Competing Activation of L1 and L2 in L3 Production—Evidence from a Case Study of Mongolian Native College Students from the Perspective of Think-aloud Approach

This article presents six case studies of L3 production performed by trilinguals in China, aiming to explore the activation status of L1 and L2 during L3 production. Six Mongolian native college students participated as research subjects, and qualitative and quantitative research methods including a think-aloud writing test and interviews, were adopted to examine which language, Mongolian L1 or Chinese L2, were more readily activated in the process of English L3 production, as well as, what roles L1 and L2 play in the process of L3 production and what factors contribute to the activation of L1 and L2. The findings show that Mongolian L1 and Chinese L2 are activated competitively in the process of English L3 production, but for different learners, they are activated to different levels and play different roles. The recency use of background languages, language proficiency and the medium of instruction are decisive factors in determining the activation levels of Mongolian L1 and Chinese L2 for Mongolian native college students in China. Language proficiency, the type and amount of exposure to L3 and the perceived language distance are the factors that contribute the most to activation of L1 and L2 during L3 production.


The Roles of the Background Languages in L3 Production
The roles of background languages in third language acquisition have been studied in depth by Williams and Hammarberg (1998) and Hammarberg (2001). In their studies, non-adapted language switches in Sarah's oral production of Swedish L3, are examined. Their study concludes that the two background languages English L1 and German L2 clearly play different roles in Sarah's conversation in Swedish L3. English L1 is mainly used for eliciting words and commenting on production; German L2 is used unconsciously with no clear communication aim. Based on this result, Williams and Hammarberg (1998) propose the notions of instrumental role for the language used for eliciting and metalinguistic functions, and supplier role for the language used more automatically. In 2006, Hammarberg conducts a study comparing the case of Sarah with another L3 learner, EE, whose L1 is German, L2 is English. The results confirm the previous findings that background languages play different roles. But in EE's case, English L2 seems to play an instrument role whereas German L1 a supplier one. Trevisiol (2006) applies the Williams and Hammarberg's model to analyze Japanese learner's code switches and word constructions in French L3. His findings show that Japanese L1 plays an instrument role and the supplier role is much less prominent.
Christina Lindqvist (2009) investigates the roles of L1 and L2 in L3 oral production and decisive factors in a case study, in which he applies Williams and Hammarberg's model, examining cross-linguistic lexeme production by six individual learners. The result shows Swedish L1/ L2 and English L1 are used as instrumental language, which is due to the interlocutors' common access to these languages. She concludes that it does not seem to matter if the instrumental language represents a learner's L1 or L2 though. The supplier language varies between individuals, and is less clear.
In conclusion, the model proposed by Williams and Hammarberg has been very useful in explaining the roles of background languages in the process of L3 production. But in other studies, the distinction between instrumental and supplier language are not as clear-cut as in Williams and Hammarberg (1998) and Hammarberg (2009). Therefore, more case studies, with different language combinations from other multilingual learners are necessary.

Factors Affecting Cross-linguistic Influence over L3 Production
Many factors are believed to influence Cross-Linguistic Influence (CLI). Researchers have reached different conclusions about which factors can condition the CLI.
Research on TLA and CLI has already identified many factors that affect learners' reliance on previously learned/ acquired languages and constrain the type and amount of influence on the target language. Jarvis (2000) assumes that several factors can be controlled in CLI studies such as age, motivation, personality, language aptitude; social, educational and cultural background; language background; type and amount of target language exposure; target language proficiency; language distance; task type area of language use; prototypically and markedness of linguistic feature. In L3 www.scholink.org/ojs/index.php/selt Studies in English Language Teaching Vol. 8, No. 2, 2020 39 Published by SCHOLINK INC. research, four factors (proficiency, typology, L2 status and recency) have often been discussed by researches, such as De Angelis and Selinker (2001), Hammarberg (2009) and Bardel and Lindqvist (2006). In addition, there are other factors that affect cross-linguistic influence over L3 production, such as language mode, personal language identity (Hammarberg, 2009) and interlocuter's common language access to certain languages (Lindqvist, 2010).
However, the influences of these factors may vary according to factors, such as type of task, language background and level of proficiency.
In the current study, the impact of language proficiency, type and amount of exposure to the target language and as well the perceived language distance will be examined in detail and in what follows these factors will be reviewed.

