2.1 Grammatical gender in Norwegian

Grammatical genders are usually defined as “classes of nouns reflected in the behavior of associated words” (Hockett 1958: 231; see also Corbett 1991: 1). In Norwegian, gender is expressed on DP-internal elements (adjectives, indefinite articles, demonstratives, possessives), anaphoric pronouns, and predicative elements (adjectives and possessives). Table 1 shows an idealized paradigm of singular gender forms for Norwegian based on the written varieties Nynorsk and Bokmål.

The distinction between feminine and masculine gender is optional in the written language Bokmål and some spoken varieties (see Section 2.3). Other dialects have additional distinctions between feminine and masculine gender on definite (weak) adjectives, plural suffixes, and the indefinite form of weak nouns. The definite suffixes and the postnominal possessives are included in the table to show that they differ in form between the three genders, although they are often not considered true exponents of gender (see Section 2.4). As we can see from Table 1, there are several instances of syncretism between masculine and feminine forms, while the neuter forms are always distinct. In Old Norse, there were separate feminine forms for indefinite adjectives and demonstratives as well, but they have disappeared over the centuries. This has resulted in a paradigm with few unambiguously feminine gender cues, making the distinction between masculine and feminine gender less salient than the one between neuter and non-neuter.

2.2 Possessives

This paper is concerned with the possessive determiners min ‘my’ and din ‘your’, henceforth referred to simply as possessives. As seen in Table 2, both forms, as well as the reflexive possessive sin, are inflected for number and gender in Norwegian.

In all varieties of Norwegian, the possessives can be either prenominal or postnominal. The postnominal construction (2a) is considered pragmatically neutral, while the prenominal one (2b) is often used in conditions of contrast or emphasis (Anderssen & Westergaard 2010: 2578). In a corpus of child language recorded in Tromsø, Anderssen and Westergaard find that the adult speakers use the postnominal possessive in 75% of the possessive constructions (ibid.: 2581). However, in the spoken language, the postnominal construction can also be stressed to give a contrastive reading, as in (2d).² This makes it challenging to elicit the two phrase types, as will be discussed in Section 4.3. Finally, possessives can also be predicative in Norwegian, as shown in (2e), but this construction is beyond the scope of this paper. The examples in (2) show how possessives are inflected relative to the masculine noun kopp ‘mug’.

2.3 Loss of feminine gender forms

Although the Norwegian three-gender system has been considered fairly stable, the use of historically masculine forms with traditionally feminine nouns is not a novelty. Starting with the dominant written language Bokmål, feminine gender is optional and often omitted. This contrasts with the other written language Nynorsk, where feminine gender is required. When it comes to spoken varieties, Fretheim (1985) shows that in the Eastern Norwegian dialects that are most similar to Bokmål, the feminine indefinite article and prenominal possessives are replaced by masculine forms, illustrated in (3a) and (3c).³ While the feminine forms are typically retained postnominally (definite article and postnominal possessives), illustrated in (3b) and (3d), this system also allows masculine forms instead. The noun bok ‘book’ in (3) is traditionally feminine.

In a corpus of speakers from Oslo, Lødrup (2011) finds that the use of prenominal feminine forms (indefinite article, prenominal possessives, adjectives) is generally low, while the postnominal definite marker -a is widely used. In total, 29% of the speakers produce at least one feminine prenominal form, but only half of these 29% produce more than one. Lødrup (2011) concludes that age is the best predictor for the use of feminine gender, where older participants tend to produce more (prenominal) feminine forms.

During the last decade, the system described by Fretheim (1985) and Lødrup (2011) has also been found in the dialects of Tromsø (Rodina & Westergaard 2015) and Trondheim (Busterud et al. 2019). Both studies use elicited production tasks to study the use of feminine indefinite determiners and definite suffixes in different age groups, and they find the same asymmetry among younger speakers in both cities: The indefinite article ei is generally replaced by en by younger speakers, while the definite suffix -a is retained for most of them. The change seems to have advanced further in Trondheim than in Tromsø: While 14/15 adult participants in Tromsø use the feminine indefinite article consistently, this applies only to one of the adult Trondheim participants, and more than 50% of them never use ei. In order to find out if the change also applies to other Norwegian dialects, especially those in smaller places and those with morphologically richer gender systems, van Baal et al. (in press) have conducted similar studies in Bodø, Mo i Rana, Stavanger, Egersund, Lyngdal, Kristiansand, and Trondheim. They find that the loss of feminine gender forms is a general tendency, but that the pace of the change differs from place to place. In some places, feminine forms are lost by most speakers in all age groups, while in other places, only (some of) the younger participants do not use the feminine forms. The authors interpret this as the beginning of a change where the end result may be a complete loss of feminine gender. Moreover, they hypothesize that the change is enabled by grammatical syncretism between masculine and feminine forms, but that the process is driven by sociolinguistic factors (see also Rodina & Westergaard 2015; Busterud et al. 2019).

