Linguistic Repertoires Embodied and Digitalized: A Computer-Vision-Aided Analysis of the Language Portraits by Multilingual Youth

Abstract

The present study sets out to investigate how multilingual youth perceive and represent their linguistic repertoires. To achieve this goal, we introduced a computer-vision-aided analytical method to deal with the obtained visual data, which comprised digital images of language portraits created by a group of young multilingual speakers. An OpenCV module is used to build and complete the graphic data processing, enabling quantitative evaluations of participants’ colored clusters and linguistic codes that express their language repertoires. In combination with oral narratives provided in their language portraits, the findings demonstrate that Macanese heritage speakers show a higher degree of “scope” than the Chinese mainland sojourners in Macao but a lower degree of “access”. Follow-up interviews further corroborated the self-perceptions of their linguistic resources across different registers. Overall, the computer-vision-aided analysis of language portraits enhances the current understanding of the “scope” and “access” of multilingual repertoires in lived experience.

1. Introduction

Moving into the 21st century, the special administrative region (SAR) of Macao in China has been transformed from “the global village into a migrant metropolis” [1] (p. 146). After its handover to China in 2009, Chinese and Portuguese have been enacted by the Basic Law of the Macao SAR as the official languages. Over the years, the territory has been characterized by “the wide use of English, Portuguese, and other foreign languages” [2] (p. 568). In its sociolinguistic contexts, Cantonese is the mainstream spoken language tightly bound up with Macao’s identity [3] (p. 78). Increasingly, Mandarin has been commonly used for intimate connections with the Chinese Mainland following the SAR government’s strategic plans to diversify its gaming and tourism sectors. Though Portuguese has official status in Macao, it is noted that “it’s getting dismissed and turned into a foreign language” [2]. The language policy of Macao features a plurilingual practice of “three written languages, four spoken languages” (“三文四语”; san wen si yu), indicating Chinese, Portuguese, and English as three written languages, and Cantonese, Mandarin, Portuguese, and English as four spoken languages. Such a linguistic diversity policy contributes to shaping Macao’s cultural distinctiveness and preserving its language ecology [4].

Previously, some studies have discussed the representation of Macanese during the colonial period [5] and investigated the identity construction of Macanese inside and outside of Macao [6,7]. Some have pointed to the social dilemmas faced by Macanese adolescents in the postcolonial era [8,9]. To date, only a few studies have focused on the Macanese ethnic group, particularly in the post-colonial era. The literature has extensively explored the language styles and linguistic features exposed by the users who live in Macao [3,10] and the interactive interplay between Asian (local Chinese) and occidental (immigrated Portuguese) communities [11]. Some investigations have discussed their social identity and lived reality in Macao in the post-colonial era [6,7]; however, thus far, little research has focused on the individual experience of this particular ethnic group in today’s multilingual Macao society [10]. An in-depth analysis is still needed to show updated information about their language use and perceptions in multilingual circumstances.

In this study, we attempt to explore how Macanese youth perceive and represent their linguistic repertoires in lived experience across different social contexts in multilingual Macao. To achieve this goal, we introduce an art-informed method of telecollaborative mediation through language portraits [12,13] as our research instrument. Additionally, we employ computer-vision-based image analysis for data analysis. Using the OpenCV framework and K-means algorithm, colored shapes and verbal annotations in language portraits can be objectively analyzed through the computational processing of digitalized images. Regarded as an interdisciplinary attempt in the present study, this method has been evidenced by several studies [14,15], which show that the method can provide an informative clue for exploring human perceptions and facilitate a quantitative evaluation of artworks and a qualitative analysis of digital drawings. We hope to demonstrate that this technology-aided analytical method provides an innovative methodological framework for investigating linguistic repertoires.

Moreover, we also triangulate the data analysis with interviews in the hope that oral narratives from the participants serve to shed further light on the subjectivity of linguistic relations across various social contexts on the basis of the current concepts of “scope” of and “access” to linguistic resources under sociolinguistic relations [16,17]. The “scope” refers to the social functions that specific linguistic resources have [17] (p. 154), and the “access” refers to the possibility of acquiring more linguistic resources in a specific social context. These two concepts link an examination of individual multilingualism in the context of socioeconomic connections with the analysis of linguistic repertoires and linguistic practice.

2. Background

2.1. Theoretical Background: From Linguistic Repertoires to Multilinguistic Repertoires

Coined in the context of language sociology, the term “linguistic repertoires” refers to “the set of skills and knowledge a person has of one or more languages, as well as their different varieties” [18]. Busch [12] argued that one’s linguistic repertoires could be shaped by subjectivity and potentiality as individual thoughts, perceptions, experiences, and emotions elaborate the assemblage of linguistic resources, thus foregrounding a subjective viewpoint toward this/these potential language(s). In parallel, Lüdi [19] and Vertovec [20] emphasized that linguistic repertoires should be seen as neither static nor clear-cut entities of languages when triggered by current perceptions in a varying space of communications.

