Error patterns in the acquisition of German plural morphology

Error patterns in the
acquisition of German plural
morphology: Evidence for
the relevance of
grammatical gender as a
cue
Bettina Spreng
University of Toronto
This paper presents some results of a study on the acquisition of
German plural morphology. The error patterns of children and
the adult control group reveal that the acquisition of plural
morphology is conditioned in a predictable manner by a) the
phonotactic shape of the noun stem and b) the gender of the
noun. An acquisition model is proposed that takes this fact into
account and also predicts the acquisition process with possible
and impossible errors. It is based on the observation that errors
demonstrate in a predictable way what children use as cues for
acquisition. It not only explains the fact that adults have similar
problems forming plurals from new nouns but also which errors
adults and children are likely to exhibit.
1
The “unpredictability” of German plural morphology
This paper presents the results of a study examining the role the category gender
plays in the acquisition of plural morphology in German. It shows the error patterns
children exhibit in the process of acquiring German plural morphology and suggests a
model of acquisition that predicts exactly these error patterns. The acquisition model I
propose takes into account the gender of the noun as a disambiguating cue for the child to
identify the correct plural suffix.
The plural suffix for each noun is to some extent determined by the phonotactic
shape of the noun stem and the gender of the noun. For some plural suffixes, one gender
is not permitted, thus making gender a distinguishing factor for these plural suffixes. In
other words, if the child knows the gender of a particular noun she should be able to rule
out certain plural suffixes that do not allow for nouns of that gender. While many studies
T W
P L
Toronto Working Papers in Linguistics 23.2: 147-172
Copyright © 2004 Bettina Spreng
BETTINA SPRENG
on the acquisition of the German plural morphology1 acknowledge that both gender and
phonotactic shape of the noun are controlling factors for determining the plural suffix of a
noun in German, the specific role of gender in the acquisition of the plural morphology
has not been independently examined.
German has three genders (masculine, feminine, and neuter) that can be identified
by the definite article for each noun but are not clearly identifiable on the noun itself. The
definite article der signals masculine gender, die signals feminine and das signals neuter.
There are five plural suffixes -e, -(e)n, -er, -s, -ø, the selection of each is primarily
determined by the phonotactic shape of the noun, i.e. number of syllables, shape of final
syllable, final segment of the stem and secondarily by the gender of the noun. Each noun
selects only one plural suffix2 while the conditions for the selection of a suffix are partly
overlapping. For instance, some plural types allow for nouns of the same phonotactic
shape but differ in the genders they allow.
Table 1.
Plural Morphology of German taken from Clahsen et al. (1992)
1
Plural
Type
-Ø(+Umlaut)
2
-e (+Umlaut)
3
-er (+Umlaut)
4
-(e)n
5
-s
Singular
Plural
English translation
der Daumen
die Mutter
das Fenster
der Hund
die Kuh
das Fest
der Leib
das Huhn
der Bauer
die Strasse
das Bett
das Auto
der Park
die Lok
die Daumen
die Mütter
die Fenster
die Hunde
die Kühe
die Feste
die Leiber
die Hühner
die Bauern
die Strassen
die Betten
die Autos
die Parks
die Loks
the thumb/thumbs
the mother/mothers
the window/windows
the dog/dogs
the cow/cows
the party/parties
the body/bodies
the hen/hens
the farmer/farmers
the street/streets
the bed/beds
the car/cars
the park/parks
the locomotive/s
A classification of plural types according to the above table however runs into
various problems, especially when one needs to consider the phonotactic shape of the
nouns. For instance, the above classification implies a phonological rule for the suffix
-(e)n. Mugdan (1977) states that noun plurals must end in a syllable containing [ə]
(orthographic ‘e’) unless pluralised with -s. Accordingly, Wiese (1996) proposes partly
based on Mugdan’s findings a rule of ə-epenthesis if the noun does not end in an open
syllable containing [ə]. Wegener (1999) argues against ə-epenthesis and proposes a rule
of ə deletion if the plural suffix follows a syllable containing [ə]. If this syllable is closed,
the last segment must be a liquid or nasal. The rule of ə-deletion has the advantage that
nouns like Abend/Abend-e ‘evening/evenings’ do not require an extra rule of epenthesis
or have to be stored as exceptions. Empirically, ə-epenthesis can also not explain why
nouns which end in full vowels take the -s plural *Oma-n/Oma-s ‘grandma’ instead of–n
as Wegener (1999) points out. However, exceptions such as Nachbar-n ‘neighbour-s’,
Ungar-n ‘hungarian-s’ constitute a problem for both proposals.
1
2
See Clahsen (1999) for a summary of the literature.
Subsequently referred to as plural type in this paper.
148
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
To avoid these issues in the study, I classified tokens where the plural suffix –n
attaches to nouns that end in a vowel (plural type V-n), in a consonant (plural type C-n)
and a plural suffix -en. Thus, I may be able to answer the question whether plural type -Ø
and -e are the same plural types or whether plural type -e and plural type -en are of the
same plural type.3
This issue also plays a role in whether to take the plural type -Ø as an extra suffix
or not, or in other words, whether umlaut is a separate plural form. In that case, I would
have had to select nouns with umlaut in the plural separated by plural suffix.
Table 2.
Tested Tokens
C-n
Plural
type:
masculine:
feminine:
neuter:
-e
-en
-er
-∅
4
-s
V-n
suffix –n after
stem final
consonant
suffix -e
suffix
-en
suffix
-er
zero suffix
suffix -s
suffix -n
after stemfinal ə
Bauer
‘farmer’
Pantoffel
‘slipper’
Muschel
‘mussel’
Angel
‘fishing rod’
Fisch
‘fish’
Bus
‘bus’
Maus
‘mouse’
Kuh
‘cow’
Schiff
‘ship’
Schaf
‘sheep’
Mast
‘mast’
Bär
‘bear’
Tür
‘door’
Uhr
‘clock’
Ohr
‘ear’
Bett
‘bed’
Mann
‘man’
Geist
‘ghost’
Käfer
‘beetle’
Kuchen
‘cake’
Hase
‘rabbit’
Löwe
‘lion’
Brücke
‘bridge’
Ente
‘duck’
Haus
‘house’
Buch
‘book’
Fenster
‘window’
Kissen
‘pillow’
Papa
‘daddy’
Uhu
‘eagle-owl’
Lok
‘locomotive’
Oma
‘granny’
Taxi
‘taxi’
Auto
‘car’
The unpredictability of the German plural system lies in the fact that each noun
takes one specific plural suffix; however the conditioning factors for each plural suffix
are by no means unambiguous. For example, the phonotactic shape of two singular nouns
that take different plural suffixes might be the same, but their gender might be different.
(1)
Singular
das Haus (n) ‘the house’
die Maus (f) ‘the mouse’
Plural
die Häus-er
die Mäus-e
‘the houses’
‘the mice’
Therefore, gender disambiguates these two nouns as to which plural suffix they
have to take. On the other hand, sometimes two nouns are the same in phonotactic shape
and gender but have different plural suffixes.
3
The conditioning of the plural types as I have classified them seems to suggest that -e and -en are of the
same plural type since they attach to the same kind of nouns whereas -o and -e differ with respect to the
gender of the nouns that take these plural suffixes.
4
There are two highly frequent feminine nouns for plural type -o: Mutter ‘mother’ and Tochter ‘daughter’.
Since feminine nouns are almost non-existent in this plural type, they were not tested and assumed to be
exceptions and to be rote-learned. Also, these only feminine nouns form the plural with Umlaut, which I
tried to avoid as much as possible.
149
BETTINA SPRENG
(2)
Singular
der Mast (m) ‘the mast’
der Geist (m) ‘the ghost’
Plural
die Mast-en
die Geist-er
‘the masts’
‘the ghosts’
For (2), the child has no means of determining from the input which plural suffix
is correct for which noun (unless the stem vowel as such provides this information), so I
claim that the plural forms of these nouns are memorised eventually as fully inflected
forms. Before that point in the acquisition process, errors occur only within the range of
possible plural types -en and -er.