As shown in Table 1, there are few distinct feminine forms in most Norwegian dialects. Both Fretheim (1985) and Lødrup (2011) report a correlation between the use of the indefinite article form ei and the prenominal possessive forms mi/di/si. That is, speakers who say ei bok ‘a.f book’ are likely to also say mi bok ‘my.f book’. Similarly, the use of the definite article -a is correlated with the use of postnominal mi/di/si, so that speakers using the form boka ‘book.def.f’ will also say boka mi ‘book.def.f my.f, my book’. As will be discussed in Section 2.4, the status of the postnominal forms as exponents of gender is disputed. Therefore, earlier studies have focused on prenominal forms. As a follow-up of the Tromsø study mentioned above (Rodina & Westergaard 2015), the same authors investigated the use of prenominal possessive forms in Tromsø (Rodina & Westergaard 2021). For most groups, they find that there is no significant difference between the use of ei and prenominal mi. Both forms are generally used more by the adult participants than the younger speakers. Most adult speakers use both feminine forms consistently, while most speakers in the groups of children and adolescents mainly use the masculine forms, en and prenominal min. As in previous studies, they find an asymmetry between pre- and postnominal forms, as all groups use both -a and postnominal mi with more than 85% of all traditionally feminine nouns.

2.4 Gender vs. declension class

As seen from the previous research in Section 2.3, the new gender paradigm is asymmetrical in that the prenominal feminine forms (ei and mi/di/si) are disappearing, while the postnominal forms (-a and mi/di/si) are retained. It has been argued that this is because the elements have a different syntactic status (Rodina & Westergaard 2015; 2021; Busterud et al. 2019; Lohndal & Westergaard 2016; 2021). In what follows, we discuss this in more detail.

4.1 Selection of places

The GenVAC data is collected from speakers of seven different dialects. The locations were chosen based on language-internal factors (the morphological richness of the gender systems of the traditional dialect) and language-external factors (the number of inhabitants, commuting patterns, etc.). The project’s main hypothesis is that the change in grammatical gender is the result of an interplay between these factors (van Baal et al. 2023).

Some linguistic changes tend to spread to bigger cities first, and then spread out locally to smaller places. This is known as urban jumping (Chambers & Trudgill 1980; Sandøy 1998). Busterud et al. (2019) argue that this model is applicable to describe the loss of feminine gender in Norwegian. In order to investigate this, the locations were selected in pairs, so that there were one bigger and one smaller place from each part of the country.⁶ The richness of gender morphology was measured with a metric developed specifically for the project. The metric consists of a list of six possible distinctions between masculine and feminine gender in different parts of the nominal paradigm (see van Baal et al. 2023 for a detailed description). The hypothesis is that a dialect with many distinctions (i.e., many gender cues) will retain feminine gender better than one with fewer distinctions. Table 3 shows the size and number of feminine cues of each of the seven places; see van Baal et al. (2023) for an in-depth discussion of the various cues and a description of how the number of cues was established for each location.

4.2 Participants

The study includes two or three age groups in each place. Group B (17–19 years) and group C (25+ years) are included in all seven places. In addition, a group A (10–12 years) is included in three of the places. Originally, a minimum of 20 participants were recruited for each group. The criteria were that they had to have grown up in the area and were native speakers of Norwegian. A few participants did not consistently distinguish between neuter and non-neuter gender. As we did not have sufficient data to investigate the lack of neuter (there were indeed few cases), and as we consider this variation to be unrelated to the change in feminine gender, the participants with less than 50% target-like production of indefinite neuter determiners were excluded from the study. The number of participants after exclusion is listed in Table 4.

