The Semantics of the Dutch Gender System

Abstract

In this paper, it is argued that, although Dutch gender assignment is not systematically organized along semantic lines in the lexicon, the gender system has a semantic basis. This semantic basis involves a distinction between masculine/common gender associated with a high degree of individuation on the one hand, and neuter gender associated with a low degree of individuation, on the other hand. This is in line with Audring (2006, 2009), who found that Dutch pronouns often show semantic agreement along these lines. It is shown that the same semantic distinction between the genders can also be found in the nominal domain. It surfaces particularly in cases where lexically stored gender does not play a role. The semantic distinction arguably goes back to Proto-Indo-European. It is argued that, since nominal gender has become an invariable, lexically stored feature of nouns, the semantic basis of nominal gender assignment has become disrupted. This causes a con-flict between lexical and semantic gender agreement in pronouns. It is suggested that the surfacing of semantic agreement in this conflict is connected with a reduced marking of lexical gender on adnominal elements.

1. Introduction

This paper focuses on the gender system of Standard Northern Dutch (subsequently referred to as “Dutch”), a language with two nominal genders, common and neuter.¹ As in the other Germanic languages with two genders, common gender in Dutch is a conflation of the former mas-culine and feminine genders. The two nominal genders are marked on determiners and pronouns, and on attributive adjectives. The personal pronoun still marks three genders: masculine, feminine, and neuter. Table 1 gives an overview of the forms marked for gender. Gender dis-tinctions only exist in the singular. In the plural, the common gender forms are used.

Table 1. Dutch gender marked forms.

Dutch nominal gender assignment is largely arbitrary. Although there exists a limited set of semantic and formal regularities in nominal gender assignment, there is generally no motivation as to why one noun is com-mon and another noun is neuter gender (see, for instance, Donaldson 1987, Haeseryn et al. 1997, Booij 2002).

In pronominal gender agreement, two types of agreement are found: LEXICAL GENDER AGREEMENT, that is, agreement based on the lexical gender of the noun, and SEMANTIC GENDER AGREEMENT, that is, agree-ment based on the properties of the noun’s referent. An example of lexical gender agreement is shown in 1.³

1. (1)

2. 

When pronouns refer to animate entities, however, they often show agreement with the referent’s biological sex rather than with the lexical gender of the noun. An example is shown in 2, where the feminine pro-noun is used, while the antecedent noun is neuter.

1. (2)

This kind of semantic agreement with animate referents is not un-common in languages which otherwise show lexical gender agreement (see Corbett 1991).

Audring (2006, 2009) shows that another kind of semantic gender agreement occurs in pronominal reference to inanimate entities. This semantic agreement involves an opposition between the masculine/ common and neuter gender forms, and appears to be based on the degree of individuation of the referent. The masculine and common pronouns, on the one hand, are typically used for referents with a high degree of individuation, that is, discrete, countable objects with clear boundaries, such as a cup or a book. The neuter pronoun, on the other hand, is typically used for referents with a low degree of individuation, that is, nondiscrete, uncountable entities with unclear boundaries, typically masses, such as sugar or water.

This agreement pattern was first described by Van Haeringen (1936, 1951), who observed that any object may be pronominalized by a masculine pronoun (1936:20), and that the neuter pronoun, instead of the masculine pronoun, may be used for non-neuter mass nouns (1951:13). Audring (2006, 2009) has supported this early observation with spoken language data from the Corpus Gesproken Nederlands [CGN, Corpus of Spoken Dutch’]. Examples 3–6 below, adapted from Audring 2006:95–99, demonstrate this semantic agreement for both personal and demon-strative pronouns. Example 3 shows the use of a masculine personal pronoun to refer to a neuter noun denoting a discrete object.

1. (3)

‘Do you need some more information about that book? Then I shouldn’t return it yet.’⁴

Example 4 shows the use of a common gender demonstrative pronoun to refer to a neuter noun denoting a discrete object.

1. (4)

Example 5 shows the use of a neuter personal pronoun to refer to a common gender noun denoting a mass.

1. (5)

Example 6 shows the use of a neuter demonstrative pronoun to refer to a common gender noun denoting a mass.

1. (6)

This type of agreement is frequent. Audring (2009:16) found that in 65% of the cases where there is a conflict between lexical and semantic gender (that is, with common gender mass nouns and neuter gender count nouns), the choice of the personal/demonstrative pronoun is not determined by the noun’s lexical gender.

Semantic agreement is found with the relative pronoun as well, though less frequently—in 25% of the cases where there is a conflict between lexical and semantic gender (Audring 2009:160). It appears that the greater the distance between the antecedent and the pronoun—both structural/syntactic distance and plain word distance—the higher the likelihood of semantic agreement (Audring 2009:165–166). Examples 7a and 7b, adapted from Audring 2006:98–99, show the use of a common gender relative pronoun with a neuter noun, and a neuter relative pro-noun with a common gender noun respectively.

1. (7)

2. 

