The cum-sine pattern in German child language: An argument for antonym decomposition

ABSTRACT

German-speaking children between ages 2 and 3 mostly use the preposition ohne (‘without’) in an adult-like way, to express the absence of something. In this article we present surprising results from a corpus study suggesting that in this age group, absence can also be expressed using the sequence mit ohne ‘with without’. We argue that this pattern becomes much less surprising if we assume that there is a conceptual representation independent of linguistic structure and that antonymic concepts such as that of absence (which we call sine in this article) are not represented as primitive concepts at this level—contrary to what might be suspected on the basis of monomorphemic prepositions such as German ohne. Instead, we argue that antonymic concepts like that of absence (sine) are composed of at least two units. Children’s non-adult-like patterns then result from difficulties in acquiring the morphological realization of this complex concept as the monomorphemic preposition ohne. To our knowledge, our results constitute the first evidence for antonym decomposition in the case of prepositions from child language.

1 Introduction

The study of antonyms has been a topic of great interest in the linguistic literature since at least the work of Bierwisch (1967). Bierwisch pointed out that in antonymic adjective pairs such as tall and short, the antonyms exhibit several asymmetries. For example, the question How tall is John? is perceived to be more neutral than How short is John?, which suggests that John is short. Similarly, She is taller than Mary is more neutral as to the individuals’ heights compared to She is shorter than Mary (Rett, 2014; Ruytenbeek et al., 2017; Gotzner et al., 2018; Moracchini, 2019).

The central question in the study of antonym pairs is whether their relation to each other is transparent in their linguistic representation. A number of influential accounts propose that one member is always structurally derived from the other (Heim, 2006, 2008; Büring, 2007; Bobaljik, 2011). Such decomposition is transparent in languages such as Hixkaryana, which exhibits antonym pairs such as kawo – kawo-hra ‘long – short’ and tiyoke – iyo-hra ‘sharp – blunt’ (Derbyshire, 1985). Heim (2006), Büring (2007), Bobaljik (2011), and Moracchini (2019) argued that this schema should be generalized across languages and proposed an abstract negative morpheme anti corresponding to Hixkaryana -hra. In any given antonym pair, one member will then be derived from the other through application of anti.¹ To illustrate, short would be analyzed as anti-long and blunt as anti-sharp. This account entails that English antonymic adjectives should be analyzed as portmonteau morphemes decomposed into anti and the corresponding positive adjective.

In this article we look beyond the adjectival domain and instead focus on the German prepositions mit ‘with’ and its antonym ohne ‘without’. We present a decompositional analysis within the Meaning First framework of Sauerland and Alexiadou (2020) and present novel acquisition data in support of our account.

1.1 Background: Meaning First approach

Within the Meaning First approach, conceptual representations are primary, whereas linguistic representations are derived from conceptual representations by a process called compression. During compression, concepts are mapped to morphemes through lexical realization and linearization. Crucially, not all parts of a conceptual representation (in the following abbreviated as CR) have to be mapped to lexical material; if a concept can be reconstructed from context,² it may remain unrealized in the linguistic representation. Applied to antonyms in general, at the CR level an antonym pair would then be represented as A – [anti A] (or equivalently [A anti] because the CR is not linearly ordered).

Concepts can be either complex or primitive, and the primitive concepts are further divided into innate core concepts and experience-based concepts.³ We follow Heim (2006) in assuming that anti is a type-shifted version of negation. This makes anti a likely candidate for being a primitive concept which is either innate, or acquired early on, with the first antonymic relationship (Austin et al., 2014; Feiman et al., 2017).

On the other hand, the concept A would in most cases not be an innate concept, and in many cases not even a primitive.

The concept A, therefore, will generally be present in a child’s mind only after the anti concept. In some cases, both anti and A may be primitives—specifically, this is plausible for the antonym pair with – without (see also footnote 5 below). We also follow Sauerland and Alexiadou (2020) in assuming that the ability to compose concepts into complex concepts is innate. Therefore, children will be able to form the complex concept anti-A (or A-anti) as soon as they have learned the concept A. On the linguistic side, however, children will first have to acquire how adults compress these concepts into linguistic representations in the relevant language.

1.2 Predictions about the acquisition of ohne ‘without’

This decompositional Meaning First account of antonyms makes testable predictions about the acquisition of antonym pairs. Namely, we expect there to be a stage during which children realize both concepts anti and A transparently, as represented at the CR level, even in languages in which this complex concept is compressed in adult language (see also Guasti et al. (2023)).