4.3 Elicitation tasks design and procedure

The data reported in this study was collected as part of a large test battery, including four picture-aided elicited production tasks. In this paper, we report the results from two of these tasks. All participants started with the determiner task (task 1). They then completed a task eliciting plural noun forms before doing the possessives task (task 3 in the GenVAC set of experiments, hence we have retained that label here).

Task 1 is based on the design used by Rodina & Westergaard (2015). The participants see a screen with two objects only differing in color. They are asked to describe what they see on the screen, and the target response consists of two nominal phrases, each including an indefinite article, an attributive adjective, and a noun (e.g., ei gul kake ‘a yellow cake’). Next, one of the objects disappears from the screen, and the participants are asked what happened. This elicits a definite nominal phrase, consisting of a determiner, a definite attributive adjective, and a noun with a definite suffix (e.g., den gule kaka ‘the yellow cake’). The procedure is repeated with 36 nouns – 8 masculine, 8 neuter, 16 feminine, half of them weak and half of them strong. Weak nouns end with an unstressed vowel (e.g., kake ‘cake’), while strong nouns end with a consonant (e.g., seng ‘bed’). The selection of nouns was based on earlier research, with some changes. The main criterion was that the noun was depictable in a clear and unambiguous way. We also aimed at nouns which had the same gender across dialects. However, gender assignment varies quite a bit in Norwegian (cf. Beito 1954), so this was not always possible. In cases where a certain noun traditionally had a different gender in a given dialect compared to the expected gender in other dialects, this was taken into consideration in the analysis. Task 1 is described in more detail by van Baal et al. (2023; in press).

Task 3 is developed for the purposes of the present study. Figure 1 shows two screens from the task.

Figure 1: The first screen (left) shows Knut and two of the practice items, fly.n (‘plane’) and tog.n (‘train’). The second screen (right) shows Knut stealing the plane from the participant.

The participants were told that the objects on the left were theirs, while the objects on the right belonged to the researcher. They were warned that the fantasy creature Knut, sitting below the objects, would steal one of the objects, and that their job was to tell the researcher what he stole and from whom. Based on previous research (see Section 2.3–4), we expected more use of feminine postnominal forms than of feminine prenominal forms. To test the use of both constructions, we elicited both of them in two separate parts of the task. The desired interactions are demonstrated in (8).

As mentioned in Section 2.2, it is difficult to design a task that controls whether the participants use the prenominal or the postnominal possessives. Eliciting contrast or emphasis to make them produce first one of the patterns and then the other, proved impossible in a pilot study. Therefore, we opted for a design where the researcher instructed the participant explicitly to produce prenominal or postnominal forms, using neuter nouns as examples. The participants were shown the screens in Figure 1, and the researcher would exemplify how they should express the event. After the example, two more neuter practice items followed for the participant to practice the procedure.

The experiment was conducted with 20 items (4 strong feminine nouns, 4 weak feminine, 4 strong masculine, 4 weak masculine, 2 strong neuter, and 2 weak neuter – a subset of the 36 items in task 1). All participants completed the task twice, first using one construction (e.g., prenominal possessives), and then using the other one (e.g., postnominal possessives). Half of the participants in each group started with the prenominal possessives and the other half with the postnominal possessives. To add more variation, the items were “stolen” from both sides of the screen, eliciting both 1^st and 2^nd person possessives (mi/min/mitt and di/din/ditt). We expected 1^st and 2^nd person forms to behave the same with respect to the production of feminine forms. Knut always stole two objects from the same person, and the two objects had the same gender. The order of the items was randomized (but fixed so each participant and each part of the task had the same order). After randomization, the order was corrected manually so that there were never more than four items with the same gender following each other.

5.1 Overall production of feminine forms

As mentioned, the data reported in this section is gathered from the GenVAC data set (Lohndal et al. 2023), and the data from task 1 (indefinite determiners and definite suffixes) is also reported in van Baal et al. (in press) but repeated here for convenience. The possessive data from task 3 is collected from the same participants, which allows us to do both inter- and intraindividual comparisons of the participants’ production of feminine forms (henceforth, fem-production).

