2. Motion expressions across languages

Talmy (Reference Talmy2000) defined a motion event as involving a Figure moving along a Path with reference to a Ground in a particular Manner, also known as voluntary motion (Hendriks et al., Reference Hendriks, Hickmann and Pastorino-Campos2022). Of these components, Path is considered the framing event and Manner the co-event, and depending on whether Path is encoded in the verb or a satellite (e.g., particle, prefix), Talmy categorised the world's languages into satellite-framed (S-languages: e.g., English) and verb-framed languages (V-languages: e.g., Spanish). English is an S-language because speakers typically express Path in a particle and Manner in the main verb, as in (1); Spanish is a V-language as Path is typically expressed in the main verb and Manner (if at all) in an adjunct, as in (2). Subsequent research noted that V-languages license satellite-framed constructions if the denoted motion does not entail crossing a spatial boundary (Aske, Reference Aske1989; Slobin & Hoiting, Reference Slobin and Hoiting1994), as in (3).

The implication of Talmy's typology for language use has been explored in numerous studies, mostly in relation to Slobin's thinking-for-speaking hypothesis (Bunger et al., Reference Bunger, Skordos, Trueswell and Papafragou2021; Filipović, Reference Filipović2011; Ji & Hohenstein, Reference Ji and Hohenstein2017, Reference Ji and Hohenstein2018; Slobin, Reference Slobin, Gumperz and Levinson1996, Reference Slobin, Gentner and Goldin-Meadow2003, Reference Slobin, Strömqvist and Verhoven2004, Reference Slobin, Hickmann and Robert2006; von Stutterheim et al., Reference von Stutterheim, Bouhaous and Carroll2017; Wang & Wei, Reference Wang and Wei2021). One recurrent observation in this research is that S-language speakers typically provide semantically denser motion descriptions (i.e., two components – Manner and Path) than their V-language counterparts (i.e., one component – typically Path only). According to the thinking-for-speaking hypothesis, such crosslinguistic differences are rooted in language-specific ways of conceptualising experience for the purpose of verbalisation. Specifically, to express Manner and Path simultaneously, S-language speakers have compact constructions at their disposal whereas V-language speakers typically have to use syntactically complex constructions (e.g., subordination) that incur greater processing load (Özçalışkan & Slobin, Reference Özçalışkan, Slobin, Özsoy, Akar, Nakipoglu-Demiralp, Taylan and Aksu-Koç2003; Slobin, Reference Slobin, Strömqvist and Verhoven2004, Reference Slobin, Hickmann and Robert2006). During online production, speakers fit their conceptualisation of an event in constructions that are most readily accessible in their language. Thus, facilitated by compact structures, S-language speakers habitually profile both Manner and Path, thereby producing semantically dense motion descriptions. V-language speakers, due to typological constraints, typically omit Manner and profile only Path – the component carried in the obligatory element of a sentence, thereby producing semantically less rich motion descriptions (Allen et al., Reference Allen, Özyürek, Kita, Brown, Furman, Ishizuka and Fujii2007; Özçalışkan, Reference Özçalışkan2015; Tusun, Reference Tusun, Shei and Li2022a; Tusun & Hendriks, Reference Tusun and Hendriks2019; Wang & Wei, Reference Wang and Wei2021). How early successive bilingual children develop such language-specific tendencies, and by implication, thinking-for-speaking patterns, will be the focus of this study.

3. Motion expression in Uyghur and in Chinese

Uyghur is a Turkic language of the southeastern branch. It is spoken in northwestern China's Xinjiang Uyghur Autonomous Region (Xinjiang) by at least 10 million native speakers (nearly half of Xinjiang's population). It is co-official with Chinese and is the lingua franca among other ethnic minorities. The language is also used in the local printing press, radio, and television broadcasting (cf. Ragagnin, Reference Ragagnin and Sybesma2016). As is typical of Turkic, Uyghur is head-final with rich agglutinative morphology. Examples (4) is the equivalent of (2) where Path is expressed in the verb, additional Path information (i.e., Goal and Source of motion) via case marking, and Manner in a converb, the functional equivalent of gerundives in European languages (Johanson, Reference Johanson, Haspelmath and König1995); in (5), Manner is expressed in the verb while Path (i.e., Goal) is expressed in a case marker. Thus, (4) is a verb-framed construction and (5) a satellite-framed construction, and usage-based studies on Uyghur (Tusun, Reference Tusun, Shei and Li2022a; Tusun & Hendriks, Reference Tusun and Hendriks2019) have shown that while Uyghur licenses satellite-framed constructions, it is a typical V-language both in terms of lexicalisation and semantic density.

