A Corpus-Based Study of Spatial Frames of References in Early Child Mandarin Chinese

Abstract

This study examines the developmental pattern of spatial frames of reference (FoRs) in Chinese preschoolers using the naturalistic data elicited from the Beijing Early Childhood Mandarin Corpus. Altogether, 2837 static spatial sentences were identified from the corpus, and 785 of them used FoRs; thus, they were analysed using the seven-class FoRs coding system developed from the literature review. The results indicated that: (1) six classes were produced by the Beijing preschoolers, including the direct reference (DR), geomorphic reference (GR), landmark-based reference (LBR), object-centred reference (OCR), relative reference (RR), and people-centred reference (PCR). But, the absolute reference (AR) frequently used by Beijing adults was not produced by the preschoolers; (2) significant age differences were found in the utterances with reference and demonstrative sentences and were also identified in the production of OCR, PCR, and RR; (3) there were no significant gender or gender x age effects in the production of FoRs, except for the no-frame sentences; and (4) there was a stable and consistent pattern of FoRs allocation by age, and the most frequently used was OCR. All these findings indicate interactions among language, culture, and cognition.

1. Introduction

Spatial frames of reference (FoRs) is the coordinate system that Gestalt theorists originally developed to compute and specify the location of objects with reference to other objects (Koffka 1935). This system has been extensively employed by psychologists, linguists, and cognitive and neuroimaging scientists (Levinson 1996, 2003; Li and Zhang 2009; Majid et al. 2004; Shusterman and Li 2016) to explore the interaction between language acquisition and cognitive development. In particular, many studies have tested the development of FoRs in children and adults with different languages (Bohnemeyer 2011; Li and Zhang 2009; Majid et al. 2004; O’Meara 2011; O’Meara and Báez 2011; Polian and Bohnemeyer 2011; Shusterman and Li 2016). However, all these studies have adopted an experimental design that lacks ecological validity; thus, the findings could not be generalised to the real world (Lozano and Mendikoetxea 2013). There are a few studies that focus on natural language data (Brown and Levinson 2000; de León 1994; de León 1995). However, the sample size is very small; thus, the results have limited generalisability. Therefore, examining the early development of FoRs using a naturalistic design such as the corpus approach is urgently needed (Monaghan and Rowland 2017). With the hypothesis that young children may not acquire the different classes of FoRs simultaneously or synchronously due to individual and age differences, this study will fill the literature gap by analysing the developmental patterns of FoRs using the naturalistic data elicited from the Early Childhood Mandarin Corpus (Wu and Li 2017). This contributes to a more comprehensive understanding of language development across different languages and cultures.

1.1. Definition of Spatial Frames of References

In linguistics, spatial FoRs refer to the coordinate system established in an individual’s mind for constructing and organising spatial relations within the surrounding world (Levinson 1996, 2003; Shusterman and Li 2016; Sekine 2009). Within one certain type of FoRs, the entity to be located (the located object; LO) is treated as the ‘figure’, and the other entity used as the reference compound (the reference object; RO) is dubbed ‘ground’ (Adamou and Shen 2017; Burigo and Schultheis 2018; Levinson 2003). In this way, FoRs provide a framework for individuals to make meaningful sense of the spatial relationship between objects. Using this abstract mental structure, adults and even young children could communicate effectively about their locations and directions. As the key concept in space cognition, FoRs were first proposed by Gestalt theorists in the 1930s and then were extensively explored in multiple areas such as psychology, linguistics, and cognitive sciences (Li and Zhang 2009; Majid et al. 2004). For instance, linguists noted that the speakers might employ different FoRs when locating themselves or other entities in the environment due to their linguistic, cultural, and conceptual preferences (Deng and Yip 2016; Levinson 1996, 2003; Pederson et al. 1998). Therefore, it is necessary to consider one or more FoRs to understand the speakers’ emphasis in their expression about the relative positions of the referent (Carlson-Radvansky and Irwin 1993; Levinson 1996; cited by Surtees et al. 2012).

1.1.1. Taxonomy of FoRs

Based on an extensive cross-linguistic exploration of over 20 languages, Levinson (2003) classified FoRs into three distinct categories: (1) Relative system. This is a viewer-centred or ‘viewpoint-centred’ system generally based on the human body (the viewers’/the speakers’) and reveals a viewpoint-dependent relation. For instance, “The tree is to the right side of the house”, indicating a relative system that locates the objects’ positions relative to the viewer (Haun et al. 2011; Li and Zhang 2009; Surtees et al. 2012). This relative FoRs system has been found in the predominant frame in most European languages when talking about locations and directions (Haun et al. 2011; Surtees et al. 2012). (2) Intrinsic system. This system presents an object-centred or ‘object-based’ framework, wherein the coordinate depends on the “inherent features” of the reference object. For example, “The yard is in front of the building”, indicating a viewpoint-independent relation. This is the second frequently used frame in European languages and the primary frame in some other languages such as Mopan (Haun et al. 2011; Li and Zhang 2009; Surtees et al. 2012). (3) Absolute system. This system relies on the constant element of the environment with fixed direction provided by quadrillage; for instance, cardinal directions such as “north”/”south”/”east”/”west”, require the speakers/viewers to coordinate their orientation with reference to the earth-based environment or a landmark. Basically, this system involves fixed angles to reveal spatial relations, as shown in “The garden is south of the town” or similar sentences (Levinson 2003; Haun et al. 2011; Surtees et al. 2012). In European languages, this system is only used for talking about locations on a geographic scale, but it is still commonly used in other languages such as Mandarin Chinese in Northern China (Haun et al. 2011; Wu and Li 2017).

Levison’s taxonomy, however, has its shortcomings in exploring the acquisition of FoRs, especially in the early stage of development. For instance, it might not always be possible to distinguish in every utterance (especially a sentence of ambiguity) between the relative FoRs and the intrinsic FoRs (Haun et al. 2011; Surtees et al. 2012). Haun et al. (2011) provided an example of this ambiguity: “The Christmas tree is in front of the Santa’s sleigh” (圣诞树在圣诞老人的雪橇前面), which could be either “the Christmas tree is at the front facet of the Santa’s sleigh” (圣诞树在圣诞老人的雪橇前面) (the Intrinsic) or “the Christmas tree is in front of the Santa’s sleigh (from the cat’s perspective, as the cat is located between the Santa’s sleigh and the Christmas tree)” (圣诞树在圣诞老人的雪橇前面(从猫的视角来看，猫位于圣诞老人的雪橇和圣诞树之间。) (the relative) (see Demo A). In addition, Surtees et al. (2012) also demonstrated an example to show the consistent situation between these two FoRs: when the speaker is shown a picture demonstrating a cat facing a Christmas tree and the cat’s face is also fronted straight onto the speaker, in this situation, if the speaker says, “the Christmas tree is in front of the cat”, this sentence can then be regarded as both belonging to relative FoRs and intrinsic FoRs (see Demo B). Therefore, to identify the exact type of FoRs in certain circumstances, it is necessary to introduce a third dimension and develop a more comprehensive and inclusive system.

