This study was financed by the Wiener Wissenschafts-, Forschungs- and Technologiefonds (WWTF, SSH11-027), by the University of Vienna and the Austrian Academy of Sciences. Furthermore, we are deeply grateful to our long-term project partner Christine Czinglar for her support in almost all phases of the project (planning, data collection, SES classification, analysis, discussion of results, publication) as well as to numerous students of Linguistics of the Department of Linguistics of the University of Vienna for their transcription work. We also thank Basilio Calderone, Markus Christiner and Jan Oliver Rüdiger for their statistical advice and the members of our advisory board (Ayhan Aksu-Koç, Rudolf de Cillia, Inci Dirim, Dorit Ravid) as well as our contact person Katrin Zell from the Viennese kindergarten authority MA10 - Wiener Kindergärten for their continuous support. We are also grateful to Sabine Laaha who developed the plural test in collaboration with the first and the last author. Last, but not least, we thank the children, their families and their kindergarten teachers for their participation.
1This contribution investigates the impact of social, dialectal and methodological factors on the acquisition of German noun plurals in monolingual and bilingual kindergarten children living in Vienna (Austria) within the usage-based theory of language acquisition (Tomasello 2003, Bybee 2010).
2Being part of non-prototypical, inherent inflection, noun plurals are known for showing high variation in many languages (e.g. Dressler 1989, Acquaviva 2004). In German, we find considerable examples of overabundance (Thornton 2011, Mörth & Dressler 2014), i.e. of competition of different plural forms with no or only slightly different semantic or pragmatic meanings (e.g. the more foreign -s plural Rikschas for rikshaws in Asia and the more native -(e)n plural Rikschen for rikshaws in Germany, cf. Mörth & Dressler 2014: 252). Compared to Standard German, the phenomenon of overabundance is even stronger in certain German dialects (e.g. Schrödl 2009, Schrödl et al. 2015 on South-central Bavarian dialects of the Austrian region of Burgenland), where two different plural forms are possible for most noun lemmas, but up to five different plural forms may coexist for certain lemmas (e.g. Wurm, Würm, Würmer, Würmmer, Wurmer of the lemma Wurm ‘worm’, see Schrödl et al. 2015: 172).
3This tendency is of course much weaker in Standard German, which – in contrast to the dialects – has clear written norms (Schrödl et al. 2015: 182), but two different plural forms for one noun lemma are not entirely unusual in Standard German as well (e.g. Ballons or Ballone for the lemma Ballon ‘balloon’; see Mörth & Dressler 2014 for further examples).
4And even if there is only one correct plural per lemma, there are often several potential plural forms that may compete with the respective correct form (e.g. Korecky-Kröll et al. 2012) due to the many plural markers (see Table 1) and the complex regularities of German plural formation. This makes the system difficult for children acquiring German as their first language (Example 1) and also for second-language learners of all ages (Examples 2 and 3). As a consequence, learners may follow different strategies: Some (usually the less advanced learners) tend to produce zero plurals, which are identical to the singular forms, but which are correct only for one specific class of nouns, namely for di- or polysyllabic nouns with a final schwa syllable such as Teller ‘plate’ (see also Table 1 below). Others (often the more advanced learners) tend to produce overt overgeneralizations of a potential, but incorrect plural marker (e.g. an -en plural overgeneralization instead of a correct -e plural, see example 3a). Or both strategies may be used interchangeably (Examples 1a/b and 2a/c) or alternate with correct plural forms (Examples 1c, 2b and 3b).
(1)
|
MOH (High SES boy with L1 German):
|
|
a.
|
Plural test at age 3;3 (overt s plural overgeneralization):
|
|
|
EXP:
|
Und
|
das
|
sind
|
drei…?
|
|
|
|
And
|
these
|
are
|
three…?
|
|
|
MOH:
|
… Vogels* .
… bird.PLs
|
|
b.
|
Plural test at age 4;8 (incorrect zero plural):
|
|
|
MOH:
|
… Vogel* .
|
|
|
|
… bird.PL0
|
|
c.
|
Spontaneous kindergarten recording at age 4;8 (correct plural):
|
|
|
MOH:
|
Vögel
|
fressen +...
|
|
|
|
bird.PLu
|
eat.3P
|
(2)
|
EDK (High SES girl with L1 Turkish and L2 German):
|
|
a.
|
Plural test at age 3;3 (overt n plural overgeneralization):
|
|
|
EDK:
|
… Apfeln.
|
|
|
|
apple.PLn*
|
|
b.
|
Spontaneous speech at age 3;3 (correct umlaut plural, dative case suffix is missing):
|
|
|
EDK:
|
Und
|
mit
|
Äpfel [*] .
|
|
|
|
and
|
with
|
apple-PLu
|
|
c.
|
Plural test at age 4;9 (incorrect zero plural):
|
|
|
EDK:
|
… Apfel.
|
|
|
|
… apple-PL0*
|
(3)
|
A. (female kindergarten teacher with L1 Polish and L2 German using an incorrect -en plural overgeneralization as well as the correct -e plural of Monat ‘month’ in one and the same recording):
|
|
a.
|
A:
|
Also
|
viele
|
Kinder
|
können
|
schon
|
die
|
Monaten*
|
goe [: gell] ?
|
|
|
|
well
|
many
|
children
|
know
|
already
|
the
|
month.PLn
|
right ?
|
|
|
A:
|
Na [: Nein]
|
da
|
sind
|
die
|
Monate .
|
|
|
|
no
|
there
|
are
|
the
|
month.PLe
|
5These examples demonstrate that there is indeed considerable variation in our data and that children as well as adults sometimes struggle to produce the appropriate plural form. But why is this the case? And what are the regularities of German plural formation?
6Table 1 shows the eight main plural markers of Standard German:
Table 1. German noun plural markers
- 1 Markers 7 and 8 (-er with or without umlaut) are sometimes treated as one and the same marker (e.g. (...)
No
|
Plural marker
|
Singular (example)
|
Plural (example)
|
Plural (English translation)
|
1
|
s
|
Auto
|
Autos
|
cars
|
2
|
(e)n
|
Katze
|
Katzen
|
cats
|
3
|
e schwa
|
Bus
|
Busse
|
buses
|
4
|
umlaut + e schwa
|
Zug
|
Züge
|
trains
|
5
|
zero
|
Teller
|
Teller
|
plates
|
6
|
umlaut
|
Apfel
|
Äpfel
|
apples
|
7
|
er (= a schwa)
|
Bild
|
Bilder
|
pictures
|
8
|
umlaut + er1
(= a schwa)
|
Haus
|
Häuser
|
houses
|
7As to type and token frequencies of German plural markers, different studies find similar tendencies (for an overview see Feldman 2005): -(e)n plurals show the highest type and token frequencies (around 50 %), followed by zero plurals and -e plurals, then -s, -er and umlaut+er plurals (-s plurals with slightly higher type frequencies, but -er and umlaut+er with slightly higher token frequencies) and finally pure umlaut plurals (less than 1 %).
8In his schema model of Standard German noun plurals, Köpcke (1998: 309) identified several factors (number of syllables, word-final phonology and gender) that lead to the following continuum (from prototypical singular to prototypical plural nouns):
- 2 The German system of definite articles consists of the following three main forms (in nominative ca (...)