Talmy (Reference Talmy2000) originally categorised Chinese as an S-language. In (6), Manner and Path are expressed in a resultative verb compound (RVC) and Talmy took the Path-encoding morpheme (V2) jin4 to be a satellite to the main verb (V1) pao3 expressing Manner, a pattern characteristic of Germanic languages. He maintained that, akin to Path satellites in English, the V2 morphemes form a closed class and the V2 slot is where semantic categories such as ‘aspect’ and ‘resulting state’ are expressed. Due to the absence of morphological marking in Chinese, however, establishing the grammatical status of the V2 morphemes is not straightforward. Moreover, unlike Germanic Path satellites, Chinese Path-encoding morphemes can function as main verbs, as in (7). It has therefore been argued that the two verbal elements in an RVC share the same grammatical status and formal significance, and that Chinese is an equipollently-framed language (E-language) (Slobin, Reference Slobin, Strömqvist and Verhoven2004). In response, Talmy (Reference Talmy, Guo, Lieven, Budwig, Ervin-Tripp, Nakamura and Özčaliskan2009) proposed a set of properties characteristic of main verbs, and recent studies testing his proposal support the claim that Chinese is an E-language (Wen & Shan, Reference Wen and Shan2021; see also Talmy, Reference Talmy2016). Importantly, numerous usage-based studies (e.g., Ji et al., Reference Ji, Hendriks and Hickmann2011c; Lamarre, Reference Lamarre2003; Wen & Shan, Reference Wen and Shan2021) have shown that Chinese speakers’ motion descriptions are semantically as dense as those of S-language speakers, and that the verb-framed option, exemplified in (7), is frequently used in Chinese. For instance, in a study based on a one-million-word corpus, Chen and Wu (Reference Chen and Wu2023) report that the verb-framed option accounts for about 24% of their data. Taking these together, I consider Chinese as an E-language with verb-framing tendencies.

4. Motion expressions in L1 and bilingual contexts

L1 research has shown that children's earliest productions reflect the typological tendencies of their ambient language (Chen, Reference Chen2008; Choi & Bowerman, Reference Choi and Bowerman1991), and, by age 3, they largely follow language-specific lexicalisation patterns (Allen et al., Reference Allen, Özyürek, Kita, Brown, Furman, Ishizuka and Fujii2007; Bowerman & Choi, Reference Bowerman, Choi, Gentner and Goldin-Meadow2003; Guo & Chen, Reference Guo, Chen, Guo, Lieven, Budwig, Ervin-Tripp, Nakamura and Özçalışkan2009; Hickmann et al., Reference Hickmann, Hendriks and Harr2018). However, children's ability to produce semantically dense motion descriptions develops over time, and as mentioned in Section 2, typological constraints play a major role. For example, children acquiring S-languages have been found to reach the adult level of semantic density earlier than those speaking V-languages (Harr, Reference Harr2012; Hickmann et al., Reference Hickmann, Hendriks and Harr2018), but children learning an E-language have been found to outperform their S-language peers. Thus, Ji et al. (Reference Ji, Hendriks and Hickmann2011a, Reference Ji, Hendriks and Hickmann2011b) compared age-matched Chinese and English children (aged 3, 4, 5, 6, 8, 10) and showed that, across age groups, the former consistently produced more high-density descriptions than the latter; in fact, Chinese children reached the adult level of density already from age 3. Ji and colleagues attributed this to the facilitative effect of readily accessible linguistic devices in Chinese (i.e., RVC). Finally, beyond the impact of typological factors, children also display certain universal tendencies: younger children experience greater difficulty encoding motion events that involve a categorical change of location, i.e., crossing a spatial boundary, as compared to a gradual change of location (Hendriks et al., Reference Hendriks, Hickmann and Pastorino-Campos2022; Ji et al., Reference Ji, Hendriks and Hickmann2011b).