Two recent advances in the concept of linguistic repertoires are particularly noticeable and relevant. First, the concept is broadened from the perspective of multilingualism. Language usage is growing in a productive and diverse way as a result of the social phenomenon of worldwide expanding mobility. These practices progressively bring a variety of linguistic resources into the context of meaning negotiations and meaning creation. Källkvist and Hult [21] regard the value of linguistic repertoires as a collective resource in favor of the idea of individual multilingualism. Building on such a multilingualism perspective, the notion of “multilingual repertoires” emanates from the original concept of linguistic repertoires, indicating an explicit sharing of knowledge of any language in discursive practices [13]. The employment of diverse linguistic resources, namely multilingual repertoires, facilitates interlanguage comprehension and privileges “cross-language transfer” under the condition of super-diversity [22]. The concept of linguistic repertoires is broadened and evaluated differently in sequential multilingual practices, as they are intricately linked with the movement of personal experience and life trajectories in multilingual social contexts [12].

Another recent development in the concept of linguistic repertoires relates to the inquiry of “How is an individual positioned to participate and pronounce their interests” [17] (p. 5) in multilingual contexts? To answer this, two more notions were brought into the discussion: scope and access. More specifically, scope is defined as “the social functions that specific linguistic resources have” [17] (p. 154). Linguistic resources are comprehended as essential elements that constitute an integral linguistic repertoire for the construction of meanings. In this sense, the term serves as the range of containment of all linguistic resources currently available for communicative needs. In different social contexts at the individual level, the scope can also be understood differently, as it is constantly influenced by linguistic resource mobilization in the flow of interpersonal communications [23]. Weirich [16] took minority people’s language use in Moldova as an example to analyze why their linguistic resources have become limited and restricted from family to public settings since the independence of the Soviet Union. The case study implies the existence of sociolinguistic inequalities due to the installed official monolingualism in local language policies. The study reviews scope as a political stance and argues for the importance of multilingualism in language policy and planning instead of waving the “first-language-only” flag. By supporting the use of all languages, minorities’ scope of linguistic resources can be more expansive, and this minority group has an opportunity to make their voices heard in society.

The concept of access refers to the “possibility that speakers can acquire other linguistic resources” [16] (p. 59). It is argued that “[a]ccess to linguistic resources depends on the opportunities a speaker learns certain languages, registers, and linguistic forms” [17] (p. 164). For example, when an individual is situated in a specific register, such as in a multilingual classroom, he/she feels motivated to talk with classmates and teachers who speak different languages. Consequently, this opens up a new opportunity to connect with other linguistic resources that he/she may not have acquired and used fluently. In this way, his/her linguistic resources are possibly reaching a higher degree of access.

Overall, these two abstract notions can be comprehended metaphorically, as in the example given by Li Wei [24] for the enlightenment of the concept of translanguaging. If we consider an individual language user as a conductor whose linguistic repertoires are constituents making up the orchestra and whose linguistic resources are all composited musical instruments, then scope refers to the kinds of resources and the many musical instruments the orchestra contains. On the other hand, access indicates the other instruments that can potentially be utilized in this orchestra to perform a more harmonious melody—that is, flexible and fluid use of all existing languages and resources. Building on these notions, we argue in this study that the concepts of “scope” and “access” can serve practically as cognitive scale tools and help to (self)evaluate one’s linguistic repertoires and competence. We concur with previous scholars that the two concepts are important to help students engage in language learning and promote critical thinking and reflective practices [25].

Previous research has shown how linguistic repertoires are unpacked and discussed in the multilingual context of the Macao region. For instance, Zhang [26] used semi-structured interviews and a language survey to understand mainland Chinese students’ perceptions and language use experience when sojourning in Macao. Wheeler [3] adopted a questionnaire survey to explore the difference in language attitudes between mainland Chinese students and local Macao students. This research adopted survey methods to deal with linguistic/multilinguistic repertoires of college students. Nevertheless, more and more researchers have noticed the limitations of using questionnaires and interviews to collect and analyze data. Young participants can get bored and fatigued and lose the patience to respond carefully because of the standardized and highly structured designs. Thus, these previous methods may have limited value in “covering complex topics and extracting students’ cognitions, such as feelings, motivations, attitudes, accomplishments, and experiences of individuals” [27] (p. 14).

2.2. A Computer Vision-Aided Analysis of Language Portraits

“Language Portraits” (LPs) is a term initially proposed by Krumm and Jenkins [28] to investigate multilingual youths’ language awareness and was further popularized by Busch [12] as a research method to elicit a subjective reconstruction of their language acquisition [29]. As a methodological approach, it is shaped by an empty body-shape silhouette. Participants can then paint the body with languages and codes they auto-categorize and prefer to illustrate as a reference to different languages or modalities. Following the theoretical underpinnings of the “visual turn” for applied linguistics [30,31], LPs help researchers to understand individual linguistic repertoires and explore their perceptions of language planning, learning, choices, and use, all associated with their personal lived experiences.