(1)
|
monosyllabic, final stop, article der or das2 (= most probably singular of masculine or neuter gender, i.e. prototypical singular): e.g. SG der Hund ‘the dog’
|
(2)
|
polysyllabic, final -er, article der or das (= probably singular of masculine or neuter gender), e.g. SG der Lehrer ‘the teacher’, but rarely GEN PL: der Bücher ‘of the books’
|
(3)
|
polysyllabic, final -e, article die (= either feminine singular or plural of any gender, i.e. neither prototypical singular nor prototypical plural, but equal probability of being a singular or a plural), e.g. SG die Runde ‘the round’, PL die Hunde ‘the dogs’
|
(4)
|
polysyllabic, final -er, article die (= rather plural of masculine or neuter gender than feminine singular), e.g. PL die Lehrer ‘the teachers’, but rarely FEM SG die Mauer ‘the wall’
|
(5)
|
polysyllabic, final -en, article die (= prototypical plural), e.g. PL die Katzen ‘the cats’
|
9Nevertheless, there are many deviations from these tendencies claimed by Köpcke’s schema model, especially if we consider local varieties of German (as mentioned above). For example, in many Austrian dialects, some of the markers listed in Table 1 are more widespread than in Standard German (Korecky-Kröll 2011, Schrödl et al. 2015), whereas others are much less frequent. Especially the zero plural, which is usually limited to di- or polysyllabic masculine or neuter nouns with a final schwa syllable (such as Teller ‘plate’ and Fenster ‘window’) in Standard German, has a much wider scope in Viennese dialect: For example, it is used for masculines and neuters instead of an -e plural due to the phenomenon of schwa apocope (e.g. der Hund – die Hund instead of Standard German die Hunde ‘dogs’), for masculines and neuters ending in full vowels instead of -s plurals (e.g. das Auto – die Auto ‘cars’) as well as in several other instances (for details see Korecky-Kröll 2011: 244ff). But also the -(e)n plural and the pure umlaut plural have a wider scope than in the standard, whereas the -e plural and the -s plural are clearly less frequent in Viennese dialect than in Standard German. Although dialect use in Vienna has been on the decline for several years (see e.g. Wiesinger 2015: 93) and is predominantly used by older people from lower socio-economic backgrounds, these forms still occur occasionally in the input to young children (3.49 % of plural word form types and 1.10 % of plural tokens of the caregiver input corpus of the present study), may influence their acquisition of noun plurals (Korecky-Kröll 2011) and thus must be considered in our analysis. Examples 4 and 5 show that kindergarten teachers sometimes use both standard and dialect plurals within the same recording:
(4)
|
J. (female kindergarten teacher with L1 German using the standard zero plural as well as the dialect -n plural of Löffel ‘spoon’ in one and the same recording):
|
|
a.
|
J:
|
Die
|
schauen aus
|
fast
|
wie
|
Löffel.
|
|
|
|
they
|
look
|
almost
|
like
|
spoon.PL0
|
|
b.
|
J:
|
Gabeln
|
oder
|
Löffeln ?
|
|
|
|
fork.PLn
|
or
|
spoon.PLn(dial) ?
|
(5)
|
S. (female kindergarten teacher with L1 German using the standard -s plural as well as the dialect zero plural of Flamingo in one and the same recording):
|
|
a.
|
S:
|
dann
|
weißt,
|
wie
|
schön
|
flamingo.PLs
|
ausseh(e)n .
|
|
|
|
then
|
(you) know
|
how
|
beautiful
|
Flamingos
|
look like
|
|
b.
|
S:
|
Flamingo
|
sind so
|
rosane
|
Vög(e)l .
|
|
|
|
flamingo.PL0(dial)
|
are (so)
|
pink
|
bird.PLu
|
10Example 6 is not even dialect, but colloquial Viennese German of a kindergarten teacher who, again in one and the same recording, uses both a zero plural and a -n plural of the colloquial diminutive Schweinderl ‘piglet’, which are both equally correct plural forms:
(6)
|
SK., female kindergarten teacher with L1 German:
|
|
a.
|
SK:
|
zwei
|
Schweinderl .
|
|
|
|
two
|
pig.DIM.PL0
|
|
b.
|
Sk:
|
Oh,
|
ich
|
brauch(e)
|
drei
|
Schweinderln .
|
|
|
|
oh
|
I
|
need
|
three
|
pig.DIM.PLn
|
11First language acquisition of German noun plurals has been a widely discussed topic for four decades: Since the 1970s, supporters of different models of language acquisition have established different classifications of noun plurals with respect to their regularity/irregularity, default status, productivity, potentiality, salience, prototypicality etc. At first, the central question to be answered by the predominantly structuralist approaches was the following (cf. Augst 1979: 221): “Which noun takes which plural marker?” From the late 1980s onwards, connectionist single-route models as well as usage-based models regarded learning as a strongly associative process based on frequent input and output patterns or schemas without assuming explicit rules (e.g. Rumelhart & McClelland 1986 on the English past tense; Köpcke 1993, 1998, Goebel & Indefrey 2000, and Hahn & Nakisa 2000 on German plurals). These usage-based models were challenged by generative dual-route models (e.g. Clahsen et al. 1992, 1996, Marcus et al. 1995) that claimed that there was only one regular default plural, the -s plural, that followed a rule and thus should be the only or at least the most frequent marker to be used for plural overgeneralizations, whereas all other plural markers were irregular and had to be rote-learned. Although this strong claim was refuted by several empirical studies and modified in later versions of the dual-route model (e.g. in Sonnenstuhl-Henning 2003, who acknowledged feminine nouns ending in e schwa in the singular as being regular because they always take a ‑n plural), the main claim of a dichotomy between regular and irregular plurals was perpetuated. Other models, such as Natural Morphology, took an intermediate position by assuming different degrees of productivity and potentiality of different plural markers, depending on their possibility to be used for the integration of loanwords (Dressler 2003, Laaha et al. 2006).
12The order of acquisition of noun plurals has been shown to be largely similar in first and early second language acquisition of German – the proportion of correct plural forms in the course of development resembles a u-shaped curve: In an early phase (often before age 2 in L1 acquisition), most children use correct rote-learned plural forms (e.g. Dressler & Karpf 1995) as well as lexical means such as numerals or quantifiers to express plurality (e.g. Wecker 2016). As soon as first productive plural patterns have been established (often around age 2 in L1), the same two or three plural markers, namely en and e (and in some studies also s), are most frequently overgeneralized in both types of acquisition (e.g. Gawlitzek-Maiwald 1994, Köpcke 1998, Szagun 2001, Behrens 2002, Bittner & Köpcke 2001, Korecky-Kröll & Dressler 2009, Korecky-Kröll 2011 for L1 acquisition and Wegener 1994, 2008, Marouani 2006, Zaretsky et al. 2013, Wecker 2016 for early L2 acquisition). This phase of overgeneralizations corresponds to the low point of the u-shaped curve. Only when children have discovered the basic regularities of German plural formation (including gender and word-final phonology), their rate of correct plurals will increase again (usually from age 3 onwards in L1). Although these basic steps are similar in L1 and early L2 acquisition, there is also a major difference: In early L2 acquisition, at least some rudimentary concept of the category of number has already been established in the L1 and might have to be restructured when the L2 comes into play. Most studies do not find a direct transfer or interference, but learners may show different degrees of sensitivity with respect to specific linguistic phenomena or categories (Wegener 1994: 288). For example, children with L1 Turkish tend to omit German plural markers on the noun because number marking is not obligatory in Turkish (Corbett 2000). In addition, children with L1 Turkish will also have more problems in discovering gender-based regularities of German plural marking because there is no gender in Turkish (Wegener 2008) or in recognizing the German umlaut as a plural marker because plural is only marked by suffixation in Turkish (Wecker 2016), excepting Turkish vowel harmony being very different from German umlaut. Or learners with L1 Russian may omit the plural marking after the numerals two to four because Russian requires the genitive singular in this particular context (Bast 2003).
