2 Literature review
2.3 Research Methods and Instruments Employed in MML Studies
2.3.2 Overview of the Research Approaches to the MML
This section offers a brief overview of the research studies and investigative techniques that have been used in the exploration of different aspects of the MML. Table 2 offers a synopsis of the six studies discussed in this section. The common trait of the studies
discussed in this section is that they started from the fundamental assumption that through spoken language cognitive linguistic processes and mental lexicon organisation can be identified (Gabrys-Barker, 2005; Libben & Jarema, 2002). The researchers who undertook the studies acknowledged that with the investigative strategies and tools currently at our disposal, and due to the intricacies and complexities of the mental processes, we can only approximate ‘objective’ truth and formulate inferences about the dynamics of the MML (Chafe, 1996; Libben & Jarema, 2002; Wimsatt, 2007).
As noted earlier, there are few available research studies which looked into multilinguality defined as the individuals’ “knowledge and use of more than two languages with fluctuating linguistic and communicative competence” (Boksay Pap, 2015, p. 87). The search for scholarly material on multilinguality in academic databases such as Google Scholar, Research Gate, Academia and JSTORE is made difficult by the fact that, as mentioned before, there is still a strong tendency in the research field to treat multilinguality under the header of bilinguality, and to consider multilinguals special bilinguals. Thus, it is quite often the case that studies investigating linguistic abilities comprising more than two languages hide under the misnomer bilingual (e.g., Grosjean & Ping, 2013; Kroll & De Groot, 2005). Alternatively, we find the opposite situation, namely that bilingual language ability is termed multilingual (e.g., Fitzgerald, 2006).
Table 2. Comparasion of Research Techniques and Tools across Studies into the MML.
Comparasion of Research Techniques and Tools across Studies into the MML.
The scarcity of MML research studies and research studies on multilinguality in general is partly determined by the complexity of the multilingual phenomena. While both bilinguality and second-language phenomena present considerable complexity, multilingual phenomena, where not two but several languages and their attributes have to be described, accounted for and kept under observation, presents a heightened challenge to researchers.
To understand why, we have to consider the two levels of attributes a language has. First, each language a person knows has several linguistic attributes, such as typology, phonology, morphology and syntax. Then, at personal level, each language has individual attributes, such as the age of acquiring, frequency of use, proficiency, and perceived typology, to name just a few. It is the individual attributes of languages, the idiosyncratic characteristics proper to each language user that can all make research of multilinguality cumbersome. These individual attributes are not static but flexible, transient and many-sided. They move along a continuum and interact with each other in various forms (Herdina & Jessner, 2002; Larsen Freeman & Cameron, 2008). This multiplicity predisposes the idea of mapping complex mental lexicon traits by just looking at a few specific attributes to be increasingly considered an illusion (Filipovic, 2014). As an apt metaphor for the complexity of individual multilinguality, I propose the kaleidoscope, a collection of objects incorporated in a system, which even at the slightest movement will show an unrepeatable view. As a result of the complexity and flexibility of multilinguality, its research has not yet produced results generalizable over large populations. This issue is often considered a problem by those researchers who adopt a positivistic approach to epistemology. Historically, following the Chomskian perspective, research into the mental lexicon has concentrated on groups rather on individual language users. This focus on groups is primarily linked to psycholinguists’ aim to link the validity of theories to the generalizability of research results and to restrict
“themselves to just one particular notion of generalizability, namely a statistical, sampling based notion” (Lee & Baskerville, 2003, p. 221). Possible solutions for the aforementioned problems would be the constant refining of research methodologies with particular consideration for multi-level studies, synergy between quantitative and qualitative research strategies, special attention to the individual language users’ evolutionary linguistic journey, and the idiosyncrasies of language use (Dewaele, 2005; Krauss, 2005; Wei & Moyer, 2008).
2.3.2.1 Experimental Research Studies Investigating the MML
This subsection discusses some research techniques employed to investigate the MML in experimental research paradigms. These techniques are the sentence completion task used by Matutin (1993) and Ecke (2001), the lexico-semantic judgment task employed by Matutin (1993), and the word translation task used by Goral, Levy, Obler, and Cohen
(2006). All three data gathering techniques are very similar to research strategies employed in memory measurement (Lockhart, 2003), and they all belong to the cued recall paradigm (Roediger III & Meade, 2003).
