lning-ar (painting-pl) can only mean ‘pictures’.5 (5)

a

b

hans ma˚lning his painting ‘his act of painting’ or ‘his picture’ hans ma˚lning-ar his painting-pl ‘his pictures’

Drawing on Grimshaw (1990) I thus assume that event-denoting categories, such as clauses, simply lack a number feature. Likewise, event-denoting nouns (‘‘complex event nouns’’) also lack a number feature. Provided the number feature is a prerequisite for pluralization in general, we conclude that pluralization is impossible due to a lack of a number feature. This explains the absence of agreement in number in (3b); the CPs making up the subject simply lack a number feature, hence agreement in plural cannot arise. The question why agreement is in neuter in (3b) remains to be explained. We shall return to the issue, but for the moment let it suffice to say that I do not assume a default analysis of the traditional kind. Instead I propose that subject–verb agreement is straightforward; this alternative is much more appealing than the idea that strange and unprototypical subjects give rise to a default type of agreement. The ‘‘deviating’’ patterns in (3b) and (4) should also be explained in terms of which features enter into the relation and how these features are expressed phonologically. It is not only event nouns that resist pluralization. It is well known that certain other nouns do not accept plural either, in particular mass nouns and substance nouns. If we add a plural suffix to nouns such as elektricitet ‘electricity’, ko¨tt ‘meat’, mjo¨l ‘flour’, and gla¨dje ‘happiness’ the result is ungrammatical or at least very strange: 5 The discussion here is somewhat simplified. Grimshaw assumes that there are two kinds of event nouns, which she calls ‘simple’ (or ‘result’) event nouns and ‘complex’ event nouns: ‘‘[T]here is a kind of secondary construal in which the complex event nominal behaves like the result nominal. I believe that this is due to a process of type shifting by which the event nominal can be treated as though it referred to an individual rather than an event’’ (p. 55). In such cases the event reading is available also when the nominalization is pluralized.

G. Josefsson / Lingua 116 (2006) 1346–1368

1351

*electriciteter, *kotter, *mjo¨ler, and *gla¨djen. (The marking of the spelling program on my computer coincides well with my linguistic intuition here.) In other cases a plural suffix (or a cardinal number modifier) can indeed be added to substance nouns, but the meaning shifts: tva˚ o¨l ‘two beer’ means two glasses or brands of beer, tre mjo¨lk (three milk) ‘three packages of milk’, mina ka¨rlekar (my loves) ‘my lovers/love affairs’ and so on.6 It is possible to do the reverse with count nouns. A word such as hund ‘dog’ is usually considered to be countable. Nevertheless (6) is well.formed: (6)

Ni fick mycket hund fo¨r pengarna. you got much dog for money-the ‘You got a lot of dog for your money.’

As a matter of fact, most so-called countables can be used as uncountables or as substances. Thus, from a principled point of view it is incorrect to speak about count and mass nouns, as though there were some constraints on their use listed in the lexicon. It is preferable to say that most nouns have the possibility of occurring in either syntactic context, even though they often display a certain inclination towards one of the options, due to their meaning. The unmarked syntactic context for hund ‘dog’ is that of a count noun, and the unmarked context for vatten ‘water’ is that of a mass or substance noun. A somewhat similar argument could be made for the male–female distinction. There is a large class of nouns that can refer to both men and women; Corbett (1991: 183–184) calls them ‘‘hybrid nouns’’; some Swedish examples would be doktor ‘doctor’, la¨rare ‘teacher’, and sjuksko¨terska ‘nurse’. If we generalize the discussion of count vs. mass/substance nouns and hybrid nouns we find that there is a semantic classification system different from – and to a certain degree independent of – the grammatical gender system. This is quite in line with what Teleman (1965, 1969, 1987) and more recently Dahl (2000) have shown. However, what I would like to do in this article is to include the distinction count vs. mass/substance nouns in a Teleman style of system as well, claiming that this is a distinction that should be made within the inanimate gender. In other words, there are two inanimate (sub)genders, which I will call THINGS and SUBSTANCES (the latter including uncountables and substances, which in the present system could be treated in the same way). Table 1 shows a sketch of Teleman’s system, as presented in Teleman (1987), and Table 2 shows the (first attempt of the) system that I propose. Table 2 shows that inanimates can either be THINGS, that is bounded entities, or SUBSTANCES, i.e. unbounded entities. Nouns are not once and for all classified as THINGS or SUBSTANCES, other than in a prototypical way; as pointed out above the noun hund ‘dog’ in (6) is syntactically treated as a SUBSTANCE, which is 6 It is interesting to note that the plural affix is avoided when substance nouns are individuated. Tva˚ kaffe ‘two coffee’ is thus fine, but not *tva˚ kaffe-n (two coffe-pl). As pointed out by an anonymous reviewer there is sometimes a variation, viz. tva˚ lemonad (two lemonade) vs. tva˚ lemonad-er (two lemonade-pl), where the former means ‘two glasses or portions of lemonade’ and the latter ‘two kinds of lemonade’. This phenomenon will not be discussed further here.

1352

G. Josefsson / Lingua 116 (2006) 1346–1368

Table 1 The semantic gender system of Swedish according to Teleman (1987) Animates

pronoun

Inanimates

MALE

FEMALE

han ‘he’

hon ‘she’

den, det ‘it’