Recall our prepositional pair from German, mit ‘with’ and its antonym ohne ‘without’. We assume that the morpheme mit realizes a concept we represent as cum.⁴ The morpheme ohne in the adult language expresses the complex concept [anti cum] or [cum anti] (recall that we assume CRs are not linearly ordered). Thus, German does not transparently express the two concepts cum and anti as two separate morphemes. In English, on the other hand, cum and anti are transparently expressed by with-out. Our assumptions for English and German adult language are summarized below:

Table

(1) German: ohne    einen Hut

English:  with- out a    hat

CR: [[cum anti] [a hat]]

‘without a hat’

The difference between German and English has a functional justification, as it minimizes the number of morphemes while accomplishing a similar granularity of disambiguation in the two languages. In English, out expresses not only anti in the context of with but is also the antonym of in.⁵ As a consequence, there are minimal pairs like (2) where out and without in English differentiate two different meanings relying on the bimorphemic expression. In German, however, the antonym of in is aus, and therefore the two monomorphemic expressions aus and ohne suffice to distinguish the two interpretations in (2).

Table

(2)     English               German

a. She is going out.          Sie geht aus.

b. She is going without.         Sie geht ohne.

In German, the transparent (and redundant) way to realize the complex concept [cum anti]/[anti cum] would be mit ohne, which is marked. It can be used only in a special ‘baby talk’ register in imitation of child speech for rhetoric effect, as in the translated book title Schutzengel mit ohne Flügel (‘Guardian angel with without wings’, by Arto Paasilinna). Another humorous use of mit ohne is possible in response to alternative questions that contrast with and without as illustrated in (3):

Table

(3) Möchtest Du den Kaffee mit  oder ohne   Zucker? – Mit   ohne  Zucker, bitte.

want   you the coffee with or  without sugar?  – with without sugar   please

Do you want your coffee with or without sugar? – (With) Without sugar, please.

How do children acquire ohne ‘without’, the antonym of mit ‘with’? As far as we are aware, the Meaning First framework is the only approach which makes specific predictions for this case. To see what these are, note first that within a Meaning First account, the question amounts to how children acquire the compression of the complex concept [cum anti]/[anti cum] into the corresponding linguistic structure. Secondly, children have to acquire that the compression of [cum anti] into ohne is obligatory; before they do, they are expected to at least sometimes articulate both of the concepts [cum anti]/[anti cum] – if they have the morphological resources to do so. Guasti et al. (2023) discussed other examples from production illustrating that children produce less compressed structures that are absent or marked in the adult language. Furthermore, the Meaning First approach predicts a corresponding difference in comprehension (Guasti et al., 2023): invoking only one concept should be faster than invoking two. However, the prediction concerning comprehension is shared with some other proposals, in particular that of van Hout (1998), who proposed the following: Learning should be easier for overt and unambiguous mappings (one-to-one) than for covert and/or conflated ones (many-to-one).

In the example at hand, this leads us to expect that German-speaking children will undergo a developmental stage at which they express the complex concept [cum anti]/[anti cum] using two morphemes that they understand to express cum and anti, respectively. The specific prediction depends on the child’s morphological understanding of ohne. Following distributed morphology (Halle and Marantz, 1993), we assume that lexical insertion targets terminals of the conceptual structure; that is, primitive concepts. Then ohne can be understood either as exponing cum whenever it appears in the context of anti at CR. Alternatively, ohne could expone anti whenever it appears in the context of cum at CR. Under both analyses, the remaining concept (either [anti] or [cum]) would not be overtly exponed.

But cum in other environments in German is exponed by mit and anti can be exponed by negation nicht ‘not’ and other expressions of antonymity (see below). Because German children furthermore may have no evidence for the linear order of anti and cum, at least the four non-adult patterns mit ohne ‘with without’, ohne mit ‘without with’, nicht ohne ‘not without’, and ohne nicht ‘without not’ may be produced by children.⁶

Table

(4) exponent of CUM  ANTI    CUM-ANTI   ANTI-CUM

allomorphy

yes      ohne  nicht    ohne nicht   nicht ohne

yes      mit    ohne    mit ohne   ohne mit

no       mit    nicht    mit nicht   nicht mit

Above we assumed that children use ohne as a contextual allomorph to either express anti or cum whenever they want to expone the complex concept [cum anti]/[anti cum]. The third row of (4) shows the articulations children are predicted to produce if no allomorphy is assumed. Both of the predicted patterns, however, are also possible in the adult language with constituent negation and receive the same interpretation as is predicted for child language except for the contrast requirement of constituent negation. Hence, it would be difficult to establish whether child uses of mit nicht or nicht mit are non-adult-like on the basis of corpora. We therefore focus on the patterns with allomorphy in what follows.