Proficiency Factor
In the field of cross-linguistic influences, language proficiency refers to proficiency in the target language and proficiency in background languages.
A notable point is that the level of proficiency in the target language is one of the most important elements for CLI. Many studies have shown that CLI decreases as the proficiency in the target language increases (Singleton, 1987;Hammaberg, 2009;Lindqvist, 2009). Most researchers maintain that CLI is more likely to occur at the early stages of acquisition, when learners' knowledge of the target language is still weak and fragmentary and the need to fill knowledge gaps in the target language is more pressing (Singleton, 1987;Taylor, 1975;Williams & Hammarberg, 1998).
According to the above-mentioned studies, language switches, foreignizations and word construction seem to be frequent phenomena at earlier stages of acquisition, when learners are not fluent in the target language.
Not all kinds of CLI will occur at all stages of learning, though. Other kinds of influence might appear in later stages of acquisition. Ringbom (1987) and De Angelis (45) found that meaning-based CLI is more important in advanced learners.
On the other hand, if the learner has already studied other languages, in general the highest level of language proficiency has the greatest impact on the target language. For instance, Ringbom (2001) examined the differences between CLI from the L1 and L2 in L3 use. His results show that meaning based CLI would only occur from languages in which the learner has native-like or very advanced proficiency. In Hammarberg's study (2009), the learner has acquired German, French and Italian before learning Swedish. Among these three foreign languages, the level of German is highest. German is most frequently used. His results show that a large amount of language switches into German appear in Swedish L3 acquisition. In addition, Ringbom (2001) notes that if the learner's second language level is high, even if it does not have formal similarity with the target language, it will influence the target language learning and become the source language of CLI. Therefore, it can be concluded that proficiency in background languages have an impact on CLI.

The Type and Amount of Target Language Exposure Factor
The type and amount of target language exposure is considered as another factor that influences the occurrence of CLI. Although it has not been discussed a lot, some researchers have studied it, such as Fouser (2001), Hoffman (1999. Hoffman examines the phenomenon that individual language learner psychologically relies on their learned/acquired languages. He notes that the time and place of using language play critical role to fluency and importance. Fouser also studies two learners one of whom is a Korean learner of English, who has lived in Japan for 5 years. In his interview, the learner reports that he tends to think in Japanese, when he speaks Korean.

Language Distance Factor
Cenoz, Hufeison and Jessner (2001) believe that learning a third language which is similar to the native language or the second language can promote the process of acquisition and make language switches happen easily. Therefore, language distance becomes a very important factor in TLA. Language distance refers two notions: the distance that a linguist can objectively and formally define and identify between languages and language families and the notion of perceived language distance that is the distance that learners perceive to exist between languages that may, or may not, correspond to the distance that actually exists between them.
The notion of perceived language distance was proposed by Kellerman in the 1970s motivating many studies on perceived language distance after that. For instance, Ringbom's (2001)  In sum, the learner's perceived language distance can affect CLI, even more than the actual language distance, especially at the semantic and functional transfer level. Even the learner's L1 or other higher-level language is irrelevant to the target language in terms of typology and CLI may occur.

Research Questions
The present study aims to investigate the following three questions: 1) Which language, L1 or L2, is more readily activated in L3 production?
2) What roles does L1/L2 play in the process of L3 production?
3) What factors contribute to the activation of L1 and L2 in the process of L3 production?

Participants
The participants of the present study are 6 female university students aged between 20 to 25 from Chi Feng University. Because this was a qualitative study with a small number of participants, the Chinese in their daily life. Ten students were asked to perform the test but two were eliminated when audio indicated that their silence pauses were of time takes up more than 20% of total writing time, and two others were excluded because their voice is too low. Table 1 shows the information of the six participants.

Instruments
The instruments of data collection were a think-aloud protocol writing test, interviews and audio-recordings of think-aloud protocol.
Writing task: Descriptive writing task is often used for elicitation by many researchers for example Mayer (1969), Kellerman (2001) and Cenoz (2001). In order to measure non-target language activation in L3 production, the researcher used a wordless picture story called "Snowman's story" by Will Hillenbrand. The book has a series of 30 pictures without any word in it. All the subjects are asked to narrate the picture story in English with think-aloud protocol.
Think-aloud protocol: Think-aloud protocol is an approach used to gather data in usability testing in product design and development in psychology and in social science, such as writing, reading, translation research, decision making, and process tracing. It is introduced by Clayton Lewis, which is developed by Ericsson and Simon later (Guo, 2007). Think-aloud protocol requires the subjects to say whatever comes into their mind while they are performing a certain task. It may contain what they are thinking and feeling and doing. This will show the cognitive process of subjects, and thought formation as explicit as possible.
Individual interview: For the sake of getting the related language background information and find the factors that influence background language activation in the process of L3 production, an open-ended interview is conducted just after each subject finishes the writing.
Both think-aloud test and interview are recorded so that the researcher can go back and refer to how they behave when doing the task.