Relevant research on bilingual speakers concerned the extent to which they think for speaking in language-specific ways, and whether, to what extent and why there is crosslinguistic influence (CLI), defined as the overuse of morphosyntactic structures in bilinguals’ one language under the influence of the other language (Serratrice, Reference Serratrice2013). Regardless of whether it is simultaneous child bilinguals (Engemann, Reference Engemann2021; Miller et al., Reference Miller, Furman and Nicoladis2018) or successive child (Aktan-Erciyes, Reference Aktan-Erciyes2020; Aktan-Erciyes et al., Reference Aktan-Erciyes, Göksun, Tekcan and Aksu-Koç2020; Aveledo & Athanasopoulos, Reference Aveledo and Athanasopoulos2016; Engemann, Reference Engemann2016) or adult bilinguals (Daller et al., Reference Daller, Treffers-Daller and Furman2011; Hohenstein et al., Reference Hohenstein, Eisenberg and Naigles2006), the general understanding is that bilinguals largely follow language-specific lexicalisation patterns, but they also exhibit CLI. To illustrate, studies almost always involved one V-language and one S-language and, compared to monolinguals, bilinguals displayed in-between encoding tendencies where they used more Manner verbs in their V-language and more Path verbs in their S-language. The dimension of bilingual speakers’ semantic density, especially with respect to how it is affected by CLI, is not well-understood, but preliminary evidence suggests that child bilinguals’ semantic density in the V-language can be augmented under the influence of the S-language, although this implicates the use of target-deviant lexicalisation patterns (Engemann, Reference Engemann2016).

Factors proposed as underlying CLI include, inter alia, structural overlap, the amount of relative exposure and language dominance. Structural overlap is important because, despite their preferred motion encoding strategies, languages tend to share lexical (and syntactic) resources (e.g., Manner verbs, Path verbs, cf. Beavers et al., Reference Beavers, Levin and Tham2010). Whilst bilinguals do capitalize on crosslinguistically shared options (Filipović, Reference Filipović2022), this seems to be modulated by language dominance and the relative amount of exposure. For example, Hohenstein et al. (Reference Hohenstein, Eisenberg and Naigles2006) found Spanish–English bilingual adults living in the U.S. to display an L2 to L1 influence, which they attributed to the sociolinguistic setting wherein English was the dominant language. Similarly, Daller et al. (Reference Daller, Treffers-Daller and Furman2011) showed that Turkish–German adult bilinguals resident in Germany tend to use lexicalisation patterns characteristic of German (S-language) when verbalising motion events in L1 Turkish whereas those living in Turkey tend to use patterns typical of Turkish in their L2 German, thereby reflecting the typical pattern of the (societally) dominant language. Two other studies (Aktan-Erciyes, Reference Aktan-Erciyes2020; Aktan-Erciyes et al., Reference Aktan-Erciyes, Göksun, Tekcan and Aksu-Koç2020) on Turkish–English child bilinguals (aged 5 vs. 7, AoO=3) in Turkey also argued for the impact of language dominance on CLI: 5-year-old bilinguals exhibited an L2 to L1 influence (more Manner verbs, fewer Path verbs) while the 7-year-olds displayed an L1 to L2 influence. Aktan-Erciyes and colleagues explained that this was because the 5-year-olds had total immersion in L2 English (8 hours daily) whereas the 7-year-olds’ quantity of L2 exposure dropped (2 hours daily) when they attended Turkish-dominant schools. Somewhat related are findings from Aveledo (Reference Aveledo2015) and Aveledo and Athanasopoulos (Reference Aveledo and Athanasopoulos2016) on Spanish–English child bilinguals (aged 5-7 vs. 8-9, AoO=3-4) in Venezuela: that only the older bilinguals showed an L2 to L1 influence due to their increased L2 exposure (16 hours weekly) compared to the younger bilinguals (8 hours weekly). Thus, shifts in language dominance, typically associated with the amount of relative exposure to a given language, shape CLI in bilinguals’ motion expression.