The elicitation tasks include nouns of all three genders. Looking at the results for the masculine and neuter nouns, we find that the traditional forms are generally intact for all groups.⁹ For the remainder of this paper, we therefore only report the forms used with traditionally feminine nouns. Table 5 sums up the production results for all participants in each of the 17 groups. For each group, the table reports the number and percentage of feminine forms compared to the total number of possible occurrences. For example, the children (A group) in Mo i Rana produced a total of 144 prenominal possessives modifying a feminine noun, and 95 of these (66.0%) had the feminine form (mi). The non-feminine forms produced with traditionally feminine nouns were, in an overwhelming majority of the cases, masculine (e.g., min kake ‘my.m cake’).

As seen from Table 5, the use of feminine postnominal forms (definite suffixes and postnominal possessives) is high for all groups except the three Stavanger groups. In most of the remaining groups, the score is at ceiling (>95%), and only Egersund-A and Kristiansand-B score below 90% on postnominal possessives and definite suffixes respectively. Considerably more variation is found in the use of prenominal forms (the indefinite determiner and the prenominal possessive), both between groups and individuals. The graphs in Figures 2 and 3 illustrate the variation between locations and age groups.

Figure 2: The use of the feminine indefinite determiner (ei) across all groups.

Figure 3: The use of the feminine prenominal possessives (mi) across all groups.

To see if there was a correlation between the use of ei and prenominal mi, we ran a Spearman’s rank correlation test.¹⁰ First, we tested the correlation between the indefinite determiner (ei) in task 1 and the prenominal possessive (mi) in task 3 for all participants. We find that there is a clear and significant correlation (rho = 0.810211, S = 1287631, p < 0.0001). We also ran the same test for the correlation between the definite suffix (-A) in task 1 and the postnominal possessive (mi) in task 3. This also came out as a significant correlation (rho = 0.6306608, S = 2484007, p < 0.0001). This means that when considering all the participants together, the use of ei and -A in the first task can predict the use of pre- and postnominal mi respectively in the third task. The phrase-internal combinations in the postnominal part of task 3, i.e., how the definite suffix (-A or -en) was combined with the postnominal possessive (mi or min), are reported in Table 6.

Of all the postnominal possessive constructions recorded in this task, there are no instances of -en + mi (boken mi ‘book.def.m my.f’) and extremely few instances of -A + min (boka min ‘book.def.f my.m’). The three occurrences of the latter combination are produced by two different participants in different locations and age groups, and they constitute only 0.11% of the total number of produced postnominal possessive constructions. We therefore conclude that these mixed constructions can be considered as production errors rather than parts of any of the participants’ grammar.

Before going into the statistical effects of age group and location, we look briefly at potential differences between items. In Table 5 and Figures 2 and 3, all feminine nouns are collapsed into one category. In order to see whether there are differences between lexical items, the results for each item across all participants are presented in Table 7.

Earlier studies (e.g., Busterud et al. 2019; Urek & Lohndal & Westergaard 2022) have suggested that the weak (vowel) ending may be a cue for feminine nouns, since nouns of this type are often combined with feminine forms. Our results also suggest that there might be a small difference in fem-production between weak and strong feminine nouns. However, this will be left to future research, and for the remainder of this paper, we treat all the feminine nouns as one category.

5.2 Differences between locations and age groups

It is clear from Figures 2 and 3 that there is variation in fem-production depending on both location and age. This was tested with a binomial mixed effects model (glmer-function of the R-package lme4, Bates & Mächler & Bolker & Walker 2015) where the predictors were location and age group. Using the Anova function in R to compare the models, we obtained the p-values reported in this section. All models include random effects for participant and lexical item. The full output of the statistical models is included in Appendix 2. As only some of the locations have three age groups, we had to run two different models. First, we ran a model with the B and C groups from all seven locations – we refer to this as the 7-BC model. Then the model was run again with the A, B, and C groups from Stavanger, Egersund, and Mo i Rana – referred to as the 3-ABC model.