The most unpredictable plural type is -s, which has led for instance Clahsen
(1999) to claim that this is the only rule-based plural type in German, whereas the other
plural types are assumed to be learned by rote. This paper will show that if compared to
the degree of predictability of the other plural types, the plural type -s is just as
predictable as the other types and does not inhabit a special status within the system.
2
Predictability of errors: a proposal for the acquisition of German
plural morphology
This paper proposes a system that takes into account the degree of predictability
for each plural type. In contrast to previous studies, it also makes use of the category
gender as a determining factor for each plural suffix. Together with phonotactic
properties of the noun stem, the feature gender serves as a means to disambiguate the
conditions for each plural type.
(3)
Plural type
a.
b.
-e
-er
c.
d.
C-n
-Ø
Singular (monosyllabic,
identical stem vowel)
Maus (f)
Haus (n)
(bisyllabic, identical final
segments and final syllable nucleus)
Bauer (m)
Fenster (n)
Plural5
Mäus-e
Häus-er
mouse
house
Bauer-n
Fenster-Ø
farmer
window
The examples in (3a) and (3b) exhibit the same phonotactic shape but belong to
different plural types. However, they are distinct in gender and (3b) does not allow for
feminine nouns. Thus, gender serves as a distinguishing factor to identify the correct
plural type of Haus.
The system I am proposing predicts not only the difficulty of each plural type in
the acquisition process but also the errors children are likely to exhibit, depending on
their stage in the acquisition process.
The phonotactic and gender conditions for each plural type are shown in the
following table.
5
All German examples are taken from the tokens of my study unless indicated otherwise.
150
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
Table 3.
a.
C-n
b.
-e
c.
-en
d.
-er
e.
-∅
f.
-s
g. V-n
Conditions for plural types
Conditions
This type allows bisyllabic masculine and feminine nouns, where the last
syllable is closed and contains a schwa. The last segment is either [r] or [l].
Bauer ‘farmer’
Muschel ‘mussel’
Pantoffel ‘slipper’
Angel ‘fishing rod’
This type allows all genders. All tokens are monosyllabic and end in a
consonant.
Fisch ‘fish’
Maus ‘mouse’
Schiff ‘ship’
Bus ‘bus’
Kuh ‘cow’
Schaf ‘sheep’
This type is identical to type -e.
Mast ‘mast’
Tür ‘door’
Ohr ‘ear’
Bär ‘bear’
Uhr ‘clock’
Bett ‘bed’
This type is identical to type -e and -en except that it does not allow feminine
gender.
Mann ‘man’
Haus ‘house’
Geist ‘ghost’
Buch ‘book’
This type allows bisyllabic masculine and neuter nouns, with a closed final
syllable containing a schwa. The last segment is [n] or [r].
Käfer ‘beetle’
Fenster ‘window’
Kuchen ‘cake’
Kissen ‘pillow’
This type allows all genders, any number of syllables and the last segment is
either a full vowel or any consonant.
Papa ‘daddy’
Lok ‘locomotive’
Taxi ‘taxi’
Uhu ‘eagle-owl’
Oma ‘granny’
Auto ‘car’
This type allows bisyllabic masculine and feminine nouns with the final open
syllable ending in schwa.
Hase ‘rabbit’
Brücke ‘bridge’
Löwe ‘lion’
Ente ‘duck’
Table 3 shows that each plural type has its own set of features that the nouns they
can attach to must bear. As already mentioned, the conditions for each plural type partly
overlap but also exclude each other. Based on the above types of conditions, I predict the
following substitution patterns children are likely to exhibit.
151
BETTINA SPRENG
(4)
Substitution patterns
-e
V-n
-en
masc./neuter
masc./neuter
final C
-er
C-n
masc. final [r]
bisyllabic, final C
-Ø
monosyllabic
masc./neuter
bisyllabic
masculine/neuter
-s
The diagram shows how the plural types can be used interchangeably. Plural
types that are connected by arrows can be substituted for each other under certain
circumstances that are depicted in the boxes. The boxes indicate restrictions on these
substitutions. For example, only bisyllabic masculine and neuter nouns can be used as
errors of plural types -ø vs. –s. The unconnected plural type V-n should show no
substitution errors with other plural types. The unrestricted arrow between plural types en and -e indicates that these plural types can be used interchangeably for all nouns for
which the plural types apply since their conditions are identical.
When the child does not know the gender of a noun, substitution errors occur
more freely, e.g. the child ignores the gender restrictions on the arrows. Thus, not only
masculine and neuter but also feminine nouns should exhibit errors between plural types ø and –s. Thus, when gender eliminates the possibility of substitution errors between
plural types, error rates should be higher than between plural types where gender is not a
distinguishing factor.
(5)
Scale of problematic cases depending on availability of gender knowledge
Gender not available > approximate gender as a category is available > exact gender is available
highest error rates
lowest error rates
Applying the system in (4) to the individual plural types, I predict the following
scenarios. The phonotactic conditions of plural type -er are identical to the phonotactic
conditions of plural type -e and -en. However, if the child knows that a noun is feminine,
the choice is restricted to plural types -e and -en since plural type -er does not allow
feminine nouns. Thus, if the child indicates masculine or neuter gender, errors should be
expected with all three plural types. However, if the child indicates feminine gender only,
responses with -en and -e can be expected.
152
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
Nouns of plural type –Ø and C-n also require knowledge of the gender of the
noun if the final segment of the singular form is [r]. In this case, the phonotactic
conditions are identical and require the knowledge that there are three distinct genders.
For example, plural type –Ø has no feminine nouns and plural type C-n allows no neuter
nouns. Therefore, if the child knows one of these phonotactically similar nouns is
feminine, no -Ø type plural forms should occur.
Gender and phonotactic shape of the plural types -e and -en are identical.
Depending on whether one assumes a rule of ə-epenthesis or ə-deletion, one of them
should be acquired more easily. If we suppose that the child has to acquire a phonological
rule of ə-epenthesis, type -e nouns should be substituted with -Ø before the acquisition of
that rule. For a rule of ə-deletion, we can expect a high rate of -Ø substitution responses
for nouns with plural type -e after the acquisition of that rule. On the other hand, we
should expect suffixation of -e, -en, and -er for nouns that take the zero suffix before the
acquisition of a ə-deletion rule.
The system in (4) implies that the system of rules for the German plural is not
only determined by the complexity of their conditioning but also by the kind of
interaction between the conditions for application. Thus, the acquisition of a plural suffix
includes not only the correct conditions for each plural suffix but also the evaluation of
each new noun against the features of previously acquired nouns. For acquisition, this
means that when the child evaluates two noun stems of different plural types that can be
disambiguated by certain properties, they are classified for their respective plural types. If
the child evaluates two nouns and there is no way to distinguish their plural types, their
plural forms are stored in memory eventually. The child may encounter one of the nouns
as singular nouns again and evaluates it again against another noun, thus disambiguating
it for a plural type.
Therefore, I claim that only nouns that have properties that do not disambiguate
for one plural type at a certain point in acquisition are stored as plural forms. Under this
view, these are the only true irregular forms. I predict that noun stems cannot be
classified for a certain plural type when the conditioning factors for two plural types
overlap. Thus, errors occur only when the plural forms of such nouns have not been
stored. This also results in a larger number of zero responses (singular utterances in plural
environments) for nouns of these ambiguous plural types.
With respect to the question of whether plural type -s is used in a different
manner, I expect the answer to be in the affirmative. Plural type -s shows very flexible
conditions compared to most other plural forms and would therefore be a preferred option
for an error response. However, I would also expect that it would be more prone to zero
responses simply because monosyllabic nouns ending in a consonant can take a large
number of plural suffixes including –s. These nouns are harder to classify and are
therefore more prone to being memorized as plural forms eventually.