Table 2 The semantic gender system in Swedish, first attempt Animates

pronoun typical examples

Inanimates

MALE

FEMALE

THING

SUBSTANCE

han ‘he’ man ‘man’

hon ‘she’ kvinna ‘woman’

den, det ‘it’ stuga ‘cottage’, hus ‘house’ ma˚lning ‘picture’

den, det ‘it’ mjo¨lk ‘milk’, sko¨rd ‘harvest’, vin ‘wine’, ma˚lning ‘the act of painting’

unprototypical, but nevertheless fully grammatical. The crucial difference between the 3rd and the 4th gender is the notion of spatial boundedness. The prerequisite for countability is that the entity/entities can be viewed as discrete objects, i.e. that they have a spatial integrity.7 This further emphasizes that semantic gender is a matter of point-ofview, i.e. that we can regard a dog either as substance, flesh, which can be measured in kilos (cf. (6)) or as an individual that can be counted. The difference between the two views is not related to the objective world but to the way we choose to conceptualize what is in our mind. Tables 1 and 2 shows that the pronouns den (it.common) and det (it.neuter) can be used either for THINGS or SUBSTANCES, depending on the grammatical gender of the noun for which they substitute in any given case. There are clear tendencies that neuter nouns denote unbounded and/or inanimate entities and common gender nouns denote bounded and/or animate referents, but there are no absolute rules.8 Below I show that the pronouns den and det have two functions; they can be either anaphoric or deictic. The distinction ‘common gender’ vs. ‘neuter’ plays a different role in each of the two cases. First, as anaphoric pronouns den/det pick up the grammatical gender and number of the noun to which they refer, their discourse antecedents: (7)

7

a

– Vad tycker du om ma˚lningen? what think you of painting.common.def ‘What do you think about the painting?’ – Den var snyggt utfo¨rd! it.common was beautifully done.common ‘It was nice.’

It should be pointed out that events are bounded or non-bounded as well. However, the kind of boundedness that I refer to here is boundedness in space, spatial integrity. Another possibility would have been to use the distinction between discrete and vs. amorphous substances. 8 See Josefsson (1997, 1998) for discussion. The idea that grammatical genders do have a meaning, though only a prototypical one, has been argued for Danish by Diderichsen (1957). (The Danish gender system is very similar to the Swedish one.) Consider also Teleman (1965: 219) and Widmark (1966: 106).

G. Josefsson / Lingua 116 (2006) 1346–1368

b

1353

– Var finns lejonet? where is lion.neut.def ‘Where is the lion?’ – Det a¨r ha¨r. it.neut is her ‘It is here.’

The use of den/det in the function illustrated in (7) is not very complicated. The pronoun den is chosen in (7a) since it corresponds to the grammatical gender and number of ma˚lningen (painting.common.def) in the previous sentence. The same applies to det in (7b). However, the pronouns den and det can also be used when no discourse antecedent is available, i.e. when there is no word determining the grammatical gender and number of the pronoun. Consider (8): (8)

a

b

[A person stands in front of a counter full of exotic fruit, nuts etc.] Seller, with a strange (probably edible) ‘‘thing’’ in his hand: – Na˚? Well ’Well?’ Buyer: – Jag tar den. I take it.common ‘I’ll take it.’ [A and B standing in front of a freshly painted boat]:9 A: – Vad tycks? what think.pass ‘What do you think?’ B: – Det var snyggt! it.neut was beautiful.neut ‘It was nice.’

In neither of the cases in (8) do we find a noun to motivate the choice of den or det. Of course we cannot find watertight proof that the speaker does not have a particular noun with a particular grammatical gender in mind when choosing den in (8a) and det in (8b). However, a much simpler and more coherent solution presents itself if we assume that the pronouns den and det in (8) simply do not stand for any words; the reference is mediated by the semantic gender. My claim is thus that the semantic gender system makes use of a set of pronouns that do not reflect the grammatical gender of a preceding (linguistic) antecedent, and that den (it.common) and det (it.neuter) are two examples of this. Instead of making reference to a word these pronouns refer directly to one of the four semantic genders THING for den and SUBSTANCE for det. 9

The noun ba˚t ‘boat’ is a common gender noun in Swedish.

1354

G. Josefsson / Lingua 116 (2006) 1346–1368

The same goes in fact for han ‘he’ and hon ‘she’, which make reference directly to the semantic genders MALE and FEMALE. In other words, conveying reference to an entity in the world can proceed by either of two channels, either by way of a preceding word (an antecedent), in which case the grammatical gender (neuter vs. common) is used to identify the word and hence the referent, or alternatively by way of semantic gender, in which case the grammatical gender of the pronoun has another function, namely to distinguish between the two inanimate genders, THINGS and SUBSTANCES. Again it is important to stress that it is not the objective properties of the referent that are of importance here. Instead it is the point-of-view taken by the speaker. Thus, if a speaker chooses the pronoun den, conceptual boundaries are imposed on the referent, whereas a choice of the pronoun det does not impose such boundaries. From a pragmatic point of view it is probable that det is the default choice of the speaker. This is the generally referring meaning of the pronoun det, as discussed for example by Falk (1993: 80).10 Table 3 shows the semantic gender system and the pronouns corresponding to each of the semantic genders. Table 3 The semantic gender system in Swedish, second attempt Animates

semantic pronoun

Inanimates

MALE

FEMALE

THING

SUBSTANCE

han ‘he’

hon ‘she’

den ‘it’

det ‘it’

If the discussion so far is on the right track it implies that neuter and common should more correctly be considered features that operate in a grammatical derivation rather than simply inherent markings on nouns. When the common vs. neuter feature is used in an anaphoric pronoun the function is to convey reference to a preceding noun with the same feature set-up — the discourse antecedent. The feature common or neuter gender on deictic pronouns, on the other hand, serves the function of making reference to an entity belonging to a specific semantic gender. In such cases, establishing a link to a possible discourse antecedent must crucially not proceed via a word/noun phrase and its grammatical features. Exactly how such discourse linking takes place and how it is constrained in different ways will not be discussed in this article. To conclude, we can now provide an account for the apparent lack of number agreement in example (3b), where the subject consisted of two conjoined clauses. Clauses are 4th gender elements and pronominalized det (it.neut) ‘it’ is chosen.11 The 10

Consider also Widmark (1966: 108) for a similar conclusion. The fact that clauses are pronominalized by det can be shown in constructions where a clause-initial element is doubled by a pronoun. In (i) below the DP Kalle is doubled by han ‘he’, and in (ii) the whole clause att han skulle fuska pa˚ examen is doubled by det ‘it’. (i) Kalle, han kan sla˚ kullerbytta. Kalle, he can make somersault ‘Kalle/he can do somersaults.’ (ii) Att han skulle fuska pa˚ examen, det hade vi aldrig missta¨nkt. that he would cheat in exam, it had we never suspected ‘We had never suspected that he would cheat on his exams.’ 11

G. Josefsson / Lingua 116 (2006) 1346–1368

1355

reason why the predicative adjective agreement is -t is that clauses lack a number feature; they are thus neither singular nor plural but belong to the same category as substances.12 The vocabulary item -t on an adjective is thus not specified for a number feature, which means that it can be inserted both in contexts where a number feature is present, such as in (1b), and when a number feature is absent, as in (2), (3b) and (4), repeated below: [Att Carl flyttar till Frankrike] a¨r tra˚kig-t. [that Carl moves to France] is sad-neut.(sg?) ‘It is sad that Carl will move to France.’