Because child language is frequently probabilistically converging toward the target adult grammar (Yang, 2002), we expect children to initially produce one or several of these patterns in addition to the adult pattern (just ohne). In sum, our account predicts that German-speaking children undergo a developmental stage where both parts of the [anti cum]/[cum anti] concept are articulated, potentially alongside adult-like productions. In the following, we show that corpus data from German-speaking children corroborate this prediction.

1.3 Relevance of acquisition data: Previous studies

Though the use of mit ohne by children is readily apparent to those having been in contact with German-speaking 2-year-old children such as the first author of this article, the phenomenon has not been explained or investigated in detail. According to our literature search, there has only been nonquantitative work on the antonym pair mit-ohne so far. The first report comes from Stern and Stern (1907), which is based on anecdotal observations of the authors’ own three children. Stern and Stern reported that around age 2;6, one of their children used ohne as if it were a generic negation.⁷ But for a different child of theirs, they observed that at age 2;9–3;0, she used ohne both adult-like and together with mit and provided the example (5).

Table

(5) ach, dann kann ich mit ohne Seife waschen.

ah,   then can   I    mit ohne soap wash

‘Ah, then I can wash without soap.’ (Stern and Stern, 1907:69)

Grimm (1975:117) reported on data from an observational study of over 100 children. She wrote that mit ohne is consistently used for a long time instead of ‘ohne’. She reported, “2-3 year-olds demand ein Brot mit ohne Honig ‘a bread with without honey’ or go into the street mit ohne Schuhe ‘with without shoes.’” Neither Stern and Stern nor Grimm provided any quantitative data or a theoretical account of the phenomenon.

From a theoretical viewpoint, we are only aware of Durkin (1978), who carried out a comprehension study looking for an asymmetry between English with and without, which revealed no evidence for such an asymmetry. However, the study is of limited value because only children aged 5 years and older were tested, and the eight children whose results could be analyzed all performed at ceiling.

In the next section, we report on a corpus study of the German with-without antonymic pair. Our study investigates different combinations that children use ohne, at which age, and frequencies, providing the quantitative data that previous studies that mention the acquisition of ohne lack. We return to the theoretical interpretation of the findings in the conclusion.

2 Methods

We report data from German-speaking children’s use of ohne in all relevant transcripts collected in the Childes database (MacWhinney, 2000). The transcripts are listed in Table 1. The many/Sza-corpus includes data from children with a cochlear implant, while all others only include typically developing children. We saw no reason to exclude children with cochlear implants, although we do not intend a systematic comparison with typically developing children, either. We accessed the transcripts via the LuCiD Language Researcher’s Toolkit (Chang, 2017), extracting all utterances containing the string ohne from all relevant German corpora.

Table 1. List of transcribed corpora of child spontaneous speech used

Child/corpus	Age(s)	Source
Leo	1;11–4;11	Behrens (2006), 10.21415/T5N01B
Caroline	0;10–4;3	10.21415/T5NS5S
Kerstin	1;9–4;0	Miller (1979), 10.21415/T56592
Simone	1;3–3;4	Miller (1979), 10.21415/T56592
many/Sza	2;1–5;9	Szagun (2001), 10.21415/T5KG7T
many/Wei	7;0–11;0	Weissenborn (1986), 10.21415/T5301C
many/Wag	1;5–14;10	Wagner (1985), 10.21415/T5ZC8K
Corinna	2;8–7;6	Lieven and Stoll (2013), 10.21415/T50S34
Cosima	1;9–7;2	Lieven and Stoll (2013), 10.21415/T50S34
Pauline	1;10–7;11	Lieven and Stoll (2013), 10.21415/T50S34
Sebastian	2;1–7;5	Lieven and Stoll (2013), 10.21415/T50S34

After extraction, we manually categorized each utterance into one of the four categories in Table 2 and added the corresponding codes to the data points.⁸ Category 0 was assigned to cases of the string ‘ohne’ occurring not as the word ‘ohne’ but, for example, as part of wohnen ‘reside’. Category 0 was excluded from analysis. Utterances where ohne was used in an adult-like way were assigned to Category 1. Category 2 was assigned to data points where ohne was immediately preceded by mit. All other data points were assigned to Category 3. This included occurrences of ohne other than category 2 that we judged either to be ungrammatical to an adult and occurrences we judged to have a double expression of negation where only one negation was intended.