Another aspect why plural type -s responses may pattern differently is that the
conditioning is rather unrestricted. For example, plural type -s is allowed for nouns of all
genders, with any number of syllables, and the final segment can either be a full vowel or
a consonant. In comparison, plural type V-n is applied only to masculine and feminine
bisyllabic nouns with an open final syllable ending in schwa. For this plural type, I do not
expect high rates of -s plural errors, unless the child is not yet able to distinguish between
full vowels and schwa. I rather predict a high rate of zero responses due to the flexible
153
BETTINA SPRENG
conditioning of plural type -s. On the other hand, I would expect a high number of -s
overgeneralizations for nouns of plural type -e which is applied to monosyllabic nouns of
all genders with the final segment being a consonant. These conditions are a subset of the
conditions of plural type -s.
Within the system I am proposing, I also expect that error rates will decrease
depending on plural type when the child knows the gender of a particular noun. As
elaborated above, certain nouns can only be identified as belonging to a certain plural
type when they have a certain gender. Knowledge of this ensures correct responses.
I also expect that error rates will decrease when children have a concept of gender
as a grammatical category but may not consider the particular gender of a noun as a
condition for a plural type yet. As previous studies have shown, children make almost no
gender errors by the age of three years (Mills 1995, Koehn 1994, Müller 1994). If
children have already identified the gender of a noun, I can certainly assume that they are
able to use properties of a noun that helps them to identify to which plural type this noun
belongs. The properties of nouns for identifying gender are relatively complex (Köpcke
1982). The properties to identify the target plural suffix are rather simple in comparison.
However, children may not yet have identified gender as part of the conditions that
determine a plural suffix. The results of the conducted study confirm that children make
no errors in gender assignment but they may avoid indicating the particular gender of a
noun (Spreng 2003). Depending on whether children use gender indicators such as
determiners in general, and depending on whether they indicate the particular gender of a
noun by those means, error rates should decrease.
For German plural morphology, a plausible acquisition model takes advantage of
the descriptive facts of the plural system by assuming that these facts will gradually
become available and accessible to the child. It is rule-based when there is a clear rule for
a certain noun without complete overlapping information from the input. In other words,
if morphosyntactic and/or phonotactic properties of the noun can determine the plural
suffix, this plural type is acquired as a rule. When the plural type cannot be evaluated,
due to completely overlapping conditions, the plural form is stored in memory. A noun
can be reevaluated when more conditioning factors become available compared to a new
noun and thus eventually be referred to a plural type. I therefore expect that
overgeneralizations occur only when the conditions for the plural type are less clear. For
instance, I expect more overgeneralizations for nouns that have the same phonotactic
shape but different plural types. However, I would expect only overgeneralizations with
the plural types whose conditions these nouns share.
Storage of full plural forms, e.g. irregular plurals in this model are therefore less
numerous than in dual route models as proposed in Clahsen et al. (1992) or Marcus et al.
(1995). The distinction between rule-based plurals and memorized plurals is not only
dependent on plural type but also on the gender of the noun. Whether a noun is stored in
memory depends on whether it can be disambiguated for a plural type during acquisition.
In other words, the conditions that determine a certain plural form have to be nonambiguous. If they are not, the plural forms are stored. If phonotactic shape is not
sufficient, gender is used as a secondary condition for further classification. If this is still
not sufficient, the plural form has to be stored.
154
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
Thus, adults’ plurals are not necessarily all rule based for each suffix. The same
evaluation process might be applied by adults when encountering a new noun. I therefore
predict error patterns for adults similar to those in child language.
The model I am proposing reduces the burden on memory considerably but it also
accounts for the error patterns in acquisition. Overgeneralization is restricted to plural
types that are possible with respect to the phonotactic and morphosyntactic properties of
the noun stem. They only occur when these properties and their relevance as determining
factors for a plural type are acquired and can be evaluated by the child. Ignorance of
these properties as determining factors results in avoidance of plural suffixation. The
evaluation process requires that the noun has been encountered more than once.
Overgeneralizations are therefore simply errors due to conflicting conditions for the
application of a rule or the non-availability for part of the conditions.
3
The study design
According to the system proposed in the previous section, repeated evaluation of
the features of a noun is a part of the acquisition process that explains the error patterns in
children as well as the error patterns in adults for nonce words. It is therefore most
important to have tokens of a similar degree of familiarity for the participants; the test
tokens came from the 200 most frequently used nouns of a child of comparable age
(Behrens 2001).6
Attention was paid to the fact that all selected nouns were simple, non-derived
count nouns. This also ensures a wide gap between familiar nouns (real nouns) and
unfamiliar nouns (nonce nouns). One other token Mast had to be used that was not in the
corpus of Behrens (2001) in order to include all plural types.
Nonce nouns were used as rhyming nonce words with initial segment
combinations that are possible in German. The nonce words were ascribed the same
gender as their real counterparts.
Table 4.
Tested Tokens: Rhyming nonce words
Plural
type
masculine
feminine
neuter
6
C-n
Fauer
Rantoffel
Suschel
Tangel
-e
-en
Sisch
Tus
Faus
Tuh
Kiff
Laf
Sast
Lär
Mür
Suhr
Nohr
Lett
-er
-∅
Rann
Peist
Säfer
Ruchen
Kaus
Luch
Renster
Tissen
-s
Lapa
Luhu
Tok
Loma
Saxi
Rauto
V-n
Mase
Töwe
Flücke
Lente
I restrict the discussion to nouns in nominative case following general practice (Clahsen et al. 1992).
155
BETTINA SPRENG
3.1
Participants
The participants were 65 monolingual children from a Kindergarten in a small
town in Southern Germany. There were 28 female and 37 male children between the age
of 3;3.7 and 7;3.27. The children all speak a Swabian variant of German, mixed with
Bavarian elements and none can read.7
3.2
Elicitation procedure
The children were tested in two sessions due to the large number of tokens. The
sessions were timed approximately 3-5 weeks apart. The children were in a separate room
with the researcher. They were shown pictures on a laptop computer, and they could
press a button to see the subsequent picture. Every session was tape-recorded and the
children’s responses were jotted down to prevent data loss due to occasional inferior
sound quality. The children were shown pictures alternatively showing a familiar item
and an unfamiliar item for which the researcher provided the name.8 After each picture of
a single item, the children saw a second picture with more than one of the same item. The
following dialogue between researcher and child accompanied each picture pair.
(6)
Elicitation dialogue
Stimulus:
Was ist da auf dem Bild?
What is there in the picture?
Answer:
ein Mann
a man
Subsequent conversation to elicit gender indicator: Was passiert auf dem Bild?
What happens in the picture?
Answer:
der/die/das sitzt auf dem Baum.
m/f/n (def. article) is sitting on a tree9
7
In the German school system, children learn to read in school, starting with grade one. Reading and
writing is not taught in Kindergarten.
8
In contrast to Berko’s (1958) WUG test, the pictures were real pictures but they were of objects that the
children were unlikely to know.
9
Definite articles are commonly used as pronouns in German.
156
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
Alternative with nonce nouns:
Stimulus:
Answer:
Stimulus:
Answer:
Das nennt man TUS
this is called TUS
Wie war das nochmal?
What was that again?
ein/eine TUS
Und was passiert da auf dem Bild?
And what happens there in the picture?
Der/die/das TUS fliegt.
The (m/f/n) is flying
Five children (2 male, 3 female) were selected randomly from the test group to
test whether this procedure would pose any problems. They were tested with the above
procedure for the first session, i.e. 34 tokens (68 pictures) (17 real, 17 nonce words).