1.1.2. The Seven-Class FoRs

Shusterman and Li (2016) noticed that some distinctions within one category of the FoRs were missing in Levinson’s taxonomy, leaving the acquisition of the related but ‘missing’ FoRs in young children unstudied. In addition, they found that the body-based intrinsic FoRs in Levinson’s taxonomy failed to differentiate egocentric from non-egocentric, i.e., “to my right” versus “to his right” (Shusterman and Li 2016). This is because the intrinsic FoRs in Levinson’s taxonomy include both body-based terms and larger FoRs such as building (Shusterman and Li 2016). It might be possible that children might treat the intrinsic FoRs of their own bodies differently from that of a room (Shusterman and Li 2016). Therefore, to understand how children initially acquire and use the FoRs system in their daily spontaneous conversations, it is necessary to further develop Levinson’s (2003) taxonomy into a comprehensive and inclusive one by adding more levels and more subtypes of FoRs.

The existing studies have focused on the features and nature of the figure object (the entity to be located), the ground object (the reference entity), and the viewpoint coordinate (the entity that defines the coordinate system) (Levinson 2003; Levinson and Wilkins 2006; Pederson 2003). To develop a more comprehensive and inclusive taxonomy of FoRs, we must consider the following major cues provided by other researchers for FoRs. First, the features of the reference entity (‘the ground’) should be taken into consideration. Shusterman and Li (2016) argued that the object-based references of the intrinsic system could be subdivided into two subtypes: (1) if the reference entities are stationary and anchored to the earth, they should be treated as an environment-based or geocentric FoR; (2) if the entities can move freely relative to the earth, they should be regarded as an object-centric FoR (Shusterman and Li 2016). Moreover, according to O’Meara (2011), an environment-based FoR could be further developed into two subtypes, including the ‘geomorphic reference (GR)’ and the ‘landmark-based reference (LBR)’. GR was defined as a frame with one or more axes formed by the gradient of the geographical environment and could be projected onto the ground or figure, and LBR was defined as a frame with the axis as a vector pointing from the ground or figure to a given landmark (Bohnemeyer 2011; O’Meara 2011; Polian and Bohnemeyer 2011). Second, the speaker’s body position should also be considered. Danziger (2010) noted that the body’s position of the speaker could be an independent classification condition apart from Levinson’s three conditions and thus proposed a new class of FoRs: the ‘direct reference (DR)’. According to Danziger (2010), DR takes the observer’s body position directly as the coordinate, which has no projection relationship with the external reference. This new class has perfectly solved the problem noted by Shusterman and Li (2016) that children might treat the intrinsic FoRs of their own bodies differently from that of a room.

According to the above analyses, scholars have generally recognised that the production and comprehension of FoRs statements might be affected by the geometric features (such as the orientation and direction) of the mentioned objects, especially the RO (Burigo and Schultheis 2018; Carlson and Van Deman 2008). Therefore, the geometric information of the RO should be included. The FoRs system has been developed from three levels to six classes, including (1) object-centred reference (OCR), (2) relative reference (RR), (3) absolute reference (AR), (4) direct reference (DR), (5) geomorphic reference (GR), and (6) landmark-based reference (LBR). Therefore, Liu (2014) adopted the above six-class taxonomy to examine the FoRs in Mandarin Chinese and found that all the classes of FoRs could be identified in Chinese adults’ language texts sampled from media reports, news, and prose.

However, the above six-class taxonomy was developed based on adult language and usage. It should be further developed to investigate young children’s spatial conversations, as they might treat different reference entities quite differently. For example, Craton et al. (1990) also noticed that listeners and other people were common in children’s perceptual space. They should also consider the subtypes of reference entities regarding the real conditions of children’s spontaneous daily talk. Although the variety of the above-mentioned reference entities has included movable objects, the speaker’s body position, landmark, the listeners, and other people could also be used as a common reference entity type in the perceptual space of the speakers and should not be neglected. For example, children may locate something with the reference of their listeners, who will be naturally treated as the reference entity (i.e., ‘Tommy, listen to me, the bear is in front of you, now give it to me, ok?’). Therefore, there is a need to include the ‘people-centred reference (PCR)’ in this study to refer to the situation, which treats both the listeners and people as an independent class of FoRs. As such, a seven-class taxonomy is proposed in this study to investigate children’s spatial frames of reference in the early years, which has included the six-class FoRs and the additional PCR class. Details of this seven-class taxonomy (Figure 1) of FoRs are presented in the coding scheme (see Appendix A), which provides definitions and example sentences for each class.

1.2. Acquisition of Spatial Frames of References in Early Years

The acquisition of spatial frames of reference could be traced back to infant and toddlerhood, a period of considerable development in this regard (Berman 1976; Bremner et al. 2006; Huttenlocher et al. 1994; Newcombe et al. 1998; Perry et al. 2009; Sekine 2009). The intrinsic reference defined by Levinson (2003) has been treated as the non-egocentric category of FoRs (O’Meara and Báez 2011; Shusterman and Li 2016), and its acquisition begins early in the infant stage when egocentric thinking dominates (Bremner et al. 2006; Huttenlocher et al. 1994; Newcombe et al. 1998). Bremner et al. (2006) found that infants could form object-centred spatial frames of reference by recognising the ‘within-object’ spatial relation as early as four months. As young as 16 months, children were able to use visible FoRs to encode the location of a hidden object and, later (aged beyond 22 months), they could encode the object’s position relative to external landmarks even when the object was moved from one side to another (Huttenlocher et al. 1994; Newcombe et al. 1998). Researchers also found that the flexibility level of using the OCR in the hiding event to search for a hidden toy differed significantly between 2.5-year-old children and 3-year-olds (DeLoache 1987, 1989). The capability of using FoRs was found to have emerged from the age of three (Schutte and Spencer 2009).

Researchers found that during the preschool years, young children experience rapid growth in their intrinsic frames of reference, along with the insufficient development of the Relative FoRs. Craton et al. (1990) indicated that the overall performance improved with age but the communication about the relative reference involved with the left-right dimension was particularly difficult for four-year-olds. Moreover, they found that although children aged four could use person references (themselves or the listener) to specify front–back relations, only children aged six could compensate for their apparent difficulty in using the terms left and right by using landmarks to specify the left-right dimension. Young children aged eight could use a combination of person and landmark references in both front–back and left–right directions. All these findings jointly indicated that young children below the age of six could not use both person and landmark references in the relative reference involved with left–right directions. Negen and Nardini (2015) found that the four-year-olds were using the intrinsic frame in some trials, which was not so consistent, suggesting that the preschool period (peak around four years) might be crucial to developing intrinsic frames. To verify this crucial period, Shusterman and Li (2016) conducted a set of experiments with four-year-olds. They found that they could interpret vague spatial denoting as having environment-based meanings similar to north and south. Furthermore, they noticed that the four-year-olds readily learned and generalised north–south meanings in their speech. This finding indicated that four-year-olds might have acquired geocentric representations in spatial language acquisition. In addition, Shusterman and Li (2016) also found that four-year-olds could acquire “left” and “right” when systematically introducing the novel words based on their bodies, and they could even extend these words to the novel (intrinsic and relative) uses. However, the four-year-olds had difficulty identifying and telling the left and right sides of a doll, even though they had no difficulties labelling its front and back sides.