In the sentence completion task, the participant is presented with a gapped sentence and asked to produce one word which best fills the gap. The gapped sentence acts as the retrieval cue for the word to be produced by creating a specific linguistic environment which is supposed to trigger relevant lexical units from the mental lexicon.
The lexico-semantic judgement task builds on the logic of single-item recognition tests (Lockhart, 2003). The participant is presented each word in sequence, and he is asked to attach a judgment to it and respond “yes”, if the word satisfies certain pre-set criteria, and
“no”, if the word does not satisfy the pre-set criteria (Roediger III & Meade, 2003).
The word translation task is a type of cued recall test, in which a prompt is presented to the participant who has to come up with the pair of the prompt-word. In the case of translation, the prompt, or cue can be an L1 word. The participant has to come up with the translation of the L1 cue word.
In addition to sentence completion, lexico-semantic judgment, and word translation, Ecke (2001) used a numeric scale judgement task and Goral et al. (2006) used natural conversation tasks as additional assessments of the MML. These two additional tasks are described later.
In one of the earliest relational studies of the MML, Matutin (1993) aimed at identifying the connections between the L2 French and the L3 English of Kirundi L1 participants in terms of the influence of a well-developed L2 lexicon on the less-developed L3 lexicon. To this end, he devised an experimental research design in which he used two tasks: a sentence completion task and a lexico-semantic judgement task. The main strengths of the data collection techniques used by Matutin (1993) is their applicability to a large number of participants with prospects to generalizability. In addition, both sentence completion and lexico-semantic judgement adhere to the requirements of ecological validity.
Ecological validity is a type of external validity which contributes to the generalizability of a study, and it concerns the testing environment and testing materials used in the research context (Cohen, Manion, & Morrison, 2007). The participants in Matutin’s (1993) study were 126 Burundian undergraduate students, speakers of L1 Kirundi, a Nigerian-Congo language spoken on large territories in the Sub-Saharan Africa (Comrie, 2001), who were highly proficient in French as an L2, and were users of English as an L3 at different language proficiency levels. Matutin (1993) hypothesized that the participants, being highly-proficient in L2 French, which is typologically closer to L3 English than L1 Kirundi is to L3 English, would
be greatly influenced by their L2 French lexicon especially when they have to access L3 English words which have L2 French cognates. The tasks centred on two types of cognates:
French-English noun and verb pairs which have identical or similar form and similar meanings (e.g., commence) and have non-cognate English synonyms (e.g., commence-start-begin) and French-English interlingual homographs, cognates which have identical forms (e.g., French venue meaning arrival, and English venue meaning site) but have different meanings in the two languages.
Matutin’s (1993) study revealed that the L1 Kirundi speakers, users of L2 French and L3 English readily transferred the interlingual homographs from L2 French into L3 English while they were more circumspect regarding the L2 French-L3 English cognates which have English non-cognate synonyms. These results are in line with the base language hypothesis (Chandrasekhar, 1978) which posits that in a multiple language speaker’s mental lexicon transfer from a language to the other will occur when the languages share similarities.
However, Matutin (1993) also remarked that transfer of L2 French cognates into L3 English appeared to decrease with the increase of L3 language proficiency. He suggested that the organization and structuring of the multilingual individuals’ mental lexicon evolve continually with the individuals’ language proficiency.
Regarding the conditions of the study, however, one limitation connected to the language mode hypothesis (Grosjean, 1998, 2001) becomes apparent. The language mode hypothesis (Grosjean, 1998, 2001) regards the degree of language activation in the mental lexicon of multiple language users. It postulates that the immediate conditions surrounding a multiple language user have a direct impact on which languages and at which degree they become activated in the language user’s mind. In Matutin’s (1993) study, the participants might have had both their L2 French and L3 English inevitably activated but not their L1 Kirundi because they were asked to carry out tasks that involved their L2 French and L3 English. However, considering that Grosjean (1998, 2001) formulated his language mode hypothesis five years after Matutin’s (1993) study, Matutin probably did not identify the inevitable activation of the participants’ L2 and L3 languages as a limitation to the study.