13Most studies on plural acquisition discussed so far follow either a source-oriented approach focussing on the singular form from which the plural must be derived (e.g. structuralist models, dual-route models, Natural Morphology) or a product-oriented schema approach focussing on prototypical plural schemas (e.g., Köpcke 1993, 1998). In contrast, Wecker (2016) and Köpcke & Wecker (2017: 86) propose an acquisition model that combines both approaches. They establish the following phases of German plural acquisition: In the beginning, learners store singulars as well as plurals holistically in their mental lexicon and match them according to their phonological and semantic similarities. Subsequently, they derive product-oriented (first-order) schemas from these stored forms and map phonological schemas to their grammatical functions (e.g. typical shapes of singulars are associated with the grammatical function of singular, whereas typical shapes of plurals are associated with the grammatical function of plural). Afterwards, learners proceed to source-oriented (second-order) schemas by relating singular schemas to their corresponding plural schemas, a strategy that will finally become predominant. Köpcke & Wecker (2017) find evidence for source-oriented as well as product-oriented strategies for plural formation in different groups of children acquiring German: In a nonce word elicitation experiment with children aged 6-10 acquiring German as their L1 or L2, the product-oriented strategy was used more frequently by the older L2 participants that had a longer exposure to German: Older L2 learners of grade 3 and 4 used fewer zero plurals for non-feminine nouns ending in -el or -er as well as more -en plurals for monosyllabic items and for items ending in an unstressed vowel than younger learners of grade 1 and 2. On the other hand, the product-oriented strategy was more dominant in the younger group of monolingual children, whereas older monolingual children rather followed the source-oriented strategy. These differences are explained by the exposure to the target language – as the L2 children only started kindergarten at age 3 or 4 or did not attend kindergarten at all, they are in clearly different stages of plural acquisition than the monolingual children. There were only marginal influences of the respective L1: German -er plurals were used slightly more frequently by older children with L1 Turkish than by older children with L1 Russian – the reason may be that Turkish plurals are formed with the suffix -lar or -ler which is relatively similar to the German ‑er suffix (at least in its written form, cf. Köpcke & Wecker 2017: 95). Children with L1 Turkish also used more zero plurals than children with L2 Russian, but more frequently with nouns that already resembled a prototypical plural schema of German.
14In section 4.6 of the present paper, a first attempt was made to apply the main hypotheses concerning product- vs. source-oriented strategies in Standard German formulated by Köpcke & Wecker (2017) to our data. Nevertheless, we must bear in mind that the model would also have to be adapted by including predictions about plural schemas in German dialects, which would go beyond the scope of the present study.
15International studies like PISA indicate that not only immigrant children that acquire German as a second language (L2), but also children from families with low socio-economic status (SES) acquiring German as a first language (L1) generally score at lower levels on language competence tests (Schwantner & Schreiner 2010). Although this is still a controversial topic, some studies on monolingual L1 acquisition show that SES affects not only vocabulary acquisition (e.g. Weizman & Snow 2001), but also the overall rate of grammar acquisition (Gathercole & Hoff 2007, Vasilyeva & Waterfall 2011). SES effects have also been found for bilingual children (Oller & Eilers 2002), but due to the many factors involved in bilingual acquisition, these effects are less clear.
16Usage-based theory can be regarded as a particularly appropriate approach when investigating SES and language background, as it stresses the importance of input factors, such as input word form type and token frequency: Children learn language from input constructions and build their own increasingly complex and abstract representations in an emergentist way (Lieven 2014: 48). We may thus predict that low SES (LSES) children, who – at least in their home environments – usually receive a poorer and less diverse input and experience fewer joint attention situations (Farrant & Zubrick 2012) that are important not only for the acquisition of meaning, but also for the emergence of grammatical constructions, will show a slower acquisition of German noun plurals than their high SES (HSES) peers.
17Kindergarten teachers are the most important source of German input for the majority of bilingual children speaking another language at home. However, studies show divergent results with respect to a potential compensating effect of kindergarten teachers’ input on SES differences: Whereas a large cohort study from Germany (Weinert & Ebert 2013) does not find compensating effects, an early French intervention program (Zorman et al. 2011) lead to positive short-term effects on children’s vocabulary and syntax development.
18We investigate the following research questions:
(1)
|
How frequent are correct plurals, incorrect zero plurals and overt plural overgeneralizations in data of L1 acquisition and early L2 acquisition of German?
|
(2)
|
What is the impact of socio-economic status?
|
(3)
|
Is there any evidence for dialectal influence?
|
(4)
|
Is the methodology of data collection relevant for the outcome?
|
(5)
|
What is the role of kindergarten teachers’ input?
|
(6)
|
Is there any evidence of the Köpcke & Wecker (2017) model of product- and source-oriented strategies in the children’s data?
|
19We present extensive data on the acquisition of German noun plurals by 56 kindergarten children that were collected over a period of approximately 1 ½ years (from children’s mean ages 3;1 to 4;8; total age range: 2;11 – 4;11) in various kindergartens in Vienna: 29 children were monolingual German-speaking, 27 were early sequential bilinguals (with Turkish as their family language and German as their early second language in kindergarten from a mean age of 2;2, age range: 1;0 – 3;3). All bilingual children’s parents spoke mainly Turkish to their children - there were no mixed couples and no child was bilingual from birth. Approximately half of the children of each language background were from higher socio-economic status (HSES), the other half from lower SES (LSES) families. Groups were also almost balanced for gender (see Table 2):
Table 2. Child participants
Language background
|
SES
|
Gender
|
N
|
Subtotals language background - SES
|
Subtotals language background
|
L1 German
|
HSES
|
female
|
8
|
15 L1 HSES
|
29 L1
|
L1 German
|
HSES
|
male
|
7
|
L1 German
|
LSES
|
female
|
6
|
14 L1 LSES
|
L1 German
|
LSES
|
male
|
8
|
L2 German
|
HSES
|
female
|
8
|
14 L2 HSES
|
27 L2
|
L2 German
|
HSES
|
male
|
6
|
L2 German
|
LSES
|
female
|
6
|
13 L2 LSES
|
L2 German
|
LSES
|
male
|
7
|
Total
|
|
|
56
|
|
|
- 3 i.e. an ISEI value that exceeded the median of our sample by at least 10 points, cf. Czinglar et al (...)
- 4 However, as education is considered more important than prestige of profession in the context of ch (...)