(2)

(3)

b

[Att Bo sa¨ljer huset] och [att Carl flyttar till Frankrike] a¨r tra˚kig-t/*tra˚kig-a. [that Bo sells his house] and [that Carl moves to France] is sad-neut/*sad-pl ‘It is sad that Bo will sell his house and that Carl will move to France.’

(4)

a

Senap a¨r gul-t. mustard is yellow-neut.(sg?) ‘Mustard is yellow.’ Men gra¨dde och mjo¨lk a¨r vit-t. but cream and milk is white-neut.(sg?) ‘But cream and milk are white.’

b

It is important to point out that -t is not a default vocabulary item — its context of insertion is specified in detail. We shall return to the question of default assignment in Section 4. The subject of the next section will be the use of -t in (4). Where does the neuter agreement come from in contexts where the subject is a common gender noun? 3. Apparent disagreement is true agreement In this section I advance the discussion introduced in Section 2 by elaborating the idea that grammatical gender features are features that are merged into the derivation, just like other features. What I would like to show is that the type of alleged disagreement, illustrated in (4) above, does not derive from the noun but from a pronominal element merged higher up in the nominal extended projection. I also intend to show that we have 12

I will not detail the source of the neuter feature that triggers predicative agreement in neuter when the subject is clausal, but two possibilities arise. One is that the complementizer carries a neuter feature, much like lexical items. The default complementizer in Swedish, att ‘that’, originates from the neuter pronoun det ‘it’. In some Swedish dialects the pronoun det and the complementizer att are still pronounced in the same way. The other option is that clausal subjects have a SemP on top (see Section 3). Such an assumption is supported by the fact that a clausal subject must be preceded by det ‘it’ when it occurs in the canonical subject position (Spec–IP). In clauseinitial position, however, det is optional. We may thus assume that det may be optionally subject to topic-drop from a sentence-initial position: (i) Da¨rfo¨r a¨r [*(det) att professorn avga˚r just nu] olycklig-t. therefore is (it/that) that professor.the resigns just now unfortunate.neut. ‘Therefore it is unfortunate that the professor is resigning right now.’ (ii) [(Det) att professorn avga˚r just nu] a¨r da¨rfo¨r ganska olyckligt.

1356

G. Josefsson / Lingua 116 (2006) 1346–1368

good reason to believe that the grammatical gender feature normally associated with the word in such constructions (common gender for senap, gra¨dde and mjo¨lk in (4)) never enters into the derivation in the first place. Before going into the details of the discussion it is important to draw a distinction between two types of alleged disagreement. Consider (9) and (10) below: Arga hundar a¨r tra˚kig-t. [[Arga hundar]DP]CLAUSE-LIKE CONSTRUCTION a¨r tra˚kig-t.  b Att ha arga hundar a¨r tra˚kig-t. to have angry dogs is boring-neut ‘It is boring to have angry dogs.’ (10) a Senap a¨r gul-t. [Senap]NOUN PHRASE a¨r gul-t 6¼ b Att ha senap a¨r gul-t. to have mustard is yellow-neut (non-sense)

(9)

a

(9) and (10) show that there are two types of alleged predicative disagreement in Swedish. In the first type, exemplified in (9), the surface DP subject could be rewritten as an infinitival phrase.13 In (10) a paraphrase of this kind is not possible. Following the Swedish Academy Grammar (Teleman et al., 1999) I will assume that the subject of the first type is in fact a larger phrase embedding a DP.14 The exact nature of this phrase is not of immediate interest for us here, but see Josefsson (work in progress) for a suggestion.15 The predicative adjective in (9) shows neuter agreement, presumably for the same reasons as infinitival and clausal subjects trigger agreement in neuter (but not number!). The clause in (10) cannot be given the same analysis as (9), as indicated by the paraphrase. So instead of looking at similarities to clausal subjects, consider the comparison with another construction type: the so-called ‘prenominal apposition’.

13

Alleged disagreement of the type shown in (9) and (though more marginally) in (10) has been the subject of vivid discussion in the literature. See e.g. Heinertz (1953), Widmark (1966, 1967), Ka¨llstro¨m (1993), Teleman et al. (1999) and more recently Enger (2003). 14 Support for this assumption includes the fact that adverbial elements can be added to the left of the finite verb in examples such as (9a) without inducing a V2-violation: (i) [[Arga hundar] [i hallen] [under sa˚dana omsta¨ndigheter]] a¨r vidrig-t. angry dogs in hall.the under such circumstances is disgusting-neut ‘To have angry dogs in the hall under such circumstances is disgusting.’ Examples such as (i) indicate that we have a larger structure, of which the DP and possible adverbial elements are parts. Furthermore, when the initial DP is a pronominal it has accusative case, which is what the proposed analysis predicts. (ii) Context: One cannibal to another; [Henne [med ketchup och senap]] vore got-t. her with ketchup and mustard was.subjuntive good-neut ‘To get [eat] her with ketchup and mustard would be good.’ See Enger (2003) for a different view on this construction. 15 Importantly I do not assume the old-fashioned idea that the subject in (9a) is an infinitival phrase that has undergone some kind of ‘‘truncation transformation’’ or pruning, yielding the surface form of a DP.