Ecke (2001), in a more recent experimental study, investigated the MML by looking at the quite common yet captivating phenomenon of occasional difficulty in word-retrieval, called tip-of-the-tongue (TOT) state. TOT states have been proved to be universal irrespective of the language(s) spoken by the participants (Brennen et al., 2007) with higher likelihood of occurrence in the case of those who use more languages (Gollan & Silverberg, 2001). When individuals experience temporary word-retrieval problems, they usually perform intensive target word searches. This is partly why TOT states offer special possibilities for studying the organization and functioning of the mental lexicon (Brown & McNeil, 1966). The
manifestation of TOT states in the case of multilingual language users is an attractive possibility for researchers to gain insights into how lexical units in the MML are connected and how lexical activation happens.
Ecke (2001) involved in his study a sample of 24 undergraduate students at a Mexican university. The participants were L1 Spanish users, their L2 English was at an intermediate towards upper-intermediate proficiency level, and they were learning L3 German at a beginning level. Ecke (2001) used a sentence-completion task slightly similar to that used by Matutin (1993). The task asked the participants to read gapped L3 German sentences and produce the missing L3 word which best filled the gap. To constrain the participants to make their choices within limited lexico-semantic boundaries, Ecke (2001) accompanied each gapped sentence with the L1 Spanish translation equivalent of the L3 German word to be filled in. There were 23 L1 Spanish words used as stimuli. By polarizing the sentence-completion task on the L1 Spanish and L3 German, Ecke (2001) reasoned that L2 English activation would not be high during the completion of the task, thus interference from L2 English would be kept at a minimum. This was an important characteristic of Ecke’s (2001) research design, as he wanted to examine closer the patterns of long-term lexical organization and functioning of the L3 German mental lexicon in connection with the L1 Spanish mental lexicon. At the time of the study, the participants were more proficient in L2 English than in L3 German, thus based on the language dominance theory (Cummins, 1976;
Genesee et al., 1995), L2 English could have better dominated the participants’ language processing and retrieval than L3 German.
To tap into the underlying processes which accompany TOT states, namely fragmentary L3 German target word knowledge, and possible associations between L3 German target words, Ecke (2001) asked the participants to indicate on a five-point numeric scale how certain they were about knowing the L3 German target word. The numeric scale made possible an efficient data collection regarding the participants’ perceived knowledge of the L3 German words. Because TOT states incorporate “failures to recall a word of which one has knowledge” (Brown & McNeil, 1966, p. 325), Ecke (2001) reasoned that beyond examining the end-product of the sentence completion task it was important to get information about the underlying mechanisms which brought about the success or failure.
Thus, beyond the data obtained in the L3 German sentence completion task, he was able to measure the participants’ self-perceived knowledge related to the target words in question.
The participants were also asked to give complementary information about the attributes of the words that popped up in their short term memory during the search for the L3 German target word. The collected data was analysed with special regard to fragmentary word retrieval, associations and non-target word retrievals or erroneous retrievals.
The results of Ecke’s (2001) study revealed a rather high number of TOT states in the retrieval of the L3 German words: of the total of 504 responses, 156 were TOT states (Ecke, 2001, p. 98). The high number of TOT states led Ecke (2001) to conclude that retrieval fluency in case of less-represented languages in the mental lexicon might be partly dependent on language proficiency. The results of the study also revealed that the L1 Spanish, L2 English, and L3 German participants’ TOT states were characterized by fragmentary target word recall, recall of non-target words, and target word associates. In the fragmentary word recall processes, the participants recalled the initial syllabic unit or the first letter of the words then the search proceeded towards the other segments of the L3 German target word. Regarding the word associates accessed in the search for the L3 German target word, the participants retrieved non-target L3 German words which had formal resemblance with the L3 German target words. These results are in line with previous monolingual and bilingual experimental studies (e.g., Brown & McNeil, 1966; Gollan & Acenas, 2000).
However, Ecke (2001) also identified some essential dissimilarities with other studies. For instance, in the case of non-target word recall, or wrong target responses, previous research studies (Ecke & Hall, 2000) found that it was either the participants’ L1 or the L2 languages which acted as sources of transfer and assistance. Ecke’s (2001) study revealed that, contrary to the researcher’s expectations, the source language of assistance and cross-lexical influence was the participants’ L3 German. Namely, the participants’ recalling of non-target words and wrong non-target responses took place intralingually, within their L3 German lexicon. This unanticipated finding, which was even more surprising as the participants were beginner learners of L3 German, seems to be in sharp contradiction with the parasitic strategy hypothesis (Hall, 1992; Hall & Ecke, 2003). Linguistic parasitism is an L2 acquisition model which posits that lexical units from an additional-language, in an early stage of learning, automatically connect to the already well-established L1 lexical units in the mental lexicon because this is the “simplest connection” (Hall, 1992, p. 45) with low cognitive demands. Based on the parasitic strategy (Hall, 1992; Hall & Ecke, 2003), the participants were expected to rely more on their L1 Spanish when they experienced L3 German TOT states. Nevertheless, the results of Ecke’s (2001) study showed that bilingual individuals with a fairly low lexical knowledge in an additional language will try to exploit that knowledge to a maximum in certain situations, for example, an L3 translation task.