Starting with the production of feminine prenominal possessives, both models show a statistically significant effect of location (7-BC: χ² = 166.93, df(6), p < 0.0001; 3-ABC: χ² = 139.42, df(2), p < 0.0001) and age group (7-BC: χ² = 146.13, df(1), p < 0.0001; 3-ABC: χ² = 52.175, df(2), p < 0.0001). The 7-BC model also reveals an interaction effect between the two predictors (χ² = 156.396, df(6), p < 0.05), indicating that the effect of age group is different across locations. In the 3-ABC model, however, there was no interaction effect (χ² = 7.3464, df(4), p ≈ 0.1187). This is probably because there are considerably fewer data point in this model (9 groups as opposed to 14 in the 7-BC model). To sum up, we find clear differences in the production of the feminine prenominal possessive between locations and age groups. The younger groups generally display lower fem-production than the older groups.

Looking at the effect of location and age group on the fem-production of postnominal possessives, there are fewer differences between groups. In the 7-BC model there is no effect of neither location (χ² = 7.8712, df(6), p ≈ 0.2477) nor age (χ² = 0.5423, df(1), p ≈ 0.4615), and there is no interaction effect (χ² = 0, df(6), p ≈ 1.0). This is expected, given that the mi production is generally high across most locations and age groups. In the 3-ABC model there is, however, an effect of location (χ² = 22.374, df(2), p < 0.0001), but not of age group (χ² = 1.2828, df(2), p ≈ 0.5266), and there is no interaction effect (χ² = 0.2751, df(4), p ≈ 0.9914). The location effect in this model is probably due to the relatively low mi production in the three Stavanger groups, compared to the groups in Egersund and Mo i Rana. In this model, the Stavanger groups constitute 1/3 of the datapoints, as opposed to 1/7 in the 7-BC model, which means that the Stavanger data have a bigger impact in the 3-ABC model and thus the difference comes out as significant. In general, as reflected in the numbers in Table 5, there is little variation in the production of the feminine postnominal possessive. The effects of location registered in model 3-ABC reveal the considerably lower postnominal fem-production in the three Stavanger groups, which is also clearly visible in Table 5.

5.3 Individual variation

The results in the previous sections outlined the general tendencies found across groups. This section is concerned with inter- and intraindividual variation within these groups. In Table 8, the participants are sorted into three groups based on their level of production of the two feminine prenominal forms, ei and prenominal mi. The threshold for consistent fem-production is set at ≥87.5, and this is the case for the remainder of this paper.¹¹

The fem/masc-mixers constitute 24.9% of the participants in the determiner task and 20% in the possessive task. These speakers use both feminine and masculine forms with traditionally feminine nouns. In other words, most participants are either consistent fem-producers or consistent masc-producers. It is also noteworthy that the mi/min mixers and the ei/en mixers are not always the same individuals: Only 39 participants (11.3%) are mixers in both tasks. This is illustrated in Table 9.

The mixers are not evenly distributed across locations and age groups. Table 10 shows all groups where more than 30% of the participants mix either ei/en or mi/min.

In some groups, the number of ei/en-mixers is very different from the number of mi/min-mixers. At the group level we find that the feminine prenominal possessive score is generally, but not always, higher than the feminine indefinite article score. The number of participants using mi consistently is higher than the number of participants using ei consistently (48% vs. 45.1% of the participants), but the number of participants using min consistently is also higher than the number of participants using en consistently (32% vs. 30.3%), indicating that there is in general less mixing in the possessive task than in the determiner task. This can potentially be explained by the fact that there are fewer items in the possessive task than the determiner task (i.e., less chance of variation), but it is also the case that the determiner task was always performed before the possessives task. At any rate, the difference is small. The effect of time, if any, seems to be that participants become more assertive, choosing one form over the other instead of using both, but whether they use more or less feminine forms in the latter task, varies between individuals.

In the remainder of this subsection, we look at how individual speakers combine the different prenominal and postnominal forms. We have already seen that the loss of feminine prenominal forms does not reflect the loss of feminine postnominal forms, as the latter are often retained. In Table 11, we see the individual production of pre- and postnominal mi across all groups, where for instance 166 participants consistently produce mi both prenominally and postnominally.

We see clearly that there is an interrelationship (although not a direct correlation) between the use of prenominal and postnominal mi: The feminine postnominal possessive is only lost when the feminine prenominal possessive is also lost. This trend is maintained also if we include the participants’ production of feminine indefinite and definite articles (ei and -A). In Table 12, we have excluded all participants who did not complete (or produce a sufficient number of forms in) both tasks. We have then merged the participants (n = 343) into one group. We have combined the prenominal forms (ei and prenominal mi) to one category, pre-F, and likewise the postnominal forms (-A and postnominal mi) into another category, post-F.