What these data show is the existence of semantic gender agreement in the pronominal domain of a language that otherwise has a lexical gender agreement system. Crosslinguistically, this phenomenon is not uncommon. The observation that semantic agreement can occur in the personal and demonstrative pronouns, and to a lesser extent also in the relative pronoun, while determiners show lexical gender agreement is consistent with Corbett’s (1991:226) Agreement Hierarchy. Corbett ob-serves that crosslinguistically, gender agreement can be represented on a hierarchy, with the likelihood of semantic gender agreement increasing the further one moves towards the right end of the scale:

1. (8)

According to this hierarchy, determiners and attributive adjectives (adnominal elements) are most likely to agree with the noun’s lexical gender, while personal and demonstrative pronouns are generally the first to show semantic rather than lexical gender agreement. This pattern is also observed in spoken Dutch.

The question is why this semantic agreement occurs in Dutch, and why it is based on individuation. In the next section, I discuss previous research showing that the semantic interpretation of the Dutch pronouns can be related to a semantic hierarchy called the Individuation Hierarchy, and that this interpretation of the pronouns is not unique to Dutch but has also been found in other Germanic languages. In this paper, I argue that individuation does not only play a role in the kind of pronominal gender agreement illustrated above, but in fact constitutes the basis of a semantic contrast that operates in the Dutch gender system as a whole. I show that the semantic agreement behavior of the pronoun relates to an existing semantic distinction between the genders, which arguably dates back to Proto-Indo-European. I propose that the surfacing of semantic agreement in pronouns is related to a reduced marking of lexical gender on adno-minal elements.

2. The Individuation Hierarchy

The observed use of masculine/common and neuter pronouns in Dutch can be described in terms of a distinction between referents on the basis of their position on the Individuation Hierarchy (Siemund 2002; Audring 2006, 2009). This hierarchy can be considered an elaboration of the Animacy Hierarchy (Silverstein 1976), in which the inanimate section of the hierarchy is further subdivided:

1. (9) The Individuation Hierarchy⁵

2. 

Entities on the left of the hierarchy have a high degree of individuation. They have a clearly bounded shape and specific characteristics. Entities on the right have a low degree of individuation. They are unbounded or have fuzzy boundaries, and are less specified. Animate entities have a high degree of individuation. Next to them on the hierarchy are bounded objects, such as a book or a table. Abstracts are intangible things, con-cepts. They can be bounded and countable, such as a problem, an idea, or a question, or unbounded and uncountable, such as pain, love, or hap-piness. Specific masses are specific instantiations of masses, such as “my mother’s soup” or “his tea.” Lowest on the hierarchy, with unbounded abstracts, are unspecific masses, such as snow, rain, or honey (Audring 2006:102–103).

In Dutch, the boundary between masculine/common pronouns and neuter pronouns is apparently situated somewhere between bounded and unbounded referents. Siemund (2002:224–225) shows that the pro-nominal gender systems of other Germanic languages can also be related to this hierarchy. The pronominal gender system of Standard English is based on a split between humans/other animates and inanimate entities, whereby the neuter pronoun is used for all inanimates. This constitutes a cut-off point that is situated further to the left than in Dutch:

1. (10)

Just as with the grammatical markings on the Animacy Hierarchy, the number of gender distinctions increases towards the left. In both English and Dutch, a further distinction between masculine and feminine gender is made on the left end of the hierarchy, with humans and other animates. Here pronominal gender varies between masculine and femi-nine depending on the biological sex of the referent.⁶

Not only in Dutch, but also in several other Germanic languages the boundary between common, or masculine and feminine, gender on the one hand, and neuter gender on the other hand, lies between bounded and unbounded referents. Siemund (2002) notes that in some regional vari-eties of English, pronouns show a mass versus count distinction. Particularly, in the English dialect of West Somerset, masculine and feminine pronouns are used for count nouns (that is, with bounded refer-ents), while neuter pronouns are used for mass nouns.⁷ This phenomenon in South Western dialects was also observed by Jespersen (1924:240):

Here and there we find a tendency to establish a grammatical distinction between thing-words (countables) and mass-words (uncountables) […] In the south-western dialects of England “full shapen things” are referred to as he, acc. en (from OE hine) and take the pronominal adjuncts theäse, thik, while “unshapen quantities” are referred to as it and take this, that.

In several varieties of Frisian as well, the choice between a masculine or neuter pronoun depends on the bounded versus unbounded nature of the referent (Wahrig-Burfeind 1989). The West Frisian vari-eties of City Frisian (Stedsk) have two nominal genders, common and neuter. With mass nouns, regardless of whether they are common or neuter, the neuter personal pronoun ‘it/’t is systematically used. Count nouns, however, are not always pronominalized by the masculine pronoun.

In the Frisian variety spoken on the island of Helgoland, a stricter bounded-unbounded distinction in pronouns is observed. In this variety, the neuter demonstrative/personal pronoun deät is always used for masses and abstracts, and the masculine personal pronoun hi/hem is systematically used for countable objects, regardless of whether the corresponding noun is common or neuter gender.