G. Josefsson / Lingua 116 (2006) 1346–1368

1357

As pointed out above, Corbett (1991) showed that so-called hybrid nouns are quite common in languages. A hybrid noun is a noun that grammatically has a certain gender, for instance feminine, but semantically denotes (or may denote) a person of a different sexual gender. One commonly cited example is the Spanish word poeta, which grammatically is feminine, but denotes (or may denote) a man.16 The presence of hybrid nouns in Swedish is not surprising, since the Swedish grammatical gender system is not based on sexual gender nor on sharing terminology with sexually based gender systems. It may be the case that there is a prototypical mapping between a certain grammatical gender and a particular semantic gender, but there is no a priori reason why we should not find unprototypical mappings. In Swedish most animate nouns are common gender nouns, but neuter nouns exist as well, for instance fruntimmer (derogatory for) ‘woman’, bitra¨de ‘clerk’, statsra˚d ‘secretary of state’, and vittne ‘witness’. On a par with common gender nouns denoting humans, nouns like bitra¨de, statsra˚d and vittne could be specified as male or female, of course. DPs denoting-animates may thus be supplied with a pronoun, han ‘he’ or hon ‘she’, immediately preceding the DP: (11)

Hon den nya professorn a¨r mycket effektiv. she def.common.sg new professor.common.def.sg is very efficient. ‘She/the new professor is very efficient.’

Note that there is no intonational break between the pronoun hon and the rest of the DP den nya professorn.17 In Josefsson (1999) I proposed an analysis of this type of pronominal doubling, in which a pronoun denoting a person is added in a particular functional projection on top of the DP. For the sake of simplicity I called the projection ‘Sem(antic) phrase’:18

16

See also Harris (1991). It is possible to make an intonational break between the two parts: hon, den nya professor. In such cases we have a different construction type though, where den nya professorn is a parenthetical expression, an appositional element. 18 FP stands for ‘‘functional phrase’’, the properties of which I will refrain from discussing here. Let it suffice to say that it hosts the attributive adjective in its specifier. 17

1358

G. Josefsson / Lingua 116 (2006) 1346–1368

Importantly the pronoun hon in (12) does not reflect the grammatical gender of the noun. Thus, it can combine with either common gender nouns, such as professor, or neuter gender nouns, such as vittne ‘witness’ and bitra¨de ‘clerk’. The choice between han ‘he’ and hon ‘she’ is not determined by grammar either, but by properties of the world. The fact that hon den nya professorn ‘she/the new professor’ is well-formed, in addition to han den nya professorn ‘he/the new professor’, is due to social progress in the society, not to linguistic factors. Josefsson (1999, work in progress) argues that the SemP is in fact also the position for 1st and 2nd person pronouns in constructions of the type jag min dumma a˚sna (I my.sg.common silly.sg.common donkey) ‘I silly donkey’, du ditt dumma no¨t (you your.sg.neut silly.neut cow) ‘you silly cow’, vi svenskar ‘we Swedes’, etc.19 The SemP in (12) provides a position for elements that are not determined by the morphosyntactic features of the noun. In a sense it is quite similar to the higher projections of the CP where speaker-oriented information may surface, according to Cinque (1999); the choice between han ‘he’ and hon ‘she’ is open to the knowledge of the speaker (does the speaker know the sex of the referent?) or her/his point-of-view (does the speaker think of the referent as male or female?). Since it is the head of the construction I assume that Semo determines the morphosyntactic properties of the whole projection, and that agreement in fact holds between the pronoun and the predicative adjective, whether or not this can be shown for all cases. The arrow below indicates this relation.

Given that the proposed parallelism holds between expressions such as the one in (13) and the construction type du ditt dumma no¨t (you your.sg.neut silly cow) ‘you silly cow’, we have quite clear evidence that an agreement relation is established between the pronominal head and the predicative adjective, without involving the features of the noun further down in the projection — the neuter of the embedded noun does not trigger agreement:20 (14)

Du ditt dumma no¨t blev ju arg/*arg-t! You your.sg.neut silly cow became MOD. PL mad.sg.common/mad-sg.neut  ‘You, silly cow, got mad!’

If the nominal extended projection contains features of a semantic pronoun such as du ‘you’ located in the topmost projection it should not be too difficult to accept that Semo in (12) should also be able to contain (the features of ) the other semantic pronouns in Table 3. A 4th gender semantic pronoun can thus be located here. In such cases we expect agreement to be in gender (i.e. neuter), but not number. I claim that this is why we can have 19

Josefsson (1999, work in progress) also argues that pronouns in general are located in SemP, and that the DP/ NP generally is null in such cases. 20 The proposed analysis requires that personal pronouns carry definiteness features independently of whether such features are present further down in the derivation, at the NP/DP level.

G. Josefsson / Lingua 116 (2006) 1346–1368

1359

agreement in neuter in examples such as (4a), which I assume to have the underlying structure in (15):

The null pronominal element in (15) corresponds to the 4th gender det (it.neut) ‘it’ — importantly without any definiteness features. I shall return later to the issue of overt vs. covert constituents in order to make it more precise. If the proposed analysis is on the right track we expect other elements to be able to be occupy Semo too. The prediction is borne out; there is a group of classifier-like elements that can have the function of assigning gender (in the sense argued for in this article) to the whole DP. An example is a¨mne, ‘substance’: (16)

A¨mnet olja a¨r *klibbig/klibbig-t. substance.neut.def oil is *sticky.common/sticky-neut ‘Oil is sticky’.

The word olja ‘oil’ is a common gender noun. Nevertheless agreement is in neuter in (16). In my view this is because the overt classifier-like neuter a¨mnet ‘the substance’, occupying Semo, triggers agreement in neuter (but not number!).21 I pointed out above that the null pronoun in (15) corresponds to det ‘it’. The question is of course why we cannot have an overt det in the same position: (17)

*Det senap a¨r gult. it mustard is yellow.neut

The answer is straightforward; the pronoun det is specified for the feature +definite. In (15) Ø must be indefinite, or at least not +definite. The insertion of vocabulary items thus proceeds as in (18). Note that the vocabulary item det is specified for [+definiteness, +neuter]. (18)

21

Insertion of vocabulary items: Xo Xo +def +def +neuter +neuter +number, plural * * det det inserted in (7b) inserted in (8b)

Xo def +neuter * Ø inserted in (15)

Another option is of course that Semo is null, yielding Olja a¨r klibbig-t (oil is sticky-neut), which is parallel to (15).