Currently, a growing number of LP-oriented studies have adopted qualitative ethnographic and biographical methods to investigate the construction of multilingual repertoires and self-identity. For instance, Prasad [32] discovered how identities and language ideology are constructed at an individual level by the influences of globalization. Yoshida and Nichols [33] and Soares et al. [34] demonstrated how youths’ experiences influence their language. Johnsen [35] examined young adults’ linguistic practices in the heritage family context, and Galante [36] investigated how college students enhance their representations based on plurilingual tasks. Kusters and Meulder [31] and Muller [37] highlighted communication in the process of LP drawings. However, the majority of related research involved the LP method in traditional artmaking by the use of pencils and worksheets contextualized in classroom settings. In multimodal digital drawings, young adults are more likely to be empowered to create a new medium of self-expression with a broad range of online resources. Technology-based drawings are the “most effective way to convey visual thoughts” [38], stimulating students’ creative thinking and enhancing their performance in terms of imagination and skills development [39]. Nevertheless, little research has adopted LPs in digital mode via computer-mediated communications, which creates a research space [40] for the present study to explore the differences among participants’ perceptions of various languages and the specific features of different language resources from their own point of view.

To date, LPs have mainly been approached and analyzed in qualitative and interpretative ways, mainly through content analysis, drawing analysis [36], narrative analysis [34], multimodal analysis [31], and storytelling [32]. Nevertheless, it is crucial for visual data to be indicatively interpreted to assure the validity of the creators’ language profiles [32,41,42]. With regard to this concern, the computer-vision technique has shown its potential capacity to indicatively evaluate graphic representations of students’ linguistic repertoires, which significantly enhances the visualization of their multilingual resources.

Computer vision (CV) is a well-researched topic and has impacted the real world. In accordance with the definition, CV is the use of images and videos to depict concepts and actual situations [43]. Digital images in the visual world reflect the nature of human cognition and the relationship between technologies and humanities. The goals of the CV approach are to present and understand concepts on the basis of images and videos. There are many algorithm tools that can be used for CV-related tasks, and one of the most commonly used is called “OpenCV” (Open-Source Computer Vision). This resourceful open-access library includes a wealth of pre-defined functions to solve different types of CV-focused tasks. In the case of dealing with digital images, K-means clustering is commonly used. Specifically, K-means refers to collecting data points aggregated together because of certain similarities. By extracting and analyzing image information through OpenCV and K-means color clustering, numerous interpretations could be given to an image in a reorganized manner [44].

Prior to our applied OpenCV and K-means analytical method, researchers had mainly relied on the popular tool of the Statistical Package for the Social Sciences (SPSS 29.0, IBM, Armonk, NY, USA) to analyze language portraits [45]. However, statistical methods require preprocessing of language portrait images by taking inputs in a spreadsheet-like manner. As

Table 1. Comparison of computer-vision and statistics tools.

	OpenCV + K-Means	SPSS
Scope	Machine learning algorithm	Statistical software 29
Input	Pictures	Numeric data
Output	Prediction	Inference
Example	Detect and classify objects	Find probability
Robustness	High	Low

shows, OpenCV-mediated analysis can directly read images as input instead of manually labeling and measuring images before being processed by SPSS. OpenCV is generated by machine learning, concentrating on the prediction of collected raw data, while SPSS focuses on population inferences from a numerical sample [46]. When dealing with a dataset whose scope is not large enough, the dominant advantage of using OpenCV for image analysis is demonstrated by more persistent robustness. Robustness refers to the “ability of a system to resist change without adapting its initial stable configuration” [47] (p. 887). Taking the example of digital LPs, when a small scope of image dataset shows irregularity of filled colors, OpenCV and K-means can precisely detect those colors positioned inside the LP silhouette and evaluate the exact percentage of different colors. However, this is impossible for SPSS.

In recent years, it has been noted that using CV methods in the fields of art and humanities, especially in language education, has become an innovative and emergent approach that has attracted attention in academia. For instance, Agbo and Agbo [48] designed a CV-aided translational system that implicated the possibility of incorporating CV in the development of English teaching and learning. Hiippala [49] developed a method that takes advantage of CV techniques that can potentially contribute to restructuring bilingual children’s books for pedagogical purposes. Regardless of the capacity of computational approaches in language research, few studies have leveraged the advantages of CV techniques to explore bi/multilingual users’ inner space. As an innovative method, the adoption of the OpenCV and K-means algorithms in linguistic analysis succeeded in converting “direct image thinking” (qualitative) into readable numeric data (quantitative), thus pushing forward the application of the “think-aloud” protocol (TAP) in linguistic studies and second-language research (e.g., Bowles [50]). It is hoped that this innovative method will contribute to bringing out multimodal meaning-making through technology affordance.

3. The Current Study

In the present study, we conduct an art-informed activity and a CV-aided analytical method to explore Macanese multilingual young adults’ linguistic repertoires and perceptions of their linguistic resources. Specifically, we aim to answer two research questions:

• Compared to the Chinese Mainland sojourners and local Macao heritage speakers, what characterizes the linguistic repertoires of the young Macanese speakers?
• How do these Macanese youth perceive their “scope” and “access” of linguistic resources across different social registers in lived experience in multilingual Macao?