In the Danish dialect of West Jutland, the same bounded-unbounded distinction is made (Ringgaard 1973:30–31). Countable objects are referred to by the common (uter) demonstrative pronoun den, while masses and abstracts are referred to by the neuter demonstrative pronoun det. What is particularly interesting about West Jutish is that the semantic distinction is not only made in pronouns, but also within the noun phrase. West Jutish is generally said to have lost its nominal gender distinctions, since all nouns combine with the determiner œ ‘the’ or en ‘a’ (for example, œ/en hus ‘the/a house’). However, the pronouns den and det are also used as demonstrative determiners, and within the noun phrase, they make the same semantic distinction as in the pronominal domain: Nouns that denote countable objects combine with den, while nouns that denote masses or abstracts combine with det, for example, den hus ‘that house’, but det mœlk ‘that milk’, det skrigen ‘that shouting’. The determiner can even vary for the same noun depending on whether its referent is concep-tualized as a countable item or a mass. The noun fisk ‘fish’, for instance, can combine with either den or det depending on whether the speaker refers to an individual fish or a type of food. It seems, therefore, that what West Jutish has lost is not gender in the nominal domain, but rather lexically stored nominal gender. Having lost that, the gender system of West Jutish is now entirely based on the semantics of individuation.

In contrast, in Dutch, lexically stored gender still plays a role, and pronominal gender can conflict with the lexical gender of a noun. In the following section, I address the question of what makes it possible for semantic agreement to surface on a large-scale in a gender system that is otherwise based on lexical gender.

3. The Rise of Semantic Agreement

Audring (2006, 2009) proposes that the observed semantic agreement in the pronominal system of Dutch is caused by a mismatch between the numbers of nominal and pronominal genders. She argues that, since masculine and feminine gender have conflated into one common gender, there exists a mismatch between the nominal and the pronominal gender system: Only two genders exist in the nominal domain—common and neuter, while three genders—masculine, feminine, and neuter—exist in the domain of the personal pronoun. According to Audring (2006:87–88), this means the following in practice:

speakers have lost the knowledge about masculine and feminine nouns, while common gender nouns in Dutch are unable to govern pronoun choice in the usual way, as there is no personal pronoun corresponding directly to a common gender noun. This problem arises whenever a speaker wants to pronominalize such a noun.

However, this claim is not uncontroversial. Although personal pro-nouns distinguish three genders, while nouns distinguish only two, in practice, this does not necessarily involve a mismatch. It should be noted that in spoken language, the feminine pronoun is, in fact, only used for semantic agreement with female referents. Audring’s findings in the CGN confirm this: Feminine pronouns are found exclusively with female referents (Audring 2009:92). It seems, therefore, that at least in spoken language, the feminine pronoun is reserved for semantic gender agree-ment and no longer plays a role in lexical gender agreement. If this is the case, then there is no mismatch between the numbers of nominal and pronominal genders with respect to the lexical gender agreement system.

Another question that arises with the mismatch analysis is how the new division of labor between the masculine and neuter pronouns is motivated by the proposed mismatch. Audring (2006, 2009) suggests that the semantic reinterpretation of these pronouns arises as a solution to the mismatch problem. However, the mismatch exists between the common gender noun on the one hand, and the masculine and feminine pronoun, on the other hand.

1. (11)

If this situation causes a reinterpretation of the pronominal genders, then a new division of labor between the masculine and feminine pro-noun is expected. This has, in fact, already occurred in the spoken language since the feminine pronoun became more and more reserved for semantic agreement with animate, female referents and the masculine pronoun came to replace it in other cases. This new division of labor between the masculine and feminine pronoun was already well on its way by the end of the 17th century (see Geerts 1966:192–210, Schönfeld & Van Loey 1970:119–120).

These historical developments can be roughly schematized as in 12.

1. (12)

Starting in the Late Middle Dutch period, deflection gradually led to the conflation of masculine and feminine nominal gender. It was in the 17th century that a mismatch had developed between the available nominal and pronominal genders because masculine and feminine nominal gender had merged into one category. According to Geerts (1966), this mismatch was gradually resolved as the feminine pronoun was in-creasingly replaced with the masculine pronoun, while the feminine form became reserved for agreement with female referents.⁸

Therefore, at this point it is not immediately clear that it is the discrepancy between the numbers of nominal and pronominal genders that causes the reinterpretation of the masculine and neuter pronoun in semantic terms. Yet there does appear to exist at least an indirect relation between semantic agreement and a reduction in the number of nominal genders. Braunmüller (2000:44), for instance, observes for the North Germanic languages that semantic gender agreement, on the basis of individuation or biological sex, only occurs in those languages/dialects that have reduced the original three-way nominal gender system to a two-way or zero gender system, such as Standard Danish, Standard Swedish (two genders) and the West Jutish dialect (zero genders), and not in varieties that still employ a three-way gender system, such as Ice-landic.

However, the distinction among the Germanic varieties is not that clear cut. Semantic agreement is also found in Germanic varieties that have not lost a nominal gender. For example, in German—a language with three nominal genders solidly in place—semantic agreement occurs, albeit to a more limited extent than in Dutch. Pronouns can show semantic agreement with animate referents. For instance, with neuter nouns like Mädchen ‘girl’, the feminine as well as the neuter pronoun are used (Corbett 1991:228, Audring 2009:193). Semantic agreement is also possible at the other end of the Individuation Hierarchy. According to Audring (2009:193), neuter pronouns can be used with non-neuter nouns with referents of low individuation. She provides the—constructed—examples in 13.

1. (13)

2. 

So, at least with referents on the outer ends of the Individuation Hierarchy, where the potential conflict between semantic and lexical gender is greatest, semantic agreement is possible in German. This suggests that semantic agreement is not an all-or-nothing phenomenon in the Germanic varieties but rather a matter of degree.