1360

G. Josefsson / Lingua 116 (2006) 1346–1368

It would not be correct to assume that the features of SemP simply override the features embedded in the DP/NP. Instead there seem to be restrictions as to what combinations of features in the SemP and the rest of the nominal extended projection are licit. In general it seems possible to have a null or overt pronoun lacking a number feature in SemP combined with a noun provided with features further down in the projection. An example is given in (19a), the structure of which is given in (19b). (19)

a

b

Moro¨tter a¨r gul-t. carrot.pl is yellow-neut  ‘Carrots are yellow.’ [[Ø.neut] moro¨tter.pl] a¨r gul.neut [[Ø.neut] carrot.pl] is yellow.neut

The meaning of (19a) is roughly ‘carrot, viewed as an aggregated substance, i.e. a nonhomogeneous substance consisting of particles, is yellow’. It contrasts with (20), where the meaning is ‘carrot, viewed as a homogenous substance, is yellow’ and (21) where the statement is about carrots as individual items; they are yellow: (20)

a

b (21)

Morot a¨r gul-t carrot is yellow-neut  ‘Carrot is yellow.’ [[Øneut] morot] a¨r gul-t [[Øneut] carrot] is yellow-neut Moro¨tter a¨r gul-a. carrots are yellow-pl ‘Carrots are yellow.’

Since there is nothing indicating the presence of a SemP on top of the DP in (21) I assume that the SemP is simply missing there. If the noun has its definite form or contains an indefinite article, the grammatical gender features must be present in the derivation.22 In such cases predicative agreement cannot be neuter (unless the noun is a neuter noun, of course, as in (22c)): (22)

a b

c

22

*Senap-en a¨r gul-t. mustard-common.def is yellow-neut En morot a¨r gul/*gul-t. a carrot is yellow/*yellow-neut ‘A carrot is yellow.’ Vinet a¨r vit-t. wine.neut.def is white-neut ‘The wine is white.’

It is possible that this is for phonological reasons only. As shown in Gross and van Riemsdijk (1981) and Sauerland (1996), the presence vs. absence of phonological forms may cause a crash at spell-out.

G. Josefsson / Lingua 116 (2006) 1346–1368

1361

The ban on neuter predicative agreement in (22a) indicates that the two parts of the nominal extended projection, the SemP and the rest, must not contain conflicting features (which seems to be an important restriction). Further evidence that the proposed analysis is on the right track is given by the presence of attributive adjectives; if the noun is provided with an attributive adjective – which presumably requires access to the information of the grammatical gender of the noun – alleged disagreement is not possible either:23 (23)

Fransk senap a¨r gul/*gul-t. French mustard is yellow/yellow-neut ‘French mustard is yellow.’

The conclusion is that grammatical gender features – neuter and common – are features that are merged into a projection, like other features. In some cases the features in question are not part of a nominal extended projection, which is fine, provided their presence is not required for phonological or other reasons. However, if the head noun in a noun phrase is a common gender noun and common gender features must be present in order for the derivation to converge – for instance for the spell-out of definiteness inflection or adjectival agreement – a null 4th gender (neuter) pronominal cannot be added, and predicative agreement in neuter is no longer an option. In other words: a noun phrase cannot contain conflicting features, e.g. both common and neuter gender features.24 To conclude: The examples in (4) are OK because the grammatical gender feature normally associated with the nouns in question (common gender in these cases) was never merged into the derivation. On top of the DP/NP (I will not take any definite stand on the structure of the projections at the DP/NP level) SemP is merged, containing a semantic 23

My judgement concerning sentences like (23) is firm, but has been questioned by some native speakers. The data has been presented twice in Lund, at the colloquium Grammatik i fokus in February 2000, and at the Grammar Seminar in May 2003. At both occasions there was a consensus in the audiences on the grammaticality judgements of (23). It is important to ensure that the test really is about the construction discussed here, and not the type exemplified in (9), where the adding of an attributive adjective is unproblematic, of course, since the DP in such cases is embedded in a clausal structure. 24 A related question is the relation between plural and grammatical gender. An anonymous reviewer points out that gender features sometimes seem to be necessary for the spell-out of plural markers. Some Swedish examples of this would be fax (fax.common) ‘fax machine’ – fax-ar (fax-pl) ‘fax machines’ and fax (fax.neut) ‘fax letter’ – fax (fax.plural) ‘fax letters’. However, the idea that gender features are a prerequisite for the spell-out of the plural marker is not necessarily the optimal solution. One reason is that there are a number of nouns which have the same grammatical gender but which differ in lexical class, hence the spell-out of the plural marker differs between lexical classes. Some examples (which all happen to be common gender nouns) are bank ‘savings bank’ – bank-er (savings bank-pl) ‘banks’ vs. bank ‘river bank’ bank-ar (river-bank-pl) ‘river banks’, and bok ‘book’ vs. bok ‘beech’, with the plural forms bo¨ck-er (book-pl) ‘books’ and bok-ar (beech-pl) ‘beeches’. In Josefsson (2001) I argue that the lexical class marker (marker of declension for nouns and conjugation for verbs) is of importance. The lexical class feature is the first feature with which a root (devoid of word class features) merges. This feature determines the global word class (noun, verb, adjective . . .) as well as subclass (i.e. lexical class). Sometimes the lexical class marker is spelled out separately, while in other cases it fuses with the number feature, thus determining the form of the plural marker. If this is correct, the form of the plural marker may be determined by the lexical class feature only, and so gender plays no role in this process. Bare plurals do thus not necessarily have a grammatical gender feature.

