From obscure echo to language of the heart: Multilinguals' language choices for (emotional) inner...

43
From obscure echo to language of the heart: Multilinguals’ language choices for (emotional) inner speech 1 Jean-Marc Dewaele Birkbeck College, University of London 1. Introduction Inner speech is a running commentary on significant aspects of ourselves and our world (Morin, Uttl and Hamper, 2011). Morin et al. (2011) carried out an investigation into inner speech frequency and content using self-report and thought sampling measures. Their findings, based on the feedback of 380 undergraduate student volunteers (89% had English as an L1), showed that inner speech is typically about the self (evaluating the self, emotions, physical appearance, and relationships) and often about others (family, friends, and intimate partner). Inner speech was reported about school, work, sports, and leisure activities (p. 1714). Self- reported inner speech was found to serve self-regulatory, problem-solving, and mnemonic functions. The researchers report that self-reports of inner speech were reliable but that there was a problem with validity: very weak correlations between other measures of inner and private speech indicated “low convergent validity among these measures” (Uttl, Morin and Hamper, 2011:1722). The authors do not seem worried by the fact that 11% of their participants were not native speakers of English, and that their responses may have differed from the rest of the participants. Also, no attempt was made to check the complete language profile of the participants, including the English first language (L1) speakers. They could have been bilingual L1 users, or early second language learners, or almost certainly second or third language learners through 1 To appear Journal of Pragmatics (2015) http://dx.doi.org/10.1016/j.pragma.2015.06.014 1

Transcript of From obscure echo to language of the heart: Multilinguals' language choices for (emotional) inner...

From obscure echo to language of the heart: Multilinguals’ language choices for (emotional) inner speech1

Jean-Marc DewaeleBirkbeck College, University of London

1. IntroductionInner speech is a running commentary on significant aspects of ourselves and our world (Morin, Uttl and Hamper, 2011). Morin et al. (2011) carried out an investigation into inner speech frequency and content using self-report and thought sampling measures. Their findings, based on the feedback of 380 undergraduate student volunteers (89% had English as an L1), showed that inner speech is typically about the self (evaluating the self, emotions, physical appearance, and relationships) and often about others (family, friends, and intimate partner). Inner speech was reported about school, work, sports, and leisure activities (p. 1714). Self-reported inner speech was found to serve self-regulatory, problem-solving, and mnemonic functions. The researchers report that self-reports of inner speech were reliable but that there was a problem with validity: very weak correlations between other measures of inner and private speech indicated “low convergent validity among these measures” (Uttl, Morin and Hamper, 2011:1722). The authors do not seem worried by the fact that 11% of their participants were not native speakers of English, and that their responses may have differed from the rest of the participants. Also, no attempt was made to check the complete language profile of the participants, including theEnglish first language (L1) speakers. They could have been bilingual L1 users, or early second language learners, or almost certainly second or third language learners through

1 To appear Journal of Pragmatics (2015) http://dx.doi.org/10.1016/j.pragma.2015.06.014

1

schooling. Does this matter? It certainly does affect validity.In her theory of language embodiment, Pavlenko (2005, 2012) explains that young children acquire their first language(s)(L1) as part of affective socialization. This is “the process of integration of phonological forms of words and phrases with information from visual, auditory, olfactory, tactile, kinesthetic, and visceral modalities, autobiographical memories, and affect” (2012:421). Words and expressions in L1(s) acquire strong affective and autobiographical dimensions. The acquisition of foreign languages (LX) is usually very different as it is much more decontextualized, i.e. the LX classroom, with a more limitedrange of speakers, situations, registers, vocabulary, and fewer “opportunities for integration of all sensory modalities and verbal conditioning (other than foreign language anxiety) and thus leads to the development of ‘disembodied’ words, used freely by speakers who do not experience their full impact” (p. 421).These questions come up in Eva Hoffman’s (1989) fascinating linguistic autobiography. She describes the upsetting experience of moving from Krakow in Poland to Vancouver, Canada with her family as a young teenager in 1949. As the Canadians could not pronounce her name “Ewa” correctly, she becomes “Eva”, a new identity that is forced on her: “Our Polish names (...) were as surely us as our eyes or hands. These new appellations, which we ourselves can’t yet pronounce, are not us. They are identifications tags, disembodied signs pointing to objects that happen to be my sister and myself” (p. 105). It is clear that her main preoccupation at the time was to be accepted by her Anglophone peers and to regain a sense of self and a new voice:

“Since I lack a voice of my own, the voices of others invade me as if I were a silent ventriloquist. They ricochet within me, carrying on conversations, lending metheir modulations, intonations, rhythms. I do not yet possess them; they possess me (...) Eventually, the voices enter me; by assuming them, I gradually make them

2

mine. I am being remade, fragment by fragment, like a patchwork quilt; there are more colors in the world than I ever knew” (p. 220)

The change turns out to be more abrupt than expected as her language preference for inner speech shifts suddenly from her Polish L1 to her newly acquired English L2:

“I wait for spontaneous flow of inner language which usedto be my nighttime talk with myself...Nothing comes. Polish, in a short time, has atrophied, shriveled from sheer uselessness. Its words don’t apply to my new experiences, they’re not coeval with any of the objects, or faces, or the very air breath in the daytime” (p. 107).

After several years in an English-speaking environment, having obtained a doctorate in literature from Harvard University, and having become a journalist, English has clearly established itself as Eva’s preferred language: “Perhaps I’ve read, written, eaten enough words so that English now flows in my bloodstream” (p. 243).The once dominant Polish has completely disappeared from herinner dialogues:“When I talk to myself now, I talk in English. ...If I tried talking to myself in my native tongue, it would be astumbling conversation indeed, interlaced with English expression. So at those moments when I’m alone, walking, or letting my thoughts meander before falling asleep, the internal dialogue proceeds in English” (p. 272).

Although Hoffman claims the shift from Polish to English forinner speech happened in a short time, this might not necessarily be completely true, nor be the case for all multilinguals. It is likely that a number of factors contribute to speed of the shift. In Eva’s case this was surely the fact of being fully immersed in Anglo-Canadian society as a young teenager, learning English both in the classroom context and using it with her new friends. A strong desire to fit in this new environment, and to regain the social status she had attained in Poland, drove her motivation to master the language and culture of the

3

cultured members of the community. As linguists, we may be forgiven for focusing on language and communication, but as Hoffman (2011) points out, her acculturation into Anglo-Canadian culture also involved an awareness of the differentconstructions of time prevailing in her two worlds: “It was not only time that moved faster in America – it pressed onwards in more stressful ways” (p. 5). This implied not just an adaptation of her own time-management “but of the self and one’s internal arrangements” (p. 6). This particular conceptual restructuring led her to become “temporally bicultural” (p.10).

Despite the fact that the question of language choice for inner speech is well-known amongst bi- and multilinguals, “The inquiry into inner speech is still in its emerging stage and thus lacks sufficient data for a chapter-long overview” (Pavlenko, 2011:242). The aim of the present paper is to help fill that gap by providing evidence on multilinguals’ preferences for the language of inner speech and, in particular, emotional inner speech. Systematic research on language choice for inner speech is warranted because it allows researchers to tackle questions of multilingual identities, as it seems that the language of a person’s inner dialogues has a special status. Research on language choice for inner speech also raises interesting questions about volition: to what extent do multilinguals actually control the language for inner speech and does thiscontrol weaken as language dominance shift from an L1 to LX?Finally, why do multilinguals who use two languages all the time and feel maximally proficient in them, still typically prefer the L1 for inner speech (Dewaele, 2011)?