All these findings jointly indicated that the relative reference of left–right might be more difficult to acquire than that of front–back in preschool years, and the age of four might be the crucial period for the development of intrinsic frames. This study will verify this finding with the empirical evidence elicited from the Early Childhood Mandarin Corpus (Wu and Li 2017).

1.3. Spatial FoRs: Where Language, Culture, and Cognition Meet

Spatial FoRs have attracted extensive attention from linguists, psychologists, and cognitive scientists. It is an ideal arena for testing the interaction among language, culture, cognition, and associated influences. For instance, Levinson (2003) suggested that different language speakers worldwide might have different habits and preferences in the use of FoRs. The following studies have provided empirical evidence from adults’ application of FoRs to support his suggestion (Adamou and Shen 2017; Le Guen 2011; Li and Zhang 2009; Shinohara and Matsunaka 2004). In particular, Shinohara and Matsunaka (2004) experimented with 49 college students to understand the adult level of linguistic expression of spatial relations of objects. They found a distinctly Japanese phenomenon called ‘figure-aligned mapping’, wherein the coordinate system on the figure object was projected onto the ground object. Later, Le Guen (2011) conducted experiments to compare the acquisition of spatial term semantics, performance on non-verbal tasks, and gestures in adults and found an obvious gender difference regarding the semantics of spatial terms and an equal preference for a geocentric FoR in nonverbal tasks. The participants from the same village with similar cultural backgrounds used various strategies in their speech, all of which showed a systematic preference for a geocentric FoR in their gestures, revealing a similar preference for FoRs (Le Guen 2011). Coincidently, some Chinese studies found that Chinese people in various areas might have different preferences for the FoRs systems in nonlinguistic spatial tasks. In 2009, Li and Zhang designed and strictly controlled an experiment to explore this regional diversity in Chinese adults’ performance on linguistic and spatial tasks. About 24 adults from Northern China and 23 from Southern China were asked to use spatial terms in the absolute (e.g., East–West) versus relative (e.g., right–left) systems. They found that the Southerners differed significantly in the response time to the terms between the absolute and relative systems. This finding indicated significant regional differences in the processing of FoR systems within the same language speakers. Recently, Adamou and Shen (2017) conducted three experiments with 52 adults and found that the geocentric responses were predominant among some participants, indicating that FoRs might be culturally defined. They argued that FoRs might be influenced by cultural and environmental factors that are independent of language. However, all these studies were conducted in laboratories with adult participants; no studies have explored young children’s preferences for using FoRs in naturalistic contexts. Therefore, this study will fill this literature gap to ascertain whether and to what extent similar interactions among language, culture, and cognition could be observed from an early age in Chinese children.

1.4. The Context of This Study

Previous studies have explored the preferences of using FoRs in young children, without considering that the use of divergent FoRs subtypes might demand different language abilities and cognition capacities. The study on Chinese adults’ use of FoRs (Liu 2014) has indicated that the various classes might not be evenly distributed in their daily production, for example, OCR was the most frequently used FoR. This difference might be affected by the convenience and clarity of different types of FoRs in a certain language and the preference of a specific cultural custom. Therefore, it is natural to assume that young children might not acquire the seven classes of FoRs simultaneously or synchronously, as there might be individual and age differences. This study will explore the developmental patterns of using FoRs in Beijing preschoolers.

In addition, the use of spatial FoRs might be jointly influenced by culture, cognition, and language. In Beijing, adults tend to use absolute spatial FoRs to indicate the location and direction in daily communication (Liu 2014; Wu and Li 2017). In contrast, Mandarin speakers in Southern China tend to use relative spatial FoRs (Li and Zhang 2009). This North–South difference in using FoRs reflects the influence of cultural customs. Therefore, we assume that Beijing preschoolers should also have a similar preference for using absolute spatial FoRs in their daily communication. They have been immersed in the language environment with relatively more input from this type of FoRs. However, if this use were not observed in the Beijing preschoolers, that would mean that their cognitive development might have constrained the influences of culture on language. Therefore, this study explored the Beijing preschoolers’ preference for using FoRs. Furthermore, the existing studies of FoRs were mostly conducted in laboratories, and their experimental conditions might have also constrained the exploration of the potential cultural influences. It might be the cognitive constraints of the tasks, rather than language or culture, that have determined which system is easier to use by young children (Li and Abarbanell 2018). Therefore, it is urgently needed to examine young children’s FoRs in some natural scenarios such as free play with toys and explore the interaction between language, culture, and cognition. The Early Childhood Mandarin Corpus has provided an ideal platform for such a study that aims to identify Chinese children’s real capability in using spatial FoRs.

Accordingly, the following questions guided this study:

• What are the basic linguistic features of spatial FoRs in early childhood Mandarin Chinese?
• What are the developmental patterns of spatial FoRs in early childhood Mandarin Chinese?
• Are the seven-class FoRs applicable to early childhood Mandarin Chinese?

2. Materials and Methods

2.1. The Corpus

2.1.1. The Sample

The Early Childhood Mandarin Corpus collected 48 h of spontaneous conversations between 192 Mandarin-speaking preschoolers (96 boys and 96 girls) in Beijing. They were representative of four age groups (ages 2–6, 3–6, 4–6, and 5–6), with twenty-four boys and twenty-four girls in each age group. These preschoolers were randomly sampled from each class of the eight participating preschools located in four major districts of Beijing: Chaoyang, Dongcheng, Xicheng, and Haidian. All the participants were native Mandarin speakers whose parents and teachers also spoke Mandarin at home and in the preschool.

2.1.2. Communication Task

The 192 participants were randomly paired (boy/girl, boy/boy, or girl/girl) and encouraged to play and talk with each other in this task. This communication task was arranged in the indoor preschool classrooms with the play settings furnished with the same set of toys, including cooking materials, ordinary furniture and electrical appliances, food and fruits, hospital materials, and vehicles. For each communication task, only one pair of participants was allowed to play in the room for 30 min. A high-definition digital camera with two separate microphones was applied to videotape the conversations during playtime; meanwhile, the researchers also observed the entire communication task uninterruptedly. Before the formal videotaping dates, the camera equipment was placed in the play area of the classroom for some days to become a familiar and natural part of the setting, to avoid making any of the children feel uncomfortable. All the recorded conversations were transcribed by experienced research assistants (RAs). The first author of this study identified all the spatial expressions and then confirmed them through a panel of Chinese linguists.