Table 2. Categories for occurrences of the string ohne ‘without’ by number and code as explained in the text, with the count of items in each category

Number	Code	Category	Count
0	–	irrelevant	159
1	ohne	ohne used adult-like	306
2	mit ohne	mit ohne	52
3	ohne+	ohne mit, ohne kein, nicht ohne	7

We decided on this categorization given our intuition and the prior anecdotal reports that the category 2 ‘mit ohne’ would be frequent, while the other non-adult patterns we listed in (4) have not been reported. All utterances in category 2 ‘mit ohne’ corroborate our hypothesis. But utterances from category 3 ‘ohne+’ may also corroborate our hypothesis, though that category may also contain other non-adult uses of ohne. We expected this category to yield fewer data points than category 2, of which each would warrant individual discussion. Including non-adult uses of ohne other than the predicted ones is important in order to evaluate our predictions since it would weaken the support for our hypothesis if such patterns were frequent in child speech.

The categorization of a few items gave rise to some difficulty. Of the category 0/irrelevant utterances, only (6) could not be clearly categorized because the word mohne does not exist but might be a contraction of mit ohne, a mispronunciation of ohne, or something else we could not reconstruct.

Table

(6) na   das Kleid  mohne

well the dress xxx

‘the dress without’ (Sza/Lara, 4;09, CI)

Among the adult-like uses of ohne, we classified as adult-like many fragments without taking into account whether the context would actually license the use of a fragment for an adult speaker. Furthermore, we included nine utterances where ohne was repeated. Most such repetitions (see (7) and (8)) were produced by very young children, but a 10-year-old also produced the repetition in (9). As repetitions are abundant in adult language as well, we regarded them as adult-like speech errors throughout.

Table

(7) die Schokolade ohne  Eis ohne   ohne  ohne   ohne   Schokolade krieg ich

the chocolate  without ice without without without without chocolate   get   I

‘I get the chocolate without ice.’ (Caroline, 2;4)

(8) ohne  ohne   sich zu stossen

without without self to bump

‘without bumping into something’ (Sza/Laura, 2;01, CI)

(9) […] und der der da   was     tut ohne  ohne   was

[…] and he  he  there something does without without something

‘ … and he does something without something’ (Wag/Regina, 10;07)

3 Results

The theoretical prediction of the the Meaning First approach (Sauerland and Alexiadou, 2020) we set out to test is that even if children only hear ohne as a single morpheme, they should nevertheless form a complex mental representation of the meaning of ohne as [anti cum]. Following Guasti et al. (2023), we assume that evidence for such a complex representation is provided by child production where ohne is not used in an adult way but in one of the predicted non-adult patterns of (4). Category 2 ‘mit ohne’ of our data classification corresponds to one of these predicted patterns, while the other three predicted patterns would all be classified in category 3 ‘ohne$+$’. The hypothesis that we test can be formulated as follows:

• Young children are more likely to undercompress and produce one of the predicted forms more frequently than older children.

We will test our hypothesis by comparing the ratio between adult-like vs. non-adult-like use of ohne between two age groups (using the chi-square test).

Figure 1 shows the distribution of the different uses of ohne by age from age 22 months (1;20) and up. No utterances including ohne occurred in younger children in our sources. Data from children older than 5;00 are sparse in our sources (see Table 1) but are adult-like with only one irrelevant exception.⁹

Figure 1: Number of occurrences of ohne ‘without’ in German child language by age and category (see Table 2). Adult-like occurrences are in green, non-adult ones in blue. Ages are binned into 4-month intervals from 22–25 months (youngest) to 142–146 months (oldest).

The histogram shows that the prediction that children decompose ohne overtly into two pieces is confirmed by the high number of ‘mit ohne’ occurrences. Three examples illustrating the ‘mit ohne’ category from three different children are shown in (10).

Table

(10) a. ich     ein Brötchen haben mit   ohne  Käse  drauf

1SG.NOM a    roll   have   with without cheese on it

‘I (want) to have a roll without cheese.’ (Sza/Falko 2;09;15)

b. und die      hatte auch ein Schokoladenbus mit   ohne   Räder

and DEF.3SG.NOM had   also  a   chocolate bus    with without wheels

‘And she also had a chocolate bus without wheels.’ (Leo 2;07;12)

c. schmeckt mir    nicht mit   ohne  Butter

tastes  1SG.DAT  not   with without butter

‘I don’t like it without butter.’ (Caroline 2;07;14)

Figure 1 also shows that most of the uses of mit ohne ‘with without’ (47 out of 52) occur between ages 2;03 and 2;09. Two of the later five occurrences of mit ohne are from a child who had received a cochlear implant. One of these utterances is shown in (11), while the other was a partial repetition of (11). We include these two occurrences in our analysis with the child’s chronological age, though Szagun (2000) used the post-implant age in her comparisons. If we made this adjustment, Eileen’s age would be reduced by 2;03 (Szagun, 2000, 41) to 2;05, which is in the age-range typical children mostly use mit ohne.