Figure 4 shows the actual level of fem-production in both prenominal and postnominal forms for each participant. Again, we see that the feminine postnominal forms are never lost by the speakers using the feminine prenominal forms.¹²

Figure 4: The correlation between prenominal and postnominal feminine forms – all participants (n = 343). Not all participants are visible in this graph; because most of them have a high production of postnominal forms, their dots are overlapping. However, the approximate distribution of participants in the different parts of the graph can be seen in Table 12.

5.4 Potential self-priming

A potential task effect is that participants may self-prime because the feminine postnominal possessives are used more frequently than the prenominal ones. That is, the use of postnominal mi could prime the use of prenominal mi, resulting in a higher production of prenominal mi among the participants starting with the postnominal possessives. This was controlled for by having half of the participants in each group start with the prenominal, and the other half with the postnominal possessives. If we combine all participants across age groups and locations, the mean prenominal mi production score is 56.8%. The participants starting with postnominal possessives use on average 56.5% prenominal mi, while the participants starting with prenominal possessives use 57.1% prenominal mi. There is in other words no overall difference between the two groups. However, the total score could conceal fluctuations between the locations, so we tested the scores for the pre- and post-beginners for each location as well. In most locations the scores are very similar. We find the biggest difference between the two groups in Kristiansand, where the post-beginners score 49.4% contra the pre-beginners at 39.3%, and in Lyngdal, where the post-beginners score 55.6% and the pre-beginners 67.5%. Firstly, the difference is small, and secondly, in Lyngdal the post-beginners actually score lower than the pre-beginners, contrary to what we would expect from a task effect. When looking at individual scores, the differences between the groups seems to be due to individual fluctuations. For example, in the Kristiansand-C group, there are only two participants with 0% prenominal mi-production. They are both in the post-beginner group, but they are also both among the participants with 0% ei-production, which suggests that their low mi-production is independent of which of the possessive sub-tasks they did first. To sum up, we do not find any indications of self-priming from the postnominal to the prenominal part of the task.

6.1 A change across dialects

Our results clearly show that prenominal feminine gender forms (ei and mi) are disappearing across Norwegian dialects. For the prenominal possessives, we see a trend that is very similar to that of the indefinite determiners: The fem-production differs both between locations and between age groups within each location. For the postnominal fem-production, there is generally little variation between locations and age groups.

The difference between age groups and locations reveals considerable variation in the nature of the change. Regarding the starting point of the language change, we see that in some locations (e.g., Stavanger), the use of feminine forms is low even in the adult group – indicating that the change has been going on for a while. In other locations, such as Mo i Rana, the decrease in fem-production has only just started in the youngest group, indicating a recent change. By comparing the different age groups within one location, we can say more about the rate of the change. For instance, in Bodø and Lyngdal we see a big difference in fem-production between the B and C groups, indicating that the change is happening very fast. In Mo i Rana and Egersund, on the other hand, the differences between the age groups are smaller, indicating that the change is happening at a slower rate. All the data is collected within the frame of one year, meaning that we are taking an apparent-time rather than a real-time approach to language change (see Labov 1966: 199ff; Cukor-Avila & Bailey 2013). This means that we cannot exclude the possibility that the level of fem-production changes throughout the lifespan, and that the younger participants will start using more feminine forms as they get older. However, no earlier studies have found such a pattern.¹³ Yet, this is an empirical question that can only be answered by monitoring dialects or speakers over time.

6.2 A systematic change

We have already established that both ei and prenominal mi are disappearing in many groups. In this section, we discuss the correlation between the use of these two forms at the individual level.

The correlation test (see Section 5.1) showed that there is a significant correlation between the use of ei (task 1) and prenominal mi (task 3). From this we can conclude that the two elements disappear in parallel. If only one of the elements disappeared, this would indicate a process of syncretism (i.e., a merger of e.g., the lexical elements en and ei, but not of the underlying feminine/masculine gender distinction).¹⁴ However, the fact that they are disappearing at the same time indicates the loss of the underlying syntactic gender distinction, not just of the surface forms.