Goral, Levy, Obler, and Cohen (2006) undertook a clinical case study to investigate whether the multilingual lexical connections and inter-language activation deteriorate as a result of aphasia. The research involved an adult multilingual speaker who knew six languages. His L1 was Hebrew, his L2 was English, and his L3 was French. He was highly proficient in these three languages, and he had been using them for different communicative
purposes. The participant also had a functional proficiency in L4 German, L5 Spanish, and L6 Italian.
Based on Damasio (1992), “aphasia is a disturbance of the comprehension and formulation of language caused by dysfuncion in specific brain regions” (p. 531). It usually affects both the receptive language modalities, auditory comprehension and reading, and the expressive language modalities, verbal expression and writing. Some of the most evident symptoms aphasic individuals produce are the semantic errors present in the expressive language modalities because aphasia “results from a break-down of the two-way translation that establishes a correspondence between thought and language” (Damasio, 1992, p. 531).
Semantic errors can usually yield valuable data about the semantic processing present in the mental lexicon (Hodgson & Lambon Ralph, 2008; M. F. Schwartz & Dell, 2016).
Previous research data collected in cases of bilingual aphasia show that the treatment of the non-dominant language facilitates the recovery of the untreated dominant language and linguistic recovery in general (Ansaldo & Saidi, 2014; Fabbro, 2001). This suggests a strong link between languages in the mental lexicon. Further results point towards the existence of a shared mechanism of language selection and control managing all the languages present in the mental lexicon (Green & Abutalebi, 2008). Research data about multilingual aphasia is scarce, and there are still several unanswered questions (Fabbro, 2001).
To investigate the lexical connections and inter-language activation in a multilingual aphasic, Goral and his colleagues (2006) used two kinds of tasks: natural conversation tasks and word-translation tasks. Both types of tasks were administered in L1 Hebrew, L2 English, and L3 French, the three languages the participant had been highly fluent in before the onset of aphasia.
The natural conversation task aimed at identifying whether all the languages became activated during the target language speech production and if so, what kind of interference patterns surfaced. The conversations in all the three languages addressed the same topics.
The researchers anticipated that during the conversation task beside the target language of the conversation the participant’s other languages would also become active and instances of cross-linguistic interference would be identifiable.
By employing the natural conversation task as a tool, Goral and his colleagues (2006) were able to detect both lexical access and cross-linguistic interference patterns. Regarding lexical access and the retrieval of words from the mental lexicon, the results revealed that while the aphasic multilingual participant could not retrieve a word in the target language of the conversation, he was often able to retrieve the word in one of his other languages. During the L1 Hebrew conversations, there was minor interference from the other two languages, in total 14, out of which 12 were L2 English interferences and 2 were L3 French interferences.
Similarly, during the L2 English conversations, interferences from the other two languages were minimal with 2 interferences from L1 Hebrew and 16 L3 French instances of language interference. In contrast, interferences from the other two languages were considerable in the L3 French conversation task, with 65 interferences from L2 English and 5 interferences from L1 Hebrew (Goral et al., 2006, p. 239). The results made evident that in the L1 Hebrew and the L3 French conversation tasks, L2 English interferences played an important role, while L1 Hebrew intervened minimally in the L2 English and L3 French conversations.