20Like in most other studies on language acquisition (cf. Ensminger & Fothergill 2003), SES was primarily assessed by the main parental caregiver’s highest educational level (cf. OECD 1999): The LSES group included ISCED-97 levels 1 to 3c or 3b (i.e. from the former five-year compulsory school in Turkey which corresponds to level 1 to the nine-year compulsory school in Austria which includes a one-year polytechnic school and corresponds to level 3c or is followed by an apprenticeship and vocational school corresponding to level 3b, but without high school diploma). The HSES group had ISCED-97 levels 3a to 6 (i.e. from high school diploma up to PhD). The prestige of the parental profession was assessed according to the International Socioeconomic Index of Occupational Status (ISEI, cf. Ganzeboom & Treiman 1996). Nevertheless, the assessment of the ISEI values affected the SES classification of only three children of our sample: Their main parental caregivers had significantly better jobs3 than one would have expected from their formal educational levels. Therefore, these three children were “upgraded” to the HSES group.4
21We used two different procedures to investigate children’s plural production:
(1)
|
A structured plural elicitation task (Laaha et al. 2006), in which children had to form plurals from given singular forms, was conducted at two data points (at mean ages 3;4 and 4;8). The test items were almost balanced for plural marker and gender. They had been selected according to their token frequencies in the maternal input to one Austrian child aged 1;6 to 3;0 as well as in the CELEX database (Baayen et al. 1995). Plurals of middle token frequency were preferred (see Laaha et al 2006: 285 for the detailed procedure of item selection and Appendix A for the list of test items with the token frequencies of the original maternal input corpus, the caregiver input of the present study and of CELEX). For the present study, we used the short version of the test which contained only half of the test items (21) of the Laaha et al. 2006 study (42). As to the procedure, children were first shown a picture card of a single item (e.g. of a rabbit) and told: “This is a rabbit.” Then they were shown a card with three rabbits and asked: “And what are these? Three...?”. The plural forms produced by the children were transcribed by using a simplified phonetic transcription (UNIBET, cf. MacWhinney 2000) and coded by using a detailed coding scheme (Laaha et al. 2006). The three main categories investigated were correct plurals, incorrect zero plurals, and overt overgeneralizations.
|
(2)
|
One-hour video recordings of spontaneous teacher-child interaction in kindergarten were conducted at four data points (at mean ages 3;1, 3;4, 4;4; and 4;8). From these one-hour recordings, 30-minute samples containing the richest caregiver-child interaction were selected, transcribed, double-checked and tagged for part-of-speech and morphological form by using the CHILDES system (MacWhinney 2000).
|
22For the statistical analysis, we used the lme4 package (Bates et al. 2015) of R (R Core Team 2015) to conduct generalized linear mixed effects analyses (glmer) of the relationship between different plural categories (correct plurals, incorrect zero plurals, overt overgeneralizations), SES and language background to discover group differences between the children, first in both procedures separately and second with procedure as a fixed factor. Data point was another fixed factor, as there were two data points for the plural test and four data points for spontaneous speech recordings in kindergarten. Several models also included individual kindergarten teachers’ cumulative plural word form type and token frequencies (log-normalized) as fixed variables. Child ID as well as Stimulus ID were entered as random variables. The dependent variable was binomial, i.e. correct plurals were coded as 1 and incorrect plurals as 0 in the analysis of correct plurals. For each model, we performed one strict analysis (in which only standard German plurals were counted as correct) and one lenient analysis (in which dialect plurals were also counted as correct). Forms which could not be identified as clear plurals of one of the categories mentioned above (e.g. diminutives, reduplications, truncations) were classified as “Other forms” (see Appendix B), but due to their relatively small numbers, we did not include them in the statistical analysis.
23As mentioned above, we differentiated between correct plurals (Correct PL), incorrect zero plurals (Incorr. Zero PL) and overt plural overgeneralizations (Overt PL overgen.) as dependent variables. We always performed one strict analysis (in which only Standard German plurals were counted as correct) and one lenient analysis (in which also dialect plurals were counted as correct) for each dependent variable, which lead to six different dependent variables in total to be investigated (Correct PL strict, Incorr. zero PL strict, Overt PL overgen. strict, Correct PL lenient, Incorr. zero PL lenient, Overt PL overgen. lenient). Table 3 shows the descriptive statistics for all groups of children in all categories of plurals of the plural test (see also Appendix B, Table 13, for raw numbers, and Table 14 for percentages).
Table 3. Plural test: means, standard deviations, minimum, maximum
Group
|
Category
|
Mean
|
Std.Dev
|
Min
|
Max
|
HSES L1
|
Correct PL strict
|
12.83
|
4.23
|
5
|
19
|
LSES L1
|
Correct PL strict
|
8.26
|
3.66
|
3
|
16
|
HSES L2
|
Correct PL strict
|
4.38
|
2.62
|
1
|
12
|
LSES L2
|
Correct PL strict
|
3.38
|
1.38
|
0
|
7
|
HSES L1
|
Incorr. zero PL strict
|
3.63
|
3.15
|
0
|
13
|
LSES L1
|
Incorr. zero PL strict
|
7.81
|
4.21
|
2
|
17
|
HSES L2
|
Incorr. zero PL strict
|
12.04
|
4.57
|
3
|
17
|
LSES L2
|
Incorr. zero PL strict
|
13.29
|
4.24
|
2
|
18
|
HSES L1
|
Overt PL overgen. strict
|
3.97
|
2.82
|
0
|
11
|
LSES L1
|
Overt PL overgen. strict
|
3.15
|
1.99
|
0
|
7
|
HSES L2
|
Overt PL overgen. strict
|
2.38
|
3.32
|
0
|
10
|
LSES L2
|
Overt PL overgen. strict
|
0.96
|
1.52
|
0
|
6
|
HSES L1
|
Correct PL lenient
|
13.7
|
3.92
|
7
|
19
|
LSES L1
|
Correct PL lenient
|
10.52
|
3.12
|
7
|
17
|
HSES L2
|
Correct PL lenient
|
7.5
|
1.91
|
4
|
12
|
LSES L2
|
Correct PL lenient
|
7.04
|
1.9
|
1
|
9
|
HSES L1
|
Incorr. zero PL lenient
|
2.73
|
2.46
|
0
|
10
|
LSES L1
|
Incorr. zero PL lenient
|
5.67
|
3.34
|
1
|
13
|
HSES L2
|
Incorr. zero PL lenient
|
9.04
|
3.21
|
3
|
13
|
LSES L2
|
Incorr. zero PL lenient
|
9.67
|
3.12
|
1
|
13
|
HSES L1
|
Overt PL overgen. len.
|
3.9
|
2.78
|
0
|
10
|
LSES L1
|
Overt PL overgen. len.
|
3.04
|
1.97
|
0
|
7
|
HSES L2
|
Overt PL overgen. len.
|
2.25
|
3.21
|
0
|
10
|
LSES L2
|
Overt PL overgen. len.
|
0.96
|
1.4
|
0
|
5
|
24As shown in Table 3, HSES L1 children have the highest mean of correct plural word form types in the plural test, followed by the three other groups of children (LSES L1, HSES L2, LSES L2). These group differences are slightly smaller in the lenient condition (i.e. when dialect forms are also counted as correct). The same order of groups is found for overt overgeneralizations, although means are much lower. In contrast, the order is reversed for incorrect zero plurals: LSES L2 children produce most of them, followed by HSES L2 children, LSES L1 children and finally HSES L1 children.