Table

(11) ich kann auch mit   ohne   Stützräder   steh’n bleiben

I  can   also  with without support wheels stand  stay

‘I can also stop without support wheels.’ (Eileen 4;08)

The remaining three utterances containing mit ohne by children older than 3 years are the following: one utterance of mit ohne Brille ‘with without glasses’ by Rig/Cosima at age 4;00;01 and the utterance (12), which was furthermore partially repeated and therefore counted twice.

Table

(12) nur  ganz kleines Schneckenhaus mit ohne    Schnecke drin

only very  small   snail house   with without snail    inside

‘only a very small shell without a snail inside’ (Sebastian 3;03;09)

Only seven utterances were assigned to category 3, one of which was outside the age range of up to 7;02 (see footnote 9). As mention above, our hypothesis predicts that at least one pattern of undercompression should occur. The pattern mit ohne represents this pattern and, hence, their occurrences support our hypothesis, which is borne out by the occurrence of mit ohne. But given our uncertainty about the young children’s morphosyntactic knowledge, at least three other patterns of undercompression may also occur (see (4)). Namely, the opposite order ohne mit and the patterns ohne nicht and nicht ohne with the interpretation of a single negative like adult ohne.

The six utterances in category 3 by children younger than 7;02 all fit to one of the other three error patterns theoretically predicted. Four of the six exhibit the ohne mit sequence,¹⁰ which even occurs twice in (13).¹¹

Table

(13) xxx willst Bohne oder nur  Bohne ohne     mit   Milch ohne   mit  Zutter

xxx want bean  or    only bean  without with milk   without with sugar

‘xxx, do you want bean or only bean without milk without sugar?’ (Sebastian, 3;05;04)

One of the two other utterances in category 3 is (14). The utterance is only coherent if the pattern nich[t] ohne is understood as exponing on [anti cum]; alternatively, the child may have omitted a negation.

Table

(14) ich erzähl alleine nich ohne  die  Oma    xxx

I  tell  alone not  without the grandma xxx

‘I will tell it alone, without grandma.’ (Corinna 3;01;21)

Finally, (15) exhibits a non-adult use of the sequence ohne keine ‘without none’ that is interpreted as a single negation, even though its negation, specifically anti, seems to be articulated twice: ohne is the negative antonym of mit ‘with’ and keine is the negative antonym of the positive indefinite ein-e ‘a-fem’. We therefore analyze (15) as involving the realizations of cum as ohne, anti as k-, and exist as eine.¹² Furthermore, the structure [[cum anti] exist] of the three concepts is more plausible than [cum [anti exist]] since contextual allomorphy generally requires a close structural relationship of the allomorph and the trigger of allomorphy (Bobaljik, 2011) and in the structure [[cum anti] exists] the two—cum and anti—are sisters. If this reasoning is correct, (15) essentially amounts to a realization of the predicted non-adult sequence ohne nicht ‘without not’ as predicted in (4) with k- instead of nicht.

Table

(15) Ohne  keine xxx

without none xxx

‘Without any (potatoes)’ (Leo 2;02;07)

In sum, we have shown evidence that all four patterns listed in (4) actually occur but not with equal frequency. This corroborates our theoretical assumptions but indicates furthermore that not all options we list in (4) to expone anti cum/cum anti are considered equally likely by German children. Specifically, children seem to have some reason to analyze ohne not as exponent of cum but as exponent of anti. Furthermore, children seem to assume that the order cum anti conforms better to their environment language than anti cum. At this point, we can only speculate on the causes of the children’s behavior and expect that research on languages other than German will be crucial to understand these better. As for the first behavior of the German children, a possible cause of why ohne expones anti is that ohne can express negative subordination, in cases such as (16) where mit cannot occur as its antonym. In such cases, ohne expresses an inner negation of while; that is, while not eating.

Table

(16) Sie  ist ohne / *mit   zu essen 42 km gelaufen.

she is without/ *with to  eat   42 km ran

‘She ran 42 km without eating.’