The cross-activation pattern that emerged from the conversations revealed that cognates, namely interlingual homographs, as elements with role in cross-lexical interference, had an important presence only in the L3 French conversations. Of all the 65 instances of L2 English interference into the L3 French conversation, 31 were L3 English-L2 French cognates (Goral et al., 2006, p. 239). In the conversation tasks in L1 Hebrew and L2 English, cognates were not significantly present. This also points towards a cognate-dependent asymmetric connection model between the participant’s L1 Hebrew, L2 English and L3 French languages, with L3 French being in a parasitic position (Hall, 1992; Hall &
Ecke, 2003) in relation to L2 English, and with L1 Hebrew being somewhat in a loose connection with L2 English and L3 French. However, it has to be noted that Hebrew does not share an important number of cognates either with French or English. Based on these observations, Goral and his colleagues (2006) posited that the organization of languages in the mental lexicon may be partly guided by the presence of cognates, as elements which can strengthen the connection between the lexicon of separate languages. Thus, between languages which share similar lexical units, the links in the mental lexicon can be assumed to be stronger than between those languages which share fewer similar lexical units or do not share similar lexical units at all. Goral et al. (2006) concluded that the ease with which a multilingual who was affected by aphasia can access the words from his different languages and recover a particular language seems to be partly dependent on the number of shared lexical items between and among languages. The results Goral et al. (2006) obtained support the base language hypothesis (Chandrasekhar, 1978) and are in line with the results of Matutin’s (1993) study regarding the effects of the interlingual homographs in the MML.
Goral et al. (2006) noticed that albeit the participant experienced several difficulties of target-word retrieval fluency, he was able to retrieve the target word after some time elapsed.
Based on this fact, the researchers reasoned that the participant’s knowledge of the words in the mental lexicon had been preserved, but due to the aphasic condition, the ability to access them was diminished. An interesting finding of Goral and his colleagues, in disagreement with Ecke’s (2001) study, was that when the participant experienced failures of L3 French lexical production, he did not perform an intralingual L3 French search for the target word,
but it was his L2 English lexicon he relied on in both searching for and in trying to supply the target word.
To complement the data gained from the natural conversation tasks and to get a deeper understanding of the aphasic participant’s multilingual lexical connections in his MML, Goral et al. (2006) employed a word-translation task in each of the participant’s L1 Hebrew, L2 English, and L3 French languages. The translation task was designed to identify lexical retrieval and measured both the accuracy and speed with which the lexical units of different languages were accessed and retrieved. The researchers predicted based on the results of the conversation task that the translation task would reveal asymmetric connections between language pairs. An asymmetric connection between the lexical units of two languages means that forward processing from L1 into L2 and backward processing from L2 in L1 occur at different speeds, which materializes in the participant’s response time, usually measured in milliseconds (Eckert, 2006). It has been shown that in the case of a bilingual backward lexical processing takes place faster than forward lexical processing. This is due to the fact that L2 words seem to develop stronger links with L1 words first, and through L1 words they become connected to representations in the conceptual store. On the contrary, L1 words are linked primarily to the representations in the conceptual store, and through the conceptual store become linked to L2 words (Potter et al., 1984; Sholl et al., 1995). This difference in the processing route of the lexical units belonging to two different languages results in longer reaction times for forward processing and shorter reaction times for backward processing.
However, a more recent trilingual translation study carried out by Francis and Gallard (2005) with fluent English-Spanish-French participants showed that accessing the conceptual store was a component cognitive process in the translation for all language pairs. Thus, they posit a symmetrical connectivity model for lexical organization in the MML in the case of fluent trilinguals.
In the word translation task, Goral and his colleagues (2006) used groups of single words and imposed language pairs for the task. The six blocks of language pairs were as follows:
L1 Hebrew into L2 English, L1 Hebrew into L3 French L2 English into L1 Hebrew, L2 English into L3 French L3 French into L1 Hebrew, L3 French into L2 English
The word translation test provided additional support for the asymmetrical lexical processing model (Potter et al., 1984; Sholl et al., 1995) and confirmed the researchers’
hypothesis regarding the asymmetric connections between the participant’s languages. As anticipated by Goral et al. (2006), the forward translation from L1 Hebrew into L2 English and from L1 Hebrew into L3 French took more time than the backward translation from L2 English into L1 Hebrew and L3 French into L1 Hebrew. In addition, the results revealed that
there was no response-time difference between the translation pairs L2 English into L3 French and L3 French and L2 English. This latter result seems to support both the base language hypothesis (Chandrasekhar, 1978) and the parasitic strategy hypothesis (Hall, 1992; Hall & Ecke, 2003) of lexical organization and connectivity.