25Table 4 shows the descriptive statistics for the spontaneous speech recordings in kindergarten (see also Appendix B, Table 13, for raw numbers, and Table 14 for percentages):
Table 4. Spontaneous speech: means, standard deviations, minimum, maximum
Group
|
Category
|
Mean
|
Std.Dev
|
Min
|
Max
|
HSES L1
|
Correct PL strict
|
5.5
|
5.44
|
0
|
23
|
LSES L1
|
Correct PL strict
|
3.62
|
3.24
|
0
|
14
|
HSES L2
|
Correct PL strict
|
2.43
|
2.98
|
0
|
10
|
LSES L2
|
Correct PL strict
|
1.06
|
1.74
|
0
|
8
|
HSES L1
|
Incorr. zero PL strict
|
0.16
|
0.46
|
0
|
2
|
LSES L1
|
Incorr. zero PL strict
|
0.2
|
0.52
|
0
|
2
|
HSES L2
|
Incorr. zero PL strict
|
0.18
|
0.43
|
0
|
2
|
LSES L2
|
Incorr. zero PL strict
|
0.08
|
0.27
|
0
|
1
|
HSES L1
|
Overt PL overgen. strict
|
0.2
|
0.64
|
0
|
4
|
LSES L1
|
Overt PL overgen. strict
|
0.23
|
0.5
|
0
|
3
|
HSES L2
|
Overt PL overgen. strict
|
0.21
|
0.56
|
0
|
3
|
LSES L2
|
Overt PL overgen. strict
|
0.02
|
0.14
|
0
|
1
|
HSES L1
|
Correct PL lenient
|
5.55
|
5.45
|
0
|
23
|
LSES L1
|
Correct PL lenient
|
3.77
|
3.32
|
0
|
14
|
HSES L2
|
Correct PL lenient
|
2.52
|
3.07
|
0
|
11
|
LSES L2
|
Correct PL lenient
|
1.12
|
1.77
|
0
|
8
|
HSES L1
|
Incorr. zero PL lenient
|
0.11
|
0.37
|
0
|
2
|
LSES L1
|
Incorr. zero PL lenient
|
0.07
|
0.32
|
0
|
2
|
HSES L2
|
Incorr. zero PL lenient
|
0.09
|
0.29
|
0
|
1
|
LSES L2
|
Incorr. zero PL lenient
|
0.02
|
0.14
|
0
|
1
|
HSES L1
|
Overt PL overgen. len.
|
0.2
|
0.64
|
0
|
4
|
LSES L1
|
Overt PL overgen. len.
|
0.21
|
0.46
|
0
|
2
|
HSES L2
|
Overt PL overgen. len.
|
0.21
|
0.56
|
0
|
3
|
LSES L2
|
Overt PL overgen. len.
|
0.02
|
0.14
|
0
|
1
|
26We can see that mean numbers of plural word form types are much lower in spontaneous speech (Table 4) than in the plural test (Table 3). Especially incorrect zero plurals, but also overt overgeneralizations are remarkably rare (means of 0.02 to 0.23).
27The following tables present the results of the generalized linear mixed effects models, first of the plural test and later of the spontaneous speech data.
Table 5. Test: plural word form types, effects of fixed factors
(z values and significance: *** 0.001, ** 0.01, * 0.05, n.s. non-significant)
Dependent variable
|
Lgbg:L2
|
SES:LSES
|
Dp:DP4
|
Lgbg:L2*SES:LSES
|
(1) Correct PL strict
|
-9.046***
|
-4.450***
|
7.494***
|
2.414*
|
(2) Incorr. zero PL strict
|
7.607***
|
3.747***
|
-4.798***
|
-1.341 n.s.
|
(3) Overt PL overgen. strict
|
-2.588**
|
-0.742 n.s.
|
-0.677 n.s.
|
-0.920 n.s.
|
(4) Correct PL lenient
|
-7.689***
|
-3.695***
|
6.177***
|
2.605**
|
(5) Incorr. zero PL lenient
|
7.158***
|
3.295***
|
-3.548***
|
-1.242 n.s.
|
(6) Overt PL overgen. lenient
|
-2.588**
|
-0.742 n.s.
|
-0.677 n.s.
|
-0.920 n.s.
|
28In the plural test (Table 5), we find significant effects of language background (Lgbg), SES and data point (Dp) as well as interaction effects between language background and SES (see Appendix B for total raw numbers and percentages as well as Appendix C, models 1–6, for statistical details):
29L1 children produce more correct plural word form types than L2 children and HSES children score higher than LSES children, although all children make considerable progress from the second to the fourth data point. There is also a significant interaction between SES and language background, showing that LSES L2 children produce more correct plurals than one would expect from their language background and socio-economic status. This result apparently reflects the fact that HSES L1 children largely outperform the LSES L1 and the HSES L2 group so that the results of the LSES L2 group still look surprisingly good.
30For incorrect zero plurals, which can most often be classified as repetitions of the given singulars, we also find significant effects of language background and SES. These forms are the most frequent category in less advanced children, i.e. in younger LSES L1 children and L2 children. In L2 children, production of incorrect zero plurals may also be reinforced by their L1 Turkish (Korecky-Kröll et al. 2016), as Turkish uses singular forms after plural numerals and quantifiers in order to avoid double-marking (e.g. üç araba ‘three car-SG’).
- 5 For productivity and potentiality of plural patterns see Klampfer et al. (2001), Laaha et al. (2006 (...)
31Overt overgeneralizations (e.g. *Bette for Betten ‘beds’), which indicate the productive application of (mostly) potential5 plural patterns, show only a significant effect of language background: L2 children use significantly fewer of these forms as they might not yet have acquired the basic regularities of German noun plural formation. Although neither SES nor data point can be identified as significant factors in Table 5, results indicate that most overt overgeneralizations are found in younger HSES L1 children (see Table 9 in 4.4). This is consistent with the usage-based model of language acquisition: Due to the exposure to their parental input, HSES L1 children receive the richest German input of all four groups of children and thus have the best opportunities to extract frequent productive patterns from their input that they may use in overgeneralizations.
32If we take dialectal variation into account (i.e. when we treat plural forms that are possible in Viennese dialect as correct forms, which increases especially the number of correct zero plurals, see “lenient” results in Table 5), the effects of language background and SES slightly decrease for correct plurals and incorrect zero plurals, but still remain significant.
33Is there an impact of kindergarten teachers’ input on the children’s results of the plural test? Although their cumulative plural word form type and token frequencies do not show significant effects, they slightly improve the fits of both models of correct plurals by diminishing their AIC values (cf. Levshina 2015: 149; for details see Appendix C, models 7-10). In the lenient condition, these slight improvements were marginally better when teachers’ plural types (instead of their plural tokens) were included.
- 6 The kindergarten of one HSES L1 girl refused to participate in the project; therefore we were only (...)
34For reasons of comparability with the plural test, we limited our analysis of spontaneous speech data to different plural types and did not perform an analysis of all occurring plural tokens. Thus, each plural type was counted only once per child and data point, even if it had been produced several times. Two different plural forms of one noun lemma (e.g., Ballons or Ballone ‘balloons’) were nevertheless counted as two different plural types. As numbers for incorrect zero plurals and overt overgeneralizations were extremely low in the spontaneous speech data (34 types of incorrect zero plurals and 37 types of overt overgeneralizations for 556 children at four data points, see also Appendix B), we limited the statistical analysis to correct plural types. Table 6 shows the main results of this analysis (for details see Appendix C, models 11-12).
Table 6. Spontaneous plural word form types, effects of fixed factors
(z values and significance: *** 0.001, ** 0.01, * 0.05, n.s. non-significant)
Dependent variable
|
Lgbg:L2
|
SES:LSES
|
Dp:DP3
|
Dp: DP4
|
Lgbg:L2* SES:LSES
|
(11) Correct PL strict
|
-2.737**
|
-1.820 n.s.
|
2.146*
|
1.601 n.s.
|
1.614 n.s.
|
(12) Correct PL lenient
|
-2.401*
|
-0.785 n.s.
|
2.227*
|
0.418 n.s.
|
1.570 n.s.
|
35Spontaneous speech data also show significant differences in type frequencies of correct plural forms between monolingual and bilingual children, but no significant SES differences, and the effect of language background is much smaller than in the test (as well as the difference between the strict and the lenient measure). Another significant effect is found for data point, as frequencies of correct plural types increase when the children get older (here especially for the third data point). In contrast to the plural test, there was no improvement of the model fit when teachers’ plural type or token frequencies were included in the above models (see Appendix C, models 13-16).