3.1. Participants

The participants are 1st-year students of the Chinese–Portuguese Translation and Interpreting under- and post-graduate programs at a university in Macao. Both programs operate in a bilingual pedagogical scheme that employs Portuguese and Chinese as the mediums of instruction. The data reported in the present study were collected at the end of the second semester of the 2021–2022 academic year. Although all the students come from either Macao or mainland China, their ethnic, sociolinguistic, and educational backgrounds differ. Eight students were recruited for the study: four undergraduates and four postgraduates. The home language use of the participants was either monolingual or bilingual, and their educational background was predominantly bilingual or trilingual. Their demographic information is reported in

Table 2. Demographics of participants.

Pseudonym	Gender	Age	Ethno-Linguistic Identity	Region of Origin	School Type Background
U1	Female	20	Mandarin Chinese	Heilongjiang	Dominantly Chinese
U2	Female	19	Cantonese Chinese	Guangdong	Dominantly Chinese
U3	Female	18	Macanese	Macao	Chinese, English, and Portuguese
U4	Female	18	Cantonese Chinese	Macao	Chinese and English
P1	Female	23	Cantonese Chinese	Macao	Chinese and English
P2	Female	24	Cantonese Chinese	Macao	Chinese and English
P3	Female	27	Macanese	Macao	Chinese and Portuguese
P4	Male	25	Macanese	Macao	Chinese, English, and Portuguese

.

Prior to their participation, the objectives, procedure, data collection process, and purpose for data collection were explained to all participants [51] (p. 210). Then, participants signed the informed consent form [52]. To preserve privacy, participants’ names were blinded and coded as “U1, U2, U3, U4” for undergraduate students and “P1, P2, P3, P4” for postgraduates. Students in the survey had the choice to respond or not to respond to the questionnaire, and they also had the right to withdraw from the study for any reason. In the interviews, participants’ confidentiality was protected, and the data was de-identified [51] (p. 211).

3.2. Tasks and Instruments

3.2.1. Background Survey

The questionnaire of the background survey asked questions regarding participants’ Portuguese name, birthplace, place of residence, age, sex, and self-reported first language, as well as their education profile before university, their study abroad experience, and their general language use of socialization.

3.2.2. Language Portrait Task

Since the study was conducted during the outbreak of the pandemic in Macao, a virtual collaboration through synchronic computer-mediated communication was adopted. Participants were enrolled in two language portrait tasks. They were first asked to design their LP based on a blank human body silhouette via “BoardMix” (official website that supports the use of online telecollaborative platform: https://boardmix.com/ (accessed on 1 July 2022) using painting tools. Then, they were allowed to decorate their LP with all iconic materials indexed in the web-based platform. A task example is provided in Appendix A.

In the end, every participant submitted two versions of language portraits: one filled with colors that contained language(s) label(s) based on the body silhouette [13], and the other illustrated with semiotic symbols on top of their digital portraits, such as icons, emojis, and images.

3.2.3. Follow-Up Interviews

Adopting the research method of previous studies applying language portraits [13,34,53], we interviewed three individual Macanese participants (coded as U3, P3, and P4 in Table 2). Each interview lasted approximately 25 min. The detailed arrangements are presented in

Table 3. Duration of language portraits tasks and interview.

Participant	Duration of LPs in Task 1	Duration of LPs in Task 2	Duration of Interviews
U3	20 min	35 min	25 min
P3	18 min	14 min	25 min
P4	15 min	25 min	25 min

so that the researchers could collect their self-explanations about the placement of languages and the choices of representative colors on individual language portraits.

3.3. Data Collection Procedures

Participants were asked to complete their language portrait tasks independently. They could ask the researcher (the first author) questions concerning task rules and technical problems. In the first LP task, participants were asked to use their tablets and electronic pencils to fill the blank body template with colors marked with language labels. The second task was based on their finished LP work. Participants were allowed to pick various stickers and face emojis provided freely by the online collaborative platform. Meanwhile, participants were also allowed to use external resources if needed. After submission of their LP works, participants were asked to fill the “scope and access” charts independently. Once the tables were completed, the researcher held semi-structured interviews with every Macanese participant. A digital screen capture method was used to record participants’ behaviors while painting the LPs to collect their drawing results. During the sessions of the interviews, the dialogues were recorded automatically by the audio-conferencing software and saved in cloud storage.

In total, eight digital LP sheets were completed by all participants in the first task, with three additional brief self-explanations in texts from three Macanese participants. Then, nine filled charts (three by each participant) of “scope and access” and verbal interviews were collected for analysis.

3.4. Data Analysis

3.4.1. Image Analysis through OpenCV and K-Means

The CV-based approach can handle a massive amount of image data, so our current work is a tentative experiment for future work on handling more images. From the participants’ language portraits, we turned the images into readable numerical data to extract information. The language portraits are in digital format and made of massive pixels. We grouped similar pixels into clusters and labeled clusters with participants’ designated languages. Clustering in machine learning is a well-researched topic, and the algorithm is a typical example of unsupervised learning [54]. As such, clustering divides data and helps to find the data’s inherent structure. The key in this process is K-means, which is the most popular clustering algorithm [55].