5. Uyghur–Chinese early successive bilingual children's acquisition of motion expressions

The above-mentioned studies have undoubtedly improved our understanding of bilingual expression of motion, but most of them included a V-language and an S-language that were genetically related (Aveledo & Athanasopoulos, Reference Aveledo and Athanasopoulos2016; Engemann, Reference Engemann2016; Hohenstein et al., Reference Hohenstein, Eisenberg and Naigles2006) while a better appreciation of the role of language-specific factors in bilingual language acquisition calls for more diverse language pairings (Serratrice, Reference Serratrice2013; Yip & Matthews, Reference Yip and Matthews2022). Additionally, most of the studies involved adult bilinguals while those on children tended not to include many age groups. We know little about how language-universal factors found to operate in monolingual child language acquisition (Hendriks et al., Reference Hendriks, Hickmann and Pastorino-Campos2022) inform bilingual language acquisition, how the aspect of ‘semantic density’, known to develop later than the acquisition of lexicalisation patterns per se in monolinguals, develops in bilingual children, how this is affected by CLI, and indeed, how CLI plays out developmentally, an issue much debated in the context of the acquisition of morphosyntax (Chondrogianni, Reference Chondrogianni, Elgort, Siyanova-Chanturia and Brysbaert2023; van Dijk et al., Reference van Dijk, van Wonderen, Koutamanis, Kootstra, Dijkstra and Unsworth2022) but relatively unexplored in the motion domain. We therefore need a developmental perspective. Finally, reflective of the field of bilingualism research, most previous studies concerned Western immigration contexts where one of the languages is societally dominant and the other the minority/heritage language, but it is doubtful that this dichotomy is readily applicable to bilingualism situations in non-Western communities. And we need more information on such communities and how affordances specific to their own sociolinguistic realities shape the acquisition and use motion expressions (Foroodi-Nejad & Paradis, Reference Foroodi-Nejad and Paradis2009; Paradis & Nicoladis, Reference Paradis and Nicoladis2007).

This study contributes to closing these gaps. First, it focuses on a hitherto unexplored language combination featuring a verb-framed language and an equipolllently-framed language that are distant genetically (Turkic vs. Sino-Tibetan), and distinct in their general linguistic profiles (agglutinating vs. isolating). Second, it adopts a developmental perspective with a view to shedding light both on bilingualism-related issues such as the role of CLI during acquisition (Hulk, Reference Hulk, Blom, Cornips and Schaeffer2017; van Dijk et al., Reference van Dijk, van Wonderen, Koutamanis, Kootstra, Dijkstra and Unsworth2022), and on what is potentially common/universal in children's spatial language development, be it monolingual or bilingual. Third, Uyghur–Chinese bilingualism presents a non-Western bilingual situation where the distinction between societally dominant versus non-dominant language is blurred, not least because, as mentioned in Section 3, Uyghur is co-official with Chinese in Xinjiang and is a regional lingua franca, and Uyghurs constitute nearly half of Xinjiang's population, and attach great importance to promoting and maintaining their language (for insights into the sociology and politics of Uyghur–Chinese bilingualism, see Elterish, Reference Elterish, Finley and Zang2015; Zang, Reference Zang, Finley and Zang2015). Within this sociolinguistic milieu, Uyghur children typically grow up speaking their L1 Uyghur, and the regional educational policy, at least in urban areas, is such that at around age 3, they attend full immersion Chinese kindergartens and subsequently full immersion Chinese schools at around age 6 (cf. Ma, Reference Ma, Beckett and Postiglione2012; Zheng, Reference Zheng2011). They are therefore early successive bilinguals (Chondrogianni & Vasić, Reference Chondrogianni and Vasić2016; Meisel, Reference Meisel, Fernández and Cairns2018) who acquire their L2 naturalistically. A relevant affordance of this unique bilingual setting, which contrasts with much of previous research, concerns bilingual children's relative exposure to and use of their two languages. Sociolinguistic research on Uyghur–Chinese bilinguals’ language use (Elterish, Reference Elterish, Finley and Zang2015, Reference Elterish, Ahn and Smagulova2016) reports that they tend to exclusively use Uyghur outside the school context, and by virtue of their schooling from kindergarten onwards, bilinguals’ exposure to L2 Chinese remains constant (about 8 hours daily). That is, bilinguals’ waking hours are somewhat naturally divided into 8 hours of Chinese immersion and 8 hours of Uyghur outside school. And in light of recent insights that bilinguals’ relative amount of language exposure and use can serve as a proxy for language proficiency and language dominance (Unsworth, Reference Unsworth, Silva-Corvalan and Treffers-Daller2016; Unsworth et al., Reference Unsworth, Chondrogianni and Skarabela2018), the sociolinguistic setting in question is arguably more conducive to more balanced bilingualism (Filipović, Reference Filipović2019), and as such, Uyghur–Chinese bilingual children’ acquisition and use of motion expressions may be different from their peers in other bilingual contexts. Against this backdrop, this study asks the following research questions:

In relation to RQ1, we predicted that, like children learning V-languages (Hendriks et al., Reference Hendriks, Hickmann and Pastorino-Campos2022 for French; Aktan-Erciyes et al., Reference Aktan-Erciyes, Göksun, Tekcan and Aksu-Koç2020 for Turkish), bilinguals from the earliest age tested would follow the adult pattern of expressing Path in the verb and Manner in the converb. Additionally, they would express additional Path information (e.g., Source, Goal) via case markers (cf. Furman, Reference Furman2012 for Turkish). However, their adult-like ability to simultaneously express Manner and Path would develop much later (cf. Harr, Reference Harr2012). In terms of RQ2, several predictions could be entertained. Given the early and systematic exposure to the L2, it was predicted that bilinguals from the earliest age tested would be fully adult-like with no L1 influence: they would predominantly use equipollently-framed constructions (i.e., RVCs) and much less frequently, verb-framed constructions (i.e., Path in verb and Manner in subordinate clause), as did Chinese monolingual children (cf. Ji et al., Reference Ji, Hendriks and Hickmann2011a). They would therefore predominantly produce semantically dense descriptions. Alternatively, given the structural overlap of verb-framing between Uyghur and Chinese, bilinguals could use such constructions more frequently than Chinese adults, thereby displaying CLI. This means that they would express only Path in the verb and additional Path information via satellites (e.g., prepositions) but would omit Manner; they would therefore produce low-density descriptions (cf. Hendriks et al., Reference Hendriks, Hickmann and Pastorino-Campos2022; Slobin, Reference Slobin, Strömqvist and Verhoven2004). In terms of how this CLI would manifest developmentally, two possibilities were considered. As per the claim that CLI is part and parcel of the bilingual experience (cf. Chondrogianni, Reference Chondrogianni, Elgort, Siyanova-Chanturia and Brysbaert2023; van Dijk et al., Reference van Dijk, van Wonderen, Koutamanis, Kootstra, Dijkstra and Unsworth2022), CLI would persist throughout childhood such that Uyghur–Chinese bilingual children across age groups would consistently use verb-framed constructions more than Chinese adults. Alternatively, in light of L2 studies showing a decrease of CLI as a function of increased proficiency (Montero-Melis & Jaeger, Reference Montero-Melis and Jaeger2020; Park, Reference Park2020), bilingual children could use verb-framed constructions more frequently than adults at the early stages, but this would decrease over time while their use of equipollently-framed constructions would increase, eventually converging on the target equipollent system. This would be compatible with the hypothesis that CLI is a developmental phenomenon (Hulk, Reference Hulk, Blom, Cornips and Schaeffer2017). In this case and considering the strong influence of language-specific factors on motion expression (Hendriks et al., Reference Hendriks, Hickmann and Pastorino-Campos2022; Ji et al., Reference Ji, Hendriks and Hickmann2011a), bilinguals’ semantic density would be higher in Chinese than in Uyghur.