The present paper is organized as follows: The literature review will start with studies on inner speech and continue with research on language preference for communicating emotion that considered the independent variables included in the present study, namely context of acquisition, age of onset, self-perceived proficiency, general use, socialization, emotionality of the LX, education level, age

4

and gender. The methodology section includes the instrumentused to collect data, the Bilingualism and Emotion Questionnaire (BEQ) (Dewaele & Pavlenko, 2001-2003), the participants, the research design, and the independent and dependent variables. The research questions will be presented subsequently. The results section contains the statistical analyses and a visualisation of the main resultsfor the L2. The findings will be linked to the existing literature in the discussion section, after which some tentative conclusions will be presented.

2. Literature review

2.1. Inner speech researchBefore embarking on a short overview of the literature on inner speech, it is important to clearly define the term. A pioneer in the field, Guerrero, defined inner speech as “theinternal (covert) and nonaudible forms of speech for oneself, such as mental rehearsal and internal self-talk” (2005:xii). Pavlenko (2011) offered a slightly more elaborated definition: “subvocal or silent self-talk, i.e. mental activity that takes place in an identifiable linguistic code and is directed primarily at the self (the dialogic function of inner speech). This mental activity constitutes an important aspect of the language/cognition interface, even though it is not representative of all ‘thinking’” (p. 242). It is not entirely clear to what extent multilinguals are able to determine a language of inner speech. In some cases, they may deliberately use -or avoid- an L1 or an LX, reflecting a resistance to acculturate, or, alternatively, a willingness to acculturate. Pavlenko (2014) argues that language choice forinner speech is mostly outside the user’s conscious control:“with the exception of deliberate mental rehearsal, we cannot force inner speech in another language any more than we can switch an inner language off” (p. 211).

Researchers working on the sociocultural paradigm have been fascinated by the genesis of L2 inner speech. Guerrero

5

(1999:48) argued that: “L2 inner speech starts out as involuntary mental playback of the L2, representing the internal projection of external speech, a sort of internal ‘echo’ of the speech being heard”.Guerrero (2004) explored the early stages of L2 inner speechdevelopment using diary and stimulated recall techniques. She followed 16 beginning ESL college students over a four-month period. Students kept diaries on the inner speech they had experienced in their L2 during class and outside the classroom. Guerrero used stimulated recall technique to obtain clarification and expansion of the diary entries. Shefound four main types of reported L2 inner speech (in order of decreasing frequency): “(1) concurrent processing of language being heard or read, (2) recall of language heard, read, or used previously, (3) preparation before writing or speaking, and (4) silent verbalization of thoughts for private purposes” (p. 90). She found that in the early stages L2 inner speech is typically the result of deliberateand spontaneous attempts to internalize external social L2 speech. She hypothesized that such inner speech activity is a crucial step towards the eventual development of the L2 asa tool for thought. She concluded that early attempts at internalizing external social L2 speech are necessary for the development of L2 inner speech but stressed that the L1 continues to be a predominant tool, even for advanced L2 learners/users.

Pavlenko (2014) included a section in her book focusing on language preferences for inner speech and the implications for thought, presenting a number of quotes from participantsin the BEQ. She found that many respondents identified inner speech with thought: Fanny, L1 Afrikaans, L2 English, L3 German, L4 Xhosa: “Afrikaans is my L1. It is the languagein which I think and feel most” (p. 212). Pavlenko (2014) also looked for evidence in the literature on internalization of L2 inner speech, where the voices of others “invade” the protagonist, such as in the extract by Hoffman quoted earlier. What starts as spontaneous playback“can be reinterpreted as appropriation of the voices of

6

others” (p. 215). Pavlenko observed that inner speech has distinct functions in the L1 and L2: inner speech in the L1 is used for self-regulation and self-evaluation, inner speech in the L2 starts as “imitation, repetition of other’sutterances, vicarious responses (...), mental rehearsal (...), and language play” before being used more for dialogic purposes (p. 218). While some authors, like Hoffman, report the loss of an L1 for inner speech, others keep the LX at bay and maintain their L1 for inner speech. This was the case of Nabokov who “maintained Russian as the language of everyday family communication, inner speech and even poetry, yet his diary was written in English, both in America and in francophone Switzerland” (p. 221). This flexibility was also reported by one of the BEQ participants: “Therèse, L1 French, L2 English, L3 German: I do use both French and English because that’s how sentences come to my mind. They come naturally in one of the languages because they are related to particular situations or experiences and also because sometimes there seems to be no exact way to say it in the other language” (p. 221).

Larsen, Schrauf, Fromholt and Rubin (2002) is one of the pioneering empirical studies into inner speech. The authorslooked at language choice for inner speech among 20 immigrants of Polish origin who had fled to Denmark where they had lived for 30 years at the time of the study. The authors compared two groups: ten “early immigrants” who wereon average 24 years old at the time of immigration, and 10 “late immigrants” who were on average 34 years old when arriving in Denmark. The group of early immigrants reported more use of Danish for inner speech, whereas the group of late immigrants reported more use of Polish. It thus seems that age at arrival had a persistent effect in language choice for inner speech.

Cook (1998) looked at language choices of 59 multilinguals for external and internal uses. The latter included mental calculation, memory tasks, dreaming, talking and singing to oneself, and praying. Descriptive statistics suggested a

7

preference for the L1 for the internal uses, and a higher frequency of use of the L2 for external (social) communication.

Two previous empirical studies have included inner speech asa dependent variable using data from the BEQ (Dewaele 2004, 2011). Dewaele (2004) looked how perceived L1 attrition affected the use of the L1 in different domains: general use, swearing, the expression of anger and feelings; inner speech and mental calculation. It also included the effect of L1 attrition on perceived emotional force of swearwords. Statistical analysis showed that language choice for these different types of emotional and non-emotional speech were similarly affected by dominance in an LX. L1 use for inner speech and general use both diminished among those who reported no longer being dominant in their L1 (perceived L1 attriters). The LX-dominant group used the L1 significantly less for inner speech than the L1-dominant and the L1+LX dominant groups, but some LX-dominant participants did use their L1 regularly. Moreover, only 14% of LX-dominant participants declared to never or rarely use their L1 for inner speech, with more than 80% of the LX-dominant participants using the L1 sometimes, frequently or all the time. In other words, LX dominance and self-perceived L1 attrition do not completely wipe out the use of the L1 for inner speech.Dewaele (2011) focused on a subsample of the BEQ, namely 386bi- and multilinguals who declared that they were maximally proficient in their L1 and an LX and used both constantly. Despite this similarity in use and perceived proficiency, participants preferred to use the L1 for communicating feelings or anger, swearing, addressing their children, performing mental calculations, and using inner speech. Theyalso reported their L1 to be emotionally stronger than theirLX and reported less communicative anxiety in their L1. Thequantitative analysis was complemented by interview data from 20 multilinguals. One participant, Michelle (Taiwanese L1, Mandarin L2, English L3), resident in the United Kingdomfor 17 years and married to an English speaker, reported

8

that her inner speech shifted from English to Mandarin depending on the linguistic identity of the people she’s thinking about:

“M: Most of the time I do talk to myself in English, but recently because I have more opportunity to speak to people in Chinese, and then, it’s very strange because when I talk to myself, to myself I’m speaking Mandarin because I think about my friends, you know, does that make sense?” (p. 43).

What this suggests is that language preferences for inner speech remain dynamic, and can shift, reflecting changes in the linguistic environment.

Other researchers have been inspired by the BEQ. Ewert (2010) focused on external and internal language use by 17 Polish multilingual students, using BEQ items, but limiting the analyses to descriptive statistics only. The most frequent language for external use was Polish, followed by English and with some Russian also (p. 165). The feedback tothe question “If you form sentences silently (inner speech),what language do you use?” (with a 5-point Likert scale) showed that Polish was the preferred language (Mean = 3.3) followed by English (Mean = 2.4) and Russian (Mean = 1.8) (p. 166).