This view also emerges from studies on dialects of Dutch. Using written questionnaires, De Paepe & De Vogelaer (2008), De Vos (2009), and De Vogelaer & De Sutter (2011) found that in East Flemish dialects with three nominal genders, the neuter pronoun is often used with mas-culine and feminine nouns referring to abstracts or masses.⁹ Comparing the results for different East and West Flemish dialects, De Vogelaer & De Sutter (2011) further observed that this type of agreement is more frequent in those dialects where the distinction between masculine and feminine gender is not marked on the indefinite article (but only, in a limited number of phonological contexts, on the definite article and attributive adjectives).

Considering this, it is possible that semantic agreement correlates not with the loss of a nominal gender, but rather with the status of the lexical gender system in general. Compared to a language like German, where semantic agreement is considerably less frequent, Dutch does not only have a reduced number of nominal genders, but it also has a reduced agreement system with respect to the number of elements that show gender agreement. Dutch indefinite determiners no longer show gender agreement, the possessive determiner shows very limited gender agree-ment, and attributive adjectives show gender agreement only when they appear in indefinite noun phrases (see table 1). Importantly, these are adnominal elements, which constitute agreement targets that most reliably express lexical gender (see the Agreement Hierarchy in section 1). The frequent occurrence of semantic agreement in Dutch may be due to the weakening of its lexical gender system, in the sense of a reduced marking of lexical gender on adnominal elements. It is possible that semantic gender agreement can gain ground only if the lexical gender system has become sufficiently reduced.

An important question that remains is why semantic agreement in pronouns occurs along the lines of individuation. Apparently, there exists a tendency in the Germanic languages to make a gender distinction between individuated and nonindividuated referents. It should be ex-plained where this semantic interpretation of the genders comes from; more specifically, why the masculine/common pronoun is connected with a high degree of individuation and the neuter pronoun with a low degree of individuation. The answer to this question may also explain why a new division of labor between masculine/common and neuter is found, while neuter gender was never involved in a mismatch between nominal and pronominal gender. In the following sections, I argue that the individuation distinction found in pronominal gender agreement re-lates to an existing semantic distinction between neuter and non-neuter gender in Dutch. In section 4, I first show that the neuter gender pronoun is more generally connected with the notion of low individuation.

4. The Semantics of Neuter Gender

It can be argued that the use of the neuter pronoun with objects of low individuation is connected to a much larger semantic role of neuter gen-der. The neuter pronoun is used not only for unbounded objects or lexically neuter nouns. It is also used with non-nominal antecedents that denote activities, processes, states, properties, events, or propositions, henceforth NONENTITIES. Consider the use of the neuter personal and demonstrative pronouns in the examples in 14.

1. (14)

2. 

In these examples, the neuter pronouns refer to activities. In 14a, the neuter demonstrative refers to the activity of ruining the garden, and in 14b, the neuter demonstrative and personal pronoun both refer to the activity of cleaning up the house.

In 15a below, the neuter pronoun refers to a state (that is, being grey) and in 15b it refers to a property (that is, happy).

1. (15)

In 16, the neuter pronouns refer to the entire proposition.

1. (16)

The neuter pronoun is also used when there is no (clear) referent at all, as in 17, where it functions as an expletive.

1. (17)

This use of neuter gender with nonentities is quite common in the Indo-European languages that have a neuter gender form available. Corbett (1991:204–205) discusses the use of neuter gender in several Indo-European languages for what he refers to as nonprototypical controllers, that is, elements with no specified gender, such as clauses and infinitival phrases. In the three-way gender system of Russian, for example, the neuter verb form is used to agree with infinitival subjects (example 18a, from Corbett 1991:204). Similarly, in Serbo-Croat, it is the neuter pronoun that agrees with predicative adjectives (example 18b, from Corbett 1991:205).

1. (18)

Corbett (1991:206) notes that this use of the neuter form cannot be explained by markedness theory, since it is not immediately clear that neuter is the unmarked form in the languages in question. This is cer-tainly the case for Dutch, where common gender appears to be the unmarked form. Common gender is always used in the plural, where gender distinctions are neutralized, which makes it the most frequently occurring form in the determiner paradigm. Furthermore, the large majority of nouns—approximately three out of four—are common gender, while only one out of four nouns is neuter (Van Berkum 1996).¹⁰ During the acquisition of nominal gender, the common determiner is also largely overgeneralized before the neuter determiner is used (Van der Velde 2004, Blom, Polişenská, & Weerman 2008). However, despite this, in several ways, default status of common gender in Dutch, it is neuter gender that is used in reference to nonentities.

Corbett (1991:205–206) proposes to explain this use of neuter gender in languages such as Russian and Serbo-Croat in terms of the distribution of the genders in the lexicon. He notes that in these languages, almost all nouns that denote humans are masculine or feminine, while nouns that denote inanimates can be masculine, feminine, or neuter. Given this distribution, Corbett (1991:206) argues, the selection of neuter gender for agreement with non-nominal antecedents could be considered the most appropriate option in semantic terms. This analysis could be used to explain the use of neuter with nonentities in Dutch as well. Here, a simi-lar distribution of common and neuter gender exists: Nouns that denote humans tend to be common rather than neuter gender (see Haeseryn et al. 1997).