3.2.1 Adult control group
It has previously been ignored in the literature that dialect speakers use both the
standard and dialectal variants. Since dialects are not written and Standard German is
mainly a written ideal, adult speakers prefer Standard German in writing as opposed to
the dialectal variant in casual conversation. In an elicitation setting with written
questionnaires, an adult control group would probably use the standard plural form as
opposed to their dialectal variant they use in spoken language. Clahsen (1999) cites a
pencil and paper experiment by Marcus et al. (1995) on adult language processing that
disregarded this problem. There is no mention of the problem that some adults might be
influenced by the fact that they had to write the forms. To avoid this problem, I tested the
adult control group only orally.
The adult control group consisted of 12 mothers who agreed to take the same test
as the children. In this area, mothers tend to stay at home and are therefore usually the
primary caregivers.
The adults were visited in their homes during a single week and were tested with
the same 68 tokens (34 real and 34 nonce words) as their children but without the
pictures. I assumed that adults do not need a concrete object to manipulate an unknown
word. They are familiar with the concept of a new word without knowing or seeing what
it describes, whereas children tend to learn new words for concrete objects they see.
The mothers were given a singular noun, alternating real and nonce words in the
same order as for the children, and were asked to give the plural form spontaneously and
as quickly as possible. I told them I would give them familiar words and words that do
not exist in German and then I would ask them what they would call it when there were
more than one.
Like the children, they were given the gender of a nonce noun by providing the
singular nonce word with the definite article. They were not shown the words and had to
provide their answers orally, and I wrote down their responses.
157
BETTINA SPRENG
4
Response profiles
4.1
Children
With respect to the types of errors children display, first I predict that plural types
that have overlapping conditions with more plural types are more prone to error. Second,
I predict that the error types are mostly errors that are possible plural types for the
particular token. Possible errors are plural types that share conditions, or, in other words,
are connected with arrows in the diagram. Detailed profiles of the children, especially of
their error types with respect to plural types and individual tokens, can confirm these
predictions.
The following sections describe the response profiles with respect to plural types,
individual tokens and error types. The tables illustrate what kind of responses the children
exhibit for each plural type and with each token. They demonstrate that if two plural
types share the same conditions, they are more likely to be substituted for each other.
Secondarily, the errors are expected ones in the sense that they share conditioning
environments or parts thereof. Errors are rare when it comes to impossible error types;
suffixes that would be determined by different properties of the noun, either phonotactic
or morphosyntactic ones. The errors are always possible substitution errors except for a
few non-existing forms. Note that the following tables disregard the minor effects of age
and session because gender is known by the children at the tested age and the session
effect is probably a secondary age effect as well (cf. Spreng 2003).
4.1.1 Type C-n
Predictions:
C-n versus -e: no substitution
C-n versus -en: no substitution.
C-n versus -er: no substitution
C-n versus -Ø: substitution occurs predominantly for masculine nouns with the final
segment [r].
C-n versus -s: substitution for bisyllabic nouns with a final consonant.
C-n versus V-n: no substitution
For this type, all the above predictions are borne out. We have a very large
number of false zero responses for masculine nonce nouns, higher than for feminine or
neuter nonce nouns as predicted. The only other error type is -s, with overall similar rates.
The real feminine nouns have, as predicted, a high rate of correct responses compared to
the masculine nouns. The rate of false zero affix responses for masculine nouns is also
higher than for feminine nouns.
158
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
Table 5.
Error type profile for plural type C-n10
plural
type
token
C-n real
C-n nonce
C-n real
C-n nonce
C-n real
C-n nonce
C-n real
C-n nonce
Bauer (m)
Fauer (m)
Pantoffel (m)
Rantoffel (m)
Angel (f)
Tangel (f)
Muschel (f)
Suschel (f)
correct
responses
53.84%
0%
33.33%
6.15%
75%
30%
93.85%
29.23%
false zero
affix
responses
35.38%
90 %
56.66%
92.31%
20%
60%
3.08&
53.84%
expected
error
responses
-s
10.77%
-s
6.66%
-s
6.66%
-s
1.54%
-s
5%
-s
6.66%
-s
3.08%
-s
10.77%
unexpected
no
response
3.33%
3.33%
3.33%
6.15%
Also according to my predictions, the only error type other than –Ø was -s. Since
all tokens have a final consonant, the error rates are low. For all tokens of this type, the
type -s was used by the children as an alternative error type instead of the zero affix
response, although very rarely. Interestingly, the error types -e and -en were not used
although the final segment of the nouns would have permitted them, thus indicating that
children can analyze the syllable structure including the structure of the final syllable
when applying the plural suffix.
4.1.2 Type -e
Predictions:
-e versus C-n: no substitution
-e versus -en: substitution
-e versus -er : no substitution except for masculine and neuter nouns
-e versus -s: substitution for nouns with a final full vowel
-e versus V-n: no substitution
For this type, I predict a high error rate with -en and -er suffixes since the
conditions are identical, except with feminine nouns. All the above predictions are borne
out. The error responses expected are -en and -s, apart from false zero affix responses.
Interestingly, apart from -s type errors, this type shows the most non-existing suffixes,
especially with the nonce word Tuh, which according to its final full vowel would be a
perfect candidate for plural type -s. The suffix attached, was -se, indicating that there was
a conflict between application of -s and the knowledge that -e does not apply to nouns
ending in a full vowel. The occurrences with other double markers such as -e -s indicate a
similar conflict. However, as predicted, the only other error type is -er and -en although
-er actually occurred with a feminine noun, indicating that gender was not used to
disambiguate the plural type for this nonce noun by five children. These were three
children from age group four, one from age group three and one from age group one,
indicating that gender is not necessarily a category that is not yet acquired at this age.
10
Note that these numbers are out of 65 responses when the tokens are from session 1 (equaling 65
children) and out of 60 when they are from session 2 (5 children did not participate in session 2).
159
BETTINA SPRENG
This result rather indicates that the gender of the noun has not been recognized as
conditioning factor for plural type.
Table 6.
Error type profile for plural type -e
plural
type
token
correct
responses
false zero affix
responses
-e real
-e nonce
-e real
-e nonce
-e real
-e nonce
-e real
-e nonce
Schiff
Kiff
Schaf
Laf
Bus
Tus
Fisch
Sisch
70.77%
6.15%
90%
1.66%
64.62%
10%
87.69%
7.69%
26.15%
81.54%
10%
90%
32.31%
76.67%
10.77%
84.62%
-e real
Kuh
61.66%
23.33%
-e nonce
Tuh
0%
69.23%
-e real
Maus
88.33%
8.33%
-e nonce
Faus
1.54%
98.46%
expected
error
responses
-s
1.54%
-s
7.69%
-e-s
-en
-e-s
-s
-e-s
-en
-s
-er
-en
-s
-s-e
-en
-er
1.54%
8.33%
1.54%
4.62%
1.54%
1.66%
13.33%
9.23%
1.54%
16.92%
1.54%
1.66%
1.66%
unexpected
no response
1.54%
4.62%
8.33%
1.54%
4.62%
1.54%
1.54%
One reason why there were more incorrect responses with a fairly well-known
word like Kuh ‘cow’ is that the phonotactic shape of the noun provides the perfect
condition for the -s plural (final full vowel). As I predicted, this token has a high number
of -s errors and its nonce counterpart does as well. It is interesting that the nonsense token
Faus had only one correct response, perhaps due to the rhyming identical set of
real/nonce words (real: Haus ‘house’ and Maus ‘mouse’, nonce: Kaus) with different
plural type and gender. I would expect a higher rate of correct responses for this token
than for Haus of plural type -er since they are distinct in that type -e allows feminine
nouns whereas -er does not. The token Maus is feminine so I expect a higher number of
correct responses than for Haus, which the results in fact show (cf. section 4.1.4).
4.1.3 Type –en
Predictions:
-en versus C-n: no substitution
-en versus -e: substitution
-en versus -er: substitution except for feminine nouns
-en versus -Ø: no substitution
-en versus -s : substitution
-en versus V-n: no substitution
160
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
For this type, I expect high substitution rates with types -e and –er with masculine
and neuter nouns since they share the same conditioning factors; a prediction that is
indeed borne out as illustrated in the table.