2.3.2.2 Non-Experimental Research Studies Investigating the MML
The interplay of languages in the MML has also been approached from a naturalistic perspective. Within the naturalistic framework, the qualitative research approach dedicates considerable attention to the participants’ individual characteristics, including their language learning histories and attitudes towards language use. They are based on the central premise that “the functions and use of the languages are subject to variation according to the speaker’s biography” (Clyne, 1997, p. 97). Very often, data collection takes place in the same or similar environments in which the participants live and use their languages, thus the ecology of language learning and production is taken into consideration. Furthermore, in contrast to experimental studies of the MML, which generally focus on the collection, measurement, and assessment of meaning behind particular variables, qualitative studies seek to uncover the underlying processes of mental lexicon organization and functioning.
Qualitative studies of the MML are descriptive studies (Trochim, 2000) which observe how different multilingual linguistic processes occur and materialize in both the expressive and receptive language modalities. They focus on collecting information that may support the creation of future hypotheses or invalidate existing theories.
An early qualitative study of the MML is that of Clyne (1997) in which he investigated the MML through the intrasentential trilingual codeswitching and conversion patterns present in multilingual speech. He used a variety of research tools to collect information about the multilinguality of his participants’ MML. In his research, he employed semi-structured interviews with questions grouped around different topics inspired by the participants’
mundane activities, and he included a picture-description session to increase the spontaneity of the data from the participants. Clyne (1997) also collected informal conversations recorded in the participants’ households.
The participants in the study were 34 fluent adult trilinguals, who were using their languages in an Australian English context. The participants’ trilinguality included a variety of language combinations. Fifteen participants were users of L1 Hungarian, L2 English, and L3 German. Fifteen participants were users of L1 Dutch, L2 German, and L3 English. Four participants were users of L1 Italian, L2 Spanish, and L3 English.
Semi-structured interviews are research instruments that support the collection of
“descriptions of the life world of the interviewee with respect to interpreting the meaning of the described phenomena” (Kvale, 1996, pp. 5–6).
The interviews Clyne (1997) used in his study had a double aim. They served as tools for gaining in-depth accounts of the multilingual participants’ language learning histories, language use and their attitudes towards languages, and as tools for detecting conversion patterns in multilingual speech. The interviews were conducted in all the languages the participants spoke. In addition to the open-ended questions, Clyne (1997) included pictures in the interviews as a means of eliciting verbal expressions focusing on content vocabulary and to facilitate comparability among individual participants within the groups and across the three groups of participants. Clyne (1997) reasoned that pictures would constitute a common starting point for conversations and, concomitantly, they would trigger different interpretations from the participants.
Self-contained pictures as research tools are a category of visual stimuli whose perception and interpretation require complex cognitive and verbal processes which rely on the individual’s knowledge representation and retrieval (DeLoache, Pierroutsakos, Uttal, Rosengren, & Gottlieb, 1998; Kress & Van Leeuwen, 1996; Mandl & Levin, 1989). Research has demonstrated that pictures are primarily processed via the conceptual store (Craik &
Lockhart, 1972; Marschank & Cornoldi, 1990; Wei, 2003; Weir & Mandes, 2017); that is, the meaning making of images is not language specific. Thus, it can take place without inducing any language modes. This has important implications for the study of language activation in the mental lexicon, especially for the study of the MML, as the process of transmutation of pictorial representation into language allows the researcher to observe how languages contribute to meaning making. A verbal response to a visual stimulus is an instance of intersemiotic translation (Aktulum, 2017; Jakobson, 1963), that is, a conversion from a pictorial code into the linguistic code (Dusi, 2015).
Picture descriptions are common tasks in foreign and second language proficiency assessment (Canning, 1998), and they have been employed in empirical research because of their power to provide information about the steps an individual follows in constructing meaning and about how lexical activation and retrieval take place (Holsánová, 2008). In MML studies, such as Clyne’s (1997), the topic-based picture description tasks can offer information about whether languages interact in the process of meaning making, and if so, what the different interaction patterns are.
To get a different view of the participants’ use of their languages, and implicitly to approach the question of MML organization and functioning from a separate standpoint Clyne (1997) collected tape-recorded informal conversations from his participants. Starting from the fundamental premise that language use is dependent on the language users’ daily
life experience (Reis, 2012), Clyne (1997) reasoned that a naturalistic environment would allow for the collection of linguistic data which was not possible during the interview sessions.