Sansault (2012, 2014) also used the BEQ as a point of departure for an interview study including 11 young Polish-French bilinguals living in Paris, and more specifically theemotional load of their languages and their linguistic practices, including inner speech. Some participants felt that their Polish language and culture was under attack and had to be defended against the overwhelming presence of French in the environment. Most participants reported a preference for Polish to express anger, but some had no clear preference between the languages. Two participants whohad started using French before the age of 6 reported that they used both Polish and French for emotional inner speech.Agata explained that her inner language choice was guided bythe context, with regular code-switching: “ça dépend de la

9

situation, même des fois en Pologne quand je suis en train de parler avec des Polonais, ça m'arrive de penser en français (...) donc ça fait soit les deux, soit le mélange” (Sansault, 2012:201) (‘it depends on the situation, even sometimes in Poland when I speak with Poles, it happens thatI think in French (...) hence sometimes one or the other language, or a mixture’).

Hammer (2015) used items from the BEQ to analyse variation in inner speech of 149 adult Polish-English bilinguals (average age: 31) who had immigrated to the UK at an averageage of 23. All participants had university degrees and werefluent in English. She found that higher levels of acculturation and larger English-speaking social networks were significantly linked to more frequent use of English L2for inner speech. Other factors linked to use of English forinner speech were perception of emotionality of the L2, and an intention to remain in the UK.

2.2. Research on factors linked to language preferencesPavlenko (2011, 2014) noted that context of acquisition is an important predictor of LX performance, as “immersion in the L2 context favors conceptual restructuring” (2011:250). This involves the modification of boundaries of categories so that the LX users “perform in accordance with language-specific constraints in each language” (Pavlenko, 2009:138).Whether a language is acquired naturalistically, in an instructed setting or in a combination of both has significant effects on the perception and the use of that language later in life (Dewaele, 2013). Languages learnt exclusively through formal instruction were less likely to be perceived as emotional, they also elicited higher levels of foreign language anxiety and participants reported lower levels of use (Dewaele, 2013). One possible reason for thisis the fact that “context of acquisition roughly equates with type and intensity of exposure to an LX and the opportunity to use that LX in authentic interactions. If contact with an LX had been limited to the classroom, stylistic range and emotion scripts would inevitably be more

10

limited compared to those LX users who have experienced and used the language in a wider variety of situations” (p. 217). Context of acquisition emerged as a significant predictor for mental calculation in the L2, L3 and L4 (Dewaele, 2007). Participants who learned an LX in an instructed setting were less likely to use that language formental calculation in their adult life. Participants preferred to do calculations in the language in which they had been taught arithmetic.

Dewaele (2009) found significant AoA effects in the BEQ database on self-perceived communicative competence, language choice for emotional speech, inner speech and for mental calculation. Early starters felt significantly more proficient in their oral and written foreign languages. The same pattern for AoA was also found on frequency use of the LX for feelings, inner speech and mental calculation in the L2 and the L3 (but less so in the L4). Early starters were more likely to use their L2 and L3 for mental calculation (Dewaele, 2007).

Self-perceived communicative proficiency reflects individuals’ evaluation of their ability to communicate withvarious interlocutors in a wide variety of settings and situations. It is based on the confidence of mastering grammar, phonology, lexis, syntax, pragmatics and non-verbalcommunication among LX users. Ożanska-Ponikwia and Dewaele (2012) considered the link between personality traits, frequency of use and self-perceived proficiency in English L2 by 102 adult Polish immigrants living in Ireland and the United Kingdom. Length of stay was positively linked with frequency of English L2 use and self-perceived proficiency. Immersion in the L2 was not the only factor to affect the dependent variables: the migrant’s inclination to seek out social interactions in theL2 played an important role. Participants scoring high on Openness, Extraversion and Self-esteem used the L2 more frequently and felt more proficient in the L2.

11

Hammer and Dewaele (to appear) looked at variation in self-perceived proficiency of Polish-English bilingual migrants living in the United Kingdom and found it to be linked to acculturation level, length of time in the United Kingdom, age of onset, frequency of L2 use and age. Participants who were highly acculturated rated their L2 proficiency significantly higher than moderately or less acculturated participants. Frequency of L2 use was linked to higher L2 proficiency levels. Language proficiency plays a role in conceptual restructuring: “advanced L2 speakers - but not low-level or intermediate ones - approximate L2 categories and preferences” (Pavlenko 2011:250). Dewaele (2007) found that participants who judged their LX written proficiency in an LX to be high were also more likely to use that LX for mental calculation. Centeno-Cortés and Jiménez-Jiménez (2004) looked at mathematical problem-solving in Spanish L2.They found that students with intermediate levels of proficiency used the L2 mainly for reading the questions butswitched to English to solve the problem, while advanced L2 students extended its use to the actual thinking process (during the reasoning stage).

Pavlenko (2004) considered the effect of LX socialization onlanguage choice for the expression of emotions in parent-child communication within multilingual families. While a majority of multilingual parents reported a preference for the L1 for emotional communication with their children, someparents reported that as a result of secondary socializationtheir LX had acquired strong emotional connotations and had become their language of preference when expressing emotionswith their children. Dewaele (2010) reported that language preferences for emotional expression, perception of the languages and Foreign Language Anxiety in the BEQ database evolve as LX users become more socialised in the LX. LX socialization was found to be a significant predictor of choosing an LX choice for mental calculation (Dewaele, 2007).

12

Hammer and Dewaele (to appear) also found that the Polish migrants who actively sought out contact with native speakers of English, established large English-speaking networks, boosted their English LX socialization process.

The first language(s) are typically perceived to be more emotional by multilinguals. A majority of BEQ participants reported that the phrase “I love you” had most emotional weight in their L1 (Dewaele, 2008). The perception of the emotional weight of the phrase in the L2 was linked to context of acquisition, age of onset, degree of socialization, network of interlocutors and self-perceived oral proficiency (Dewaele, 2008). Pavlenko (2012) suggestedin her literature review that this effect stems from the wayearly languages are learnt. They are picked up during childhood and have rich emotional connotations, i.e. they are learnt with full involvement of the limbic system and stored in implicit memory. In contrast, languages learnt later in life are often perceived to be disembodied (Pavlenko, 2012). They have typically been acquired throughclassroom instruction, with a stronger reliance on declarative memory, with fewer emotional connotations, whichwould explain why they produce weak physiological responses and a general feeling of detachment.

Dewaele’s (2013) analysis of the BEQ database showed that the pentalinguals rated their L1 significantly more positively on a number of characteristics, including emotionality. They also perceived L1 swearwords as being more powerful than LX equivalents, with values declining forlanguages acquired later. Although LXs could gain in emotionality, a gap often remained between L1 and LX.

To sum up, it seems that research into language preference for inner speech is still in its infancy. Moreover, no researchers have, to our knowledge, compared language choicefor inner speech with that for emotional inner speech. Thismight be an important distinction, as the literature suggests that articulated emotional speech often elicits

13

different language choices compared to articulated non-emotional speech.

3. Method

3.1. The instrumentThe data have been gathered through the Bilingualism and Emotion Questionnaire (BEQ) (Dewaele and Pavlenko, 2001-2003)2. The web questionnaire contained 35 closed and open questions concerning emotional and non-emotional language use in different situations in up to five languages. Data were collected through the snowballing technique, i.e. non-probability sampling. The BEQ was an open-access survey, advertised through several listservs, targeted emails to teachers and students, and informal contacts asking them to forward the link to colleagues and friends. The researchers had no control about who received the call for participation. Participants filled out a short sociobiographical questionnaire with questions about gender,age, education, language history and present language use. The research design and questionnaires received ethical clearance from the research institution. The use of online questionnaires is becoming increasingly popular in applied linguistics because it allows researchers to reach a large audience and collect data very efficiently. The resulting samples may not represent a random sample of the general population but then, this of less importance as participantsmust of necessity be multilinguals with sufficient metalinguistic awareness and willingness to participate in the survey (Wilson and Dewaele, 2010). Reliability is not a major problem in internet-based research as “it is hard to see the attraction of falsifying answers to an on-line language history questionnaire” (p. 108). Gosling et al. (2004) demonstrated that web-questionnaire results generalize across presentation formats and are consistent

2 The BEQ is no longer online but can be found as an appendix in Pavlenko (2005), Dewaele (2013) and be downloaded from the IRIS database (http://www.iris-database.org)

14

with results from pen-and-paper methods. Wilson and Dewaele (2010) concluded that the advantages of web-based studies far outweigh the drawbacks.

