A Small-Scale Study on the Relationship between First Language Attrition and Language Attitudes in Polish Speakers in Sweden

ABSTRACT

This study investigated the relationship between overall language attitude and first language attrition amongst educated adult Polish speakers in Sweden. The attritional data consisted of two variables – meta-textual indicators of uncertainty and the overt/covert use of first-person personal pronouns. The attitude value was obtained through an attitude questionnaire. There was a moderate negative correlation between the attitude value and meta-textual indicators of uncertainty. The correlation between the attitude value and the use of first-person pronouns was not significant. In line with previous research on first language attrition and its explanatory correlates, this study shows that some aspects of first language attrition might be influenced by the attitudinal factor. Thus, the study adds to a relatively limited body of research on Polish adult language development in Sweden and its explanatory correlates.

KEYWORDS:

• language attitudes
• Polish
• first language attrition
• Polish immigrants
• Sweden

1. Introduction

This small-scale study¹ investigates the potential relationship between language attitudes and Polish first language (hence L1) attrition. The study adds to a relatively limited body of research on Polish adult language development in Sweden and its explanatory correlates (cf. Lubińska, in preparation). The study is placed within the theoretical framework of L1 attrition and language attitudes that are presented in more detail in next sections. Consequently, the research question in this study is as follows: What is the relationship between language attitudes and L1 Polish attrition in a group of L1 Polish speakers in Sweden?

This article is organized as follows: After this introduction, the theoretical framework of the study is presented. Next, previous research on language attitudes and L1 attrition is summarized, followed by two sections that present the operationalisation of Polish L1 attrition (section 4), and the conceptualization of language attitudes (section 5) respectively. Next, the methodology is described, followed by a section on results. The final section comprises a discussion and conclusion.

2. Theoretical framework

As mentioned, the theoretical framework of the study is L1 attrition and language attitudes. L1 attrition is understood here as “[ … ] the changes to linguistic skills or language proficiency [ … ]” (Seton and Schmid 2016, 338; cf. also e.g., Andersen 1982; Gardner 1982; Köpke and Schmid 2004; Montrul 2008; Sharwood Smith 1983a, 1983b, 1989, 2007; Schmid and De Bot 2004). The change might take place when a person moves from L1 environment to L2 environment, and most probably stays under conditions of reduced L1 use (e.g., Van Els 1986; Hyltenstam and Viberg 1993; Field 2004; Seton and Schmid 2016). Importantly, it is widely accepted that L1 attrition, or at least some aspects of it, might be reversible under favourable conditions, e.g. return to L1 environment or reactivation exercises (Hyltenstam et al 2009; Pallier et al 2003; Stolberg and Münch 2010; Ventureyra and Pallier 2004; Ventureyra et al 2004; Seton and Schmid 2016).

Interestingly, L1 attrition has also been discussed within the framework of cross-linguistic influence that distinguishes five different types of L2 influence on one’s L1: borrowing, restructuring, convergence, shift, and attrition (Pavlenko 2000, 2004; Pavlenko and Jarvis 2002). However, Pavlenko (e.g. 2004) proposes that cross-linguistic influence can be called L1 attrition given that (i) it is empirically proved to be permanent, and (ii) it can be also observed in a monolingual L1 context, i.e. when a person exhibiting this influence continues to retain it in her/his linguistic repertoire outside the L2 context. According to this view, restructuring, convergence and shift can be called L1 attrition, only when they met the two criteria mentioned before. Obviously, this view is different from the one presented in the first paragraph of this section, and puts pressure on the experimental evidence for L1 attrition. This will be elaborated on in section 5.

The second framework of the study is that of language attitudes. The concept of attitudes belongs to the field of social psychology, and attitudes are the single most researched topic within this discipline. For this study, the following working definition of an attitude, which originates from social psychology (Giles et al 1987; Baker 1992; see also O’Rourke 2011), is employed: “an attitude [is] a summary evaluation of an object [ … ]” (Bohner and Wänke 2002, 5; cf. also Aiken 2002; Fiske 2010; Garrett 2010; McKenzie 2010).