Recorded conversations are “acoustic observations of participants’ momentary environment within the natural flow of their lives” (Mehl & Robbins, 2012, p. 177) and are a form of participatory action research in which participants are asked to document aspects of their lives through recorded-voice material (McIntyre, 2008). The strengths of recording are that it is a less obtrusive method of data collection than direct observation, and it is a way of obtaining real-time data in a natural environment. Its main disadvantages are connected to the psychological discomfort they place on the participants, the effects of evaluation apprehension (Cottrell et al., 1968), and the identity of the participants (Mehl & Robbins, 2012).
Clyne (1997) observed some relevant language connections patterns emerging from both the interview and the conversation data. The participants in the L1 Dutch-L2 German-L3 English group and the participants in the L1 Italian-L2 Spanish-L3 English group in their speech acts performed a close monitoring of their languages with the intention of
“differentiating (…) languages clearly” (Clyne, 1997, p. 103). Namely, they seemed to avoid intentionally cognates belonging to typologically close languages and strived to select lexical units exclusively belonging to the target language. This phenomenon was more visible in the case of the participants who were L1 Dutch, L2 German and L3 English language users, as these three languages belong typologically to the same language group.
The most conspicuous indications for the careful monitoring of language production were hesitations before those lexical units whose form is shared by two languages, and occasional false starts in which first syllables of words from the non-target but typologically close language were present. These results are in line with the language inhibitory control hypothesis (Green, 1998; Poarch & Van Hell, 2012) which attributes break downs in communication, materializing as intrusions from the unwanted language(s) and hesitancies in speech, to the load placed on language processing by the cognitive process of inhibitory control (Poarch & Van Hell, 2012).
Clyne’s (1997) findings also support Matutin’s (1993) results regarding the participants’ conscious effort in avoiding the use of cognates. Both Matutin’s (1993) and Clyne’s (1997) research findings regarding multilinguals consciously monitoring their language production and paying special attention to avoidance of particular words proffer a view of a MML in which languages connect based on the shared characteristics of the lexical units and can be kept apart with some effort. Clyne (1997) noted the significance of language proficiency as a factor which seems to have direct influence on the participants conscious monitoring of language production. This effort, Clyne (1997) noted, becomes less visible as the language user gains more proficiency in each of his non-native languages.
Clyne’s (1997) collected data also indicated that in the case of languages that are less stable in the MML, being in the incipient phase characteristic to the process of new-language learning, a parasitic relationship documented by Hall (1992) and Hall and Ecke (2003) seemed to emerge between a support-language and a dependent language. This support language-dependent language relationship emerged as a consistent pattern between typologically close languages such as L1 Dutch and L2 German, or L1 Italian and L2 Spanish. Interestingly, Clyne (1997) remarked, no such links manifested between the L2 German and L3 English of the L1 Hungarian-L2 German-L3 English trilinguals. The parasitic connections became visible in the case of lexical units which shared similar characteristics (formal, phonologic and semantic) across two languages. In connection with shared characteristics of lexical units across languages, Clyne (1997) also identified a special kind of concerted transfer, which he termed “trilingual identification” (p. 110). He observed that if there are shared similarities between the lexical units of two of the languages in the MML, these features may transfer to the third language. Most interestingly, trilingual identification seemed to operate even in the case of typologically distant languages. This is a both captivating and quite unique finding, and it has not been documented since Clyne’s (1997) study.
In addition to trilingual identification, Clyne (1997) distinguished some trilingual code-switching instances in the data he collected. He argued that this very rare phenomenon seemed to have been triggered by the similarities certain lexical units share across particular languages. He observed that code-switching in which lexical units from three different languages were present always occurred between typologically close languages. Namely, this special type of code-switching manifested in the L1 Dutch-L2 German-L3 English context.
Clyne (1997) concluded that in the MML connections between languages at the level of the lexical units appear to be determined firmly by language typology, and beyond typology, by shared similarities across languages. He also remarked that the inhibitory control at work during language production is more prominent in the case of less well-established lexica, but it becomes less conspicuous along with improved language proficiency. He noted that purely cognitive and linguistic considerations might not be enough to answer the question of why language inhibition and selection in the case of multilingual individuals happen, suggesting that these processes may be also guided by personality factors and a general tendency towards the most effective behaviour.
Another research study which accorded special attention to language typology in the organization of the MML was carried out by Cenoz (2003). The researcher used an L3 non-timed intersemiotic translation task to observe how typologically less related languages