As for the order, the children’s almost exclusive use of the order mit ohne is surprising from a semantic perspective as negation expressed by anti must take scope over the possession concept cum to express not having something. However, in the adult language, mit displays greater morphosyntactic variability, occurring for example as pre-verbal particle (for example, mit-arbeiten – ‘co-work’, mit-teilen – lit. ‘co-share’, ‘communicate’), which are impossible with ohne, and therefore children might place mit in a higher position. We note that adults also prefer the order mit ohne to ohne mit in cases where they find such child-like uses acceptable, though the acceptability may be entirely in child speech imitation. In the following analysis of number of occurrences we focus on the pattern mit ohne ‘with without’ because it is the most frequent non-adult pattern.

The overall picture that emerges is that children frequently produce mit ohne during the third year of life but hardly ever afterwards. Instead, the older children almost always produce ohne alone; that is, not preceded by mit ‘with’. Specifically, the rate of occurrences of ohne preceeded by mit (category mit ohne) of all occurrences of ohne in the categories ohne and mit ohne in the children up to 3;00 was 25.3% (47 of 186, while excluding 4 in category ohne$+$) compared to 4.6% (5 of 172 excluding 3 in category ohne$+$) in the older children in our sample. At the peak from 26 to 29 months of age, even 41% (29 of 70) of all utterances of ohne are non-adult (category mit ohne or ohne+). To confirm the significance of the generalization, we performed a chi-square test with the two binary variables age (below 3 years vs. 3 years and older) and mit-ohne (category 1 ‘ohne’ vs. category 2 ‘mit ohne’; see Table 2). The test indicates that the interaction between the two variables is statistically significant (chi-square(1) = $22.292$, $\mathit{p}<.00001$).

Since adults may sometimes use mit ohne as we mentioned above, we also computed two frequencies that could be compared to the child data. First we looked at books published in German since 1800 and digitized by Google. We found that the frequency of ohne in our source varied over time from slightly above 0.1% to 0.05%. But the frequency of mit ohne was only ever 0.000016%; that is, less than one in 3,000 occurrences of ohne was preceded by an occurrence of mit, compared to 1 in 4 in speech of the children under 3;00. Secondly, we performed a similar ratio computation for two publicly available corpora that represent spoken adult German from DWDS: the corpora gesprochene Sprache (spoken language) and also Filmuntertitel (film subtitles), since the former is rather small. We found that 0 out of 1,541 and 6 out of 32,575 occurrences of ohne were preceded by mit; that is, less than 1 in 5,000.

The data in Figure 1 suggest that the use of ohne in child language follows a U-shaped curve similar to overregularization found with inflectional morphology. A robust finding in the acquisition literature on irregular forms is that children at some stage of development produce overgeneralizations of the regular morphological process such as goed for ‘went’ and heared for ‘heard’ ((Ervin and Miller, 1963) and others). In such cases, children’s development exhibits a U-shaped trajectory: initial productions are adult-like, then overregularizations occur, and only later do children return to adult-like productions (Marcus et al., 1992). Figure 1 seems to exhibit three similar stages, an initial adult-like ohne-stage up to age 2, then a 6-month stage when mit ohne is produced, followed by the adult-like stage starting from age 3 in this case. While the U-shaped pattern is similar for both phenomena, their linguistic character is not comparable: there is no general rule in German grammar of inserting mit (or another preposition) before prepositions, while English past tense -ed is used to express past tense with almost all verbs. Therefore, children’s use of mit ohne itself does not constitute an overregularization. Consequently, the observed U-shaped pattern is surprising and deserves further scrutiny.

We therefore looked more closely at the productions of ohne in the 1;8–2;1 (22–25 months) age bin, which we initially categorized as adult-like. These utterances were all from Pauline ($\mathit{n}=4$) and Leo ($\mathit{n}=14$), except for one utterance by Laura/Sza (2;01;14), a child with a cochlear implant. All four of the relevant utterances by Pauline and seven by Leo consisted of only the word ohne. These one-word utterances may be due to the child being in a one-word stage of language development where the child’s cognitive resources are too limited to produce sentences consisting of more than a single word. They therefore do not support the claim of a true U-shaped curve since the early, seemingly adult-like utterances of ohne can reasonably be explained by independent constraints at the one-word stage.

All 11 other occurrences of ohne by children who are at most 25 months old are listed in (17).