1362

G. Josefsson / Lingua 116 (2006) 1346–1368

pronoun with the features of the 4th semantic gender. This pronoun carries the feature setup +neuter, and possibly also definite.25 The null pronominal element has overt lexical correspondents with classifier-like properties. Predicative adjectival agreement is triggered by the features of the SemP. 4. A default alternative? In the traditional literature the use of agreement in what appears to be 3rd person, neuter, singular is usually thought of as default inflection. The question is of course whether the ideas presented in this article are just another version of a default approach. The answer is no. First of all, the idea of a default alternative in the traditional sense is a form that is inserted when nothing else really fits. In this sense, a default approach implies that the insertion of -t on a predicative adjective takes place when there is no agreement. In this article I have tried to show that the reverse holds: predicative adjectival agreement on -t is just as much agreement as other types of agreement; the features of the subject are reflected on the features of the predicative adjective. If the subject has a number feature, such a feature will have to be found on the predicative adjective as well; if the subject lacks number features, no number features are present on the predicative adjective. To show that this is actually the case it is crucial to make a clear distinction between the phonological form, i.e. the vocabulary items, and the feature content of the terminal nodes of the syntax. From a semantic point of view we find that the presence vs. absence of number features corresponds directly to a particular interpretation, i.e. bounded entities (‘‘things’’) vs. substances. The possibility of adding a SemP on top of the DP/NP allows us to capture the distinction between sentences such as (19a) and (20a), where adding a pronominal element of the 4th semantic gender renders the interpretation ‘‘aggregated substance’’ to the expression; the plural feature of the lower part is thus responsible for the interpretation ‘‘aggregated’’ (i.e. non-homogeneous, consisting of smaller parts), whereas the overall interpretation is derives from the pronoun in SemP. There is another sense in which the proposed analysis goes against the traditional idea of -t being the default inflection of predicative adjectives. If some adjectival inflection were to be appointed as the default one, it should not be -t but -Ø. The reason is the following: In a DM framework the notion of default vocabulary insertion is captured in terms of the Retreat to the general case (Halle and Marantz, 1993). If -t was the default inflection it would thus be used in the general case, and this is not what we find. First, there are good reasons to believe that the personal pronouns han ‘he’, hon ‘she’ jag ‘I’, and du ‘you’ lack a grammatical gender feature. One reason is that han ‘he’ and hon ‘she’ may be used as prenominal appositions also with neuter DPs. If han and hon had a grammatical gender feature (presumably common gender) we would expect a clash of features, causing ungrammaticality. Consider (24): (24)

25

a

Hon bitra¨de-t var sjuk/*sjuk-t iga˚r. she clerk-def.neut.sg was sick-common.sg/*sick-neut.sg yesterday ‘She/the clerk was sick yesterday.’

Another possibility is that this projection also contains person features. Consider Josefsson (1999) and Platzack (2003) for a discussion.

G. Josefsson / Lingua 116 (2006) 1346–1368

1363

Han statsra˚d-et var sjuk/*sjuk-t iga˚r. he secretary of state-def.neut.sg was sick-common.sg/sick-neut.sg yesterday ‘He/the secretary of state was sick last night.’

b

A second piece of evidence indicating that personal pronouns may lack a grammatical gender feature comes from the properties of right dislocated DPs. Consider (25)–(26): (25)

a

b

c d (26)

a

b

Den a¨r sen, buss-en. it.common.sg is late.common.sg, bus-def.common.sg ‘It is late, the bus’ Det a¨r sen-t, busshelvete-t. it.neut is late-neut.sg bushell-def.neut.sg ‘It is late, the damned bus.’ *Den a¨r sen, busshelvete-t. it.common is late.common.sg bushell-def.neut.sg *Det a¨r sen-t, buss-en. it.neut is late-neut.sg, buss-def.common.sg Han/hon var sen, den dumma a˚sna-n. he/she was late, def.common.sg stupid donkey-def.common.sg ‘He/she was late, the stupid donkey.’ Hon/han var sen, det dumma no¨t-et. she/he was late, def.neut.sg stupid cow-def.neut.sg She/he was late, the stupid cow.’

(25) shows that there must be a concord in grammatical gender between the pronoun and a right dislocated DP. (26) shows that han ‘he’ and hon ‘she’ may be linked to right dislocated DPs of either grammatical gender. This clearly indicates that han ‘he’ and hon ‘she’ lack a grammatical gender feature. It is quite possible that pronouns corresponding to han ‘he’ and hon ‘she’ in other languages contain a null noun provided with grammatical genders buried deep down in the lexical part of the extended projection, but this is simply not a plausible analysis for Swedish han ‘he’ and hon ‘she’. As pointed out above, the pronouns in question just convey reference to individuals belonging to a semantic gender, a female referent for hon and a male referent for han. If neuter gender was a default gender in Swedish – in the sense the gender that is assigned if no other grammatical gender feature is present – we would also expect that neuter would be added to the noun phrase in which the 1st and second personal pronouns appear, thus triggering predicative agreement in neuter.26 This does not happen, though:27

26

Consider also Trosterud (2001) who argues convincingly that neuter cannot be the default gender in Norwegian in other senses as well. Trosterud’s argumentation is applicable to Swedish as well. 27 Sauerland (1996) proposes that -e is the ‘‘Elsewhere alternative’’ for adjectival agreement in Norwegian, which would correspond to the default alternative. Norwegian and Swedish are fully comparable in the cases under discussion, with the difference that -e in Norwegian corresponds to -a in Swedish. However, Sauerland does not count definiteness as a part of the feature set-up, but as property of the context of insertion, creating weak and strong contexts. In my view such a procedure is neither minimal nor intuitive; we achieve a more coherent and unified solution if definiteness counts as a feature on par with number, gender, etc.

1364

(27)

G. Josefsson / Lingua 116 (2006) 1346–1368

Jag a¨r arg/*arg-t. I am mad/*mad-neut ‘I am mad.’