We now describe the K-means algorithm and how we utilized the algorithm for our data analysis. First, we decided on the number of clusters. After several trials, we found that the setting of 11 clusters achieved the best result. We randomly put a centroid on the language portrait graph F and assigned each pixel to the cluster with the nearest centroid. If the clustering result could be improved, we updated and refined the centroids by moving them toward the means of their clusters. After we finished the clustering, the total number of pixels in each category was obtained by traversing the clusters and calculating the total number of image pixels and the proportion of pixels in each category. By setting the threshold of the pixel ratio and color variance, we filtered distracting colors such as white, black, and gray. Finally, the statistical results are presented intuitively in the form of pie charts by using the Python drawing library [56]. The analysis procedure is visually depicted in Figure 1 (in which “廣東話” means “Cantonese”, “日本語” means “Japanese”, “普通話” means “Mandarin” and “Português” means “Portuguese”). The analytical process is illustrated by the K-means flow chart retrieved in the Supplementary Materials.

3.4.2. Discourse and Register Analysis

Discourse and register analysis were conducted in the present study as a qualitative method to interpret participants’ perceptions, beliefs, and emotions. On the one hand, discourse analysis involves identifying the relationship between language and the situational context in which language is used [57]. On the other hand, register, as a concept in systemic functional linguistics (SFL), is defined as “a configuration of meanings that are typically associated with a particular situational configuration of the field, mode and tenor” [58] (p. 39). Register analysis was adopted in this study because it helped categorize different uses of languages in specific contexts.

Widdowson [59] argues for register analysis as “instances of linguistic usage” and discourse analysis as “instances of communicative use” (p. 39). In light of this view, we consider the combination of both discourse and register analytical methods as a cognitive approach for better interpretations of participants’ verbal discourse. In so doing, we aimed to take into account the importance of both sociological and textual analysis. Therefore, discourse and register analysis were conducted in the present study as a qualitative method to interpret participants’ perceptions, beliefs, and emotions about language resources and use.

4. Results

In this section, we present the results in three main themes and provide a detailed description and interpretation of the study’s main findings. More specifically, our discussion aims to answer the two questions proposed earlier in the article.

Figure 2 shows the raw data from participants (a, b, c, d, e, f, g, h). In each figure, the left image is the portrait drawn by participants, and the pie chart is the proportion of different colors. We divided the pie charts into three categories: Chinese mainland students, Macao resident students, and Macanese students. The “LP counting” approach helps us to explore the language relationship established between the center and the periphery [45]. As such, the unequal distribution of percentages for different languages is visually linked with the frequency of occurrence and dominance level. CV-aided image analysis reflects, in a quantitative manner, the linguistic repertoires in two characteristics: richness and openness.

4.1. The Richness of Macanese Youths’ Linguistic Repertoires

Richness indicates the language variety and diversity one’s linguistic repertoires possess [60]. The higher multiplicity of linguistic resources is expressed in one’s language systems, the higher heterogeneity is demonstrated in the linguistic repertoires and embedded in their language practice [61]. By counting several named languages labeled in LPs, we can analyze the richness of linguistic resources in three groups. As Figure 3 shows, we can generate the following findings: the percentages of Mandarin and English in Chinese mainland PFL students’ language portraits are given more shares than Cantonese and Portuguese. Moreover, in the item “Others”, no other language is considered one part of their linguistic repertoires. The percentages of Mandarin (33.02%) and English (30.74%) are similar, which is understandable because Mandarin is the official spoken language in mainland China, and English is included in the curriculum reform of China in the nine-year compulsory education system and three-year high school teaching curriculum. In addition, the two languages (Chinese and English) are mandatory subjects for most Chinese students, as required by the National College Entrance Examination (known as the Chinese “Gaokao”). On the other hand, in PFL Macao students’ (of purely Chinese blood) self-portraits, Mandarin is mentioned less frequently (17.33%) than Cantonese (21.67%), English (28%), and Portuguese (27.81%). We also noted that, despite Cantonese being the most used language in Macao Chinese youths’ families, its percentage (21.67%) is lower than the other two languages, Portuguese (27.81%) and English (28%). For Macanese participants, since Portuguese is their heritage language and Cantonese is their mother tongue, Cantonese (36.69%) and Portuguese (30.09%) assume most proportions in their LPs, as expected.

4.2. The Openness of Macanese Youths’ Linguistic Repertoires

Openness highlights the degree of acceptability regarding the use of different languages and accessibility to multilingualism [62]. A speaker who has higher openness is more likely to engage in the communication process in multilingual circumstances, thus leading to the “gradual development of shared and commonly understood ways of communication” [63] (p. 7). We calculated the variance among these five categories of data (shown in Figure 3) and found that Macao Chinese residents value different languages with less discrepancy in terms of use frequency and importance since the variance (s2 = 88.48847) of their LPs is significantly lower than those of the other two groups. Moreover, the variance of Chinese Mainland participants (s2 = 173.54192) is slightly lower than that of the Macanese group (s2 = 181.94372). In other words, Macanese participants present a biased variance, as shown in Graph 1 by the unbalanced columns, whereas Cantonese (36.69%) and Portuguese (30.09%) register higher percentages. In contrast, English (18.83%) is relatively low, and Mandarin (10.23%) is the lowest.