However, a different approach is possible. Instead of departing from the distribution of the genders in the lexicon, one could consider the use of neuter gender with nonentities to reveal something about the seman-tics of neuter gender itself. Returning to the Individuation Hierarchy discussed above, it can be argued that neuter gender is more extensively connected with the meaning of low individuation than the present hierarchy reveals. The larger semantic domain of neuter gender becomes more apparent if the Individuation Hierarchy is extended on the right side with nonentities—the category that can be considered the least individuated in the system:

1. (19)

Under this approach, the observed division of labor between masculine/common and neuter pronouns in Dutch can be said to reside within a broader system of semantic categories: While common gender is used for referents of high individuation, neuter gender is used for a wide range of referents of low individuation, from abstracts/masses to all other possible referents that are not bounded in nature, such as events, activities, and propositions.¹¹

Considering this function of neuter gender, it is interesting to note that there are several Indo-European languages which otherwise have a two-way masculine-feminine gender system, that still use a neuter form with nonentities (Corbett 1991:214–215). Corbett’s examples include Portuguese, the Surselvan dialect of Romansh, and Spanish. Below is a Spanish example from Corbett 1991:214.

1. (20)

The neuter pronoun ello ‘it, that’ refers to the activity of going to football matches. This kind of reference is the only function of the neuter form in Spanish, for there is no class of neuter nouns in the language. It seems that, while the original three-way gender system of Romance has been reduced to two, neuter gender has been preserved for this type of agreement (Corbett 1991:215). With the proposed extension of the Individuation Hierarchy in mind, it can be said that the Spanish gender system operates on this hierarchy, with the boundary between neuter and feminine/masculine gender located between masses and nonentities.

So far, it has been shown that the individuation distinction observed in Dutch pronominal gender agreement is connected to an existing semantic distinction between neuter and non-neuter pronouns. In the following section, I show that this semantic gender distinction is not restricted to pronouns, but also exists in the nominal domain.

5. The Individuation Distinction in the Nominal Domain

It is generally assumed that Dutch nominal gender assignment is not based on any semantic principle, but that it is arbitrary. Although this is largely the case, as discussed further in section 6.2, in certain cases the semantics of individuation appear to play a role in nominal gender assignment.

5.1. Nominalized Adjectives

The individuation distinction is reflected in the choice of the determiner that appears with nominalized adjectives. In Dutch, adjectives can be nominalized by adding a determiner to the inflected form of the adjective. This process is fully productive, that is, any adjective can be nominalized in this way. If there is a nominal antecedent, the determiner-adjective construction could be analyzed as an elliptical phrase, where the gender of the determiner simply corresponds to that of the elided noun, as in 21.

1. (21)

However, in spoken language, the determiner can also agree with the degree of individuation of the referent rather than with the lexical gender of the elided, or antecedent noun. If, as in 21b above, the elliptical phrase refers to a bounded object, the determiner can also be common gender, regardless of the neuter gender of the corresponding noun.¹²

1. (22)

If the determiner-adjective construction occurs without a corre-sponding noun in the linguistic context, the individuation distinction becomes even more apparent. In this case, individuation alone appears to determine the form of the determiner. The neuter determiner is always used if the referent is a nonentity (examples 23a–b), while the common determiner is always used if the referent is animate (examples 23c–d).¹³

1. (23)

2. 

These nominalized adjectives can be considered ad-hoc, nonlexicalized noun formations that do not have a lexically stored gender. In this case, gender assignment occurs along the lines of individuation.

In the sections below, I show that the individuation distinction can, in some cases, be found with lexically stored gender as well.

5.2. Double Gender Nouns

Lexically stored nominal gender shows much arbitrariness. Compare, for example, de trui ‘the sweater’, de jurk ‘the dress’ but het hemd ‘the shirt’; de stoel ‘the chair’, de kast ‘the cabinet’ but het bed ‘the bed’. These objects are the same in terms of individuation, and they do not seem to differ systematically in any other way. Yet the nouns take different genders.

Nevertheless, the individuation distinction can be observed with lexical gender in some cases. It is visible with nouns that can be either common or neuter gender, with a concomitant difference in meaning. In such minimal pairs, the neuter variant often has a less individuated referent than the common gender variant. The most straightforward ex-amples are found among nouns denoting materials. When a noun can denote both a material and an object made of that material, it is usually neuter when it refers to the material and common when it refers to the object (Haeseryn et al. 1997). In the minimal pairs in 24, the neuter variant is a mass noun referring to a material, while the common gender variant is a count noun referring to an object.

1. (24)

The neuter noun zout ‘salt’ shows a similar gender alternation. In spoken language, when a speaker refers to the salt in a saltshaker, typi-cally at the dinner table, a common determiner is sometimes used. This gender alternation is illustrated in 25.

1. (25)

(Example 25b taken from http://www.onzetaal.nl/advies/zout.php, accessed on January 24, 2009.)