Table 7.
Error type profile for plural type -en
plural
type
toke
n
correct
responses
-en real
Bär
78.46%
false
zero
affix
responses
16.92%
-en nonce
Lär
4.62%
86.15%
-en real
-en nonce
Uhr
Suhr
73.33%
9.23%
20%
81.54%
-en real
Ohr
89.23%
4.62%
-en nonce
Nohr
9.23%
84.62%
-en real
Bett
29.23%
32.31%
-en nonce
Lett
0%
85%
-en real
-en nonce
Mast
Sast
0%
1.66%
68.33%
80%
-en real
Tür
81.54%
12.31%
-en nonce
Mür
26.67%
46.67%
expected
responses
-s
-e
-s
-e
-s
-s
-e
-s
-e
-s
-e
-e
-er
-e
-er
-e
-s
-e
-s
-e
-s
-e
error
3.08%
1.54%
4.62%
3.08%
6.66%
3.08%
1.54%
1.54%
4.62%
4.62%
1.54%
32.31%
6.15%
10%
5%
23.33%
3.33%
8.33%
1.54%
4.62%
8.33%
15%
unexpected
no response
1.54%
4.62%
8.33%
6.66%
3.33%
One exception to the general pattern is the real noun Mast which has more than
50% incorrect responses. This is probably because it was not taken from the 200 most
used nouns since there were not enough of this type in the list (Behrens 2001). This noun
is probably less familiar and would therefore pattern similar to the nonsense nouns. Most
of the nonsense nouns in this type have no correct response except Mür which has an
unusually high number of correct responses for a nonce word. In that regard, it obviously
seems to pattern with real nouns. It also shows a high number of -e error responses and -s
error responses as I would expect with respect to their conditioning just like its
corresponding real noun Tür ‘door’. The error types that were used are -s, -e, -er, and no
response. The high ratio of -e error types was expected since both plural types have
identical phonotactic and morphosyntactic conditioning. Another interesting real word is
Bett ‘bed’, which received an unusually high number of error responses as a real noun. It
shows the dialectal influences which allow -er as a plural type for Bett, which occurred
once in the adult control group. The error type -er was also used for the corresponding
nonsense word Lett. The usage of the error types -er, -e and -en was expected since the
conditioning factors are almost identical for the plural types -er, -e, and -en. Only plural
type -er does not allow feminine nouns. As I would thus expect, error type -er was not
used with the feminine nouns Tür ‘door’, Uhr ‘clock’, Mür and Suhr. Instead, the
phonotactically and morphosyntactically possible error types -e, and -s were used.
161
BETTINA SPRENG
4.1.4 Type –er
Predictions:
-er versus C-n: no substitution
-er versus -e: no substitution
-er versus -en: substitution except for feminine nouns
-er versus -Ø: no substitution
-er versus -s : substitution
-er versus V-n: no substitution
The conditions for this type overlap phonotactically with types -e and -en; I would
therefore expect these as error types, a prediction that again is confirmed. Apart from
plural type -s, which is also expected, all other error responses are of the type false zero
affix response, -e and -en.
Table 8.
Error type profile for plural type -er
plural
type
token
correct
responses
-er real
Haus
69.23%
false
zero
affix
responses
23.08%
-er nonce
-er real
-er nonce
Kaus
Buch
Luch
0%
90.77%
0%
95.38%
6.15%
90%
-er real
Mann
78.33%
13.33%
-er nonce
Rann
0%
81.54%
-er real
-er nonce
Geist
Peist
80%
1.54%
13.33%
93.85%
expected
error
responses
-e
6.15%
-e-s
1.54%
-e
4.62%
-er-s 3.08%
-e
3.33%
-s
6.66%
-e
6.66%
-s
1.66%
-e
4.62%
-s
3.08%
-e
6.66%
-e
3.08%
unexpected
no response
-t
9.23%
1.54%
1.54%
As mentioned for type -e, the token Maus conflicts with the -er type token Haus,
which has a high number of false zero affix responses despite the fact that it is a wellknown noun. An explanation might be that it is considered plural due to the final [s]. The
error types are -e, -s, two double plurals (Haus-e-s) and (Büch-er-s), one non-existing
suffix (Rann-t), false zero affix responses for real nouns and no responses. Note that
again the only error types are phonotactically possible ones.
162
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
4.1.5 Type –Ø
Predictions:
-Ø versus C-n: substitution for masculine nouns with the final segment [r]
-Ø versus -e: no substitution
-Ø versus -en: no substitution
-Ø versus -er: no substitution
-Ø versus -s : substitution
-Ø versus V-n: no substitution
This type’s conditions overlap with the ones for type C-n, so I expect errors of
that type. I also predict a high ratio of correct responses since the omission of the suffix
obviously serves as a general strategy, whether the plural type is known or not.
Contrary to expectation, errors of C-n type were not found. For tokens that end in
[n] this seems fairly obvious since German does not allow a sequence of two adjacent [n].
For the other tokens, this supports the idea that tokens like Bauer ‘farmer’ in plural type
C-n are in fact exceptions to a schwa epenthesis rule just like Nachbar-n ‘neighbours’
(cf. Wegener 1999) and should not be included as regular plural types. The conditioning
for plural type C-n would therefore allow only final segment [l].
We see from the table that the only error types here are no responses and -s. Since
the tokens are bisyllabic (which are also allowed in -s type plurals) and the final syllable
can be taken as mimicking a plural suffix (see also Köpcke 1998) the only other error
type to be expected is type –s.
Table 9.
Error type profile for plural type -Ø
plural
type
-Ø real
-Ø nonce
-Ø real
-Ø nonce
-Ø real
-Ø nonce
-Ø real
-Ø nonce
token
Kuchen
Ruchen
Kaefer
Saefer
Kissen
Tissen
Fenster
Renster
correct
responses
96.66%
98.46%
96.66%
93.33%
100%
98.33%
88.33%
76.67%
false zero affix
responses
expected error
responses
-s
3.33%
unexpected
no response
1.54%
-s
-s
3.33%
3.33%
3.33%
-s
-s
-s
1.66%
11.66%
18.33%
5%
4.1.6 Type -s
Predictions:
-s versus C-n: no substitution
-s versus -e: substitution
-s versus -en: substitution for monosyllabic nouns with final consonant
-s versus -er: substitution only for monosyllabic nouns with final consonant
-s versus -Ø : substitution
-s versus V-n: no substitution
163
BETTINA SPRENG
This type follows the general pattern with high numbers of correct responses for
real nouns and low numbers of correct responses for nonce words. The error types are the
phonotactically possible types -en and -e as well as non-responses, and an extremely high
number of false zero affix responses for real nouns. It seems that even with this rather
flexible plural type, the default error type is the zero affix response even for real nouns.
Contrary to the predictions as proposed in Clahsen (1999), the overall error rate is not
lower than for the other plural types. It is distinct in the high number of false zero affix
responses for real nouns. An explanation might be that due to the flexible conditioning,
the choice of plural suffixes is rather open and the children have difficulties in deciding
which suffix it should be. This means they have not learned the rule for -s plural yet.
However, considering the age of the children in this study, this seems not to be
the case when comparing it with the ages of children in studies that wanted to prove
exactly that (Clahsen et al. 1992). Note also that the also possible error type -er was not
used but the phonotactically also possible type -en was preferred.