2.2. Coding

2.2.1. The Seven-Class FoRs Coding Scheme

Based on a systematic review of the existing coding schemes (Bohnemeyer 2011; Danziger 2010; Levinson 2003; Levinson and Wilkins 2006; Liu 2014; O’Meara 2011; Pederson 2003; Polian and Bohnemeyer 2011; Shusterman and Li 2016), we have developed the seven-class frames of reference coding book: (1) object-centred reference (OCR): refers to the object-based reference within the intrinsic system, the reference entity is movable and not stationary, for example, ‘the vase is in front of the sofa’, in which the reference entity ‘sofa’ is a movable object; (2) relative reference (RR): Levinson’s (2003) definition is adopted, for instance, ‘the vase is to the right of the sofa’ (‘the right of the sofa’ is projected on the speaker’s viewpoint); (3) absolute reference (AR): specifically refers to the reference involved with cardinal directions (east/west/north/south), for example, ‘the vase is north of the sofa’; (4) direct reference (DR): refers to the coordinate that takes the speaker’s body position directly as the reference, i.e., ‘the vase is in front of me/us’; (5) geomorphic reference (GR): refers to a frame with one or more axes formed by the gradient of the geographical environment and could be projected onto the ground or figure, i.e., ‘the vase is upstream from the sofa’; (6) landmark-based reference (LBR): refers to a frame with the axis as a vector pointing from a ground or figure to a given landmark, i.e., ‘the vase is toward the church from the sofa’; (7) people-centred reference (PCR): refers to the coordinate that takes the others’ body position (such as the listeners or people other than the speakers themselves) as the reference, i.e., ‘the little vase is in front of the boy’. For details about this coding book, please refer to Appendix A.

2.2.2. Coding Procedure

This study followed systematic coding procedures that included three major steps.

First-step coding. Based on the reference status, all the static spatial sentences were differentiated descriptively into three types: (1) no reference sentences (no reference entity appeared in the sentence); (2) demonstrative reference sentences with demonstrative denoting words such as ‘这里 zheli [here]/那里 nali [there]’, which serve as the reference. Using demonstrative words indicates a tendency to take the speaker’s own body as the origin of FoRs, although the reference entity is vague (Deng and Yip 2015); and (3) with reference sentences (sentence with a certain reference entity).

Second-step coding. The system of the frame was analysed exhaustively to identify whether there is no frame, intrinsic, relative, or absolute system in each chosen sentence. Normally, most of the sentences could be easily recognised and exclusively classified; for instance, the sentence ‘the doll is in front of the table (洋娃娃在桌子前面)’ could be classified intuitively into the intrinsic system as the spatial expression of ‘in front of the table’ is determined by the front facet of the table that is intrinsic and has nothing to do with any other references. However, there were very few cases that might not be easily coded, such as the sentence ‘the cat is in front of the dog (貓在狗的前面)’, there might be an intrinsic or relative system depending on the perceptual space. Therefore, the first author had to watch the video clips to understand the perceptual space’s reality and confirm the speakers’ classification of these undetermined sentences.

Third-step coding. The exhaustive seven-class FoRs codes were used to differentiate all the sentences with reference, respectively. As some children might repeat a sentence repeatedly during their play communication, the same sentence repeated was treated as one token during the coding procedure.

2.2.3. Inter-Rater Reliability

All the static sentences were examined by the first author. About 20% of them were also coded by a trained research assistant to examine inter-rater reliability. Before the formal coding, the first author trained the research assistant to understand the seven-class FoRs system. Inter-rater agreement was calculated for each of the three steps of the coding procedure separately.

During the first-step coding, the two coders’ agreement in the identification of reference status resulted in a Cohen’s kappa value of k = 0.98 for ‘no reference sentence’, a Cohen’s kappa value of k = 1.00 for ‘with reference sentence’, and a Cohen’s kappa value of k = 0.96 for ‘demonstrative reference sentence’.

During the second-step coding, the two coders’ agreement in recognition of the frame system resulted in a Cohen’s kappa value of k = 1 for ‘no frame sentence’, a Cohen’s kappa value of k = 0.95 for ‘absolute sentence’, a Cohen’s kappa value of k = 0.89 for ‘intrinsic sentence’, and a Cohen’s kappa value of k = 0.92 for ‘relative sentence’.

During the third step of coding, the two coders’ agreement in the coding of seven-class FoRs resulted in a mean Cohen’s kappa value of k = 0.94. For each of the FoRs levels, the two coders’ agreement resulted in a Cohen’s kappa value of k = 0.93 for OCR, k = 0.95 for PCR, k = 0.96 for RR, k = 0.93 for DR, k = 0.95 for GR, and k = 0.90 for LBR (no AR sentence identified in the whole dataset). According to statistical criteria (Bakeman and Gottman 1997; McHugh 2012), the strength in agreement values 0.81–1.00 are regarded as an ‘almost perfect agreement’; thus, it could be confirmed that the above kappa values indicate satisfactory inter-rater reliability of the coding in this study. Given this strong inter-rater reliability, the first author felt confident and comprehensive enough to independently code the rest of the target sentences. The other two authors subsequently reviewed the coding results.

2.3. Data Analysis

First, we calculated the frequency and the related proportion of each class of FoRs used per child and the headcounts and the related percentages of FoRs users to explore the distribution of the seven classes of FoRs among Chinese preschoolers aged 2–6 to 5–6. Second, we calculated the types, grammatically complete sentence ratio (WSR), and mean length of utterance (MLU) for each age group to explore the linguistic features of the language materials produced by these Chinese preschoolers. In this study, MLU was calculated by the number of Chinese characters used in each sentence, which has been widely used by Chinese psycholinguists (Tse et al. 2002). Third, to explore the developmental trend of using FoRs among these preschoolers, we conducted two independent sets of one-way ANOVA to examine the age and gender effects. A set of MANOVA was applied to test if there is any potential age (4) by gender (2) interaction effect. Fourth, to confirm whether there is a link between age and the number of FoR users, a chi-square test was also conducted. Fifth, to examine whether there is a stable allocation pattern of seven classes of FoRs used by these preschoolers among different age groups, the mean of the FoRs ratio (proportion) against the total static spatial sentence of four age groups and the means of the FoRs ratio (proportion) against the total static spatial sentence of each age group were calculated. A stable allocation pattern was examined by conducting a set of one-way ANOVA to see if there is an age difference in the proportion of using each class of FoRs.

3. Results

3.1. Descriptive Statistics of the Language Materials

Altogether, 2837 static spatial sentences were elicited from the corpus, including various categories of spatial expressions such as localisers, demonstrative deictic, and existential sentences, all of which were produced by 168 preschoolers (N_{aged 2;6} = 42; N_{aged 3;6} = 42; N_{aged 4;6} = 42; N_{aged 5;6} = 42), 87.5% of the full sample (N_total = 192) in the corpus. Within all the 2837 static spatial sentences, 306 (10.78%) were no reference, 1735 (61.16%) were demonstrative reference, and 796 (28.06%) were sentences with reference. The 796 sentences with reference included 11 sentences (1.38%) with no frame and 785 sentences (98.62%) with FoRs, and the latter were treated as the target data to be coded in this study using the FoRs coding scheme.