The conclusion is that neuter is not the default gender in Swedish. The question still remains whether there is any vocabulary item that serves as the ‘‘default’’ choice of adjectival inflection, in the sense ‘‘used in the general case’’, i.e. when no more specified inflectional item can be inserted. To investigate this question we must determine which features each piece of adjectival inflection is marked for. For this reason the discussion must be extended to both predicative and attributive adjectives. It is a well-known fact that adjectival predicative agreement is the same regardless of whether the subject is definite or indefinite. In other words, predicative adjectives are not specified for definiteness. (28)

a

b

Hunden a¨r opa˚litlig. dog.the is unreliable ‘The dog is unreliable.’ En ra¨dd hund a¨r ofta opa˚litlig. a scared dog is often unreliable ‘A scared dog is often unreliable.’

The form used when the adjective is in predicative position is the same as the one used in attributive position, provided the noun is indefinite. (This is sometimes called strong inflection.) When the noun is definite, however, a particular form of the adjective (weak inflection) is used in the attributive position, cf. (29b,c): (29)

a

b

c

en opa˚litlig hund a unreliable.sg.common.indef dog.sg.common.indef ‘an unreliable dog’ den opa˚litlig-a hund-en the unreliable-def dog-sg.common.def ‘the unreliable dog’ det opa˚litlig-a lejonet the unreliable-def lion ‘the unreliable lion’

The conclusion is that definiteness features are never encoded on predicative adjectives but only on attributive ones. I do not have a full explanation as to why this is the case, I will just take the facts at face value, assuming that definiteness is never part of predicative adjectival inflection. In a feature-based framework the agreement properties of (29) are easily captured by the assumption that the presence of the feature +definite on the adjective gives rise to vocabulary insertion of -a, regardless of the rest of the feature set-up. If the terminal node contains the plural feature (regardless of the predicative/attributive distinction, definite/ indefinite or common/neuter distinction) -a is inserted too. Note that hund ‘dog’ is a common gender noun, whereas lejon ‘lion’ is neuter:

G. Josefsson / Lingua 116 (2006) 1346–1368

(30)

1365

Hundar/lejon/hundarna/lejonen a¨r opa˚litlig-a. dogs/lions/dogs.the/lions.the are unreliable-pl ‘Dogs/lions/the dogs/the lions are unreliable.’

In conclusion, -a is the vocabulary item that is inserted as adjectival inflection if the syntactic node contains the feature +definite or +plural. I have argued that -t is inserted in positions where the definiteness feature is absent (i.e. in predicative position), but only in those cases where the noun carries a neuter feature but not a plural feature. This leaves us with the following set-up of features, where all features are marked for binary oppositions:28 (31)

A conflict may arise with the combination [+neuter, +definite] and [+neuter, +plural]. Provided we assume a hierarchy where the feature of grammatical gender is below number and definiteness, the conflict can be resolved; -a wins the competition in such cases. In the remaining cases, i.e. where definiteness features are not present (i.e. in predicative position) and where the subject lacks a grammatical gender feature, i.e. when the subjects are personal pronouns like han ‘he’, hon ‘she’, jag ‘I’ or du ‘you’, the adjective ends in -Ø: (32)

Han/hon/jag/du a¨r arg-Ø. he/she/I /you is/am/are mad

The conclusion is that, if anything, the default adjectival inflection in Swedish is -Ø, i.e. the inflection used if neither plural, definiteness nor grammatical genders are present, not -a and certainly not -t!29

28

There is one more option: In cases where the noun is male the inflection is -e: den glade gossen the happy.common.sg.def.male boy ‘the happy boy’ The vocabulary item -e is more specified than -a, and wins therefore the competition. 29 It is difficult to actually prove that -Ø is not an expression of the feature + singular, since bare, common gender nouns cannot be used as subjects together with predicative adjectives: (i) *Hare a¨r vit pa˚ vintern. rabbit is white during winter.the In sentences like (i) either an indefinite determiner or a definite and/or plural noun is required:

(i)

En hare/haren/harar/hararna a¨r vit(a) pa˚ vintern. a hare/hare.the/hare.pl./hare.pl.def are white during winter.the ‘The hare is white during the winter.’ As pointed out by an anonymous reviewer the postulation of two different -a markers is not desirable; a more principled account would clearly by preferred. However, an exact formulation of the rules for insertion of vocabulary items in the context of predicatives is not the main goal for me here; the main point is that -t is not the elsewhere alternative, and hence it cannot be considered a default alternative. (ii)

1366

G. Josefsson / Lingua 116 (2006) 1346–1368

(33)

5. Summary and discussion I have argued that Swedish has two gender systems, a semantic one and a grammatical one. The semantic gender system is split into animates and inanimates. The animate gender is further split into MALES and FEMALES; the inanimate gender is split into THINGS and SUBSTANCES, which comprise uncountables and unbounded entities. Pronouns can make use of the grammatical gender system, in which case a pronoun reflects the grammatical gender of a discourse antecedent (i.e. a phrase), but equally pronouns can make use of the semantic gender system, in which case a pronoun refers to one of the four semantic genders. What confuses the picture is that two of the pronouns, den (it.common) and det (it.neut) are used in both systems. The 4th semantic gender has been a focus of this paper. What is particular about this gender is that it comprises uncountable entities and hence lacks the grammatical number feature. As a result a noun phrase that belongs to this category cannot trigger predicative agreement in number, not even if the subject consists of conjoined phrases. The grammatical genders, neuter and common, are traditionally viewed as markings on nominal roots or stems. I have suggested that grammatical gender features are better viewed as features that enter into derivations just like other features, by the operation External Merge. The fact that a certain stem or root is usually combined with a certain grammatical gender feature should be viewed as a sort of idiomatic relationship. To use a metaphor: just because A and B is an old couple that are frequently seen together does not mean that one cannot occasionally go out without the other. This is what I have claimed has happened when common gender nouns like senap ‘mustard’ and gra¨dde ‘cream’ seem to trigger predicative agreement in neuter; the common gender feature never enters the derivation, so the root has to go out alone. The source of the neuter agreement is the 4th gender pronominal with properties similar to classifiers merged in the top of the nominal extended projection. An even more radical solution would be to consider grammatical gender features in a broader perspective, taking both nouns and pronouns into consideration. An interesting question is what grammatical gender is good for in general; many languages, including English, do very well without it. The most obvious answer if we look at Swedish is that a grammatical gender feature is needed for the spell-out of definiteness features on nouns; the form of the definiteness/indefiniteness morphemes depends on the grammatical gender of the noun. Perhaps this is the only function for grammatical gender, considering nouns in isolation. For pronouns, however, the function of grammatical gender features is to facilitate the establishing of discourse links, either between a pronoun and a preceding word carrying the same grammatical gender feature – this would be grammatically mediated discourse linking – or between a pronoun and a semantic category – semantic

G. Josefsson / Lingua 116 (2006) 1346–1368

1367

discourse linking. Grammatical gender features would thus not have any real meaning per se, only a function: in relation to phonology and discourse linking.