In a context like 25a, where zout has an unbounded interpretation, the noun is always neuter, while in a context like 25b the noun can take a common determiner. This use of the common determiner is not readily explained by some kind of ellipsis from a compound, as a likely can-didate would in that case be zoutvaatje ‘saltshaker’, which is neuter. It appears that in this context, the preference for common gender is due to the individuated nature of salt in a saltshaker.¹⁴

5.3. Nouns from the Same Semantic Domain

Besides gender alternations for the same noun, the gender of nouns that belong to the same semantic domain can also show the individuation distinction. Zubin & Köpcke (1986) provide an elaborate description of nominal gender in German, in which they examine the semantic differ-ences between neuter and non-neuter nouns in several semantic fields. They show that nouns referring to superordinate categories—that is, overarching terms that subsume more specific categories—are usually neuter, while nouns referring to more specific categories—that is, more basic level terms—tend to be masculine or feminine. This gender distinction between neuter and non-neuter can be observed in Dutch as well. Some examples are listed in 26.

1. (26)

Zubin & Köpcke (1986:144) observe that neuter gender is associated with a lesser differentiation in the lexical hierarchy and a corresponding lack of precise characteristics, while masculine and feminine gender (common gender in Dutch) are linked to a greater differentiation in the lexical hierarchy and hence more precisely defined characteristics. This distinction relates to the Individuation Hierarchy: The referents of super-ordinate terms are less specific than those of basic level terms. Therefore, superordinate terms are associated with a lower degree of individuation than their corresponding basic level terms. Thus, the gender distinction between the two categories follows the individuation distinction.

The examples of the individuation distinction in lexical gender make up only a limited proportion of the lexicon, however. Therefore, they cannot be said to represent a consistent pattern in lexical gender assignment.¹⁵ Nevertheless, their existence, together with the gender assignment pattern observed in productive nominalizations, indicates that the semantic agreement pattern found in pronouns is not an isolated phenomenon within the Dutch gender system. Rather, the behavior of pronouns reveals a semantic principle that operates in the gender system as a whole.

Possibly, the semantics of individuation have always played a role in the gender system. The individuation distinction in the nominal domain may have been inherited from an earlier stage of the language when nominal gender classification had a more transparent semantic basis. In general, it is likely that the assignment of nouns to different genders has not always been arbitrary, but was originally based on some semantic distinction between their referents. In the following sections, I discuss the possible semantic basis of the Indo-European genders, and how it became disrupted.

6. The Semantic Basis of Nominal Gender and Its Disruption

6.1. The Original Meanings of the Indo-European Genders

Although the Indo-European genders are traditionally labeled masculine, feminine, and neuter, it is not at all certain that gender classification in the Indo-European languages originally involved a distinction on the basis of biological sex. Fodor (1959) notes that these gender labels were adopted in the heroic age of grammar writing for the modern European languages. The 19th century grammarians were proud to discover that the same three genders existed in the modern European languages as in the honored classical languages. Driven by the idea of anthropomor-phism, they thought it “reasonable and consequently superb that natural gender [that is, biological sex, MK] was reflected in language through such subtle grammatical devices” (Fodor 1959:3).

One notable proponent of the natural gender basis was Jacob Grimm. In Grimm 1831, he claimed that the gender system was originally a reflection of biological sex, which was subsequently extended to nouns with inanimate referents through the human imagination. This would have involved the personification of inanimate objects, which were classified as having either typically masculine (such as big, firm, active), feminine (such as small, weak, passive) or sexually indifferent (such as general, underdeveloped) characteristics.

Brugmann (1889, 1897), however, opposed this view and claimed that the association of the genders with biological sex was only a secondary phenomenon. He argued that the class of nouns labeled “feminine,” that is, the nouns ending in -ā and -ī, was not made up of nouns that referred to female animate beings, but of collective and abstract nouns. This view still stands today. Furthermore, evidence has now converged on the view that Proto-Indo-European originally had only two genders, common and neuter, and that the third, feminine, gender developed later (Matasović 2004:33, 165; Schwink 2004:12).¹⁶ There-fore, it is unlikely that biological sex formed the semantic basis of the Indo-European genders.

At the same time, there are indications that the genders were connected with different degrees of individuation. Lehmann (1958:188–192) reconstructs a paradigm of nominal suffixes in Proto-Indo-European that gave rise to the genders. Making use of insights from phonological theory, Lehmann argues that the long vowel endings -ā, -ī, and -ū resulted from a merger of different short vowels with the same ending, the laryngeal -h. Thus, three nominal endings could be distinguished that gave rise to the genders: -s (“masculine”), -h (“feminine”), and -m (“neuter”). According to Lehmann, these suffixes were systematically connected with certain meanings. They corresponded to the respective meanings “individual,” “collective,” and “resultative,” later including “mass.”¹⁷

Every noun could be inflected with each of these suffixes, which modified the noun’s meaning in a specific way. For example, the “feminine” form himā (< himah) had the collective meaning ‘season of cold and frost’ or ‘winter’, the “masculine” form, himás, referred to a singular occurrence of cold or frost, while the “neuter” form, himam, meant ‘snow’, which could be paraphrased as “the result of cold”. Thus, the same noun could have different forms associated with systematic differences in meaning.

Following Lehmann 1958, Leiss (2000) argues that remnants of this meaningful gender alternation can still be found in Old High German. In Old High German, many nouns occur in two or three different genders, which is usually interpreted as arbitrary variation leading up to a gender change. Leiss argues, however, that these different genders can often be related to the categorical meanings count (masculine), collective (femi-nine), and mass (neuter). An example of such gender variation is the Old High German noun that means ‘wood’ or ‘string of wood’. The mass meaning was expressed by the neuter form witu ‘wood’, while the col-lective meaning was expressed by the feminine form witta ‘string of wood, headband’.