The trend reflected in this statistical correlation can also be seen in Table 9 (Section 5.3). Although we find all combinations of ei/en and mi/min production, there is a clear tendency towards preferring the same pattern in both tasks, i.e., either

1. using ei and mi consistently, or

2. mixing both en/ei and mi/min, or

3. using en and min consistently.

In total, 75.8% of the participants use one of these three patterns. The remaining 24.2% change the level of fem-production through the set of tasks, but as mentioned in Section 5.3, there are instances of both higher and lower fem-production in the possessive task compared to the determiner task: many of the determiner-mixers settle as consistent mi-users or consistent min-users in the course of the experiment. This could be an indication that the speakers in question are alternating between two competing grammars, one three-gender system and one two-gender system (see also the discussion in Section 6.4 below).

In Table 10 (Section 3.5) we see that the highest number of mixers belong to the groups where the gender change is well underway. In the locations where the change has only just started (Mo i Rana, Egersund, Bodø), only the youngest groups have a high number of mixers, as the older participants are still consistent fem-users. In Stavanger, on the other hand, the adult group has many mixers, but not the children and high-school groups, because the process of feminine gender loss is already completed in the younger participants.

6.3 When three genders become two

Earlier studies (Rodina & Westergaard 2015; 2021; Busterud et al. 2019; van Baal et al. in press) have found an asymmetry in the loss of feminine forms in Norwegian: While the feminine prenominal forms are often vulnerable and subject to change, the feminine postnominal forms have been described as generally robust. The current study confirms this pattern.

As presented in Table 5 (Section 5.1), the feminine postnominal possessives are retained to a similar degree as the definite suffixes, i.e., most groups perform at ceiling (>95%). At the group level, there are five exceptions, and all these five groups produce less than 95% feminine forms of both definite suffixes and postnominal possessives. Egersund-A (92.1% -A and 84.3% mi) and Kristiansand-B (87.6% -A and 94.1% mi) are both close to or above 90%, suggesting that we might see the beginning of a change in the postnominal forms, but still they must be considered relatively stable at this point. In the three Stavanger groups, on the other hand, we see more solid evidence that an actual change is going on.

In answering RQ3, the remainder of this section focuses on individual speakers and how they combine different forms. The speakers who do not use the feminine postnominal forms will be particularly relevant for the following part of the discussion. Table 12 and Figure 4 (Section 5.3) show the individual fem-production of all the participants. Although there is bound to be some variation in a sample consisting of so many individual speakers, there seems to be a clear implicational scale: Speakers who have maintained the feminine/masculine distinction on prenominal elements also use the feminine postnominal forms (-A and -A+mi). These latter forms may be vulnerable for speakers who have lost the feminine/masculine gender distinction on the prenominal forms, but at this point, most speakers still use -A and postnominal mi. This indicates that the (potential) loss of feminine postnominal forms depends on the absence of feminine prenominal forms.

The form of the postnominal possessives, although separated from the noun in writing, is clearly conditioned by the form of the definite suffix: There are no instances of -en + mi (boken mi ‘book.def.m my.f’) and virtually none of -A + min (boka min ‘book.def.f my.m’) within the same phrase (see Table 6). In other words, we see that postnominal mi always follows -A, and never any other suffix. This is also what is found in earlier research (Lødrup 2011; Svenonius 2017; Rodina & Westergaard 2021). As mentioned in Section 2.4.3, the question of how the postnominal possessives should be analyzed may be one which cannot be solved empirically. The data from the current study confirms that mi always follows -A, but recall that -A includes different realizations, such as -a, -e, and -o. In Svenonius’ (2017) analysis, mi is phonologically conditioned by the vowel /a/. This only works in dialects where the -A suffix is realized as -a. In the results from Egersund, we see that mi can also follow /e/ and /o/, but only in non-neuter phrases. Neuter nouns also end in /e/ in the definite form, but they are never combined with mi. In dialects with other realizations than -a of the feminine definite suffix, the phonological rule stated in (7) must be modified so that it is specified for the relevant vowel qualities. The syncretism between the definite suffix used on neuter nouns and strong feminine nouns in the Egersund dialect, /e/, shows, as Svenonius (2017) also emphasizes, that mi is only an allomorph of the common gender form min, leaving the neuter/non-neuter distinction intact regardless of the phonological rule.