3.2. ParticipantsA total of 1579 adult multilinguals (1040 females, 419 males; mean age = 34, SD = 12) contributed to the web questionnaire database used in the present study. The participants spoke a total of 77 different L1s. Anglophone native speakers represent the largest group, followed by native speakers of Spanish, French, German, Dutch, and Italian. The remaining participants share another 71 languages. Participants could only claim one L1 but because information on age of onset of acquisition was collected, 223 participants grew up with two languages from birth, and 28 with three languages3.The most frequent L2 is English, followed by French, Spanish, and German. French is the most frequent L3, followed by English, German and Spanish. The same languagesare the most frequent L4s: German, French, and Spanish. The most frequent L5s are Spanish, German and Italian.The mean age of onset of learning the L2 was 9 years (SD = 6); 14 years (SD = 7) for the L3; 18 years (SD = 7) for the L4 and 22 years (SD = 8) for the L5. The L2 was defined as the second language to have been acquired (though participants could indicate that age of onset was zero, in which case it was a second L1), the L3 the third, etc. Participants are generally highly educated with 155 having only a high school diploma, 418 a Bachelor’s degree, 452 a Master’s degree, and 424 a doctoral degree. This high proportion of highly educated, mostly female multilinguals is typical for on-line language questionnaires (Dewaele, 2010).

3 This slightly boosted the means for L2, but did not distort the general picture, as this group constituted a minority (18%).

15

3.3. Research designThe independent variables selected in the present research design represent language learning history of the LX (context of acquisition, age of onset), general use, socialization in LX, perception of L1 and LX (emotionality and self-perceived proficiency) and sociobiographical variables (education level, age and gender).The effects of the independent variables will be calculated separately for every dependent variable in every LX and the L1 - when applicable.

3.4. Independent variables

3.4.1. LX context of acquisitionThree types of contexts of acquisition were considered and ordered according to the amount of extra-curricular contact with the target language: 1) instructed context (i.e. formalclassroom contact only); 2) mixed context (i.e. classroom contact and naturalistic contact); and 3) naturalistic context (i.e. no classroom contact, only naturalistic communication outside school).

3.4.2. Age of onset of acquisition of an LX (AoA)Participants were asked at what age they started learning their languages. Information was collected for the L2, the L3, L4 and the L5.Five AoA groups were created for each language: those who started learning the LX between birth and age two, between three and seven, between eight and twelve, between thirteen and eighteen and those who learned the language at age nineteen or older.

3.4.3. Self-perceived proficiency in L1 and LXSelf-perceived oral proficiency was measured through a 5-point Likert scale.Questions focused on self-perceived competence in speaking: “On a scale from 1 (least proficient) to 5 (fully fluent) how do you rate yourself in speaking your L1/L2/L3/L4/L5?”

16

Possible answers included: (1) Minimal, (2) Low, (3) Medium,(4) High, (5) Maximal.

3.4.4. General frequency of use of L1 and LXParticipants were asked how frequently they used each of their languages. This included the following choices on a 5-point Likert scale: 1) Yearly or less, 2) Monthly, 3) Weekly, 4) Daily, 5) All day.

3.4.5. LX socializationThe variable ‘‘socialization in the LX’’ is a derived variable based on the difference in the general frequency ofuse of the L1 and an LX (the L2, L3, L4 or L5). The subtraction of the score for the L1 and the score for the LXgives a value that reflects the difference in frequency of use of the L1 and the LX. For example, if a participant indicated a much more frequent use of the L1 than the L2, itwould mean a very weak degree of socialization in the LX. If, on the other hand, the LX was used much more often than the L1, it would suggest a much stronger level of LX socialization. The resulting scale presents the following range: 1) very weak, 2) weak, 3) moderate, 4) strong.

3.4.6. Emotionality of the L1 and LXEmotionality of participants’ languages was operationalised through the following item: “Here is a subjective statement about the languages you know. Please mark to what extent it corresponds to your own perceptions. (My L1/L2/L3/L4/L5 is …emotional).” Possible answers on a 5-point Likert scale included: 1 = not at all, 2 = somewhat, 3 = more or less, 4 = to a large extent, 5 = absolutely. The aim of the question was to have a general measure of the attitude towards a language, ranging from detached and disembodied toembodied and emotional (Hoffman 1989, Pavlenko 2012).

3.4.7. Sociobiographical informationParticipants were asked about their highest diploma (rangingfrom High school, Bachelors, Masters or PhD), which represented education level. Age groups were created

17

(teenagers, those in their twenties, thirties, forties, fifties and sixties and above). Twenty-one participants didnot provide information about their gender.

A Spearman rank correlation analysis of the independent variables for the L2 of participants shows that many of themare significantly correlated with each other (see appendix).The strongest correlation emerged for general use and L2 socialisation, with variables sharing 60% of the variance, which means that despite the overlap, they are still separate dimensions. General use and proficiency shared 38% of variance, but few of the other variables shared more than10% of variance.

3.5. Dependent variablesSelf-reported frequency of language choice for inner speech and emotional inner speech are the dependent variables in the present study. The distinction between them could be seen as a nested design, with “inner speech” representing general inner use, and emotional inner speech referring to more specific inner use of the language.

Data were obtained through the two following questions in the BEQ:

1) If you form sentences silently (inner speech), what language do you typically use?

2) What language do you express your deepest feelings in (when you are alone)?4

Information was collected for all languages on a 5-point Likert scale: (never = 1, rarely = 2, sometimes = 3, frequently = 4, all the time = 5). The dependent variables are thus two separate sets of numerical values reflecting frequency of use of a language for inner speech and emotional inner speech.4 This is based on the assumption that people typically do not use stretches of overt speech when they are alone, they may use, at most, the odd word.

18

A series of one-sample Kolmogorov-Smirnov tests revealed that the values for inner speech and emotional inner speech in the five languages are not normally distributed (Kolmogorov-Smirnov Z values vary between 0.2 and 0.43 for inner speech, all p < .0001; and 0.19 and 0.33 for emotionalinner speech, all p < .0001). For inner speech and emotional inner speech the distributionof participants across 5 frequency categories (ranging from ‘‘never” to ‘‘all the time”) are skewed towards the high endof the continuum for the L1 and L2, and are skewed towards the low end of the continuum for the L3, L4 and L5.

As a consequence, Spearman Rank tests were used as nonparametric equivalents to Pearson’s r tests, Mann-Whitneytests were used instead of t-tests and Kruskall Wallis testswere used instead on ANOVAs. The disadvantage is that thesetests are less powerful and do not allow to measure how muchvariance each independent variable explains (as MANOVAS or multiple regression would allow) nor does it permit to investigate interactions.

The BEQ also contained a number of open questions on language preferences for emotion speech, some of which contained spontaneous mentions of inner speech.The BEQ data were complemented by interview data collected from 20 multilinguals living in the UK (12 females, 8 males) who filled out the BEQ before being interviewed by a research assistant on the topics covered by theBEQ. The aim of these interviews was to find out the reasonsbehind language choices in emotional interactions. The resulting corpus consists of 115,000 words. This group of 20participants had a similar socio-demographic make-up as the participants in the BEQ. Extracts from the interviews will be used to give participants a voice and to illustrate some of the statistical patterns uncovered in the BEQ.