Table

(17) a. so ohne (‘so without’, Pauline 1;11;06)

b. do ohne (‘[s]o without’, Pauline 1;11;06)

c. den ohne (‘the one without’, Pauline 1;11;06)

d. oben ohne (‘above without’, Leo 2;00;19)

e. oben oben ohne (‘above above without’, Leo 02;00;26)

f. ohne Gleise (‘without tracks’, Leo 2;01;10)

g.Glocke läuten ohne Rauch raus (‘bells ring without smoke out’, Leo 2;00;22)

h. ohne Gleise (‘without tracks’, Leo 02;01;25)

i. Frösche ohne Auto fahren (‘frogs without car drive’, Leo 02;01;25)

j. ohne Frösche (‘without frogs’, Leo 02;01;25)

k. ohne ohne sich zu stoẞen (‘without without hit oneself’, Laura/Sza 2;01;14)

Of these, the first two with the particle so (once mispronounced or mistranscribed as do) are akin to one-word sentences because the ohne explicates so, as in the English ‘Like this – without’. The other nine occurrences of ohne, however, constitute genuine sentences of two or more words.

We furthermore checked whether the three children who produced these adult-like occurrences of ohne went on to produce the non-adult-like mit ohne later in their development as expected from a U-shaped developmental curve. For Laura/Sza, there were no other occurrences of ohne. For Pauline, there are five additional occurrences of ohne between age 2;02 and 2;12 of which all are adult-like.¹³ In the Leo corpus, there are 31 occurrences of mit ohne between age 2;02 and 2;07. Leo’s data do indeed conform to a true U-shaped pattern, with 7 early, adult-like uses of ohne followed by 31 uses of mit ohne and only adult uses after age 2;07. Unfortunately, the corpus data are too scarce to establish whether this finding is due to chance; at this point all we can say is that the available data weakly suggest a U-shaped pattern, which would be surprising from a theoretical perspective, as discussed above. More data are needed to investigate this issue further, but our main conclusion is unaffected by this uncertainty.

4 Discussion

We argued that child production data provide new evidence for a decompositional analysis of at least some antonyms. Specifically, we argued in favor of such a decompositional analysis for the case of German ohne as the prepositional antonym of mit; unlike its English counterpart, ohne is monomorphemic. We showed that German 2-year-olds frequently produce the complex mit ohne instead of just ohne. The children’s behavior is creative and non-adult-like, and we find the same behavior across different children who were not in contact with each other. As far as we can see, our data provide strong support for the assumption that ohne cannot correspond to a single mental concept but must be decomposed into two pieces, one corresponding to what we called the cum concept (English with), and another corresponding to a negation or antonymity concept, which we have denoted as anti.

The result of our study also provides evidence for the Meaning First approach of Sauerland and Alexiadou (2020), which views conceptual structure and language as separate but closely linked by a relation of compression. Under this approach, children are expected to diverge from adults with respect to how they compress conceptual representations into linguistic structures. Before adult-like compression is fully acquired, children will make errors of both over- and undercompression, the latter in cases where nonpronunciation of a concept is obligatory for adults. We argued that despite being morphologically simple, the German proposition ohne in fact corresponds to the complex concept [anti cum]/[cum anti] at the level of conceptual representation. Undercompression of this complex concept predicts that children will produce patterns like mit ohne—a novel prediction which was corroborated on the basis of corpus data.

In contrast to German ohne, English without is morphologically complex, thus corresponding more closely to what we assume is a universal conceptual representation across languages, in line with the more general assumption within the Meaning First framework that conceptual representations do not vary across speakers of different languages. Given the universal complex concept [anti cum]/[cum anti], then, the morphological difference between English and German must be due to the fact that English articulates the cum concept while German compresses it. This difference seems to be rooted in morphological properties of the two languages. Namely, the exponent of the anti concept as part is out in English but ohne in German. But while out occurs in other contexts as well, ohne is restricted to exponing the complex [anti cum]/[cum anti]. The presence of cum can therefore be directly inferred from any occurrence of German ohne, while English out does not license this inference. In order for German children to arrive at the adult-like production of ohne, then, they first have to learn that ohne is limited to [anti cum] contexts and that the realization of the cum concept in such contexts is therefore redundant and can be left out.

Two general predictions of our result concern other languages and other antonym pairs.