Procedures of Data Collection
Four steps were employed in the procedure of data collection in this study. First, the subjects in this study were strictly selected from Mongolian native college students, and they were technically trained before the test in order to get valid data. Second, the participants were asked to narrate the wordless picture story in English in a way of think-aloud, with about 150 words. Each participant was tested individually in a classroom. When students were doing the think-aloud task, the role of researcher was to keep silent and keep distance from the participants to serve as an observer who hints the students to speak out aloud what they are thinking about. The think-aloud writing process of each participant was recorded by the researcher with an audio-recorder as data base for describing the activation of non-target language.
Third, an open-ended interview was conducted with each of participant right after their think-aloud writing. The interview included two parts. The first part of questions were based on language background. The second part of questions were used to get their opinions about activation of L1 and L2 occurring in L3 production. The interview was also recorded. The six participants were further selected out of the 10 for the final data analysis, because their silence time was less than 20% of the total writing time. Finally, all the audiotapes of the six participants were transcribed into text by the researcher for further statistical analysis.

Data Analysis
When the data collection is complete, the author starts to analyze the data in relation to the three research questions. In transcriptions of the recordings, one kind of cross-linguistic influence, non-adapted language switches, has been identified. It is proposed by Williams and Hammarberg in 1998. Non-adapted language switches refer to utterances of previously learned languages, usually a word or a short sentence, which are not adapted to the target language in any way. All the language switches occurring in the data, are examined.
The study begins with the first research questions. All the non-adapted language switches from L3 to L1 and from L3 to L2 are selected in order to see which language, L1 or L2, is activated more readily in L3 production.
To answer the second research question, non-adapted language switches occurring in the data are further classified into four types depending on which functions the switch to have in L3 production, according to Williams and Hammarberg's (1998) categorization. Below are definitions of four types, followed by examples and keys from the data of this study.
The first type is EDIT. It contains terms, such as "no, sorry", which are used to introduce a self-repair and also includes interactive feedback signals such as <yeah, what >.
e.g.,  Table 2 presents the frequency of language switches occurring in Chun and Qing's cases. It can be seen that switches into Chinese L2 occur in high frequency in both cases, with a range of 85% to 87%.

Result and Discussion of Research Question 1
While switches into L1 occur in a very low frequency, with a range of 13% to 15%. It indicates that Chinese L2 is more readily activated in these two cases. As can be seen in Table 3, in the cases of Hu and Wu, switches into Mongolian L1 are not as frequent as Chinese L2, ranging from 26% to 37%; switches into Chinese L2 is much more frequent, covering from 63% to 74% of the occurrences. This result resembles in Hu and Qing's cases, that Chinese L2 is readily activated too in the process of their English L3 production. According to De Bot's Model (1992), he believes that the language choice is made at conceptual level Then, how can the six subjects' language switches be related to the production of L3? According to De Bot model, the language choice is made at the conceptual level and is governed by language node, but the activation of element from another language occurs at different stages of production process. For these six subjects, of course, English L3 is the most highly activated as their selected language. But

Result and Discussion of Research Question 2
In order to test what roles L1 and L2 play in L3 production, language switches produced by six participants are examined in more details.
As to what roles these non-adapted language switches play in L3 production, Hammarberg and Williams (1998) divides them into two categories. They propose that the switches of EDIT, META and INSERT types are made to complement and support the speech production of L3, they are playing the role of instrument; WIPP switches are produced more automatically on the basis of supplier role. The results provided below follow this category.