6. The study

6.3. Coding and analysis

All the responses were transcribed into CHAT format (CHILDES; MacWhinney, Reference MacWhinney2000) and were first segmented into clauses, with a clause defined as a unit containing one verb and its arguments (Hickmann et al., Reference Hickmann, Hendriks and Harr2018). Thus, responses exemplified in (8) and (9) were segmented into two clauses, a subordinate clause and a matrix clause. When occasionally participants gave more than one response for an item, two criteria were applied hierarchically, i.e., richness and relevance of Path. For example, based on ‘richness’, R2 in (10) would be selected as the ‘target response’ because it simultaneously expressed Manner and Path. However, when responses contained either Manner or Path (1.4%), as in (11), we chose, as per ‘relevance’, R2 as the ‘target response’ (cf. Talmy, Reference Talmy2000). In all cases, R1 was marked as ‘potential target response’, but was not included in our analysis.

Each target response was coded in terms of the semantic information expressed in various linguistic devices (information locus) and the total number of motion components expressed (semantic density). In relation to information locus, following previous studies (e.g., Hendriks et al., Reference Hendriks, Hickmann and Pastorino-Campos2022; Wang & Wei, Reference Wang and Wei2021), two loci were identified: the main verb (the verb locus); and the satellite, defined as all other devices outside the main verb (the OTH locus). Both V1 and V2 elements of an RVC in Chinese were coded as two verbs, and the OTH locus included dative/ablative case markers, converbs, adverbials (e.g., fei1su4de ‘quickly’) and prepositional phrases (e.g., cong2 you4bian1 ‘from the right side’). In terms of motion information in the verb locus, responses in Uyghur fell into two categories– those encoding Path and those encoding Manner (see 12-13), while those in Chinese fell into three categories, i.e., Path, Manner, or Path+Manner (see 14-16). With respect to the OTH locus, responses across the two languages were categorised into those expressing Path (see 17 and 21), Manner (see 18 and 22), Path+Manner (see 19 and 23), and Zero – a residual category for responses with no satellite devices (‘bare verb constructions’ à la Hohenstein et al., Reference Hohenstein, Eisenberg and Naigles2006) and thus no spatial information in this locus (see 20 and 24). For semantic density, only semantic information from distinct categories was considered (irrespective of the linguistic devices used) such that multiple mentions of Path within one response counted as density 1 (SD1, see 12 and 21) while one mention of Path and one of Manner counted as density 2 (SD2, see 13 and 23).

For the statistical analyses, our independent variable was age whereas the dependent variables were the mean occurrence of Path verbs, Manner verbs, Path satellites, Manner satellites as well as SD1 and SD2 responses. The count data were analysed by fitting generalised linear mixed-effects models with a Poisson distribution, using R (R Core Team, 2013), the glmer() function in the lme4 library. We first fitted a model to the same dataset with the fixed effects in question, against a reduced model without the fixed effects in question. We then compared the relative goodness of fit of the two models using a likelihood ratio test via the anova() command, which revealed the relative fits (expressed as log likelihood) of the two models to test the statistical significance of the fixed effect removed in the reduced model. For all models fitted, random intercept for participant and item were included. Planned contrasts with Bonferroni adjustment were specified where more than two factors were compared (cf. Appendix S2). We report the chi-square statistics, degrees of freedom and p value for the tests. All model outputs are provided in the ‘Appendix S3’.

7.1 Information in the verb locus in Uyghur and in Chinese

Figure 1a shows information expressed in the verb locus in Uyghur by ageFootnote ⁴. A two-way packaging (Path, Manner) x age (4yrs, 6yrs, 8yrs, 10yrs, adults) analysis revealed a significant interaction (χ²(4)=68.998, p < .001), suggesting that the two lexicalisation patterns varied by age. Further analyses found an age effect only for Manner (χ²(4)=29.14, p < .001) as 4- and 6-year-olds encoded this component more frequently than adults (β_4yrs-AD = –o.81, SE = 0.32, Wald z = –2.53, p = .011; β_6yrs-AD = –1.26, SE = 0.27, Wald z = –4.69, p = <.001). That is, children fully established their L1 verb-framed pattern from age 4 while their early tendency to encode Manner dropped to the adult level at age 8.