Moreover, in a reconstructed lexicon of Proto-Indo-European nouns, Matasović (2004:133–134) observes semantic tendencies in gender assignment that operate along the lines of individuation. Nouns with referents at the leftmost end of the Animacy Hierarchy (humans, animals, and most plants) are consistently common (masculine/feminine) gender, while nouns with referents at the rightmost end (fluids and masses) are consistently neuter. Nouns with referents located between these two extremes show a less consistent assignment pattern. Interestingly, how-ever, nouns denoting abstracts are always assigned either feminine or neuter gender.

Considering the above, the semantic basis of gender classification in Indo-European likely involved a distinction between the meanings coun-table/individual, mass, and collective/abstract. The gender distinction on the basis of biological sex would have developed later, in addition to these original meanings.

1. (27)

Thus, each of the Indo-European genders corresponded to a different degree of individuation. Masculine gender (countable entities) was associated with the highest degree of individuation, while neuter gender (masses)—with the lowest. Feminine gender (collective/abstract entities) was associated with a degree of individuation that lies between the two extremes.

These semantic distinctions still appear to play a role in the Dutch gender system today. Particularly in the choice of pronouns, the individu-ation distinction, as well as the biological sex distinction, are involved. As discussed in the previous section, the semantics of individuation can be observed in lexical gender as well, although to a limited extent. The gender of many nouns does not follow the individuation distinction and appears to be arbitrary in terms of semantics. In the following section, I argue that this situation may have resulted from a gradual, though large-scale, disruption of the semantic basis of gender classification in the lexicon.

6.2. Disruption

The fact that nowadays the gender of most Dutch nouns is arbitrary shows that there is no longer a semantic principle underlying lexical gender. The original semantic classification has somehow become largely disrupted in the nominal domain. One factor that must have been particularly important in this process is that nominal gender has become an invariable, lexically stored feature of the noun rather than a variable feature that modifies the noun’s meaning.¹⁸ It is possible that once nomi-nal gender becomes lexicalized, the semantic principle underlying gender classification becomes susceptible to disruption. In time, a noun tends to change its meaning, and if its gender remains the same, the connection between the meaning of the noun and its lexical gender can be lost.

Importantly, meaning shifts along the Individuation Hierarchy are not uncommon. One common type of meaning change involves the development of a more concrete, or specific meaning from a more ab-stract, or general one (narrowing). Another common change proceeds in the opposite direction and involves the development of a more abstract, or general meaning from a more concrete, or specific one (widening) (Williams 1986:170–177). These semantic changes are particularly dis-ruptive in a gender system based on individuation, as they involve shifts along the Individuation Hierarchy, from a lower to a higher degree of individuation in the case of narrowing, and from a higher to a lower degree of individuation in the case of widening.

An example of the shift towards a higher degree of individuation is the development of the neuter noun brood ‘bread, loaf of bread’. Nowadays, it can either be a mass noun, denoting the substance bread, or a count noun, denoting a loaf of bread. This noun dates back to Old Germanic, possibly Proto-Germanic (Philippa et al. 2007). It is likely that it has always been neuter, for according to the dictionary of Middle Dutch (CD-ROM Middelnederlands 1998), it was neuter in Middle Dutch and in Middle High German. According to Philippa et al. 2007, the noun originally meant ‘food’ or ‘bread in general’. Only later did it develop the meaning ‘a piece of bread’. Thus, the neuter gender of the noun originally corresponded with the low degree of individuation of its referent. However, since the noun has also come to denote a unit of bread, while retaining its gender, there is now a discrepancy between the individuated meaning and the neuter gender of the noun. It is possible that over time, many more nouns have changed their meanings or have developed additional meanings along similar lines, while maintaining their lexically stored gender.

A more drastic disruption of the connection between semantics and gender occurs when a productive suffix systematically connected with a specific gender changes its meaning. A salient example is the diminutive suffix -(t)je, which goes back to Old Dutch -kī and Proto-Germanic -kīn- (Philippa et al. 2007). This suffix is systematically connected with neuter gender. It is likely that the diminutive originally denoted a young child (see Jurafsky 1996). With this denotation, the neuter gender corresponds with the absence, or irrelevance, of a specified sex for the referent. How-ever, gradual metaphorical extension of this original meaning has led to several additional meanings, including partition. The suffix has often been used to derive count nouns from mass nouns, and it is still moder-ately productive in this function. Some examples of the partitive meaning of the Dutch diminutive suffix are given in 28.

1. (28)

Although the partitive meaning of the suffix follows quite naturally from its original semantics—that is, from child to smallness to partition—it creates a structural violation of the connection between meaning and gender. The concrete, countable meaning of the partitive nouns on the right in 28 in no way corresponds with the semantics of neuter; yet the diminutive suffix systematically connects them with neuter gender.

The example of the diminutive suffix shows that the connection between gender and form is very strong. This connection is maintained even when the meaning of a morpheme changes and no longer motivates its gender or even contradicts it. This persistence of gender-form pairings can be an important factor in the gradual disruption of the semantic basis of nominal gender classification.