As seen in this and previous studies, there is, inarguably, an asymmetry in how feminine prenominal (ei and mi) and postnominal (-A and mi) forms behave in a language change. A similar asymmetry is found in L1 acquisition of the same forms. The suffixes are acquired before the indefinite articles, and it is often assumed that nouns and their definite suffixes are acquired together (Anderssen 2006: 179ff; Rodina & Westergaard 2015). The asymmetry in how the prenominal elements and the suffixes are both acquired and lost has often been presented as evidence for the distinction between gender (expressed by the prenominal elements) and declension class (expressed by the definite suffixes) (see Busterud et al. 2019; Rodina & Westergaard 2015; 2021; Lohndal & Westergaard 2016; see section 2.4). Recall that according to this analysis, the definite suffixes are never exponents of gender, and the -A suffix is therefore not affected by the loss of feminine gender. This analysis is clear and consistent: It predicts a difference between indefinite articles and definite suffixes in the case of a gender change: Because the feminine definite suffix is not a gender exponent, it is not affected by the gender change. What this analysis does not account for, however, is the close relationship that nevertheless exists between the -A suffix and the feminine prenominal forms.

This relationship is better accounted for in different analyses: Following Svenonius (2017), the definite suffixes may be considered gender exponents in Norwegian, as long as there is additional evidence to postulate the existence of a grammatical gender feature in the grammar. This is also in line with Enger (2004) and Dahl (2000). These analyses adopted a slightly altered version of Hockett’s (1958: 231) gender definition, extending it to “the behavior of associated morphemes” (rather than just “associated words”). Following Svenonius (2017), the -A suffix can only mark feminine gender in a system that has other feminine markers, i.e. three-gender systems. In two-gender systems, which lack other feminine markers, -A must be an exponent of common gender, the relevant gender distinction being that between common and neuter gender, which are separated by their determiners and adjectives as well as their definite suffixes. The alternation between -A and -en in two-gender systems is analyzed as declension class. This contrasts with the neuter form -et which always corresponds to a neuter gender feature and is therefore an exponent of neuter gender both in two- and three-gender dialects.

We believe that Svenonius’ (2017) analysis captures the patterns found in the current study better than the analyses based on Hockett’s definition. Firstly, it is clear that the definite feminine suffix -A (and the postnominal possessive form mi) can disappear, but that this is entirely dependent on the absence of a feminine/masculine gender distinction marked on freestanding elements. Secondly, Svenonius’ analysis captures the important insight that not all speakers of Norwegian necessarily share the same conceptualization of grammatical gender. In a three-gender system with a one-to-one mapping between gender and definite suffix form, each suffix can easily be treated as an exponent of each gender. In systems where the mapping is not one-to-one (that being the hybrid systems described in this study, or the system of some traditional three-gender systems that have several declension classes (i.e., different suffixes) within one gender, like in traditional Egersund dialect), the speakers must resort to the notion of declension class in order to systematize the alternation between different suffixes. However, the fact that -A and -en must be exponents of different declension classes in one system, does not prove that they cannot be exponents of different genders in another system. It remains to be seen how the postnominal forms evolve over time in groups and individuals who have lost the prenominal feminine/masculine distinction.

6.4 Future research

Svenonius (2017) distinguishes between two-gender and three-gender systems. The majority of our participants (65%) fall into one of these groups, as they either always or never use ei and prenominal mi. However, the remaining 35% of the participants are neither clear three-gender speakers nor clear two-gender speakers. This is to be expected in a changing system, and we see clearly from Table 10 that the mixers mainly belong to the groups that are in the middle of the change. Earlier studies (Rodina & Westergaard 2021; Klassen & Lundquist & Westergaard 2023) have also reported that individuals tend to mix feminine and masculine forms, and the proportion of mixers varies a lot depending on participant selection. At the group level, this makes sense. The question is how this should be understood at the individual level. How is gender conceptualized within an individual who uses feminine gender forms only to some extent? Is the F feature still present in their syntax, but only assigned to a limited class of words (i.e., only to a subset of the traditional feminine nouns)? Or do these individuals have two competing systems, one two-gender system and one three-gender system, which they use interchangeably? Preliminary findings suggest that the latter explanation is more likely, but this question will be addressed further in future research.