Twenty bilingual children, acquiring Russian and German, aged from 4–6 years were selected via a Russian Facebook group. In these Facebook groups, members favor culture of their countries of origin. Inclusion criteria were that L1 Russian is the dominant language for the children at home (that is spoken to at least 60%). Both of the parents or at least one of them and additionally other caregivers and relatives speak Russian, and contact with L2 German started before the age of 25 months [14]. The children are simultaneously bilingual and started to acquire L2 at the age of 4.5 months on average. All children are exposed to Russian as their dominant language. They attend kindergarten, and score within the mean range in nonverbal cognitive ability [15]. Exclusion criteria were the presence of a hearing and/or vision impairment, a developmental disorder, a DLD (especially ICD-10, F80), and/or a chronic disease.

The children were screened in both of their languages in order to evaluate their respective language proficiencies and to subsequently correlate these results to results of the language-specific nonword repetition tests.

Regarding the evaluation of Russian, children were screened with “The Russian language proficiency test for multilingual children (SRUK)” [16]. This test is a linguistically and psycholinguistically grounded screening procedure. Bilingual children’s proficiency can be examined in the domains language production and comprehension. The test is based on preliminary norms of 167 German–Russian bilingual children between the ages of 3;0 and 6;11 years and has to be administered by a Russian native speaker. The following subtests were selected for the present study: language production (nouns and verbs), language comprehension (nouns and verbs), and comprehension of grammatical structures.

Regarding the evaluation of German, children were screened with “Patholinguistische Diagnostik bei Sprachentwicklungsstörungen (PDSS)” [17]. This procedure is linguistically and psycholinguistically grounded as well. The subtest selected were word production (nouns and verbs), word comprehension (nouns and verbs), and comprehension of syntactic structures.

NWR is influenced by phonetic and phonological development. Regarding the exclusion of children who do not have age adequate phonological abilities, the following screening procedure was performed: 5 words which were pronounced as part of the subtest active lexicon from SRUK [17] were transcribed according to International Phonetic Alphabet (IPA) from a Russian native speaker and linguist and classified according to their age-adequatedness from an experienced clinical linguist in cooperation with a speech pathologist.Footnote 1 All children with the exception of child 3 have age adequate phonetic and phonological abilities. Child 3 had to be excluded because of a phonological disorder (ICD-10, F.80.0). The case study of this child has been published [18].

Regarding NWR in the respective languages, language-specific nonwords for Russian and German, developed by Mathieu et al. [19], were used for the present study. The nonwords were developed with regard to similarity with real words in the respective languages, concerning item length, phonotactic probability, phonological complexity, phonological similarity of the elements, and language-specific prosodic patterns [19, p. 13]. The nonwords for the respective languages were controlled for length (words with three, four, and five syllables in both languages, and in German additionally two-syllabic words), prosody (accentuated first vs accentuated last syllable) and syllable structure (consonant–vowel vs consonant–consonant–vowel).

A computer game (smartphone/tablet application) which was developed by the second author was used to elicit the nonword repetitions in a playful way inspired by the software used by Mathieu et al. [19]. In this game the children have to repeat the names of monsters.

Russian nonwords [19] are shown in Table 1.

German nonwords [19] are shown in Table 2.

All children (except child 3 who was excluded from the analysis) could be classified as having typical language development. As the aim of the NWRT was to measure the ability to repeat syllables, NWR abilities of the children were evaluated with tolerance. This means that phonetic errors and typical phonological processes are not counted. Consequently, the obtained raw scores per child could vary from 0–28 (see also Mathieau et al. [19]). For a descriptive analysis of the language abilities in the respective languages, and thus, in order to answer research question 1, the results of the language screenings are shown in percentage ranks (for the PDSS), and in raw scores (for SRUK). Results of SRUK were analyzed by transforming the raw scores into four categories according to Gargarina et al. [16]: within normal range, above normal range, conspicuous, and very conspicuous based on provisional norms. This was in order to perform a descriptive analysis of the screening results in relation to the biographic data. Then, a statistical analysis of the relation between the NWRT and language development in both languages was performed.

For statistical analyses, and thus, in order to answer research questions 2–5, one-sided correlation tests were performed in R [20]. The tests were one-sided, because the hypotheses postulate a positive and not a negative relation between the nonword repetition tests and language development at the diverse linguistic levels. Correlation tests were performed between the NWRT in Russian and the NWRT in German, and a set of three correlation analyses was performed between each of the NWRT and all three subtests of language development in one language, i.e., the comprehension of verbs and nouns, the comprehension of grammar, and the production of verbs and nouns. As all variables are interval scaled, Pearson’s product moment correlation coefficient was calculated for variables that were normally distributed. Tests of normality were performed with the Shapiro–Wilk test. Correlation tests involving variables that were not normally distributed were performed by calculating Spearman’s rho (ρ). Moreover, p-values of all correlation tests were Bonferroni corrected in order to account for type 1 errors, because of the multiple tests in which the variables of the NWRT were involved. In this case, the significance level of p > 0.05 was divided by 3, i.e., the number of correlation tests performed within each set of correlation analysis, leading to a corrected significance level of 0.05/3 = 0.017.