At the same time, this strong connection between gender and form can be responsible for the development of more formal—phonological and/or morphological—gender systems. For example, although the Romance languages, like the Germanic languages, no longer have a transparent semantic basis for nominal gender assignment, many of them do show a systematic connection between gender and form. In Spanish, for instance, nouns ending in -a are usually feminine, while nouns ending in -o are usually masculine (Mel’cuk 1974). Thus, Spanish can be con-sidered to have a formal gender assignment system.

In this respect it is interesting to note that even in gender assignment systems that are predominately based on form, there always remains a semantic core to nominal gender assignment (Corbett 1991:8, Corbett & Fraser 2000:294). In their typology of gender systems across the world, Corbett (1991) and Corbett & Fraser (2000) broadly distinguish between two types: (predominantly) semantic systems, in which nominal gender is assigned completely or largely according to meaning, and predom-inantly formal systems, in which gender is assigned largely according to the morphological or phonological properties of a noun. Interestingly, while there exist purely semantic gender systems, there appear to be no systems based solely on form.¹⁹ Semantics always play a role in gender assignment, and importantly, the formal and semantic assignment criteria always overlap to some extent (Corbett 1991:63). This observation supports the idea that gender systems are primarily based on semantics, while formal regularities may develop in tandem. If the semantic basis for gender assignment becomes opaque, the nominal form itself may eventually become the basis for gender assignment, resulting in a formal gender system.

However, although some formal regularities exist in nominal gender assignment, the Germanic languages have not developed transparent formal gender systems. The development of a formal system could have been prevented by the Germanic nominal stress pattern. Since in Proto-Germanic nominal stress became fixed on the initial syllable, there has always been erosion of nominal endings (Barber 2000:92). This has led to the loss of many transparent formal gender distinctions on nouns. Consequently, lexical gender is now largely arbitrary in the Germanic languages, with respect to form as well as semantics.²⁰

7. Semantic versus Lexical Gender Agreement

Since the semantic basis of nominal gender assignment has become disrupted, the lexical gender of nouns no longer always reflects the properties of their referents. This situation causes a conflict between lexical and semantic gender agreement in pronouns. Such a conflict of course only exists if the genders do, in fact, correspond to particular referent properties. It seems that the Dutch genders have retained their semantic interpretation, that is, masculine/common gender is associated with high individuation and neuter gender with low individuation, despite the fact that these meanings have become opaque in the lexical domain. What could have played a role in the retention of the semantic interpretation of the genders is that lexical gender is not always involved in pronominal reference. This is the case in deictic reference, that is, in the absence of a linguistic antecedent, or when the antecedent is not ex-pressed nominally, as in reference to nonentities. When no nominal antecedent is present, lexical gender does not have to interfere and pronouns can agree semantically.

In the presence of a nominal antecedent, however, lexical and semantic gender are in competition for agreement if the gender of the noun is not in accordance with the properties of the referent. Which of these two genders wins must somehow depend on the amount of pull that semantic and lexical gender exert on the pronoun. Several factors appear to play a role here: the place of the referent on the Individuation Hierarchy, the distance between the pronoun and the antecedent, and, more generally, the amount of lexical gender marking in the nominal domain.

The pull from semantic gender appears to become greater as the degree of individuation of the referent is more saliently high or low. As discussed in section 3, semantic agreement is less common in German than in Dutch, but it can occur with referents on the extreme ends of the Individuation Hierarchy. Audring (2009:167–168) also observes for Dutch that semantic agreement is more likely to occur with animates and unspecific masses than with referents that take a position in between these two extremes.

The pull from lexical gender seems to depend, first of all, on its degree of overt expression in the language in general. As discussed in section 3, lexical gender agreement takes precedence over semantic gen-der agreement in languages that have many adnominal elements marking lexical gender. Second, the pull from lexical gender appears to depend on how far the pronoun is separated from its nominal antecedent. As noted earlier, the likelihood of semantic agreement increases the further a pro-noun is separated from the antecedent noun (Audring 2009:165–166). Apparently, the longer ago a noun was expressed, the less weight its lexical gender has in the competition for gender agreement. It could be said that it is generally the saliency of lexical gender, in the agreement system in general and in the actual context of the agreeing pronoun, that determines the pull from lexical gender in the competition between lexical and semantic agreement.

8. Conclusion

I have argued that the semantic agreement behavior observed in Dutch pronouns reveals a semantic principle that operates within the Dutch gender system as a whole. The gender distinction that is made between referents based on their degree of individuation was shown to relate to an existing semantic interpretation of the Dutch genders that arguably goes back to Proto-Indo-European. These semantics come to the fore most clearly in cases where no lexically stored gender is involved, such as in nominalized adjectives and in pronominal reference to non-nominal ante-cedents. It was suggested that ever since nominal gender became an invariable, lexically stored feature of nouns, the semantic basis for nominal gender assignment has been susceptible to disruption through meaning changes. This analysis explains why lexical nominal gender is now largely arbitrary. This situation creates a conflict between lexical and semantic gender, which currently surfaces in Dutch and several other Germanic languages, where pronouns show variation between lexical and semantic gender agreement. Considering the distribution of semantic gender agreement in the Germanic languages, it was suggested that there is a relation between semantic gender agreement in pronouns on the one hand, and the number of adnominal elements marking lexical gender, on the other. Semantic agreement may only gain ground if the marking of lexical gender within the nominal domain has become sufficiently reduced.