Some early studies have found asymmetries in the acquisition of adjectival antonym pairs (Clark, 1972), but more recent work such as Tribushinina et al. (2013) is less conclusive on the matter. Outside of adjectives, some cases exhibit asymmetries which are in line with a decompositional account along the lines outlined above (Kotzor, 2021). Specifically for quantificational determiners, Katsos et al. (2016) reported that across more than 30 languages, 5-year-old children accomplish higher rates of correct understanding with some compared to its negative antonym no. Some recent work (Nicolae and Yatsushiro, 2022; Driemel et al., 2023; Bill et al., 2019) indicates that German children seem to sometimes overtly decompose the negative indefinite kein ‘no’ into two pieces, though this is absent from the adult grammar.

Secondly, we expect undercompression errors similar to German to arise with the exponent of [anti-cum]/[cum anti] in all other languages where the exponent is not at least bimorphemic. We have not completed our systematic examination of this prediction yet but have collected anecdotal evidence in favor of this prediction from several languages including Dutch (Jaqueline van Kampen, p.c.) and Portuguese (Elaine Grolla, p.c.). We plan to explore both predictions in future work.

Research Ethics

No ethical review was required since the results we report were achieved by reusing data from the cited sources that are in the public domain.

7 Declaration of interest

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of this article.

Acknowledgments

We are grateful to Jesse Snedeker, Angeliek van Hout, Elaine Grolla, Jacqueline van Kampen, our colleagues within the LeibnizDream project, our editors, Sudha Arunachalam and Krista Szendroi, and two anonymous reviewers for helpful comments on this article. A CC BY license is applied to the Author Accepted Manuscript (AAM) arising from this submission.

Disclosure statement

No potential conflicts of interest are reported by the authors(s).

Additional information

Funding

This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 856421).

Notes

¹ Heim (2006) calls the morpheme little but reduces its semantics to negation. Bobaljik (2011) uses the symbol ‘$\leftrightarrow$’.

² The relevant context can be purely linguistic in the form of an antecedent, but context may also include conventions within the linguistic community (Guasti et al., 2023).

³ Sauerland and Alexiadou (2020) used the term primitive meaning not complex as in formal logic. They furthermore assumed that the concept expressed by an open class word like pencil or road is generally complex consisting of at least a category concept (for example, nominality or verbality) and a concept capturing the root’s idiosyncratic meaning following much work in morphology (Alexiadou et al., 2014)

⁴ While this nomenclature might suggest that cum is a primitive concept, we can remain agnostic as our account would not be affected if cum was internally complex.

⁵ Note that without in English can also express a meaning similar to outside (in Scottish English, the order outwith is used in this case), while in German ohne does not exhibit this ambiguity and only the word außerhalb expresses this meaning of without.

⁶ Three of the four bimorphemic patterns with nicht in (4) are possible sequences in the adult language but cannot convey the meaning of ohne: two convey contrastive negation as in mit nicht Zucker, sondern Salz ‘with not sugar but salt’ and nicht mit Zucker, sondern ohne ‘not with sugar, but without’, and one conveys double negation as in nicht ohne Gefahr ‘not without danger’. The fourth sequence, ohne nicht ‘without not’, is not acceptable in modern Standard German, but a preliminary search (Google) shows that ohne nicht ‘without not’ was used about 200 years ago with the meaning of plain ohne. For example, Man legt keinen Garten an ohne nicht vorher einen Plan darüber gesehen, ohne den Boden untersucht zu haben (Carl von Proff. 1803. Ideen über die Organisation einiger untern Staatsgewalten, und verschiedene darauf Bezug habenden Gegenstände, mit Rücksicht auf das Herzogthum Berg. p. 108).The existence of the transparently decomposed pattern in at least a historical stage of German adult language corroborates our contention that ohne is structurally complex, but we leave serious historical work for future research.

⁷ The two examples provided – ohne Hanschuh ‘without gloves’ and ohne Vater ‘without father’ (Stern and Stern, 1907, 104) – could also be analyzed as bare prepositional phrases.

⁸ The categorization was done initially by the first author and checked by the other two authors.

⁹ The exceptional utterance was mit was ohne ‘with something without’ by Teresa/Wag at age 9;07, which seems meaningless to us even in the context of the transcript. Since the child is classified as typically developing and the data point is not predicted by our hypothesis, we have to put it aside as a speech error.

¹⁰ Leo (2;04;25) once produced the sequence ohne mit ohne as a fragment utterance, which we assigned to category 2 on the assumption that it was a self-correction of Leo’s.

¹¹ The discourse situation in (13) is similar to (3) above, but the order of ohne and mit is fully ungrammatical for adults.

¹² We assume that the concept exist expresses existential quantification.

¹³ Pauline’s later uses of ohne are also adult-like, which is unsurprising because they are from age 5;07 and older.