19

4. Research Questions1) Are languages that have been acquired early used more frequently for inner speech and for emotional inner speech?2) In there a link between frequency of use of an LX for emotional inner speech and for inner speech?3) Is the LX used less frequently for emotional inner speechthan for inner speech?4) What is the relationship between context of acquisition, AoA, self-perceived proficiency in L1 and LX, general use ofL1 and LX, LX socialization, L1 and LX emotionality, education level, age group, gender and frequency of use of the L1 and LX for inner speech and for emotional inner speech?

5. ResultsLXs were used significantly less frequently than the L1 for emotional inner speech. The same pattern emerged for inner speech (table 1). A Friedman’s ANOVA tests for related samples revealed that this effect is highly significant (table 2). The decline from L1 to L5 is monotonic for both dependent variables (see figure 1). The 486 pentalinguals5 use the L1, on average, frequently for inner speech and emotional inner speech. The L2 is used, on average, betweenfrequently and sometimes for inner speech. It is closer to sometimes for emotional inner speech. The L3 is used between rarely and sometimes for inner speech, and rarely for emotional inner speech. The L4 and L5 are, on average, rarely used forinner speech, and even less frequently for emotional inner speech. A typical comment on the preference for L1 for inner speech and emotional inner speech was that by Carlo (Portuguese L1,

5 Their sociobiographical profiles do not differ from that of the complete sample. Mean age was 36.8 (SD = 11.5) and 69 per cent were women (n = 336). Levels of education were equally high: 7 percent had a high school degree (n = 34), a quarter had a BA (n = 117), a third had a MA (n = 156) and a third had a PhD (n = 179).

20

Spanish L2, English L3, French L4): “L1 because it is more direct into my line of inner thought”.However, Corrado (Italian L1, English L2, Russian L3, FrenchL4), reported that an LX in which he had high but not maximal proficiency, allowed him to express his feelings when alone just as well as his L1: “I would use mainly Italian, i.e. my first language, even though I could also use Russian (...) I consider this language a great medium toexpress my inner feelings.”

Table 1: Friedman’s ANOVA tests for related samples (L1, L2,L3, L4, L5)

Variable Nd.f. χ2 p

Inner Speech 378 4 767.7 .0001Emotional Inner

Speech 409 4 852.2 .0001

21

Table 2: Friedman’s ANOVA tests (mean ranks) (L1, L2, L3, L4, L5)

Language Inner Speech Emotional innerspeech

L1 4.5 4.6L2 3.6 3.5L3 2.8 2.7L4 2.2 2.2L5 2.0 2.1

Figure 1: Mean frequency for inner and inner emotional speech of all languages in the complete database

A series of Spearman Rank correlations revealed a strong positive relationship between frequency of use of inner speech and for emotional inner speech in all languages: L1 (d.f. = 1409), rho = .54, p < .0001; L2 (d.f. = 1329), rho = .73, p< .0001; L3 (d.f. = 967), rho = .70, p < .0001; L4 (d.f. = 665), rho = .65, p < .0001; L5 (d.f. = 352), rho = .62, p < .0001. These figures suggest sufficient validity.

A Wilcoxon Signed Ranks Test revealed that the values for frequency of use for inner speech and for emotional inner

22

speech were significantly different within each of the 5 languages (see table 3). The L1 was used significantly more frequently for emotional inner speech than for inner speech but opposite patterns emerged for the LXs, with higher frequency for inner speech than for emotional inner speech. The z-scores range between 6 and 14, meaning there is a likelihood of less than 0.001% of arising by sampling error,assuming the null hypothesis to be true (one-tailed hypothesis).

23

Table 3: Intra-language difference in frequency for inner speech and emotional inner speech (Wilcoxon Signed Ranks Tests, Z values)

Ranks NMean Rank Sum of Ranks

L1 IES-IS Negative Ranks 191 282 53869Positive Ranks 353 267 94371Ties 840Total 1384

L2 IES-IS Negative Ranks 516 378 194901Positive Ranks 197 303 59640Ties 575Total 1288

L3 IES-IS Negative Ranks 415 263 108588Positive Ranks 91 216 19683Ties 462Total 968

L4 IES-IS Negative Ranks 210 137 28720Positive Ranks 50 104 5211Ties 405Total 665

L5 IES-IS Negative Ranks 115 68 7855Positive Ranks 21 70 1461Ties 217Total 353

Test Statistics

L1 IES-ISL2 IES-

ISL3 IES-

IS L4 IES-ISL5 IES-

ISZ -6.02c -12.95b -14.10b -10.14b -7.31b

Asymp. Sig. (2-tailed) 0 0 0 0 0

b Based on positive ranks.c Based on negative ranks.

24

The next research question deals with relationship between independent variables and frequency of use of a language foremotional inner speech and inner speech. Kruskall Wallis tests were carried out separately (see tables 4 and 5). Mean ranks are reported for every test but the figures visualise the patterns of mean values (for the L2 only)6.

Table 4: The link between independent variables and frequency of use of inner speech in the LX (Kruskall Wallis Chi2)  L1 L2 L3 L4 L5Context of acquisition (d.f. = 2)

N/A 58*** 73*** 79*** 42***

Age of onset (d.f. =4) N/A 13*** 4 3 3

Self-perceived proficiency (d.f. = 4)

105*** 335*** 335*** 201*** 106***

General use (d.f. = 4)

216*** 389*** 306*** 196*** 103***

LX socialization (d.f. = 4) N/A 262*** 184*** 92*** 49***

LX emotionality (d.f. = 4) 19** 41*** 61*** 27*** 18**

Education level (d.f. = 3) 0.4 4 2 2 1

Age group (d.f. = 5) 9 3 9 9 11*

Gender (d.f. = 1) (Mann Whitney test, Z)

-0.6 -3.3** -0.3 -1.8 -1.1

* p < .05. ** p <.001. *** p < .0001

6 This is because non-parametric tests calculate mean ranks rather than mean values, but it is easier to interpret the latter in graphs.

25

Table 5: The link between independent variables and frequency of use of emotional inner speech in the LX (Kruskall Wallis Chi2)  L1 L2 L3 L4 L5Context of acquisition (d.f. = 2) N/A 84*** 70*** 63*** 34***

Age of onset (d.f. = 4) N/A 30*** 17** 2 6

Self-perceived proficiency (d.f. = 4) 120*** 326*** 328*** 260*** 136***

General use (d.f. = 4) 164*** 327*** 267*** 210*** 132***

LX socialization (d.f.= 4) N/A 296*** 200*** 125*** 84***

LX emotionality (d.f. =4) 59*** 21*** 54*** 31*** 13*

Education level (d.f. =3) 0.3 7 10* 3 1

Age group (d.f. = 5) 2 3 5 4 4

Gender (d.f. = 1) (MannWhitney test, Z) -0.3 -3.7*** -0.5 -1.2 -0.5

* p < .05. ** p <.001. *** p < .0001

Context of acquisition was significantly linked to the use of LX for inner speech and for emotional inner speech (see tables 4 and 5). Participants who had acquired the L2 only through formal instruction used that language significantly less for inner speech and for emotional inner speech compared to those who had learned the L2 in a mixed context or naturalistically. The Kruskall Wallis test shows gradually higher mean ranks for inner speech (formal instruction: 580, mixed context: 623, naturalistic context: 685). A separate Kruskall Wallis for emotional inner speechshows a similar pattern: formal instruction: 543, mixed context: 602, naturalistic context: 703 (see figure 2 for the mean values). Silvio, an Italian resident in the UK for 10 years who had learned English at school in Italy, expressed a clear

26

preference for Italian L1 for inner speech and emotional inner speech. He explained that Italian was unlikely to ever be replaced by English L2 because of his lack of emotional nuance in English: “I don’t have access to the full semantic field, so I don’t know where to locate different feelings, different moods, or different ideas and concepts (…), so it’s as if my keyboard was made out in Italian, yeah, my Italian keyboard, language, is made up of I don’t know 88 keys, and my English language keyboard is just two octaves, right? And I have to play the same melodies.”