Table 10. Error type profile for plural type -s
plural
type
token
correct
responses
-s real
-s nonce
Lok
Tok
44.62%
15%
false
zero
affix
responses
43.08%
76.66%
-s
-s
-s
-s
-s
real
nonce
real
nonce
real
Papa
Lapa
Taxi
Saxi
Uhu
64.62%
10%
43.08%
21.54%
35%
35.38%
81.66%
53.85%
67.69%
61.66%
-s
-s
-s
-s
-s
nonce
real
nonce
real
nonce
Luhu
Auto
Rauto
Oma
Loma
13.85%
90%
30%
78.33%
12.31%
80%
10%
70%
21.66%
87.69%
expected
error
responses
-en
9.23%
-e
3.33%
-en
1.66%
-en
1.66%
-e
-en
-e
1.66%
1.66%
1.54%
unexpected
no response
3.08%
3.33%
6.66%
3.08%
10.77%
4.62%
4.1.7 Type V-n
Predictions:
V-n versus C-n: no substitution
V-n versus -e: no substitution
V-n versus -en: no substitution
V-n versus -er: no substitution
V-n versus -Ø: no substitution
V-n versus -s: no substitution
For phonotactic reasons, I expect only null and -s error types according to the
model proposed in (4), the latter only if the final schwa is not distinguished from full
vowels. When the gender is known, I would not expect substitution with plural types that
164
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
do not allow feminine nouns, such as -er. I also expect that plural type -s is only used
when the condition on stem final full vowel is not known or the distinction between full
vowels and schwa is not known. The data confirm this prediction and show no great
variation between error types. The real nouns show zero affix responses and the other
error types are -s, no response and one non-existing suffix with the nonce noun Mase-r.
The occurrence of plural type -s indicates either that in these cases the distinction
between schwa and full vowels is not acquired or the condition for that distinction is not
yet acquired.
Table 11. Error type profile for plural type V-n
plural
type
token
correct
responses
V-n real
V-n nonce
V-n real
V-n nonce
V-n real
V-n nonce
V-n real
V-n nonce
Brücke
Flücke
Ente
Lente
Hase
Mase
Löwe
Töwe
86.66%
58.33%
96.92%
51.66%
96.66%
16.66%
83.08%
4.62%
false zero
affix
responses
13.33%
38.33%
3.08%
48.33%
3.33
75%
13.85%
86.15%
expected
error
responses
-s
3.33%
-s
5%
-s
7.69%
unexpected
no response
-r
1.66%
3.08%
1.54%
1.66%
Again, no impossible error types occur in the data. Gender was also not a deciding
factor for error types since the error types used are completely flexible in which gender
they allow.
4.2
Summary of error type profiles for children
The response type profiles show that the only errors are within the phonotactic
and/or morphosyntactic possibilities of a plural type. In other words, if the conditions of
plural types overlap, these plural types serve as the only options for errors. Gender serves
as an additional distinguishing factor when there is a complete overlap of phonotactic
conditions. Error response -e is possible with plural types -e and -en. However, only
masculine and neuter tokens show error responses with -er since only this plural type
disallows feminine nouns.
Since this is only possible for a distinction between types that disallow one
gender, gender in general cannot serve as a distinguishing factor. However, when it is
necessary and available, the child uses it. These profiles also show that false zero affix
responses are used for unknown words. For real words that are less known, the number of
false zero affix responses increases such as in the less known real token Mast.
The zero affix response is certainly used when the noun is not known (nonce
nouns). However, it would be hasty to conclude that this is a default plural in the
theoretical sense. It is certainly not used as a means of indicating plural when the suffix is
not known. In particular, the high number of false zero affix responses for nonsense
words indicates that this is either a means of avoiding a non-response or it is a legitimate
plural form in the child’s grammar. As we will see in section 4.5, adults use the zero
165
BETTINA SPRENG
response far less often than children do. Similar to the adult control group, the children
cannot generalize from the phonotactic shape of the known word to the nonsense word,
even when they are given the gender. This indicates that since the conditions are not
unambiguous for each plural type, more exposure is necessary to evaluate the correct
plural form, thus supporting the proposal put forward in section 2. Phonotactic shape
helps children to identify the plural type of known words but it does not enable them to
generalize this information for nonsense words instantaneously. This is true for the
phonotactically flexible plural types -en, -e, and -s as well as for more restricted types
such as C-n, -er, and
V-n.
4.3
Schwa epenthesis or schwa deletion: plural types C-n, -e, –en,
-er, and V-n
As mentioned in section 2.1, there is an ongoing debate as to whether for these
plural types there is ə-epenthesis (Wiese 1996) or ə-deletion (Wegener 1999).
The question can be answered by considering the response profiles for the plural
types C-n, V-n, -e, -en, and -er. Before the acquisition of a rule of schwa epenthesis,
tokens of type -en should have error responses of type C-n which is phonotactically
possible in German. However, there are no instances of this error response. On the other
hand, before the acquisition of a rule of ə-deletion, I expect error responses of -en for
plural type C-n. We have no instance of this error response either.
However, the double markings -e-s in plural types point to an overgeneralized
application of a rule of ə-epenthesis after [s] and [ʃ]. This rule seems to be stable at the
tested ages. Although this seems to be a clear indication for a ə-epenthesis rule in the
children’s grammar, the data basis is too small to provide solid support for such a claim.
(7)
type -er:
type -e:
*Haus-e-s
*Bus-e-s
*Fisch-e-s
*Sisch-e-s
4.4
Error type zero suffix
‘house’
‘bus’
‘fish’
nonce word
Plural type -s has the highest susceptibility to false zero responses, which
contradicts the argument that overt plural marking of nouns that have a ‘plural-like’ final
syllable is usually avoided (Köpcke 1998). The final segments of -s plural type nouns are
actually the opposite of plural-like. On the other hand, type V-n is the lowest, which
would be supporting evidence for the above argument. These results support a proposal
that defines the interaction of morphosyntactic and phonotactic conditions in more detail
instead of treating them as bundles of properties that determine the plural suffix for a
noun. The ‘plural-likeness’ of the final syllable is obviously not a sufficient criterion to
predict the overtness of a plural suffix.
166
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
4.5
Adult control group: Response profiles
The following table shows the response profiles according to error responses for
each plural type.
Table 12. Adult responses split according to real/nonce and plural type11
resp
type
C-n
real
C-n
nonce
en
real
en
nonce
e
real
e
nonce
er
real
er
nonce
Ø
real
Ø
nonce
s
real
s
nonce
V-n
real
V-n
nonce
C-n
en
e
41
1
(85.4%) (2.1%)
33
(68.8%)
69
(95.8%)
42
(58.3%)
2
(2.8%)
19
(26.4%)
72
(100%)
29
28
(40.5%) (39.3%)
er
Ø
s
V-n
e-s
ren
se
r
Total
6
(12.5%)
8
7
(16.7%) (14.6%)
1
(1.2%)
1
(1.4%)
2
(2.4%)
48
(100%)
12
22
6
2
(25.0%) (45.8%) (12.5%) (4.2%)
47
(97.9%)
43
(86.1%)
10
2
2
(13.9%) (2.8%)
(2.8%)
26
3
1
3
(36.1%) (4.2%) (1.4%) (4.2%)
1
5
(2.1%) (10.4%)
48
(100%)
48
(100%)
72
(100%)
10
72
(13.9%)
(100%)
72
(100%)
9
2
1
1
72
(13.1%)
(2.4%) (1.2%) (1.2%)
(100%)
48
(100%)
4
2
48
(8.3)
(4.2%)
(100%
1
48
(2.1%)
(100%
5
48
(13.9%)
(100%
58
72
(80.6%)
(100%)
39
72
(54.2%)
(100%)
48
48
(100%)
(100%
2
1
39
48
4.2% (100%
(2.1%) (81.3%)
As predicted, adults also have problems generalizing plural types to nonce words
especially for plural types where the conditions overlap. The types -en and -e have a wide
variety of error responses, confirming that conflicting conditions make it difficult to
identify the expected plural type. Plural type -er also has a low success rate for nonce
words, even lower than the error responses for this type, -en and -e. These results were
expected as shown in the diagram in (4).
The number of non-existing suffixes and double marking that adults use to form
plurals is rather astonishing. Two instances each of -e-s in plural types -en and -er, one
instance each of -ren and of –se in plural type -e and two instances of -r in plural type
V-n were observed. These are all phonotactically possible suffixes in German. With the
exception of –ren they are all suffixes that exist in German. One explanation for -ren
11
There are no missing cases.