INSERT EXPLICIT ELICIT
From the results of these cases, it can be concluded that to the four subjects of Chun, Qing, Hu and Wu, both their Mongolian L1 and Chinese L2 are activated in English L3 production. In aspect of instrumental role, with Chun and Qing, L2 seems to dominate the instrumental role. In case of Wu and Hu, Mongolian L1 dominates the instrumental role in English L3 writing. In aspect of supplier role, four of them make use of their Chinese L2 frequently, which is in line with many previous findings from other studies, by Hammarberg and Williams (2009).
But in the cases of Ha and Ao, the result turns out to be much different from the other four subjects.
The former four subjects activate their Chinese L2 in a higher frequency, whereas Ha and Ao always activate their Mongolian L1 in a higher frequency. The result also shows that Mongolian L1 is used both as instrumental and supplier language in Ha and Ao.
Therefore, L1 and L2 play different roles for different learners.
With Chun and Qing, Chinese L2 is more readily activated as instrument. It is apparent in META and INSERT switches. While to Hu, Wu, Ao and Ha, Mongolian L1 is readily activated in an instrumental role. It is apparent in all types of switches. This different choice for instrumental language can be explained by many interacting factors.
As to factors affecting the choice of instrumental language, there are many factors that appear in the literature, such as the learners' knowledge about the interlocutor and her/his linguistic competence, language mode, proficiency in the target language, personal identification, the status of language as a contact language, and the language that has been established as a language for communication. In Hammarberg's study (2009), for the two participants, English is most frequently activated as an instrument language. The factors seems susceptible of activating instrumental language are the learners' knowledge about the interlocutor and her/his linguistic competence, trilingual mode that they adopt in the project, and the learners' proficiency in target language in his study.
But in present study, all the participants are set in monolingual mode that they are tested individually in a classroom. There is no interlocutor to communicate with. Therefore, the factors concerning interlocutors are excluded in this study. The most crucial factor seems to be the medium language used for learning English. Because, in China, a major chance for students to expose to English is the classroom. And they tend to depend strongly on a medium language. This medium language serves as an instrument in learning English. For Chun and Qing, Chinese L2, which is used as a medium language for their English learning, serves as instrument in L3 production. They all learn English through medium of Chinese L2. They tend to think in Chinese as they speak English. It seems that language and culture. Since they had a strong identity towards their mother tongue and culture, despite they are able to express themselves fluently in Chinese. Because they three are all from Xilingol. They have strong awareness of protecting their mother language and their culture. While, Chun, Hu and Qing have no strong Mongolian identity.
The interactive effect of these two factors makes Chinese or Mongolian become activated as an instrumental language, regardless of L1 or L2.
With respect to the activation of the supplier language, four factors including learner's proficiency, recency, typology, and L2 status, have been mostly discussed in the literature (De Angelis & Selinker, 2001;Hammarberg, 2009;Lindqvist, 2009). Hammarberg and Williams propose that interaction of these four factors determine the activation of a supplier language in Multilingual learners' L3 production. Meanwhile, they point out that the impact of these factors may change according to different situations of learners. In their case studies, the decisive factor is its L2 status, because the fact that the learner consider German L2 as a "foreign language" in her mind. So she prefers using German L2 to make up the L3 words that she does not know.
In current study, with Chun, Hu, Wu, and Qing, Chinese L2 dominance is very strong as a supplier, which is in line with the findings of Williams andHammarberg (2009 [1998]), De Angelis and Selinker (2001). In aspect of recency effect of background languages, they all learn Mongolian and Chinese within both natural and school context, and both are often used in their daily life. But Chinese is more closely related to English for them, because they learn English with the help of Chinese, rather than Mongolian. In the aspect of proficiency, their Chinese competence is stronger than Mongolian, though Mongolian is their mother tongue. They are from Tongliao City, and they speak Mongolian, which is strongly assimilated by Chinese. In aspect of typology, although Mongolian, Chinese, English belongs to different language phylums, another kind of typology called psychotypology, proposed by Kellerman (1983), which refers to the learner's perception of distance between languages. Many studies have shown that pyschotypology is an important factor for cross-linguistic influences. In this study, Chun, Hu, Qing and Wu consider that L2 Chinese is more similar to English. They know that both Chinese and English share some similarities in sentence structure and word order. But Mongolian follows SOV order. Therefore, in the case of Chun, Hu, Wu and Qing, psychotypology, proficiency and recency interacts and makes Chinese L2 frequently activated as a supplier.
But with Ha and Ao, Mongolian L1 is readily activated as a supplier. The proficiency level of Mongolian L1 is much higher than in Chinese L2, because Mongolian is the most frequently used in their daily life. Also Mongolian L1 is used as media language in their English class, and they both think that Mongolian is close to English. Thus, the proficiency, recency and pyschotypology of Mongolian L1 scores high for Ha and Ao, which makes Mongolian L1 activated as a supplier.