With the variance analysis based on OpenCV data, we can access a holistic understanding of the openness of the three types of groups towards the languages mentioned above, which further probes into the language hierarchy at the individual level [64]. As the formation of individual language systems is closely related to personal lived experiences and prescriptive language policy, it is understandable that a higher value towards spoken languages would be distributed by priority than those rarely used in primary socialization.

Data retrieved from interviews also lent support to findings on the Macanese young adults’ lower openness to languages, showing a strong inclination to Cantonese and Portuguese.

Excerpt 1: (from Isabel) 我係澳門出世同埋長大嘅，呢度係一個雙語制度嘅地方，自然識聽同埋講廣東話同葡文。(⋯) 我對呢兩種語言接受度更高，因為係我嘅母語同埋接觸最多嘅語言。

[Literal translation: I was born and grew up in Macao, this is a region administered in a bilingual system, so naturally, I can understand and speak Cantonese and Portuguese. My acceptance of these two languages is higher because they’re the first languages I have met and started to learn.]

Isabel explained her language portrait in terms of the attachment of importance to Cantonese and Portuguese because of familiarity and dominance in her daily life within the social-cultural milieu of Macao. In this sense, the openness of one’s linguistic resources is influenced by various factors, such as the immersed living environment, the family language policy, and the language translation policy.

4.3. The Scope and Access of Linguistic Resources by Macanese Youths

To respond to the second research question, we applied a qualitative discourse and register analysis [58,59] to determine two of the Macanese participants’ perceptions of the availability and accessibility of linguistic resources and to trace their perceptive changes across different registers (from the spoken and informal register to the academic and formal register).

Contextualized in the family and public places, in

Table 4. Isabel’s perceived degrees of scope and access to linguistic resources in family and public places.

	Cantonese	Mandarin	Portuguese	English	Other: ______
Scope	High	Low	High	Medium
Access	High	Low	High	Medium

, Isabel (coded as P4) perceived the scope of her Cantonese and Portuguese as “high” and English as “medium”. In her opinion, Mandarin registers a low level due to less frequency and less need for usage. Meanwhile, she also evaluated the degree of access to the four linguistic resources in coincidence with her choices in the scope.

When the setting is changed to the classroom of her specialist subjects, her perceptions experienced a change. In the classes where Portuguese and Chinese are the primary medium of instruction and professional knowledge, her Portuguese scope became lower since she has no required register to present professional knowledge in translation classes in Portuguese. Under the full support of the program trainer, she felt motivated to elevate her access to extend Portuguese linguistic resources.

Excerpt 2: Acho estou competente em falar português com o meu avô, meus colegas na empresa e nos serviços públicos, mas na aula de aula de interpretação, não estou! (…) A dificuldade de dominar bem a terminologia especialmente nas áreas económica e industrial é enorme. (…) Apesar disso, a professora era muito simpática e sempre me presta a paciência em esclarecer dúvidas na interpretação. De forma qualquer, estou no progresso de ser um intérprete qualificado.

[Isabel: I think I’m competent in speaking Portuguese with my grandfather and colleagues in the company and public services, but in the interpreting class, I’m not! (…) The difficulty of dominating the terminology well, especially in economic and industrial areas, is quite significant. (…) Even so, my teacher was kind and always responded to my questions by interpreting with patience. No matter what, I’m on the way to being a qualified interpreter].

Across different registers, individual perceptions of scope and access can be changed because linguistic capacity turns out to be limited, from a spoken register to an academic register, where high formality and textuality are needed. Another finding refers to Isabel’s perceived access to Mandarin. In the family context and socialization, she hardly speaks Mandarin, but in the classroom, where a number of her classmates come from mainland China, she has more chances to hear and use it.

Excerpt 3: O meu Mandarim é malíssimo… Todas as minhas colegas sabem…Pá, (…) Quando a professora falava com os outros em Mandarim, estava distraída em outra coisa. Mas a minha colega,/amiga, deve ser, ajuda-me em perceber o que estão a conversar e até ensinar-me mandarim no tempo livre. (…)

[Isabel: My Mandarin is the worst… All of my classmates know this. (…) When the teacher spoke with others in Mandarin, I was distracted by something. However, my classmate, a friend she should be, helps me comprehend what they are talking about; she even teaches me Mandarin daily].

Hence, Isabel described her Mandarin as an accessible language in the determined situation (as shown in

Table 5. Isabel’s perceived degrees of scope and access to linguistic resources in classrooms.

	Cantonese	Mandarin	Portuguese	English	Other: ______
Scope	High	Low	Low	Medium
Access	Medium	High	High	Medium

); the help of her classmates allowed her to have the opportunity to have a voice [22] and to feel belonging to the group. The upgrade of access to particular linguistic resources promotes individual inclusion in the social community and enables the proliferation of multilingual communication [65].

By revisiting Julietta’s (coded as U3) scope and access in the family setting (see

Table 6. Julietta’s perceived degrees of scope and access to linguistic resources in the family setting.

	Cantonese	Mandarin	Portuguese	English	Other: ______
Scope	High	Low	Low	High
Access	High	Low	Low	High

), findings showed that not all Macanese youths affirm a high scope of their heritage language. Julietta also has the condition to learn and dominate Portuguese because her father is a native Portuguese speaker, but she perceived her scope as “low”. This seemingly paradoxical finding lies in the dominant American culture in her family, which directly influenced her language use and choice.