167
BETTINA SPRENG
might be a generalization of syllable structure, an analogy effect. The token was Tuh [tu:].
The plural form for Tür ‘door’ would be Tür-en with a shift of the syllable boundary to
the left after the stem vowel. A plural form Tu-ren seems a possible solution, confusing
syllable boundary with morpheme boundary. Application of plural to unknown words
seems to function by comparing phonotactic shapes of singular nouns.
The significant difference between real and nonce plurals for each plural type
indicates different means to form the plural for frequent and familiar nouns as opposed to
unknown nouns. However, since the individual differences are significant for real nouns,
it seems that there is a difference for each plural type in how it is accessed, pointing
strongly to a rote access strategy for frequent, well-known and often used nouns. Since
the participants of the adult control group are also the primary caretakers of the tested
children, a possible explanation is that these words not only have a high frequency in the
child’s vocabulary but also in the mothers’ and are therefore accessed as whole forms.
4.6
Interpretation
I expect significant differences in substitution error values between plural type
that have no connection or only restricted connections in the diagram. Within plural
types, significant differences may be due to lack of specific gender agreement of one of
the genders. The diagram in (4) repeated here as (8) gives us the expected substitution
patterns.
(8)
Substitution Patterns
-e
V-n
-en
masc./neuter
masc./neuter
-er
final C
C-n
masc. final [r]
bisyllabic, final C
monosyllabic
masc./neuter
-Ø
bisyllabic
masculine/neuter
-s
For plural types with only two genders, there is, as expected, no substitution
between masculine nouns of type C-n with masculine nouns of type -Ø but, as predicted,
no significant difference between these two types with neuter nouns. Feminine nouns of
type C-n are not substituted for neuter and masculine nouns of type -Ø. We have no
substitution with type V-n nouns and no substitution with type -er nouns. For type -er, we
168
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
have no substitution with type V-n nouns. Where gender plays the most significant role is
in the significant difference between C-n feminine and -er neuter as well as C-n feminine
and -er masculine, both for real and nonce nouns. Type -Ø nouns also are not substituted
with type V-n nouns. The significant differences between -er neuter and masculine V-n
feminine also show that feminine gender plays an important role in distinguishing these
types, since type -er does not allow feminine gender. The gender effect thus also plays a
role for types that are not related phonotactically. Nonce nouns in this category of plural
types with two genders show the same effects except that there are no significant
differences to -Ø type nouns for plural type C-n. This is probably due to the fact that the
high number of zero affix responses in nonce nouns overall. There is an additional effect
between C-n feminine and V-n feminine, although both types allow for feminine gender,
thus indicating that phonotactically, the difference between nouns of that plural type is
sufficient.
Type -er effects are the same as for real nouns except for an additional effect
between -er masculine to -Ø masculine, distinguishing these types from each other
phonotactically.
The effects for type -Ø are different from real words, distinguishing type -Ø
masculine from V-n masculine and V-n feminine and type -Ø neuter from V-n feminine.
Recall that type -Ø has no feminine gender and type V-n has no neuter gender. The fact
that even -Ø masculine and V-n masculine are distinct shows that the factor gender plays
a significant role even when the nouns are not well known. These effects are probably
due to the fact that nonce nouns were presented with their gender. However, type -Ø is
problematic because of the high number of counted correct responses for nonce nouns
and it is phonotactically completely different from type V-n.
For plural types that allow three genders, the effects for real nouns are less than
expected. Although plural type -e could be substituted with types -en, -s, (unrestricted),
and -er (restricted), there are significant differences to all genders in type -s but only
when the genders are different. There are significant differences to neuter and feminine of
type -en, again only when the gender of type -e is different from the gender of type -en.
This points to a strong gender effect even when the particular gender should not make a
difference. For non-specific gender agreement, this may point to the importance of the
presence of the abstract category gender.
For nonce nouns, we have fewer significant effects, showing that the gender effect
is less important for these nouns. Other than for the nouns with two genders, gender plays
a lesser role for nonce nouns because it could not yet be learned.
To sum up, the substitution patterns follow the predictions. This especially true
for real nouns, not only the particular gender as such but also the category gender in
combination with the phonotactics of a noun play a role in distinguishing the plural types.
5
Conclusion
The error patterns of the study presented here could shed some light on when and
how gender acts as a disambiguating factor for identifying plural types for children, thus
introducing a model of the German plural system that illuminates the interaction between
these different kinds of conditions.
169
BETTINA SPRENG
Gender functions as a secondary condition for plural types with overlapping
conditions in a predictable and systematic manner. Gender even plays a role for plural
types with all three genders, indicating that gender as an abstract category aids the child’s
acquisition of the plural.
This study confirms the findings of previous studies (Behrens 2001, Köpcke
1998) that plural type -s does not behave differently from the other plural types once the
conditioning factors and their interactions for each plural type are clearly defined. The
fact that the plural type -s is used as error type for almost every plural type is because its
conditioning factors overlap partly with most of the plural types. Errors occur only with
plural types that share conditioning factors. Phonotactic shape of the singular noun is
hereby the main distinguishing factor for children. However, gender is used as a
secondary disambiguating factor whenever possible. Since adults treat real and nonce
words equally with respect to plural type, I conclude that they apply more or less
productive rules for each plural type. Frequency effects can be observed with respect to
real and nonce words for adults and children. For adults, this points to a frequency effect
with respect to the real nouns. Since they are very frequent in their children’s vocabulary,
rote access might also be the strategy for adults.
The error type profiles show clearly that only errors occur that are possible with
respect to the overlapping of conditioning factors between plural types. No instance was
found where children would use a plural type for another that does not share at least one
conditioning factor. Phonotactic shape was the determining factor but recourse to gender
was apparent whenever possible. In light of this, the high number of null plural responses
can be explained as avoidance strategy rather than a default plural rule, especially for
unknown words. Since the performance for nonce words was most affected by this, I
suggest that overgeneralization cannot be used as a sole indicator for the presence of a
linguistic rule. Since it was found (Marcus 1995) that overgeneralizations are in general
more infrequent than previously assumed, the easy conclusion that overgeneralizations
equal linguistic rules must be abandoned.
Since rules are productive only relative to each other, the difference in
performance between real and nonce words both for adults and children can be explained
by training. Accessibility of information in new words is probably limited, both for adults
and children. However, the performance with respect to plural types is equal, which
means, there are ‘easier’ plural types and ‘harder’ ones both for children and adults.
Easier plural types are plural types with clear conditions, nouns of high frequency and
only phonotactic shape to deal with. The more complex the conditions are, the harder it
becomes to identify the plural type. Whenever there is no distinguishing factor, the whole
plural forms is stored in memory as the results for adults demonstrates for the
overlapping plural types -en and -e.
References
Albright, A. (1999). The default is not a unitary rule. paper presented at the 73rd LSA meeting 1999 in Los
Angeles.
Altmann, G. & Raettig, V. (1973). Genus und Wortauslaut im Deutschen. Zeitschrift fuer Phonetik,
Sprachwissenschaft und Kommunikationsforschung, 297-303.
Brown, R. (1973). A First Language: The Early Stages. Cambridge, MA: Harvard University Press
170
ACQUISITION OF GERMAN PLURAL MORPHOLOGY
Bartke, S. (1998). Experimentelle Studien zur Flexion und Wortbildung. Pluralbildung und lexikalische
Komposition im unauffälligen Spracherwerb und im Dysgrammatismus. Tübingen: Niemeyer.
Behrens, H. (2001). How to learn a minority default: the acquisition of the German -s plural, ms.
Berko, J. (1958).The child's learning of English morphology. Word 14, 150-177.