36Sections 4.2 and 4.3 revealed some common tendencies, but also some considerable differences between the results of the formal plural test and the spontaneous speech data. The following analysis includes Procedure (test vs. spontaneous speech) as a fixed factor and thus examines the effect of the methodology of data collection (see Table 7 and Appendix C, models 17-18):
Table 7. Correct plural word form types in the test and in spontaneous speech: effects of fixed factors (z values and significance: *** 0.001, ** 0.01, * 0.05, n.s. non-significant)
Dependent variable
|
Lgbg:L2
|
SES: LSES
|
Procedure: Test
|
Dp:DP3
|
Dp: DP4
|
Lgbg:L2* SES:LSES
|
(17) Correct PL strict
|
-9.292***
|
-4.680***
|
-8.909***
|
2.829**
|
2.859**
|
2.497*
|
(18) Correct PL lenient
|
-8.345***
|
-4.043***
|
-7.317***
|
2.384*
|
1.727 n.s.
|
3.086**
|
- 7 Although data point is still included as a factor in this model and its t values and significances (...)
37Indeed, we find clear effects of the procedure: All children have fewer correct plurals and higher error rates in the test than in spontaneous speech, but some groups of children may show greater differences within a certain setting than others. Therefore, we tested several additional models in order to examine potential interaction effects between Procedure (Proc) and children’s background variables (i.e. language background and SES):7
Table 8. Correct plural word form types in the test and in spontaneous speech: interaction effects of Procedure and children’s language background and SES (z values and significance: *** 0.001, ** 0.01, * 0.05, n.s. non-significant)
Dependent variable
|
Lgbg: L2
|
SES: LSES
|
Proc: Test
|
Lgbg:L2* SES:LSES
|
Lgbg:L2* Proc:Test
|
SES:LSES* Proc:Test
|
Lgbg:L2* SES:LSES* Proc:Test
|
(19) Correct PL strict
|
-3.073 ***
|
-1.308 n.s.
|
-5.837 ***
|
1.466 n.s.
|
-3.013 **
|
-1.668 n.s.
|
-0.207 n.s.
|
(20) Correct PL lenient
|
-2.752 ***
|
-0.369 n.s.
|
-4.604 ***
|
1.796 n.s.
|
-2.067 *
|
-1.844 n.s.
|
-0.814 n.s.
|
38As shown in Table 8, there are is a significant interaction effect between Procedure and children’s language background for correct plurals: In comparison to their spontaneous speech results, L2 children produce significantly fewer correct plurals in the plural test than one would expect merely from their background.
39As mentioned in 4.3, frequencies of incorrect zero plurals and overt overgeneralizations were too low in spontaneous speech to obtain statistically reliable results and thus were excluded from the general linear mixed effects analysis. Nevertheless, percentages of incorrect zero plurals and of overt overgeneralizations reveal some differences in strategies employed by the four groups of children when dealing with the test situation (see Table 9): On the one hand, HSES L1 children use slightly more overt overgeneralizations than incorrect zero plurals in the test when they are not sure about the correct plural form (23.64 % of overt overgeneralizations vs. 21.73 % of incorrect zero plurals at the second data point and 14.29 % vs. 13.02 % of zero plurals). On the other hand, L2 children of both SES and to a lesser extent also LSES L1 children have a strong tendency to produce incorrect zero plurals by repeating the singular forms given by the experimenter (up to 71.35 % in LSES L2 children at the second data point). In contrast, group differences in spontaneous speech are much smaller than in the plural test.
Table 9. Proportions of incorrect zero plurals and overt overgeneralizations out of all plural types by group of children and data point (strict coding; for the complete table see Appendix B, Table 14)
Language background
|
L1
|
L2
|
SES
|
HSES
|
LSES
|
HSES
|
LSES
|
Incorrect zero plurals
|
Spontaneous
|
|
|
|
|
DP1
|
3.03%
|
13.04%
|
0.00%
|
20.00%
|
DP2
|
4.41%
|
3.57%
|
8.00%
|
0.00%
|
DP3
|
2.83%
|
2.50%
|
4.92%
|
16.67%
|
DP4
|
1.11%
|
5.80%
|
8.06%
|
0.00%
|
Test
|
|
|
|
|
DP2
|
21.73%
|
42.91%
|
62.00%
|
71.35%
|
DP4
|
13.02%
|
33.10%
|
56.70%
|
69.00%
|
Overt overgeneralizations
|
Spontaneous
|
|
|
|
|
DP1
|
7.58%
|
8.70%
|
9.09%
|
0.00%
|
DP2
|
4.41%
|
3.57%
|
12.00%
|
14.29%
|
DP3
|
0.94%
|
6.25%
|
3.28%
|
0.00%
|
DP4
|
2.22%
|
5.80%
|
9.68%
|
0.00%
|
Test
|
|
|
|
|
DP2
|
23.64%
|
14.55%
|
10.50%
|
3.24%
|
DP4
|
14.29%
|
15.86%
|
12.37%
|
6.27%
|
- 8 This analysis is restricted to the correlations of spontaneous speech and test results of data poin (...)
40Interestingly, despite the differences between the results of the two procedures described above (see Tables 7 to 9), we find some significant correlations between the test and the spontaneous speech data that indicate that both procedures still represent similar aspects of language development (see Table 10):8
Table 10. Pearson correlations (two-tailed) of plural types in the test and in spontaneous speech (significance: *** 0.001, ** 0.01, * 0.05, n.s. non-significant)
Children’s measures (test – spontaneous)
|
DP2: Pearson r
|
DP2: p
|
DP4: Pearson r
|
DP4: p
|
Correct PL strict
|
0.620
|
< 0.001***
|
0.388
|
0.003**
|
Incorr. zero PL strict
|
-0.083
|
0.573 n.s.
|
0.169
|
0.216 n.s.
|
Overt PL overgen. strict
|
0.142
|
0.334 n.s.
|
0.156
|
0.255 n.s.
|
Correct PL lenient
|
0.565
|
< 0.001***
|
0.374
|
0.005**
|
Incorr. zero PL lenient
|
-0.090
|
0.552 n.s.
|
0.092
|
0.502 n.s.
|
Overt PL overgen. lenient
|
0.136
|
0.361 n.s.
|
0.099
|
0.472 n.s.
|
41But the significant correlations are limited to correct plurals: Children that use more correct plural types in the test also use more correct plural types in spontaneous speech. These correlations are stronger at data point 2, i.e. when the children are still younger. For incorrect zero plurals as well as for overt overgeneralizations, we do not find any significant correlations, which seems indeed to be an effect of the procedure: If a child knows a plural form, he or she may freely use this form in spontaneous speech as well as reproduce it correctly in the test. However, if children are not entirely familiar with a certain plural form, they will probably just not use it in spontaneous speech, but in the test, they will be more likely to repeat the given singular form (which leads to an incorrect zero plural in most cases) or produce an overt overgeneralization.
42As described in sections 4.2 and 4.3, we obtained divergent results with respect to the impact of kindergarten teachers’ input on children’s plural production: We did not find significant effects of teachers’ input on children’s results of the plural test, but the fits of the respective models were slightly improved when including teachers’ input plural frequencies as a variable, which was not the case for the spontaneous speech data. For the plural test, plural type frequencies yielded marginally better results than plural token frequencies. In order to examine these results more closely, we conducted additional correlation analyses between children’s output and teachers’ input. As shown in Table 11, only a few correlations yield significant effects: Most effects are found for children’s correct plural types at data point 2 (i.e. at mean age 3;4). Here we find some significant correlations of teachers’ spontaneous speech plural type frequencies not only with children’s plural test results, but also with children’s spontaneous speech results. Younger children whose teachers use more spontaneous plural types are thus more likely to use more correct plurals in both settings.