As shown in

Table 1. Descriptive statistics of linguistic features.

	Utterances
FoRs Category	Aged 2–6		Aged 3–6		Aged 4–6		Aged 5–6
[Type (MLU)]	Types (MLU)	WSR	Types (MLU)	WSR	Types (MLU)	WSR	Types (MLU)	WSR
DR [30 (6.85)]	4 (6.75)	75%	5 (6.40)	60%	9 (7.00)	78%	12 (7.25)	83%
GR [9 (5.80)]	1 (3.00)	0	1 (5.00)	0	2 (8.00)	50%	5 (7.20)	80%
LBR [26 (8.38)]	4 (8.25)	50%	7 (7.43)	71%	9 (8.33)	78%	6 (9.50)	100%
OCR [131 (7.45)]	17 (7.05)	65%	24 (7.04)	63%	37 (7.54)	65%	53 (8.15)	89%
PCR [18 (8.41)]	0 (-)	0	2 (8.00)	50%	7 (7.57)	57%	9 (9.67)	100%
RR [41 (8.52)]	5 (7.40)	80%	9 (7.33)	67%	10 (10.40)	90%	17 (8.94)	94%
AR [- (-)]	-	-	-	-	-	-	-	-

Note. WSR = whole sentence ratio—refers to the ratio of grammatically complete sentences.

, types, grammatically complete sentence ratio (WSR), and mean length of utterance (MLU) by age were calculated. The results show that six out of seven categories of FoRs were produced by the Beijing preschoolers, including the DR, GR, LBR, OCR, PCR, and RR. No AR was produced. In particular, there were noticeable age-related increases in the six FoR categories produced by the 168 participants. Among them, four categories of FoRs including DR, LBR, OCR, and PCR reached the peak of MLU at ages 5–6 (Type_DR = 12, MLU_DR = 7.25; Type_LBR = 6, MLU_LBR = 9.50; Type_OCR = 53, MLU_OCR = 8.15; Type_PCR = 9, MLU_PCR = 9.67), and, in the other two, GR and RR reached peak MLU at ages 4–6 (Type_GR = 2, MLU_GR = 8.00; Type_RR = 10, MLU_RR = 10.40). Among the six classes of FoRs, the most frequently used type was the OCR (type = 131), followed by the RR (type = 41), DR (type = 30), LBR (type = 26), PCR (type = 18), and GR (type = 9). Interestingly, there were four categories of FoRs used with a longer MLU, including the RR (MLU = 8.52), PCR (MLU = 8.41), LBR (MLU = 8.38), and OCR (MLU = 7.45) sequentially, while the other FoRs categories were used with a relatively shorter MLU, including the DR (MLU = 6.85) and GR (MLU = 5.80).

Next, it is noticeable that the percentages of grammatically complete sentence expression in each category of FoRs increased dramatically after ages 4–6, with a low level at ages 2–6 or 3–6 and a peak level at ages 5–6 (Table 1). Specifically, the sample demonstrated a capacity to produce grammatically complete sentences of the RR category between ages two and six (WSR_RR2;6 = 80%) but low capacity in using a grammatically complete sentence to express GR, LBR, and PCR at a young age (WSR_GR2;6 = 0; WSR_LBR2;6 = 50%; WSR_PCR2;6 = 0). A high capacity for using a grammatically complete sentence in each category of FoRs was observed between ages five and six (WSRs > 80%), especially for the LBR and PCR (WSRs = 100%). Among the incomplete sentences, the most missing part was the ‘subject’, which could be commonly observed from all age groups in all the FoRs categories (see example a, b, c), while the missing or inappropriate verbs could be identified from some FoRs categories by young age (such as GR, RR; see example d, e). These findings suggested that the linguistic demand across different categories of FoRs might not be equal for the same age group.

Example a. Child 1 at ages 2–6
(某物)在	我	面前
(sth) be	my	front
(sth) is in front of me.
Missing component: subject
FoRs category: DR

Example b. Child 2 at ages 4–6
(某物)在	蛋糕	后面
(sth) be	cake	back
(sth) is at the back of the cake.
Missing component: subject
FoRs category: OCR

Example c. Child 3 at ages 2–6
(某物) 向着	门
(sth) towards	door
(sth) is towards the door.
Missing component: subject
FoRs category: LBR

Example d. Child 4 at ages 3–6
树	要(应该)	在	上坡
tree	should be	at	uphill
The tree should be on the uphill.
Missing component: appropriate verb
FoRs category: GR

Example e. Child 5 at ages 5–6
你	怎么	在门	右边	(站着)
you	why	door	right	(standing)
Why (are) you (standing) on the right of the door?
Missing component: appropriate verb
FoRs category: RR

3.2. Production Patterns of Spatial FoRs in Early Years

Among the 785 FoRs sentences, 435 (54.65%) were from the intrinsic system, 304 (38.19%) were from the relative system, and only 46 (5.78%) were from the absolute system. Among the seven-class FoRs, the most frequently used one was the ‘object-centred reference’ (OCR = 410, 52.23%), followed by the ‘relative reference’ (RR = 226, 28.79%), the ‘direct reference’ (DR = 78, 9.94%), the ‘landmark-based reference’ (LBR = 35, 4.46%), the ‘people-centred reference’ (PCR = 25, 3.18%), and the ‘geomorphic reference’ (GR = 11, 1.40%). However, no ‘absolute reference’ (AR = 0) was produced by the Beijing preschoolers (

Table 2. The frames of reference identified in the static spatial sentences uttered by the Chinese preschoolers (N = 168).

Step 1. Reference Status		Step 2. General FoRs		Step 3. Seven-Class FoRs in This Study
No reference [0]	306 (10.78%)
Demo reference [99]	1735 (61.16%)
With reference [1]	796 (28.06%)	No frame [N]	11 (1.38%)
		Absolute [A]	46 (5.78%)	Geomorphic reference [GR]	11 (1.40%)
				Landmark-based reference [LBR]	35 (4.46%)
				Absolute reference [AR]	0 (0%)
		Intrinsic [I]	435 (54.65%)	Object-centred reference [OCR]	410 (52.23%)
				People-centred reference [PCR]	25 (3.18%)
		Relative [R]	304 (38.19%)	Direct reference [DR]	78 (9.94%)
				Relative reference [RR]	226 (28.79%)
total	2837	total	796	total	785

Note. Figure status: with figure = 1721, no figure = 1116; the number or alphabet within the ‘[x]’ refers to the code.

).