Figure 1 Information expressed in the verb locus across age groups

Figure 1b represents information expressed in the verb locus in Chinese by age. A two-way packaging (Path, Manner, Path+Manner) x age (4yrs, 6yrs, 8yrs, 10yrs, adults) interaction analysis was significant (χ²(8)=306.11, p < .001), indicating that children's lexicalisation patterns varied by age. Further analyses found age effects for Path (χ²(4)=35.932, p < .001), Manner (χ²(4)=54.398, p < .001) and Path + Manner (χ²(4)=68.828, p < .001). Follow-up analyses revealed that 4-, 6- and 8-year-olds encoded Path more frequently than adults (β_4yrs-AD = –0.78, SE = 0.17, Wald z = –4.40, p < .001; β_6yrs-AD = –0.59, SE = 0.17, Wald z = –3.35, p < .001; β_8yrs-AD = –0.63, SE = 0.21, Wald z = –2.96, p = .003), and 4- and 6-year-olds encoded Manner more frequently than adults (β_4yrs-AD = –1.58, SE = 0.33, Wald z = –4.79, p < .001, β_6yrs-AD = –1.45, SE = 0.30, Wald z = –4.76, p < .001). Finally, only 4- and 6-year-olds used the Path+Manner pattern less frequently than adults (β_4yrs-AD = –0.98, SE = 0.15, Wald z = 6.24, p < .001; β_6yrs-AD = 0.58, SE = 0.11, Wald z = 4.87, p < .001), the steady increase of this pattern within the four child groups was significant at each age level (β_4yrs-6yrs = –0.40, SE = 0.16, Wald z = 2.47, p = .013; β_6yrs-8yrs: = 0.34, SE = 0.13, Wald z = 2.53, p = .011; β_8yrs-10yrs: = –0.35, SE = 0.10, Wald z = –3.34, p < .001). That is, children's verb-framed pattern dropped to the adult level at age 10 while their equipollently-framed pattern (i.e., RVC) increased to the adult level from age 8. The Chinese equipollent framing system (both verb- and equipollently-framed lexicalisation patterns) was fully established at age 10.

7.2 Information in the OTH locus in Uyghur and in Chinese

Figure 2a illustrates information expressed in OTH locus in Uyghur by age. A two-way packaging (Path, Manner, Path+Manner, Zero) x age (4yrs, 6yrs, 8yrs, 10yrs, adults) analysis revealed a significant interaction (χ²(12)=198.91, p < .001), reflecting that children's packaging strategies varied by age. Further analyses found age effects only for Path (χ²(4)=77.53, p < .001) and Path + Manner (χ²(4)=46.80, p = .001) such that all the child groups expressed Path more frequently than adults (β_4yrs-AD = –0.92, SE = 0.12, Wald z = –7.46, p < .001; β_6yrs-AD = –0.91, SE = 0.33, Wald z = –3.00, p < .001; β_8yrs-AD = –1.01, SE = 0.12, Wald z = –8.30, p < .001; β_10yrs-AD = –0.74, SE = 0.14, Wald z = –5.23, p < .001), and Path + Manner less frequently than adults (β_4yrs-AD = 0.98, SE = 0.13, Wald z = 7.40, p < .001; β_6yrs-AD = 0.75, SE = 0.12, Wald z = 6.15, p < .001; β_8yrs-AD = 0.60, SE = 0.11, Wald z = 5.17, p < .001; β_10yrs-AD = 0.34, SE = 0.11, Wald z = 3.08, p = .002). That is, children did not fully establish the lexicalisation pattern for the OTH locus in Uyghur even at age 10.