Figure 2 Context of acquisition and mean frequency of use ofthe L2 for inner speech and emotional inner speech

AoA only had a significant effect on the use of L2 for innerspeech and the use of the L2 and L3 for emotional inner speech (see tables 4 and 5). Early starters used the L2 more frequently for emotional inner speech compared to laterstarters. The mean rank is highest for the youngest AoA group (i.e. those who started learning before the age of three): 738, with a systematic decrease for older AoA groupswith mean ranks of 711, 625, 611 and 597 respectively (see

27

figure 3 for the mean values). A similar difference emergedfor L2 inner speech (with mean ranks of 723, 596, 564, 545 and 582 respectively. Johanna (Dutch L1, Italian L2, French L3, German L4) reported frequent use of Italian for inner speech and emotional inner speech and linked this preference to early contact with Italian: “I went to school during my first years in Italy and was not able to speak a word of Italian on my first school day. But today (and ever since I masteredthe language) I really love Italian and feel comfortable and‘at home’ when speaking it”.

Figure 3 Age of onset and mean frequency of use of the L2 for inner speech and emotional inner speech

Participants who felt more proficient in speaking the L1 andLX were more likely to use the language for inner speech andfor emotional inner speech (see tables 4 and 5). Little difference exists between those with minimal, low and mediumlevels of self-perceived proficiency in the L2 for inner speech (mean ranks of 437, 331 and 405 respectively) and emotional inner speech (mean ranks of 369, 250 and 397 respectively), but it increases markedly for those with a high and maximal self-perceived level of proficiency for inner speech (mean ranks of 615 and 858 respectively) and

28

emotional inner speech in the L2 (mean ranks of 591 and 846 respectively) (see figure 4 for the mean values).This pattern was confirmed by Chris, a participant with German L1, high proficiency in English L2, French L3 and basic proficiency in Spanish L4: “I talk to myself in German, English, French, or sometimes it’s in Spanish”.

Figure 4 Self-perceived oral proficiency and mean frequency of use of the L2 for inner speech and emotional inner speech

General frequency of use of the L1 and LX was very strongly related to the use of that language for inner speech and emotional inner speech (see tables 4 and 5). Frequency of overall use of an L2 is linked to a linear increase in innerspeech (with mean rank of 294 for yearly use of the L2, increasing to 401 for monthly use, 531 for weekly use, 689 for daily use and 876 for all-day use of the L2. A similar pattern emerged for emotional inner speech (with mean rank of 291 for yearly use, increasing to 448 for monthly use, 496 for weekly use, 682 for daily use and 839 for all day use of the L2 (see figure 5 for the mean values).Klaus, a German L1 participant resident in the UK for 17 years (with English L2, French L3, Russian L4 and Spanish L5), explained that his emotional inner speech would be

29

linked to the context of the emotions: “if it is something, for example, that upsets me about my parents, then it would be in German, but if it is something that affects work then I would use English, but it depends on the context and the environment.” Klaus mentions one exception for one type of inner speech however: “when I count, I very very often countin German, even if it is at work, if I add some numbers I use German, everything else is English in my office.”

Figure 5 General use of the L2 and mean frequency of use of the L2 for inner speech and emotional inner speech

A relatively similar pattern emerges for LX socialization (see tables 4 and 5). The steepest increase in frequency of L2 use between those with very weak and weak levels of L2 socialization for inner speech (with mean rank of 519 for ‘very weak’ and 795 for ‘weak’ L2 socialisation) and for emotional inner speech (with mean rank of 502 for ‘very weak’ and 780 for ‘weak’ L2 socialisation). Frequency of usegrows more steadily at higher levels of socialization for inner speech (with mean rank of 864 for ‘moderate’ and 958

30

for ‘strong’ L2 socialisation) and for emotional inner speech (with mean rank of 841 for ‘moderate’ and 915 for ‘strong’ L2 socialisation) (see figure 6 for the mean values).One participant, Cristina, of Catalan origin but strongly socialized in LX English after 17 years in the UK, made the following comment: “English is now what is more readily available when I’m speaking, the first word that comes to mymind is English, and even when I’m talking to myself, kind of you know inner speech, I don’t necessarily think it’s Catalan”. She attributes this to the fact that Catalan is linked to “a different self, a different person (…). When I use Catalan, I can feel that I’m attached to almost the person that I was before, because my language hasn’t developed and it hasn’t developed with my personality”.

Figure 6 L2 Socialisation and mean frequency of use of the L2 for inner speech and emotional inner speech

Higher rates of perceived emotionality correspond with increased use of the L1 and LX for inner speech and for emotional inner speech (see tables 4 and 5). In the case ofthe L2 it appears that if it is perceived to be emotional, it is more likely to be used for inner speech (Mean rank

31

“not at all” = 516, “somewhat”: = 543, “more or less”= 648,“to a large extent” = 674, “absolutely” = 726). The patternis very similar for emotional inner speech (Mean rank “not at all” = 472, “somewhat” = 521, “more or less” = 615, “to a large extent” = 630, “absolutely” = 732) (see figure 7 forthe mean values).Silvia, a participant with Italian as an L1 and English as an L2, resident in the UK for 13 years, expressed a preference for the less emotional L2 in inner speech. For her, the spontaneous use of English allows her to be more rational and clearer: “I use English before Italian gets there, English gets there first, for reasons of clarity probably because if I can express an inner speech in Englishit means that I’m rationalizing it a lot more than if I did it in Italian, I don’t choose it, I don’t choose to do it inany language, usually it comes out in English first.”Another participant, Julia (Russian L1, English L2, French L3, German L4), explained that while emotional inner speech is typically in her L1, she does intersperse it with emotionvocabulary from her L2 and L3. In other words, the L1 is the base language but there is occasional code-switching of emotion words: “Russian is preferable, sometimes I use emotional terms (most often in inner speech) from English and French.”

Figure 7 L2 Emotionality and mean frequency of use of the L2for inner speech and emotional inner speech

32

The effects of the education level, age group and gender aremore limited and scattered. More highly educated participants made more frequent use of the L3 for emotional inner speech than those with lower levels of education. Younger participants made more frequentuse of the L5 for inner speech. Female participants used the L2 more frequently for both inner speech and emotional inner speech.

6. DiscussionThe analyses have revealed that the well-established patternthat languages that have been acquired earlier in life are used more frequently (Dewaele, 2010; Ewert, 2010; Pavlenko, 2005; 2012) apply equally for specific use of the language for inner speech and emotional inner speech. While the L1s were used more frequently for emotional inner speech than for inner speech, LXs are used significantly more frequently for inner speech than for emotional inner speech. While monolinguals never face a language choice to express themselves, multilinguals do have that choice, and

33

it even affects their frequency of use of the L1. One possible explanation for the higher frequency of use of the L1 for emotional inner speech than for inner speech is that multilinguals are exposed to many languages, with the resulting multilingual code-switching in inner speech. However, they stick more to the L1 for inner emotional dialogues.The difference is inverted, and becomes wider for the LXs, with inner speech having higher values than emotional inner speech. This is a clear indication that it takes time beforemultilinguals internalize the LXs and start using an LX for emotional inner speech (where the L1 remains a favorite choice). One possible interpretation for the higher use of the LX for inner speech than for emotional inner speech could be the fact that users may have less control over language choice for inner speech than for emotional inner speech (cf. Pavlenko, 2014). Inner speech may be a general reflection of the linguistic soundscape of the individual (though that individual obviously has the capacity to decideon using a specific language), while emotional inner speech is more likely to be the result of a volitional act, a conscious language choice to formulate something to one-self. Observations by participants confirmed this pattern, although some claimed to use LXs in equal measure.The strong positive correlations between frequency of use ofinner speech and emotional inner speech in all languages suggest that the measures are strongly linked, and it suggests a sufficient degree of validity of the two measures.