3.2.1. The Pattern of Linguistic Features

To investigate the developmental patterns of using different types of FoRs of the preschoolers, we calculated the frequency of each category of FoRs for each age group with one-way ANOVAs. The results indicated that with the growing volume of static spatial sentence expressed from ages 2–6 to ages 5–6, there was a rapidly increasing trend of using the no-reference sentences (M_2;6 = 0.952; M_3;6 = 1.500; M_4;6 = 2.333; M_5;6 = 2.500), demonstrative reference sentences (M_2;6 = 4.905; M_3;6 = 6.071; M_4;6 = 9.286; M_5;6 = 21.048), and reference sentences (M_2;6 = 2.024; M_3;6 = 3.191; M_4;6 = 5.071; M_5;6 = 8.667). In specific, among the seven-type FoRs production, the most rapidly increasing trends of using frequency were found in the usage of OCR (M_2;6 = 0.857; M_3;6 = 1.476; M_4;6 = 2.571; M_5;6 = 4.857) and RR (M_2;6 = 0.595; M_3;6 = 0.952; M_4;6 = 1.524; M_5;6 = 2.310). Among the rest of the five types of FoRs, slightly increasing trends across ages were also found (Figure 2,

Table 3. Descriptive statistics of the static spatial utterances produced by the four age groups (N = 168).

	Aged 2–6 (n = 42)			Aged 3–6 (n = 42)			Aged 4–6 (n = 42)			Aged 5–6 (n = 42)
	Sum	Mean	SD	Sum	Mean	SD	Sum	Mean	SD	Sum	Mean	SD	F	p
Static spatial sentences	331	7.881	10.600	452	10.762	10.455	701	16.691	11.267	1353	32.214	20.646	25.554	0.000
No reference sentence	40	0.952	1.710	63	1.500	2.340	98	2.333	2.138	105	2.500	2.452	4.665	0.004
Demonstrative reference	206	4.905	7.999	255	6.071	6.961	390	9.286	6.992	884	21.048	14.002	25.623	0.000
Reference sentence	85	2.024	2.434	134	3.191	3.549	213	5.071	4.571	364	8.667	6.842	16.435	0.000
Subtypes of reference sentence
No frame sentence	3	0.071	0.261	1	0.024	0.154	2	0.048	0.216	5	0.119	0.395	0.943	0.421
Direct reference	15	0.357	1.144	17	0.405	1.037	20	0.476	0.890	26	0.619	0.882	0.554	0.646
Geomorphic reference	1	0.024	0.154	1	0.024	0.154	4	0.095	0.370	5	0.119	0.328	1.386	0.249
Landmark-based reference	5	0.119	0.328	10	0.238	0.617	7	0.167	0.437	13	0.310	0.680	1.021	0.385
Object-centred reference	36	0.857	1.389	62	1.476	1.798	108	2.571	2.596	204	4.857	5.020	14.029	0.000
People-centred reference	0	0.000	0.000	3	0.071	0.342	8	0.191	0.594	14	0.333	0.612	4.242	0.006
Relative reference	25	0.595	1.127	40	0.952	1.481	64	1.524	1.838	97	2.310	2.342	7.624	0.000

Note. Sum = total frequency.

).

The one-way ANOVA results indicated that significant age effects were found in producing the no-reference sentence (p < 0.01), reference sentence (p < 0.001), and demonstrative sentences (p < 0.001) among the four age groups. Significant age effects were also identified as expressing OCR (p < 0.001), PCR (p < 0.01), and RR (p < 0.001), suggesting an obvious age-related increasing trend in using these three types of FoRs. In addition, another set of one-way ANOVA tests was conducted, and the results indicated that there were no significant gender differences in producing all the target categories of sentences.

To identify the potential interaction effects between those variables, we conducted two sets of MANOVAs, using the age (4) by gender (2) model. The dependent variables for the first set of MANOVAs were the frequencies of the reference sentence, no-reference sentence, demonstrative reference sentence, and no-frame sentence. The second set of MANVOAs focused on the six types of FoRs used, including DR, GR, LBR, OCR, PCR, and RR. The analyses revealed a significant age-by-gender interaction in the use of no-frame sentences among these preschoolers: F (1, 3) = 2.764, p < 0.05, η² = 0.049, power = 0.660. However, no significant age x gender interactive effect was found in any category of FoRs.

3.2.2. Developmental Patterns

To investigate the user preference for FoRs among children aged from 2–6 to 5–6, the headcounts and the related percentages of the children using each FoR category were also calculated. The increasing trends of users in each FoRs category across four ages were identified: DR (Aged 2–6 = 14%; Aged 3–6 = 19%; Aged 4–6 = 31%; Aged 5–6 = 43%); GR (Aged 2–6 = 2%; Aged 3–6 = 2%; Aged 4–6 = 7%; Aged 5–6 = 12%); LBR (Aged 2–6 = 12%; Aged 3–6 = 17%; Aged 4–6 = 14%; Aged 5–6 = 21%); OCR (Aged 2–6 = 43%; Aged 3–6 = 62%; Aged 4–6 = 69%; Aged 5–6 = 83%); PCR (Aged 2–6 = 0%; Aged 3–6 = 5%; Aged 4–6 = 12%; Aged 5–6 = 26%); RR (Aged 2–6 = 33%; Aged 3–6 = 45%; Aged 4–6 = 57%; Aged 5–6 = 79%); except the AR FoRs (no production) (Figure 3). This result indicated that not only the tokens of each FoRs category increased with age but also the number of children using each FoRs category increased from ages 2–6 to ages 5–6. To examine whether there is an interaction between age and the number of FoRs users, a chi-square test was conducted and the results indicated a significant relation between the target variables, X² (15, N = 168) = 25.431, p = 0.044. This result indicated that older children were more likely than younger children to use FoRs.

3.3. Distribution Patterns of FoRs: Missing, Allocation, and Preferences

Altogether six out of seven FoRs were identified in Beijing preschoolers’ utterances, including DR, GR, LBR, OCR, PCR, and RR. However, no AR expression was observed even though Beijing adults frequently use AR to express space and direction (Wu and Li 2017). All these findings jointly suggest that the proposed seven-class FoRs coding system is comprehensive and inclusive. As shown in

Table 4. Age differences in the ratios (proportions) of different types of frames of references (N = 168).

	Aged 2–6 (n = 42)			Aged 3–6 (n = 42)			Aged 4–6 (n = 42)			Aged 5–6 (n = 42)
	Rsum	M	SD	Rsum	M	SD	Rsum	M	SD	Rsum	M	SD	F	Sig.	Msum
R_DR	1.455	0.035	0.103	1.404	0.033	0.109	1.009	0.024	0.047	0.773	0.018	0.025	0.399	0.754	1.161
R_GR	0.100	0.002	0.015	0.333	0.008	0.051	0.322	0.008	0.028	0.095	0.002	0.007	0.454	0.715	0.213
R_LBR	0.519	0.012	0.044	1.033	0.025	0.070	0.313	0.007	0.020	0.396	0.009	0.022	1.283	0.282	0.565
R_OCR	5.689	0.135	0.224	5.848	0.139	0.169	5.515	0.131	0.118	6.136	0.146	0.108	0.063	0.979	5.797
R_PCR	0.000	0.000	0.000	0.125	0.003	0.014	0.299	0.007	0.023	0.518	0.012	0.026	3.567	0.015	0.235
R_RR	3.046	0.073	0.172	3.865	0.092	0.145	3.332	0.079	0.095	3.085	0.073	0.069	0.211	0.889	3.332

Note. R_sum = ratio (Proportion) of target FoRs against total static spatial sentence; M_sum = the mean R_sum of four age groups. R_OCR: ratio of the object-centred reference; R_RR: ratio of the relative reference (RR); R_AR: ratio of the absolute reference; R_DR; ratio of the direct reference; R_GR: ratio of the geomorphic reference; R_LBR: ratio of the landmark-based reference; R_PCR: ratio of the people-centred reference.