Table 11. Pearson correlations (two-tailed) of children’s output measures (test and spontaneous speech) and teachers’ cumulative input plural type and token frequencies (significance: *** 0.001, ** 0.01, * 0.05, n.s. non-significant)
Children’s measures
|
Teachers’ measures
|
DP2: Pearson r
|
DP2: p
|
DP4: Pearson r
|
DP4: p
|
Correct PL strict (test)
|
PL TYP
|
0.348
|
0.015*
|
0.260
|
0.056 n.s.
|
Correct PL strict (test)
|
PL TOK
|
0.169
|
0.251 n.s.
|
0.112
|
0.414 n.s.
|
Incorr. zero PL strict (test)
|
PL TYP
|
-0.121
|
0.411 n.s.
|
-0.269
|
0.047*
|
Incorr. zero PL strict (test)
|
PL TOK
|
0.031
|
0.832 n.s.
|
-0.136
|
0.322 n.s.
|
Overt PL overgen. strict (test)
|
PL TYP
|
0.126
|
0.393 n.s.
|
0.068
|
0.620 n.s.
|
Overt PL overgen. strict (test)
|
PL TOK
|
0.045
|
0.761 n.s.
|
0.081
|
0.558 n.s.
|
Correct PL lenient (test)
|
PL TYP
|
0.393
|
0.006**
|
0.253
|
0.062 n.s.
|
Correct PL lenient (test)
|
PL TOK
|
0.244
|
0.095 n.s.
|
0.103
|
0.455 n.s.
|
Incorr. zero PL lenient (test)
|
PL TYP
|
-0.135
|
0.361 n.s.
|
-0.290
|
0.032*
|
Incorr. zero PL lenient (test)
|
PL TOK
|
0.007
|
0.958 n.s.
|
-0.141
|
0.305 n.s.
|
Overt PL overgen. lenient (test)
|
PL TYP
|
0.134
|
0.369 n.s.
|
0.079
|
0.569 n.s.
|
Overt PL overgen. lenient (test)
|
PL TOK
|
0.080
|
0.591 n.s.
|
0.079
|
0.564 n.s.
|
Correct PL strict (spont.)
|
PL TYP
|
0.302
|
0.037*
|
0.228
|
0.093 n.s.
|
Correct PL strict (spont.)
|
PL TOK
|
0.191
|
0.195 n.s.
|
0.074
|
0.591 n.s.
|
Incorr. zero PL strict (spont.)
|
PL TYP
|
0.185
|
0.208 n.s.
|
-0.334
|
0.012*
|
Incorr. zero PL strict (spont.)
|
PL TOK
|
0.089
|
0.546 n.s.
|
-0.348
|
0.009**
|
Overt PL overgen. strict (spont.)
|
PL TYP
|
0.043
|
0.771 n.s.
|
-0.010
|
0.939 n.s.
|
Overt PL overgen. strict (spont.)
|
PL TOK
|
0.096
|
0.517 n.s.
|
-0.460
|
0.739 n.s.
|
Correct PL lenient (spont.)
|
PL TYP
|
0.284
|
0.051 n.s.
|
0.213
|
0.119 n.s.
|
Correct PL lenient (spont.)
|
PL TOK
|
0.175
|
0.234 n.s.
|
0.063
|
0.646 n.s.
|
Incorr. zero PL lenient (spont.)
|
PL TYP
|
0.411
|
0.004**
|
-0.244
|
0.072 n.s.
|
Incorr. zero PL lenient (spont.)
|
PL TOK
|
0.261
|
0.074 n.s.
|
-0.306
|
0.023*
|
Overt PL overgen. lenient (spont.)
|
PL TYP
|
0.043
|
0.771 n.s.
|
-0.029
|
0.831 n.s.
|
Overt PL overgen. lenient (spont.)
|
PL TOK
|
0.096
|
0.517 n.s.
|
-0.059
|
0.671 n.s.
|
43At data point 4 (i.e. at mean age 4;8), we mostly find significant negative correlations between teachers’ spontaneous plural type and token frequencies and children’s use of incorrect zero plurals: Older children whose teachers use many plurals are thus less likely to use the avoidance strategy of producing incorrect zero plurals in the test as well as in spontaneous speech.
44As described in section 1.2, Köpcke & Wecker (2017) claim that children use product-oriented as well as source-oriented strategies to deal with the different categories of Standard German noun plurals. In the following tables (Table 12 – 16), their five main hypotheses (Köpcke & Wecker 2017: 98-100) were applied to our data. As raw numbers of some categories were very low, data of all groups of children as well as both procedures (test and spontaneous speech) were collapsed. Hypotheses (H) a-c concern the product-oriented strategies, whereas hypotheses d-e deal with the source-oriented strategies:
45H(a): “-ø will be used for non-feminine items ending in -el, -er or -en to an increasing amount.“
Table 12. Hypothesis (a): Percentages and raw numbers of children’s plural types
Non-feminines
|
No zero plural
|
Zero plural
|
Total
|
-el
|
33.55% (105)
|
66.45% (208)
|
100.00% (313)
|
-er
|
16.38% (38)
|
83.52% (194)
|
100.00% (136)
|
-en
|
9.56% (13)
|
90.44% (123)
|
100.00% (232)
|
Total
|
22.91% (156)
|
77.09% (525)
|
100.00% (681)
|
46Hypothesis (a) was confirmed by our data, at least when percentages were considered: Zero plurals were used most frequently for non-feminine nouns ending in -en, followed by -er and -el, although nouns ending in -el had higher raw numbers than the two other categories.
H(b): “-ø will be used in increasing number for items of all genders ending in -ø (monosyllabic), unstressed full vowel (-v), or -e.“
Table 13. Hypothesis (b): Percentages and raw numbers of children’s plural types
All genders
|
No zero plural
|
Zero plural
|
Total
|
-ø (monosyllabic)
|
63.49% (847)
|
36.51% (487)
|
100.00% (1334)
|
-v
|
43.48% (140)
|
56.52% (182)
|
100.00% (322)
|
-e
|
74.14% (301)
|
25.86% (105)
|
100.00% (406)
|
Total
|
62.46% (1288)
|
37.54% (774)
|
100.00% (2062)
|
47Hypothesis b was not fully confirmed by our data: Zero plurals were least frequently used for nouns ending in -e, although they should have been used for them most frequently. But as predicted, zero plurals were used more frequently for nouns ending in unstressed full vowels than for monosyllabic nouns (at least when considering percentages).
H(c): “-ø will be used for feminine items ending in -el or -er in increasing number.“
Table 14. Hypothesis (c): Percentages and raw numbers of children’s plural types
Feminines
|
No zero plural
|
Zero plural
|
Total
|
-el
|
100.00% (9)
|
0.00% (0)
|
100.00% (9)
|
-er
|
50.00% (1)
|
50.00% (1)
|
100.00% (2)
|
Total
|
90.91% (10)
|
9.09% (1)
|
100.00% (11)
|
48Although percentages seem to confirm hypothesis (c), raw numbers are very small (9 types for ‑el and 2 types for -er). Therefore it is questionable whether hypothesis (c) can indeed be confirmed by our data.
H(d): “-ø will be used more frequently for non-feminine items ending in -el or -er than for feminine items ending in -el or -er.