A series of analyses on the effect of various independent variables on the use of languages for inner speech and emotional inner speech showed that general use, socialization and self-perceived proficiency are most strongly linked to frequency of use in the L1, L2, L3, L4 and L5 (see tables 4 and 5). These three independent variables are linked, as increased LX socialization means more use of the LX, and hence a boost in LX proficiency (seeappendix). In other words, an LX that is used in daily

34

interactions will percolate down to become the preferred language for inner speech and, gradually, for emotional inner speech (cf. Hammer, 2015). Nancy Huston, who moved toParis from Canada as a young adult and maintained that French did not have the emotional power of English, providedevidence of this phenomenon when producing a single unexpected high frequency French swearword (“merde”) after uttering a string of English swearwords, in answer to a radio journalist’s question about her language choice in thehypothetical case of accidentally hurting herself (Dewaele, 2010).

The strong positive relationship between proficiency, inner speech and emotional inner speech is not surprising. The relationship also exists in the L1, suggesting that a shift in language dominance also reduces the use of the L1 for inner speech and emotional inner speech (Dewaele, 2004). Asfar as the LXs are concerned, it thus seems that an LX user does not need to be very advanced to experience spontaneous inner echoes of what has been heard during everyday social interactions (Guerrero, 1999, 2004). A certain level of proficiency in the LX is needed to engage in important inneremotional dialogues and for thinking (Pavlenko, 2014). Indeed, the choice is probably dictated by a need for efficiency. If language knowledge is insufficient for the formulation of sophisticated thoughts, it will be avoided. Since there are no interlocutors, there is no concern about social desirability. Multilinguals will thus prefer a language in which they can express themselves easily. A language that has particular emotional connotations may be preferred over one that lacks these connotations, or whose connotations have faded and are linked to a former self, as one participant observed. The use of an LX for emotional inner speech might be more resistant to change because of the absence of social pressure to accommodate to the language of the interlocutors. The choice of using more emotional inner speech in LX could also be interpreted as anindication of conceptual restructuring and of a repositioning of the self (Pavlenko, 2011, 2014). The

35

multilingual using more LX for emotional inner speech might have embraced a new sense of identity. It is also striking that the difference between the values of inner speech and emotional inner speech becomes smaller at higher levels of proficiency, just as it was the case forthe L1.

Context of acquisition of the LX and perceived emotionality of the L1 and LX also had significant effects on frequency of inner speech and emotional inner speech. Participants who had learned their LX only through formal instruction were less likely to use that LX later in life. On the other hand, those who had learned the LX at school but had also been using it outside school, or who had learned the LX completely outside the school grounds were more likely to opt for the LX for inner speech and emotional inner speech. It thus seems that the consequences of learning of an LX in a decontextualised way linger on for many years, and the LX continues to feel more disembodied (Pavlenko, 2005, 2012) compared to those who learned the LX in richer authentic contexts. Perceived emotionality of the L1 and LX may be linked to frequency of use of the languages, but it is quitedifferent in nature from context of acquisition. Indeed, the causality is less obvious, as perceived emotionality of languages shifts over a lifetime (Dewaele, 2013). A language that has more perceived emotionality might be used more frequently to talk to oneself and others about emotional matters, but more general use of the language alsoincreases the opportunities to hear it in emotional interactions which may boost its emotional resonance (cf. Hammer, 2015). Finally, the positive link between perceivedemotionality of the L1 and the use of that language for inner speech and emotional inner speech suggests that those for whom the L1 had lost emotional resonance also used it less frequently for inner speech and emotional inner speech.

AoA had a significant effect for the choice of L2 and L3 used for emotional inner speech and for the choice of L2 forinner speech. In other words, those who had started using

36

an L2 early in life used it more frequently for both inner speech and emotional inner speech. A similar effect emergedfor the use of L3 for emotional inner speech. This pattern largely reflects other studies on language preferences for the communication of emotion (Dewaele, 2009, 2010, 2013; Pavlenko, 2005).

The sociobiographical variables had much weaker and more scattered effects and should therefore be interpreted carefully. Education level only had an effect on emotional inner speech in the L3, with more highly educated participants reporting more use. Age group had a similar limited effect in the use of the L5 for inner speech. The effect of gender was limited to the L2, with female participants using it significantly more for inner speech and emotional inner speech than male peers. As this effect is not repeated in the other languages, it is difficult to draw any firm conclusion. Could it be that the female participants opened their heart more quickly to the L2 – as Eva Hoffman seems to have done? This could be an indication that the adoption of an LX for inner speech is more idiosyncratic. LX users may have relatively little choice in which languages they use for their overt speech in everyday interactions but they have more freedom in the choice of their inner speech (cf. Cook, 1998). The inclusion of participants’ voices added a useful emic perspective to the statistical analyses. Participants often highlighted the dynamic nature of their language preferencesfor inner (emotional) speech. Their choices were often determined by topic and context in the present, and some realized that these preferences had shifted over the years, so that different language choices made them feel different (Dewaele, 2015). Another point raised by participants is that code-switching can occur in inner speech, just as it does in articulated speech.While most participants expressed a preference for the L1 ininner (emotional) speech, some explained that the use of an LX allowed them to avoid the emotionality linked to the L1

37

and be more rational, which corroborates recent research on the Foreign Language Effect (Costa et al, 2014).Morin et al. (2011) assumed that the different language profiles of their participants could be ignored in the analysis of inner speech frequency and content. The presentstudy suggests that frequencies vary in function of the language. Although no information was gathered about content of inner speech, it is very likely that it is equally variable. Grosjean (2010) explained that multilinguals do not use their languages in equal measures across different domains with different interlocutors, and that “Different aspects of life often require different languages” (p. 29). Based on participants’ feedback from theBEQ and the interviews, it seems that language choice for inner speech, and increasingly for emotional inner speech, varies equally according to the topic and the environment.

7. ConclusionThe present study is the first systematic study of multilinguals’ language preferences for inner speech and foremotional inner speech. Multilinguals reported using the L1 most frequently for inner speech and for emotional inner speech. Languages learnt later in life (L2, L3, L4, L5) wereused gradually less frequently. A significant difference was found between the two types of inner speech: LXs were used more frequently for inner speech than for emotional inner speech. This suggests that it takes a while before anLX becomes internalized to the point of becoming a multilingual’s “language of the heart” (Dewaele, 2008). TheL1 can also cease to be the language of the heart for those who do not use the L1 much anymore, who no longer feel very proficient in it and for whom the emotionality of the L1 fades. Increased use of the LX for inner speech and emotional inner speech is linked to higher levels of self-perceived proficiency, general use and LX socialization (cf.Hammer and Dewaele, to appear; Hammer, 2015). Other variables, such as context of acquisition and LX emotionality are also linked to use of the LX for inner speech and emotional inner speech. An early start in

38

learning some LXs was linked to more frequent use for inner speech and for emotional inner speech. I would argue that ashift towards increased use of an LX for inner speech, and especially for emotional inner speech, is a sign of conceptual restructuring (cf. Pavlenko, 2011). It is also an indication that an LX can evolve from an obscure echo of social interactions (Guerrero, 2005) to a language of the heart, i.e. the LX can attain embodiment and take over -or share- the role of the L1 in inner (emotional) speech.

AcknowledgmentMany thanks to the participants of the BEQ and to the interviewees for filling out the questionnaire and sharing their insights. Special thanks also to María C. M. de Guerrero and to the anonymous reviewers for the excellent feedback on an earlier version of this paper.

ReferencesCenteno-Cortés, Beatrice, Jiménez Jiménez, Antonio F., 2004.Problem-solving tasks in a foreign language: the importance of the L1 in private verbal thinking. International Journal of Applied Linguistics 14, 7–35.