, the percentage of the target FoRs’ categories over the total static spatial sentences was calculated for each child in this study. To explore the possible patterns of allocation in different FoRs produced by the four young ages, we calculated the means for each age group and the total sample, respectively. As shown in Figure 2, similar patterns of allocation were found in each of the seven categories of FoRs across the four ages: M_DR = 0.018~0.035; M_GR = 0.002~0.008; M_LBR = 0.007~0.025; M_OCR = 0.131~0.146; M_PCR = 0~0.012; M_RR = 0.073~0.092; M_AR = 0. For the four age groups, the most frequently used category of FoRs was OCR (M_sum = 5.797), and the second most frequently used FoRs was RR (M_sum = 3.332) (Figure 4).

The above results indicated that the four age groups shared the same pattern of allocation of FoRs categories; thus, no age differences were found in the pattern. Furthermore, the one-way ANOVA results indicated that there were no significant age differences in all the FoRs categories (p_s > 0.05) except FoRs in the PCR category (p < 0.05), indicating a similar stable pattern in the six categories out of the seven FoRs. Specifically, an obvious increasing trend of the proportion of the PCR per capita used in the four age groups was also found (M_aged2;6 = 0; M_aged3;6 = 0.003; M_aged4;6 = 0.007; M_aged5;6 = 0.012). This result indicated that the 5–6 age group had a significant increase in the percentage of PCR, indicating a preference for using PCR expressions among children aged 5–6.

4. Discussion

As the first corpus-based exploration of the naturalistic use of FoRs in Mandarin-speaking preschoolers, this study has confirmed the applicability of the seven-class coding system and identified the linguistic features of FoRs sentences, developmental patterns, and distribution patterns. This section discusses these major findings and their theoretical contributions and practical implications.

4.1. Linguistic Features of Spatial FoRs in Early Child Mandarin

First, this study has elicited 2837 static spatial sentences from the corpus and found that more than one-quarter of them (27.7%) used FoRs. The mean length of the utterance of these FoR sentences ranged from 5.80 to 10.40, with a relatively longer MLU in the RR, PCR, LBR, and OCR. In addition, the percentages of grammatically complete sentences for each class of FoRs increased dramatically after ages 4–6, with a low level at ages 2–6 or ages 3–6 and a peak level at ages 5–6. The RR category has seen the highest percentage of grammatically complete sentences during the preschool years. In contrast, the GR, LBR, and PCR FoRs were observed with the lowest percentages of grammatically complete sentences. It is important to note that high percentages of complete sentences were observed in each FoRs class at ages 5–6 (WSRs > 80%), especially for LBR and PCR (WSRs = 100%). This result indicated that MLU and the percentages of complete sentences, which are intercorrelated with each other in this study, varied across ages and FoR classes. Among the incomplete sentences, the most missing part was the ‘subject’, which could be commonly observed from all age groups in all FoRs categories. Moreover, the missing or inappropriate verbs could be identified from GR and RR use (see Example D and E). All of these findings suggest some interaction between the linguistic features, FoRs classes, and age groups. Chinese preschoolers might have individual and age differences in using FoRs, which require varying linguistic features. This finding implies that experiments on FoRs should consider the influences of varying linguistic features required by different FoRs in addition to the consideration of cognitive requirements (Galati et al. 2010; Iachini and Logie 2003).

Moreover, the preschoolers in this study did not use the word ‘left/right’ in expressing the relative class of FoRs. Two reasons might cause the absence of ‘left/right’ in this case. First and cognitively, Chinese preschoolers might not be able to identify others’ left and right before the age of five because of their egocentric thoughts (Zhou and Boehm 2001). Therefore, they tended not to use this concept. Second and linguistically, the task in this study was a naturalistic situation; thus, it did not forcefully ask the young children to use the concept. Therefore, follow-up experiments will help to confirm the real cause of this absence.

4.2. Developmental Trends of Spatial FoRs in Early Child Mandarin

This study found significant age effects in the frequency of OCR, PCR, and RR produced by Chinese preschoolers aged 2–6 to 5–6, indicating a significant age effect. Moreover, it is observed that the total frequency and the number of users (headcounts) increased with age. The chi-square test identified a significant link between the FoR classes and the age group, indicating that older children were more likely than younger children to use FoRs. This finding bridged the gap of the acquisition pattern during the preschool stage, in addition to previous studies on primary students, which focused more on understanding how the acquisition of FoR expressions extends in the school years (Haun et al. 2011; Li and Abarbanell 2018; Ruggiero et al. 2016; Surtees et al. 2012).

First, there was a significant age difference in the production of OCR, which found the most frequently used FoRs class. OCR is a dominant frame for spatial descriptions in Mandarin Chinese (Liu 2014); thus, the age-related increase found in this study implied a developmental key stage of the preschool period regarding the acquisition of OCR, which is consistent with previous studies (Craton et al. 1990; Sekine 2009). Moreover, previous research has also revealed that infants as young as four months old could form OCR (Bremner et al. 2006). This study has extended its development to the period of preschool years. In addition, this study also found a great improvement in using OCR with a grammatically complete sentence between ages 4–6 and 5–6, indicating significant progress in expressing OCR in communication. This finding has provided empirical evidence to prove the critical period of age five in developing the intrinsic frames (Negen and Nardini 2015; Shusterman and Li 2016).

Second, this study revealed that the second frequently used RR also had a noticeable age-related growth among these preschoolers. In particular, the ratio of a grammatically complete sentence and MLU of RR reached a high level at a very young age (WSR for 80% and MLU for 7.40 at ages 2–6, see Table 1). This result indicated that RR might have been acquired by the Chinese preschoolers, who are younger than American primary students in Surtees et al. (2012). However, the latter was an experimental study that tested primary (6- to 11-year-old) and college students’ use of different FoRs when making judgments about descriptions of social and nonsocial scenes. Future studies with comparable designs should be conducted to see whether this difference is still valid between Chinese and American children.

Third, this study has also found a significant age effect in the production of PCR. In particular, no children aged 2–6 produced any utterances of PCR; however, from ages 3–6, it was frequently used. This might be due to the development of peer play: coordinating with peers in the play context is challenging for children under three years old, as the late development of joint commitment leads to a preference of parallel play by the two-year-olds who may monologue but not interact with the peers (Brownell and Carriger 1991; Hamann et al. 2012). In this study, although the two-year-olds produced utterances, they might not have much to do with the interactive conversation involved with the peers using the words “you” or “other people”. This study also found that the PCR expressions were identified from ages 3–6, who may start to realise their peers and interact with each other so that the PCR expressions were used, which is consistent with the previous studies of children’s peer play (Brownell and Carriger 1991; Gräfenhain et al. 2009).