Bittner, D. (2000). Gender classification and the inflectional system of German nouns. Barbara Unterbeck
(part I), Matti Rissanen, Terttu Nevalainen & Mirja Saari (part II) (Eds.) Gender in Grammar and
Cognition. Berlin, New York: Mouton de Gruyter, 1-23
Carstairs, A. (1994). Inflection Classes, gender, and the principle of contrast. Language 70, 4, 737-788
Clahsen, H., Eisenbeiss, S. & Sonnenstuhl-Henning, I. (1997). Morphological Structure and the Processing
of Inflected Words. Theoretical Linguistics 23, 3, 201-249.
Clahsen, H. et al. (1992). Regular and Irregular Inflection in the Acquisition of German Noun Plurals.
Cognition 45, 225-255.
Clahsen, H., Marcus, G. & Bartke, S. (1996). Compounding and Inflection in German Child Language.
Yearbook of Morphology 1995, 1-34.
Clahsen, H. (1992). Overregularization in the acquisition of inflectional morphology: A comparison of
English and German. Gary F. Marcus, Steven Pinker, Michael Ullman, Michelle Hollander, T.
John Rosen, Fei Xu: Overregularization in Language Acqusition. Monographs of the Society for
Research in Child Development Serial no. 228, vol. 57, no. 4, 166-178
Clahsen, H. (1999). Lexical entries and rules of language: A multidisciplinary study of German inflection.
Behavioral and Brain Sciences 22, 991-1060.
Dresher, B. E. (1999). Charting the Learning Path: Cues to Parameter Setting. LI 30:1, 27-67.
Ewers, H. (1999). Schemata im mentalen Lexikon: Empirische Untersuchungen zum Erwerb der deutschen
Pluralbildung. J. Meibauer & M. Rothweiler (Eds.): Das Lexikon im Spracherwerb. Tübingen:
Francke, 106-127.
Ferenz, K. S. & Prasda, S. (2002). Singular or Plural? Children's knowledge of the factors that determine
the appropriate form of count nouns. Journal of Child Language 29, 49-70.
Gawlitzek-Maiwald, I. (1994). How do children cope with variation in the input? The case of German
Plurals and Compounding. Rosemarie Tracy & Elsa Lattey (Eds.) How Tolerant is Universal
Grammar? Tübingen: Niemeyer, 225-265.
Goebel, R. & Indefrey, P. (2000). A recurrent network with short-term memory capacity learning the
German -s plural. Peter Broeder & Jaap Murre (Eds.): Models of Language Acquisition: Inductive
and deductive approaches. Oxford: Oxford University Press, 177-200.
Halle, M. & Marantz, A. (1993). Distributed Morphology and the Pieces of Inflection. Hale, Ken & Keyser,
Samuel Jay (Eds): The View from Building 20: Essays in Linguistics in Honor of Sylvain
Bromberger. Cambridge, Mass: MIT Press, 111-176.
Indefrey, P. (1999). Some problems with the lexical status of nondefault inflection. Behavioral and Brain
Sciences 22, 1025: Commentary to Clahsen (1999).
Klampfer, S., Koretzky-Kroell, K., Dressler, W. (2001). Morphological potentiality in children's
overgeneralization patterns: evidence from Austrian German noun plurals. Wiener Linguistische
Gazette 67-69, 25-43.
Koehn, C. (1994). The Acquisition of gender and number morphology within NP. J. Meisel (Ed.): Bilingual
First Language Acquisition: French and German grammatical development. Amsterdam: John
Benjamins, 29-51.
Köpcke K.-M. (1982). Untersuchungungen zum Genussystem der deutschen Gegenwartssprache.
Tübingen: Niemeyer.
Köpcke, K.-M. (1993). Schemata bei der Pluralbildung im Deutschen. Versuch einer kognitiven
Morphologie. Tübingen: Narr.
Köpcke, K.-M. (1998). The acquisition of plural marking in English and German revisited: schemata versus
rules. Journal for Child Language 25, 293-319.
Köpcke, K.-M. & Zubin, D. A.. (1983). Die kognitive Organisation der Genuszuweisung zu den einsilbigen
Nomen der deutschen Gegenwartssprache. Zeitschrift für germanistische Linguistik 11, 166-182.
Lust, B. et al. (Eds.). Syntactic Theory and First Language Acquisition: Cross-Linguistic Perspectives, vol.
1: Heads, Projections and Learnability, Hillsdale, N.J.: L. Erlbaum Associates.
171
BETTINA SPRENG
MacWhinney, B., Leinbach, J., Taraben R. & McDonald, J. (1989). Language Learning: Cues or Rules?
Journal of Memory and Language 28, 255-277.
Marcus, G. F. (2000). Children's overregularization and its implications for cognition. Peter Broeder & Jaap
Murre (Eds.) Models of Language Acquisition: Inductive and deductive approaches. New York:
Oxford University Press, 154-176.
Marcus, G. F. et al. (1995). German Inflection: The Exception that Proves the Rule. Cognitive Psychology
29, 189-256.
Mills, A. E. (1986). Acquisition of the natural-gender rule in English and German. Linguistics 24, 31-45.
Mills, A. E. (1995). The Acquisition of German. Paul Fletcher & Brian MacWhinney (Eds.). The
Handbook of Child Language. Oxford, UK; Cambridge, Mass., USA: Blackwell, 141-254.
Mugdan, J. (1977). Flexionsmorphologie und Psycholinguistik. Tübingen: TBL Verlag Gunther Narr.
Müller, N. (1994). Gender and Number Agreement within DP. J. Meisel (Ed.) Bilingual First Language
Acquisition: French and German grammatical development. Amsterdam: John Benjamins, 53-88.
Nakisa, R., Plunkett, K. & Hahn, U. (2000). Single and Dual-Route Models of Inflectional Morphology.
Peter Broeder & Jaap Murre (Eds.) Models of Language Acquisition: Inductive and deductive
approaches. New York: Oxford University Press, 201-222.
Neef, M. (1998). A case study in declarative morphology: German case inflection. Wolfgang Kehrein &
Richard Wiese (Eds.) Phonology and Morphology of the German languages. Tübingen: Niemeyer,
219-240.
Pinker, S. & Prince, A. (1994). Regular and Irregular Morphology and the Psychological Status of Rules of
Grammar. Susan D. Lima, Roberta L. Corrigan & Gregory K. Iverson (Eds.) The Reality of
Linguistic Rules. Amsterdam; Philadelphia: J. Benjamins, 321-351.
Plunkett, K. (1995). Connectionist Approaches to Language Acquisition. Paul Fletcher & Brian
MacWhinney (Eds.). The Handbook of Child Language. Oxford, Cambridge, Mass.: Blackwell,
36-72.
Rumelhart, David E. & McClelland, James (1996). On Learning the Past Tenses of English Verbs.
Geirsson, Heimir, & Losonsky, Michael (Eds.) Readings in Language and Mind. Cambridge, MA:
Blackwell Publishers, 507-526.
Spreng, B. (2002). Plural in German. Proceedings of WECOL 2000, vol. 12.
Spreng, B. (2003). The Role of Gender in the Acquisition of the German Plural Morphology. Ms,
University of Toronto.
Tracy, R. (1986). The acquisition of case morphology in German. Linguistics 24, 47-78.
Veit, S. (1986). Das Verständnis von Plural- und Komparativformen bei (entwicklungs) dysgrammatischen
Kindern im Vorschulalter. G. Kegan et al. (Eds.) Sprachwissenschaft und Psycholinguistik:
Beiträge aus Forschung und Praxis. Opladen: Westdeutscher Verlag, 217-286.
Wegener. H. (1999). Die Pluralbildung im Deutschen: Ein Versuch im Rahmen der Optimatitätstheory.
Linguistik online 4,3.
Wiese, R. (1996). The Phonology of German. Oxford: Clarendon Press.
Wurzel, W. U. (1984). Flexionsmorphologie und Natürlichkeit: Ein Beitrag zur morphologischen
Theoriebildung. Berlin: Akademie Verlag.
172