Excerpt 4: 屋企比較祟尚美國文化，因此，雖然我既父母都識講葡文，但係我地更多以中英兩語交流。

[Julietta: My family adores the US culture, so, although my parents speak Portuguese well, we talk with each other more often in Chinese and English].

Language maintenance or shift patterns are driven by both social and cultural factors [66]. In Julietta’s case, the predominance of the US culture enhanced her ability to use and acquire the English language.

Family is a crucial site for language maintenance, which in turn will influence the younger generations’ heritage language use [67]. Julietta also considers the low access to Portuguese the result of her family language policy. In Julietta’s home, the languages addressed bilingually are Cantonese and English, influencing her language choices in schooling.

As with Isabel, when the setting changed, Julietta’s perceptions shifted when she attended university and received systematic Portuguese instructions. As

Table 7. Julietta’s perceived degrees of scope and access to linguistic resources in the family setting.

	Cantonese	Mandarin	Portuguese	English	Other: ______
Scope	High	Low	Low	High	Medium
Access	High	Low	High	Medium	Medium

shows, Julietta considered high access to Portuguese because one of her teachers speaks Portuguese and Cantonese fluently, which helped her acquire Portuguese grammar and lexical knowledge. Moreover, she purposefully marked one more language—Japanese—in the table. She explained why Japanese was accessible in her linguistic resources and cited a case happening in a Portuguese classroom:

Excerpt 5: 日文係葡文課堂雖唔係經常出現，但結合日文講解真係可以幫到我領會葡文嘅一些語法點。例如，一次上堂時，老師講este，esse同埋 aquile，點都唔明係邊個方位嘅，梗老師話：“Julietta，esse就係これ，este就係それ，aquile就係あれ嘛?！”咁我就理解咗.

[Julietta: Japanese does not frequently occur in Portuguese classrooms, but sometimes it can be helpful for my uptake of Portuguese grammatical knowledge. For example, once I wondered about the meaning of “este, esse and aquile” in Portuguese class, and my teacher said: “Julietta, esse is ko re, este is so re, aquile is a re, ok?” Instantly, I comprehended it].

This experience shared by Julietta illustrated that even individual preferences for additional language learning could also positively affect linguistic relations. In a specific register, new linguistic resources can be joined in the macro scope of one’s language repertoires. Inviting multiple languages in the channel of communication brings more possibilities for enriching language resources so that they can be employed in language practices. Thus, it is crucial to value and acknowledge personal linguistic resources [68]. The shared knowledge of other languages between the teacher and the student has resulted in satisfactory learning experiences, where the practice enriches and develops one’s cross-linguistic comprehension and accessibility to language learning.

5. Discussion

The present study explored Macanese youth’s multilingual repertoires in LP inquiry and mapped the way they perceive their linguistic resources across different registers. By virtue of CV-aided analysis of colored clusters and labeled linguistic codes in LPs, their linguistic repertoires could potentially be identified in two measurable dimensions: richness and openness.

In terms of richness, our findings showed that Macanese young adults significantly represent a self-perceived multilingual identity when digitally drawing language portraits. Regarding openness, divergence was found among perceptions of the three groups, which can be explained by the different education policies between the Macao region and mainland China. The former adopts bi/trilingual instructions, and the latter promotes the national language (Putonghua Mandarin) as the medium of instruction in nearly all middle and high schools. Another influence in attributing higher values to English and Chinese in perceptions of mainland Chinese youth may be the notoriously grueling college entrance exam (“Gaokao”), as these two languages are equally weighted with math as the dominant subjects in the exam. In addition, the computational results indicated a high level of richness of the Macanese youth’s linguistic repertoires, suggesting their awareness of multiple linguistic competences and their accurate perception of their communicative resources [34]. In contrast, we found a relatively lower openness for the same group, which could result from the way different languages interact in their lives [68]. Follow-up interviews have justified the findings and pointed out that this may be because of their predominant language used exclusively for homebound and casual communications.

To address the second research question, we further refined the concepts of access and scope as two measurable indicators of linguistic resources to practically unpack participants’ perceptions of the relationship with various languages. In line with Ou et al. [69], we found that the transformation of registers, educational backgrounds, and heritage language maintenance may influence the relationship between scope and access from the individual perspective. Macanese youth intuitively perceived that the scope of linguistic resources can be affected by different levels of formality and complexity in a given social context [17]. The access can be developed to various degrees with the aid of instructors and classmates. As postulated by Weirich [17], scope and access capture the relation of speakers with regard to their linguistic resources as currently (un)available and potentially (un)learnable, depending on the social context in which they are practiced. Our findings suggest that the interaction of scope and access tends to be dynamic when speakers need to adapt to new circumstances where the sociolinguistic situation is ethnically multilingual.