Result and Discussion of Research Question 3
The results show that the proficiency of target language, type and amount of exposure to the target language, perceived language distance are the main factors that contribute to the activation of Mongolian L1 and Chinese L2. The results will be presented below.

Language Proficiency
The information about the six subjects (Table 1) shows that the proficiency in target language (English L3) of Chun, Hu, Wu and Ao is at the pre-intermediate level, Ha and Qing are at pre-advanced level.
The different level of proficiency in the target language cause different degree of activation, which had been shown by other researches as well. In this study, the researcher compares the amount of background language activation in English L3 production (See Table 11). The result shows that Ha and Qing, whose English proficiency is at pre-advanced level, has less frequency than the other four participants. The finding suggests that the activation of background languages (Mongolian L1 and Chinese L2) decreases as proficiency in the target language (English L3) increases. As to the proficiency in background languages, the result shows that the level of proficiency of all the subjects' L1 and L2 is much higher than their target language, English. In specific, with Chun, Hu, Wu and Qing, their proficiency level in Chinese L2 is higher than that in Mongolian L1. But to the other subjects, like Ha and Ao, their Mongolian L1 proficiency is higher than that of Chinese L2. In addition, they all think that the most important factor activating L1 and L2 is the higher proficiency in L1 and L2.
All these data shows that proficiency is the decisive factor of activation of L1 and L2.
Language switches, as a type of CLI, occur frequently at earlier stages of acquisition, when learner is not fluent in target language (Lindqvist, 2009). On the other hand, only the strongest background languages seem to be activated easily. Learners' high level of proficiency in L3 makes them continue their production in L3, but if learners are at low level of proficiency in L3, when they meet lexical difficulties, they always turn to resort to one or two background languages, since they require all possible resources when speaking their L3 (Lingqvist, 2009

Type and Amount of Target Language Exposure
The result shows that Chun, Hu, Wu and Ao can mainly expose to target language in their English classroom, only about three hours in a week. Since they are non-English majors, their English class is too big to have chances to practice in class. They also mention that, because of students low proficiency in English, their English teacher often switches into Chinese, sometimes into Mongolian during English class. So when the learners speak English, they often think in Chinese or Mongolian.
But with Ha and Qing, the situation is different. Ha and Qing are English Majors. According to the data from the interview, they have at least twelve English classes each week. So they expose to English more often, as least 10 hours each week. They report that in their English classes, their teacher does not often switch into Mongolian or Chinese when he/she speaks English.
On the other hand, all the participants mainly expose to English in their classrooms, where their English teachers closely connect his or her English to Chinese or Mongolian. Therefore, when they speak or write in English, they are accustomed of thinking in the similar way as they do in the classroom.

Perceived Language Distance
In present study, the data from interview indicates that all the subjects do not know that English, Mongolian and Chinese respectively belong to different language family. But they perceive that there exist some similarities and relatedness between these languages. Chun, Hu, and Qing think English is much closer to Chinese since they have common in word order in a sentence, while Mongolian is not perceived as very close to English. To Wu, both Mongolian and Chinese are perceived to be close to English. She thinks that Chinese and English have similarities in word order, while Mongolian and Chinese also have some similarities in tense, word-formation, etc. Ha and Ao believes firmly that Mongolian is closer to English. They think there are many similarities between Mongolian and English, from phonetic characteristics to word-formation of derivation. That is why they often activate their background languages in L3 production.
In addition, this study suggests that cross-linguistic influences can also occur between languages that are more distant in terms of language family.

Conclusion
This study confirms the findings from other L3 research that background languages have great influences on L3 learning. First of all, Mongolian L1 and Chinese L2 are activated competitively in the process of English production, but for different learners, they are activated to different levels. The findings can be explained by their different language learning background.
Moreover, L1 and L2 play different roles for different learners. The roles L1 and L2 play depend on the factors of medium language of instruction and identity of language and culture, level of proficiency in background language, recency and typology. Therefore it is not proper to say that L1 always plays as an instrument, while L2 always a supplier role in all cases. In fact, many factors condition the activation of L1 and L2, but the main contributing factors to the activation of background languages are the language proficiency, the type and amount of exposure of target language and perceived language distance.

Note
This article is granted by the research program of foreign languages of 2019 "13th Five-Year Plan" Education Science Research of Inner Mongolia (NGHWZ201936) titled as "A Study of the Media Language in the English Teaching of College Mongolian Native Students".