Table 15. Hypothesis (d): Percentages and raw numbers of children’s plural types
Non-feminines
|
No zero plural
|
Zero plural
|
Total
|
-el
|
33.55% (105)
|
66.45% (208)
|
100.00% (313)
|
-er
|
16.38% (38)
|
83.52% (194)
|
100.00% (136)
|
Total non-feminines
|
26.24% (143)
|
73.76% (402)
|
100.00% (545)
|
Feminines
|
|
|
|
-el
|
100.00% (9)
|
0.00% (0)
|
100.00% (9)
|
-er
|
50.00% (1)
|
50.00% (1)
|
100.00% (2)
|
Total feminines
|
90.91% (10)
|
9.09% (1)
|
100.00% (11)
|
Total
|
27.52% (153)
|
72.48% (403)
|
100.00% (556)
|
49As shown in Table 15, hypothesis (d) was clearly confirmed by our data.
H(e): “-en will be used more frequently for feminine monosyllabic items ending than for non-feminine monosyllabic items.“
Table 16. Hypothesis (e): Percentages and raw numbers of children’s plural types
Monosyllabic nouns
|
Other plural
|
-en plural
|
Total
|
Feminines
|
92.27% (215)
|
7.73% (18)
|
100.00% (233)
|
Non-feminines
|
92.38% (1018)
|
7.62% (84)
|
100.00% (1102)
|
Total
|
92.36% (1233)
|
9.09% (102)
|
100.00% (1335)
|
50Compared to monosyllabic non-feminine nouns, percentages of monosyllabic feminine nouns taking ‑en plurals were marginally higher, but the difference was only 0.11 %. Furthermore, in raw numbers, non-feminine monosyllabic nouns were much more frequent than feminine monosyllabic nouns. Therefore it is again questionable whether hypothesis (e) can indeed be confirmed by our data.
51To summarize, most of the hypotheses were at least not disconfirmed by our data, except for H (b), which predicted nouns ending in -e to be most frequently used with zero plurals (which were indeed used least frequently with zero plurals). Overall, the model highlights the importance of noun gender and shows that the use of zero plurals is not random, but occurs more frequently with nouns that have a more plural-like structure already in the singular. Nevertheless, there are three problematic points concerning the application of the model to our data: First, some categories showed extremely low type frequencies making it difficult to confirm the hypotheses. Second, not all plural categories of Standard German were covered by Köpcke’s & Wecker’s (2017) model: For example, there were no predictions for umlaut plurals, ‑s plurals or plurals of polysyllabic nouns not ending in schwa syllables (e.g. Schmetterling ‘butterfly’). Third, the model would also have to be adapted by including predictions about plural schemas in Viennese dialect in order to fully account for the variation in our data.
52In accordance with previous studies, our results show that the acquisition of German noun plurals is a great challenge for young children acquiring German: We find high numbers of errors (first more omission errors, i.e. incorrect zero plurals, later more commission errors, i.e. overt overgeneralizations which appear earlier in the L1 children and later in the L2 children who are in earlier phases of their plural development in comparison to the L1 children). Furthermore, we find considerable variation of plural forms. For example, the plurals of Haus ‘house’ produced by the children were correct Häuser, incorrect zero *Haus as well as five different overt overgeneralizations, namely *Häus, *Häuse, *Häusers, *Hause, *Hausen. Examples with similar degrees of variation are Zug ‘train’ (correct Züge, incorrect zero Zug, overt overgeneralizations *Zuge, *Zugen, *Zuger, *Zügen) and Maus ‘mouse’ (correct Mäuse ‘mice’ , incorrect zero Maus, overt overgeneralizations *Mäuser, *Mause and *Mausen). Two other forms, namely the diminutive Mausi that is produced with a dialectal zero plural and the lexical replacement Ratten ‘rats’, were classified as a separate category (“Other forms” in Appendix B).
53Our results demonstrate that even children acquiring German as their L1 show considerable difficulties in producing correct plural forms, but for children acquiring German as their L2, this challenge is still greater: Children’s language background (in our case L1 German vs. L1 Turkish and L2 German) turned out to be the most significant participant factor in almost all measures investigated. One specific challenge for children acquiring Turkish as their first language concerns the procedure of the plural test that may evoke influence from the L1: The plural form was elicited by using the numeral drei ‘three’ which would require a singular form in Turkish. For example, one HSES L2 boy used the correct Standard German plural Stifte ‘pencils’ in a spontaneous sentence during the plural test, but when asked for the three pencils on the test picture, he still used the zero form Stift.
54Socio-economic status has also been shown to be a very important factor in acquisition of German noun plurals, although it has a slightly smaller impact than language background. But LSES L1 children also score significantly lower than their HSES peers, especially in the test setting. We also find some interaction effects of language background and SES: In the plural test, HSES L1 children produce more correct plurals than can be explained by their language background and SES, and LSES L2 children also score surprisingly high in comparison to the two other groups. On the other hand, L2 LSES children use fewer overt overgeneralizations than all other groups of children.
55We also found strong effects of methodology: All children had smaller rates of correct plurals as well as higher error rates in the highly metalinguistic plural test than in spontaneous speech, but there were also some interaction effects of the procedure with children’s background variables. Compared to their spontaneous speech results, L2 children produced significantly more incorrect zero plurals and fewer overt plural overgeneralizations in the plural test – for LSES children, there was a similar trend, but no significant interaction effect. On the other hand, HSES L1 children used more overt overgeneralizations in the test. These results provide evidence for different strategies employed by different groups of children, namely an avoidance strategy for the more vulnerable groups, but a strategy of guessing according to frequent plural patterns (as claimed by usage-based theory) for the most advanced group.
56In order to account for plural variation in Standard German vs. Viennese dialect, we conducted two versions of each analysis, one based on the strict coding (in which only plurals that are correct in Standard German were counted as correct forms) and another one based on the lenient coding (in which all plurals that are allowed in Viennese dialect were also counted as correct forms). Although both sets of results did not show great divergences, the lenient coding slightly attenuated the effects of language background and SES in most cases. Dialect-standard variation thus also plays a certain role, but does not have a great impact on the results. Accordingly, the hypotheses of the Köpcke & Wecker (2017) model on Standard German plurals also apply to our data in a large part, but should also be adapted to Viennese dialect as well as extended to further categories that are not covered by the present model.
57Kindergarten teachers’ input is also relevant for children’s performance in the plural test: The models that include teachers’ cumulative type or token frequencies show better fits than the corresponding models that do not include these variables. We also find some significant correlations of teachers’ plural type frequencies and children’s correct plurals in the plural test at the second data point. Furthermore, we find some negative correlations concerning children’s avoidance strategy to use incorrect zero plurals: Children whose teachers use more plurals produce significantly fewer of these zero plurals. In accordance with previous studies (e.g. Bybee 2001), we also found that teachers’ plural word form type frequencies played a more important role than their plural token frequencies, suggesting that it is more important for children’s grammatical development to hear more different forms than to hear the same forms again and again. Nevertheless, the stronger correlations with teachers’ plural type frequencies may also be due to the fact that we limited our analyses to children’s type frequencies and that type frequencies may correlate better with type than with token frequencies. Overall, our results suggest that kindergarten teachers’ spontaneous input is also highly relevant for children’s grammatical development.
58While both language background and SES have been shown to be highly relevant factors in our sample, the testing procedure also plays an important role: Fast real-time access to German plural forms is clearly easier in spontaneous speech than in a test that requires high metalinguistic awareness and creates a situation of high insecurity of selection.
59Our results also lead us to some considerations concerning future research: When investigating vulnerable populations such as young children, LSES children or second language learners, we strongly recommend the usage of mixed-methods designs in research projects: Researchers should not only rely on tests, which may be especially challenging for these populations, but also on spontaneous speech recordings or on interview data in order to get a more comprehensive picture of the participants’ linguistic proficiencies.