References Cinque, G., 1999. Adverbs and Functional Heads. A Cross-Linguistic Perspective. Oxford University Press, Oxford. Corbett, G., 1991. Gender. Cambridge University Press, Cambridge. ¨ ., 2000. Animacy and the notion of semantic gender. In: Unterbeck, B., Rissanen, M., Nevalainen, T., Saari, Dahl, O M. (Eds.), Gender in Grammar and Cognition II: Manifestations of Gender (Trends in Linguistics. Studies and Monographs 124). Mouton de Gruyter, Berlin and New York, pp. 577–593. Diderichsen, P., 1957. Elementær dansk Grammatik [Elementary Danish Grammar], 2 udg Gyldendal, København. Enger, H.-O., 2003. Pannekakesetninger er semantisk kongruens [Pancake sentences are semantic agreement]. Maal og Minne 1, 81–106. Falk, C., 1993. Non-Referential Subjects in the History of Swedish. Department of Scandinavian Languages, Lund University. Grimshaw, J., 1990. Argument Structure. MIT Press, Cambridge, MA. Gross, A., van Riemsdijk, H., 1981. Matching effect in free relatives: a parameter of core grammar. In: Beletti, A., Brandi, L., Rizzi, L. (Eds.), The Theory of Markedness in Generative Grammar. Proceedings of the 1979 GLOW Conference, Scoula Normale Superiore di Pisa, pp. 171–216. Halle, M., 1997. Distributed morphology: impoverishment and fission. In: Bruening, B., Yoonjung, K., McGinnis, M. (Eds.), MIT Working Papers in Linguistics 30, Papers at the Interface, pp. 425–449. Halle, M., Marantz, A., 1993. Distributed morphology and the pieces of inflection. In: Hale, K., Keyser, S.J. (Eds.), The View from Building 20: Essays in Honor of Sylvain Bromberger. The MIT Press, Cambridge, pp. 111–176. Harris, J., 1991. The exponence of gender in Spanish. Linguistic Inquiry 22 (1), 27–62. ¨ rter a¨r gott—Karakteristiskt fo¨r latinet a¨r rika bo¨jningsmo¨jligheter’. Om ursprunget till Heinertz, N.O., 1953. ‘A det predikativa adjektivets neutralform i dylika uttryck [‘Peas is good—characteristic for Latin is rich possibilities of flexion’. About the origin of the neutral form of the adjective in such expressions]. Moderna spra˚k 47, 257–276. Josefsson, G., 1997. On the Principles of Word Formation in Swedish [Lundastudier i nordisk spra˚kvetenskap A 51]. Lund University Press, Lund. Josefsson, G., 1998. Minimal Words in a Minimal Syntax. John Benjamins, Amsterdam. Josefsson, G., 1999. On the semantics and syntax of Scandinavian pronouns and object shift. In: van Riemsdijk, H. (Ed.), Clitics in the Languages of Europe. Mouton de Gruyter, Berlin and New York. Josefsson, G., 2001. The meaning of lexical classes. Nordic Journal of Linguistics 24, 218–231. Josefsson, G., 2000. Gender and related categories in a feature-based grammar. Manuscript. Department of Scandinavian languages, Lund University. Ka¨llstro¨m, R., 1993. Kongruens i svenskan [Agreement in Swedish]. Nordistica Gothoburgiensia 16 [Acta Universitatis Gothoburgiensis]. Department of Swedish, Go¨teborg. Kamiya, M., 2001. Dimensional approach to derived nominals. University of Maryland Working Papers in Linguistics 11, 134–162. Muromatsu, K., 1998. On the Syntax of Classifiers. Ph.D. dissertation, University of Maryland. Platzack, C., 2003. The Person Phrase. Manuscript. Department of Scandinavian Languages, Lund University. Sauerland, U., 1996. The Late Insertion of Germanic Inflection. Manuscript. MIT. Teleman, U., 1965. Svenskans genusmorfem [The gender morphemes in Swedish]. Arkiv fo¨r nordisk filologi 80, 217–230. Teleman, U., 1969. On gender in a generative grammar of Swedish. Studia Linguistica 23, 27–67. Teleman, U., 1987. Hur ma˚nga genus finns det i svenskan? [How many genders are there in Swedish?]. In: Teleman, U. (Ed.), Grammatik pa˚ villova¨gar. Svenska spra˚kna¨mnden. Stockholm, Liber, pp. 106–114.

1368

G. Josefsson / Lingua 116 (2006) 1346–1368

Teleman, U., Hellberg, S., Andersson, E., 1999. Svenska Akademiens grammatik [The Swedish Academy Grammar]. Svenska Akademien och Norstedts Ordbok. Trosterud, T., 2001. Genustilordning i norsk er regelstyrt [Gender assignment in Norwegian is rule governed]. Norsk Lingvistisk Tidskrift 19, 29–58. Vangsnes, Ø.A., 2001. On noun phrase architecture, referentiality, and article systems. Studia Linguistica 55, 249–299. Widmark, G., 1966. Den inkongruenta neutrala predikatsfyllnaden och dess plats i svenskans genussystem [The disagreeing neutral predicative and its position within the Swedish gender system]. Arkiv fo¨r nordisk filologi 46, 91–134.