Figure 2 Information expressed in OTH locus across age groups

Figure 2b displays information expressed in OTH locus in Chinese by age. A two-way components (Path, Manner, Path+Manner, Zero) x age (4yrs, 6yrs, 8yrs, 10yrs, adults) interaction analysis was significant, (χ²(12)=99.566, p < .001), showing that the relative frequency of the four patterns varied by age. Further analyses specified the age effects to Manner (χ²(4)=23.06, p < .001), Path+Manner (χ²(4)=19.79, p < .001) and Zero (χ²(4)=23.29, p < .001) such that 4- and 6-year-olds expressed Manner (β_4yrs-AD = 1.21, SE = 0.33, Wald z = 3.60, p < .001; β_6yrs-AD = 1.69, SE = 0.39, Wald z = 4.32, p < .001) and Path+Manner less frequently (β_4yrs-AD = 2.91, SE = 1.11, Wald z = 2.61, p = .008; β_6yrs-AD = 1.80, SE = 0.73, Wald z = 2.44, p = .014) while all child groups produced Zero more frequently that adults (β_4yrs-AD = –0.41, SE = 0.10, Wald z = –4.06, p <.001; β_6yrs-AD = –0.37, SE = 0.10, Wald z = –3.57, p <.001; β_8yrs-AD = –0.21, SE = 0.10, Wald z = –2.06, p = .039; β_10yrs-AD = –0.21, SE = 0.10, Wald z = –2.00, p = .045). That is, children converged on the adult pattern in the OTH locus by age 8, although they continued to produce more motion constructions without satellite devices, i.e., Zero, than adults.

7.3 Semantic density in Uyghur and in Chinese

Figure 3a depicts semantic density in Uyghur by age group. A two-way density (SD1, SD2) x age (4yrs, 6yrs, 8yrs, 10yrs, adults) analysis showed a significant interaction (χ²(4)=132.93, p < .001), indicating that semantic density varied by age. Further analyses identified age effects for both SD1 (χ²(4)=54.53, p < .001) and SD2 (χ²(4)=31.38, p < .001) such that all child groups produced SD1 descriptions more frequently than adults (β_4yrs-AD = –0.82, SE = 0.10, Wald z = –7.63, p <.001; β_6yrs-AD = –0.63, SE = 0.13, Wald z = –4.84, p <.001; β_8yrs-AD = –0.80, SE = 0.12, Wald z = –6.25, p <.001; β_10yrs-AD = –0.63, SE = 0.12, Wald z = –5.01, p <.001) but SD2 descriptions less frequently than adults (β_4yrs-AD = –0.69, SE = 0.10, Wald z = 6.62, p <.001; β_6yrs-AD = –0.36, SE = 0.10, Wald z = –3.48, p <.001; β_8yrs-AD = 0.58, SE = 0.12, Wald z = 4.77, p <.001; β_10yrs-AD = –0.34, SE = 0.09, Wald z = 3.67, p <.001). That is, children stopped short of the adult frequency for SD2 descriptions even at age 10.

Figure 3 Semantic density across age groups

Figure 3b shows semantic density across age groups in Chinese. A two-way density (SD1, SD2) x age (4yrs, 6yrs, 8yrs, 10yrs, adults) analysis revealed a significant interaction (χ²(4)=362.5, p < .001), suggesting that semantic density varied by age. Further analyses found age effects for both SD1 (χ²(4)=106.66, p < .001) and SD2 (χ²(4)=85.22, p < .001). Specifically, the stepwise decrease of SD1 descriptions was significant at each age group (β_4yrs-6yrs = –0.17, SE = 0.08, Wald z = –2.12, p =.033; β_6yrs-8yrs = –0.38, SE = 0.11, Wald z = –3.20, p =.001; (β_8yrs-10yrs = –0.60, SE = 0.19, Wald z = –3.19, p =.001; β_10yrs-AD = –0.63, SE = 0.27, Wald z = –2.33, p =.019) and all child groups produced SD1 descriptions more frequently than adults (β_4yrs-AD = –1.79, SE = 0.20, Wald z = –8.81, p <.001; β_6yrs-AD = –1.62, SE = 0.20, Wald z = –7.90, p <.001; β_8yrs-AD = –1.23, SE = 0.24, Wald z = –5.01, p =.001; β_10yrs-AD = –0.63, SE = 0.27, Wald z = –2.33, p =.019). Meanwhile, the increase of SD2 descriptions was significant from age 4 to 10 (β_4yrs-6yrs = –0.39, SE = 0.15, Wald z = –2.51, p =.011; β_6yrs-8yrs = –0.38, SE = 0.12, Wald z = –3.11, p =.001; β_8yrs-10yrs = –0.28, SE = 0.09, Wald z = –3.21, p =.001). At age 10, children attained the adult level of frequency for SD2 descriptions.