In terms of methodology, this study paved the way for future research analyzing LPs in a digital mode. Though previous studies have advocated LPs as a methodological tool, they mostly followed the traditional way of administering LP activities as either individual or group work [35,67,69]. Then, Kusters and Meulder [31] noted that LPs conducted with pencils and papers might be an obstacle for young participants who might be resistant to drawing and generate negative feelings towards languages. In classroom settings, computer-mediated activities allow students to develop digital literacy, enhancing digital tool knowledge and critical thinking, and stimulating social engagements [46]. Digital LPs used in the present study can effectively reduce the unwillingness to participate because of personal, limited painting skills. The online platform offers available pictorial resources that allow participants to express themselves without technical difficulty. As a result, this approach has improved students’ and teachers’ digital literacy, empowering collaborative innovations in language-related tasks.

6. Conclusions

Previous studies have advocated LP as a methodological tool and followed the tradition of qualitative and narrative research to further understand multilingualism in a visible and interpretable way [34,35]. In this paper, we adopted language portraits [12] as an art-informed research instrument to unfold Macanese youths’ linguistic repertoires. From a methodological viewpoint, this study is one of the first to shed light on the affordances of computer-vision techniques. The computer programming results allowed us to analyze linguistic repertoires on the basis of LPs in a quantitative approach, which we call “LP counting”. Employing the CV-aided method as an innovative interdisciplinary approach for image analysis, we analyzed the LP by quantifying the proportion of different colored clusters so that linguistic repertoires could potentially be identified in two main dimensions: richness and openness. The results revealed that students whose ethnic category is Macanese have a richer linguistic resource than those raised in mainland China, while their openness toward different languages is lower. We argue that this CV-aided analytical system can be used for future research in art-based pedagogical activities. As a methodological shift in the field of language and perception studies from traditional methods to information and technology-aided analysis, the findings support the evaluation of learners’ visual productions that reflect their prior knowledge and current thinking, and further help to improve teachers’ understanding of multilingual students’ perceptions and attitudes in pursuit of appropriate and customized language teaching methods. Moreover, this visual aid deepens a positive relationship between educators and students since CV-based solutions assist in multimodal interactions, sharing ideas and opinions, and improving peer-to-peer and teacher-to-student cooperation in online teaching and learning [70]. With these benefits and concerns in mind, it is expected that by empowering systems to perceive, process, and interpret visual data in real time, computer-vision use in education will hold great promise in the future.

From a theoretical perspective, we argue that scope and access [17], as two interconnected fundamental notions in linguistic relations, can be practically understood. To date, the literature remains unclear on how to treat the concepts as methods. In this paper, we designed a “scope and access” self-evaluation chart so they can be deliberated with complete subjectivity. By tracing the subjective point of view, scope and access could be demonstrated in a dynamic of linguistic relations because of the restructuring of their linguistic repertoires under a new social circumstance [17] or in a different register. Our findings illustrate how Macanese youths perceive their scope and access in different social contexts and registers. Additionally, our analysis further explored the reasons for a movement in their perceptions of the accessibility and availability of a particular language.

The data gathered through digital language portraits and then analyzed in this case study provided several methodological implications for further image-informed research. First, we propose conducting and developing a computer-assisted analytical method in order to holistically explore learners’ linguistic repertoires in linguistically and ethnically diverse classes. This is in line with the Council of Europe [18] that outlined the need for instructors to obtain more profound knowledge of multilingual students’ input and output of available language(s) and take into consideration the importance of commonly valued linguistic and cultural resources. While it might be possible to use and collect LPs in a large-scale multilingual classroom, it is indeed time-consuming for the researcher to capture all learners’ linguistic resources within a short time. In second/foreign language education, analyzing LPs via computer-vision models helps to paint a rich picture, reflecting the importance of the target language as perceived by the student. It also provides the possibility for the teacher to consider the varieties and particular languages used or highly valued by the minority of them, subverting the dominant one-language-only norm in pedagogical classroom practice. Moreover, CV techniques can decipher linguistic labels and recognize pictorial meanings in the LP paradigm, which helps teachers assess students’ proficiency in language learning and gauge their involvement in both traditional and virtual classrooms [68]. The second methodological implication is the need to use technological tools in communicative tasks. In our proposed digital LP method, participants were able to engage experimentally in remote collaborative activities. Our results indicated that the online platform mediated in this study could potentially cultivate a constructive educational context. The multimodality of networking tools creates space for educators and learners, empowering meaningful language teaching and learning supported by flexible and unobtrusive negotiations.

Several limitations of this study leave room for future work. First, the sample size in our study is small, which may influence the generalizability of our results. Scholars could recruit more participants to verify the accessibility of the CV analytical method in LP research. Regarding the CV-based approach, although it can process a considerable amount of image data, our current work is only a tentative experiment for future work that can handle more images. Second, due to time limitations, the designed algorithm could not detect where the user colors in the graph of the LP. Busch [12] noted that the body positions where languages are marked are also important clues for eliciting personal language values and emotionality. In future work, we may leverage a supervised learning algorithm for the LP tasks. For example, we may manually label the area in a graph colored by users, and then the computer can learn to determine the coloring area for the labeled data. Third, in our study, the activities were done through an online collaborative platform, and students exploited digital visual resources in their LPs; however, we did not analyze their creative works in task 2 in the current study (due to the space limit here). Our further work will focus on how the participants have used the affordances of digital tools to express their linguistic repertoires and language identities.