Cook, Vivian, 1998. Internal and external uses of a second language. Essex Research Reports in Linguistics, 100-110.

Costa, Albert, Foucart, Alice, Arnon, Inbal, Aparici, Melina, Apesteguia, Jose, 2014. ‘Piensa’’ twice: On the foreign language effect in decision making. Cognition 130, 236–254.

Dewaele, Jean-Marc, 2007. Multilinguals’ language choice formental calculation. Intercultural Pragmatics 4, 343-376.

Dewaele, Jean-Marc, 2004. Perceived language dominance and language preference for emotional speech: The implicationsfor attrition research. In Schmid, M.S., Köpke, B., Kejser, M., Weilemar, L. (Eds.), First Language Attrition: Interdisciplinary Perspectives on Methodological Issues. Amsterdam-Philadelphia: Benjamins, pp. 81-104.

Dewaele, Jean-Marc, 2008. The emotional weight of ‘I love you’ in multilingual’s languages. Journal of Pragmatics 40: 1753–1780.

39

Dewaele, Jean-Marc, 2009. The effect of age of acquisition on self-perceived proficiency and language choice among multilinguals. Eurosla Yearbook 9, 245-268.

Dewaele, Jean-Marc, 2010. Christ fucking shit merde!’ Language preferences for swearing among maximally proficient multilinguals. Sociolinguistic Studies 4, 595-614.

Dewaele, Jean-Marc, 2011. The differences in self-reported use and perception of the L1 and L2 of maximally proficient bi- and multilinguals: A quantitative and qualitative investigation. International Journal of the Sociology of Language 208, 25-51.

Dewaele, Jean-Marc, 2013. Emotions in Multiple Languages. Basingstoke: Palgrave Macmillan (2nd ed.).

Dewaele, Jean-Marc, 2015. Why do so many bi- and multilinguals feel different when switching languages? International Journal of Multilingualism, DOI:

10.1080/14790718.2015.1040406Dewaele, Jean-Marc, Pavlenko, Aneta, 2001–2003. Webquestionnaire Bilingualism and Emotions. University of London.

Ewert, Anna, 2010. An educational language community: External and internal language use by multilingual students. In: Arabski. J., Wojtaszek. A. (Eds.), Neurolinguistic and Psycholinguistic Perspectives on SLA. Bristol UK: Multilingual Matters, pp. 159-174.

Gosling, Samuel D., Vazire. Simine, Srivastava, Sanjay, John, Oliver P., 2004. Should we trust web-based studies? A comparative analysis of six preconceptions about internet questionnaires. American Psychologist 59, 93–104.

Grosjean, François, 2010. Bilingual. Life and Reality. Cambridge. Mass. & London: Harvard University Press.

Guerrero, Maria C. M. de, 1999. Inner speech as mental rehearsal: the case of advanced L2 learners. Issues in Applied Linguistics 10, 27–55.

Guerrero, Maria C. M. de, 2004. Early stages of L2 inner speech development: what verbal reports suggest. International Journal of Applied Linguistics 14, 90–112.

Guerrero, Maria C. M. de, 2005. Inner Speech – L2: Thinking Words in a Second Language. New York: Springer.

40

Hammer, Kate, 2015. Life in a new language: An acculturationperspective on language shift in bilinguals. Unpublished PhD dissertation. Birkbeck. University of London.

Hammer, Kate, Dewaele, Jean-Marc, 2015. Acculturation as thekey to the ultimate attainment? The case of Poles in the UK. In: Forsberg Lundell, F., Bartning, I. (Eds.), Cultural Migrants and Optimal Language Acquisition. Bristol: Multilingual Matters, pp. 178-202.

Hoffman, Eva, 1989. Lost in translation: A life in a new language. New York: Penguin Books.

Hoffman. Eva. 2011. Time. London: Profile Books (2nd ed).Larsen. Steen F., Schrauf, Robert W., Fromholt, Pia, Rubin, David C., 2002. Inner speech and bilingual autobiographical memory: A Polish-Danish cross-cultural study. Memory 10, 45-54.

Morin, Alain, Uttl, Bob, Hamper, Breanne, 2011. Self-reported frequency. Content, and functions of inner speech. Procedia-Social and Behavioral Journal 30, 1714-1718.

Ożanska-Ponikwia, Kate, Dewaele, Jean-Marc, 2012. Personality and L2 use. The advantage of being openminded and self-confident in an immigration context. Eurosla Yearbook 12, 112-134.

Pavlenko, Aneta, 2004. Stop doing that, ia komu skazala!: Emotions and language choice in bilingual families. Journal of Multilingual and Multicultural Development 25, 179-203.

Pavlenko. Aneta. 2005. Emotions and Multilingualism. Cambridge MA: Cambridge University Press.

Pavlenko. Aneta. 2009. Conceptual representation in the bilingual lexicon and second language vocabulary learning.In: Pavlenko, A. (Ed.), The bilingual mental lexicon: Interdisciplinaryapproaches. Clevedon, UK: Multilingual Matters. pp. 125-160.

Pavlenko, Aneta, 2011. (Ed.) Thinking and Speaking Two Languages. Bristol, UK: Multilingual Matters.

Pavlenko, Aneta, 2012. Affective processing in bilingual speakers: Disembodied cognition? International Journal of Psychology 47, 405-428.

41

Pavlenko, Aneta, 2014. The Bilingual Mind and what it tells us about language and thought. Cambridge – New York: Cambridge University Press.

Sansault, David, 2014. Représentations affectives dans le répertoire langagier de lycéens bilingues polono-français.LingVaria 9, 227-237.

Sansault, David, 2012. Migrations et émotions: les langues pour le dire. Dimensions affectives des langues dans le répertoire langagier de lycéens bilingues franco-polonais.Unpublished MA dissertation. Université de Sorbonne Nouvelle – Paris 3.

Uttl, Bob, Morin, Alain, Hamper, Brean, 2011. Are inner speech self-report questionnaires reliable and valid? Procedia - Social and Behavioral Sciences 30, 1719-1723.

Wilson, Rosemary, Dewaele, Jean-Marc, 2010. The use of web questionnaires in second language acquisition and bilingualism research. Second Language Research 26, 103–123.

BionoteJean-Marc Dewaele is Professor of Applied Linguistics and Multilingualism. He investigates individual differences in psycholinguistic, sociolinguistic, pragmatic, psychological and emotional aspects of bi- and multilingualism. He has published over 160 papers and chapters, 5 edited books and the monograph Emotions in Multiple Languages (2nd ed. in 2013). Heis Vice-President of the International Association of Multilingualism, Convenor of the AILA Research Network Multilingualism and formerpresident of the European Second Language Association. He is General Editor of the International Journal of Bilingual Education and Bilingualism. He is recipient of the ‘Robert C. Gardner Award for Outstanding Research in Bilingualism’ from the International Association of Language and Social Psychology (2016).He is father of a trilingual daughter and holds a black beltin karate (Shodan).

AppendixCorrelation matrix independent variables in L2

L2

Context acquisition AoA

Proficiency

Generaluse

Socialis-ation

Emotio-nality

Education

42

AoA -.442p 0N 1515

Proficiency Rho .332 -.271p 0 0N 1549 1525

General use Rho .231 -.134 .620p 0 0 0N 1545 1521 1559

Socialisation Rho .114 -.103 .443 .783

p 0 0 0 0N 1438 1454 1449 1445

Emotionality Rho .132 -.060 .151 .065 -0.015

p 0 0.022 0 0.011 nsN 1501 1480 1512 1509 1403

Education Rho 0.007 .084 0.04 -.086 -.093 0.023p ns 0.001 ns 0.001 0 nsN 1433 1449 1444 1440 1454 1398

Age Groups Rho 0.036 .098 .094 0.048 -0.034 .072 .387p 0.16 0 0 0.06 ns 0.005 0N 1553 1530 1563 1559 1453 1514 1450

43