4.3. Preferences of Using Spatial FoRs among Chinese Preschoolers

This study found a stable allocation pattern by age, and the most frequently used FoRs was OCR. This finding is consistent with the study of Chinese adults (Liu 2014), providing empirical evidence to support the conclusion that the use of FoRs might be jointly affected by the linguistic features and the nature of the communication task. However, this study also found that children’s preferences of spatial FoRs mildly differed from adults (Liu 2014). In particular, the second most frequently used FoRs was RR, and no AR was found from Beijing preschoolers. In contrast, Liu (2014) found that AR was the second most frequently used FoR by Beijing adults. This difference in AR production might be caused by the different language materials used in the two studies: Liu (2014) used written text, whereas this study used natural and spontaneous spoken language. According to Halliday (1985), people may shift flexibly in choosing lexical terms due to register variation corresponding to various situations. Conventionally, the AR involved with terms describing the cardinal directions such as north/south/west/east occurs more prevalently in written Chinese than spoken Chinese. However, Beijing adults have been observed to produce more AR in spatial descriptions in their daily life, especially in wayfinding tasks (Wu and Li 2017). Therefore, it is natural to expect that Beijing preschoolers should also produce AR in their natural conversation in this study. But this study found that Beijing preschoolers never used the absolute spatial FoRs in a free play context, indicating that their cognitive level might have constrained the influences of culture on language. In addition, the absence of AR in this study might be because the language task in this study did not offer cues for AR expressions. Either of the interpretations could be acceptable in this study. Therefore, future studies using more spatial descriptions with cardinal directions across ages should be conducted to confirm the real cause.

In addition, this study found a noticeable difference between adults’ and children’ use of RR. Liu (2014) found that the Chinese adults demonstrated preferences in using the OCR and AR but not in using the RR. The reduced use of RR in adults has been discussed by other researchers (Janzen et al. 2012; Shusterman and Li 2016). As Shusterman and Li (2016) concluded, “to reduce miscommunication and ambiguity, perhaps it is simpler if languages do not make use of relative frame of reference expressions when intrinsic or geocentric frames of reference are available, sufficient, and unambiguous for indicating directions and locating objects”. In other words, as Chinese is a language that heavily uses intrinsic frames of references (especially the intrinsic OCR, see Liu 2014), the Chinese adult speakers may relatively reduce the relative FoRs to avoid miscommunication and ambiguity. In contrast, the preschoolers in this study preferred to use both OCR and RR, with the statistically significant age-related increasing trends. This phenomenon might be related to the fact that although the Chinese children were rapidly developing their OCR expressions, they still had to depend on the RR expressions. In the same vein, there might be a critical period after the preschool stage, when the children learn to overcome their dependence on using RR. However, when and how the Chinese children abandon such a kind of dependence deserves further investigation.

Last, this study also found that the four age groups followed the same stable ratio of FoRs production (except for the PCR): the children aged between five and six had a significant sharp increase in PCR, while the other three age groups just demonstrated minor differences. More interestingly, there was no PCR expression identified in the group aged between two and six. PCR refers to the ‘people-centred reference’ developed in this study to refer to a situation that treats both the listeners and peoples as an independent class of FoRs. The absence of PCR in the two-to-six-year-old group indicates that the younger children might not be able to take other’s perspective to mention the direction and location. According to Piaget, young children do not have the perspective-taking ability until age five (Newcombe and Huttenlocher 1992). Therefore, it might be the limited cognition that had ‘limited’ the use of PCR in the two-to-six-year-old age group. However, more experiments with younger children should be conducted to confirm this.

5. Conclusions, Limitations, and Implications

This study has proposed a seven-class FoRs coding system and found that the Chinese preschoolers produced all the classes (except for the absolute reference) in the corpus. The absolute reference has been widely used by Beijing adults (Li and Zhang 2009) but not used by the Beijing preschoolers in this study, indicating that either the cultural influences might not have taken effect in the early years of language performance or that they might be constrained by young children’s cognitive development. In addition, significant age differences were found regarding the no-reference sentence, reference sentence, and demonstrative sentences. Significant age effects were also identified in the production of OCR, PCR, and RR. However, there were neither significant gender differences nor significant age x gender interactive effects in the production of FoRs, except for the use of no-frame sentences. Last, this study found a stable and consistent pattern of allocation of FoRs in the early years: the most frequently used was OCR. This finding indicated that there might be a constant distribution pattern in a given communication task. All these contributed to a more comprehensive understanding of production patterns of FoRs in early childhood Mandarin Chinese.

This study has some limitations. First, the corpus collected Chinese children’s utterances from their naturalistic conversations and thus might have only documented a particular portion of all the linguistic evidence about the spatial FoRs we expected to gather. In other words, the lack of specific classes of FoRs in the corpus does not necessarily indicate the children’s inadequate knowledge of them. Therefore, the best solution might be the combined use of naturalistic and experimental data in a cyclic fashion, as suggested by Lozano and Mendikoetxea (2013). Second, as a free play session for the young children participating in the study was limited to 30 minutes, during which they were accompanied only by limited number of toys and a counterpart from the same age group, they might not have talked as much spatial language as they would be in a more relevant language context, given that context shapes language use (Hoff 2006). Third, the cross-sectional nature of the data collected in the corpus might have also constrained us from making consolidated conclusions about children’s acquisition trajectory of spatial expressions. Future studies may consider employing a longitudinal design to further investigate how children progress through increasing ages of developing spatial FoRs and their developmental patterns. Fourth, limited by the nature of communication tasks in this corpus, no significant age differences were found in some of the FoRs classes (i.e., DR, GR, LBR). This non-significant finding might have two conflicting interpretations. The positive one is that the young children might have already commanded the related expressions of DR, GR, and LBR before ages two to six, whereas the negative one is that they might not have acquired the skills before ages five to six. Therefore, future studies are needed to examine the use of DR, GR, and LBR in children younger than two to six or older than five to six.

Accordingly, the findings of this study have implied the future directions on this topic as follows. First, a fusion of both corpus-based and experiment studies should be conducted to further explore the developmental trends of FoRs and the interaction between language, culture, and cognition. Second, longitudinal studies should be conducted to explore how young children develop their knowledge and skills using spatial FoRs in their younger ages (under ages two to sic) or primary school years. Extending this age range will allow us to ascertain whether the absence of DR, GR, and LBR during ages 2–6 to 5–6 might be a negative or positive indicator. Last but not least, the objects involved in this corpus-based study might have different geographic properties, thus causing different levels of linguistic difficulty. Therefore, future studies should compare the geographic properties of the referent objects (ground), which will be analysed and presented in another article.