Language attitudes belong to widely investigated factors in language studies; the term language attitudes serves as an umbrella term that encompasses a range of language-related areas (Agheyshi and Fishman 1970; Baker 1992; Portolés Faromir 2015). In particular, language attitudes have long been an important factor to investigate in studies on multilingualism; they are assumed to determinate the growth or decay of languages both at the individual and at the group levels (e.g., Zaręba 1979; see also Andersen 1982; Baker 1992; Bradley and Bradley 2002; Pavlenko 2006; Köpke 2007; Schmid 2011; Pauwels, 2016). The importance of language attitudes in multilingualism has been summarized as early as in 1982 by Grosjean (1982, 127) as follows: “In the end, language attitude is always one of the major factors in accounting for which languages are learned, which are distressed, and which are preferred by bilinguals.” Furthermore, the importance of language attitudes in L1 attrition is described by Andersen (1982, 114) in the following way: “[ … ] socio-affective factors favor or inhibit the operation of the brain on the linguistic material, often to the point of making the socio-affective factors the most crucial [ … ]”. Recently, Selton and Schmid (2016, 341) point out that “Investigations of language attrition have often struggled to find adequate descriptive or explanatory frameworks to account for the fact that one attriter may exhibit more cross-linguistic influence than another.” The concept will be discussed in more detail further in this paper.

3. Previous research on language attitudes and L1 attrition

Despite the recurrent view of the relationship between the attitudinal factor and L1 attrition, existing studies show inconclusive results (Schmid 2011). There have been six studies: Silva-Corvalán (1994), Waas (1996), Yağmur (1997), Hulsen (2000), Schmid (2004), Ben-Rafael and Schmid (2007) and Cherciov (2013). All these studies operationalized and measured attitudes in different ways.

Two of the studies, Hulsen (2000) and Yağmur (1997), applied a defined framework, namely, the theory of ethnolinguistic vitality. In both studies, attitudes were conceptualized as a subjective perception of the ethnolinguistic vitality of the migrant group and the majority group. Hulsen (2000) investigated the Dutch group in New Zealand and found no significant correlation between Dutch vitality and language processing. Yağmur (1997) investigated Turkish in Australia and did not find any significant correlation with the linguistic variables.

The five remaining studies - Silva-Corvalan (1994), Waas, (1996), Schmid (2004), Ben-Rafael and Schmid (2007) and Cherciov (2013) – applied no specifically defined framework. Silva-Corvalan (1994) studied Spanish speakers in the USA and found a relation between neutral or negative attitudes towards maintenance of Spanish and L1 attrition. Waas (1996) studied German in Australia and operationalized attitudes as participation in the activities of the German group in Australia. Ethnic ties were shown to have a significant impact on how the participants estimated their knowledge of German, but not on their actual German language proficiency.

Schmid (2004) studied German Jews who fled Germany during the Nazi era and for which German was their mother tongue. These Jews fled during different phases of persecution. She found evidence that the individuals who fled during the last and most cruel phase of persecution made more systematic errors in their German, and they also showed signs of an English accent and reduced linguistic repertoire. In addition, Schmid (2004) found that individuals who fled during the second phase did not make significantly more errors than the comparison group, even though their language system was influenced by attrition to some extent. Schmid (2004) also observed that individuals who fled during the first and least cruel phase retained their German almost entirely free from attrition signs, even 60 years after immigration. Based on these results, Schmid (2004) argues that the level of persecution affected these individuals’ attitudes towards the German language. This in turn affected the degree of attrition.

Ben-Rafael and Schmid (2007) investigated a French- and Russian-speaking group in Israel and found that pragmatic versus ideological motivation to emigrate and the ensuing attitudes were clearly linked to L1 attrition.

Finally, Cherciov (2013) studied Romanian migrants in Canada, using a mixed quantitative-qualitative approach. The quantitative analysis revealed only partially significant results, which Cherciov interpreted as an evidence for a limited impact of the attitudinal factor on L1 attrition. The individual qualitative analyses, however, revealed important links between attitudes and language proficiency of the migrants, which led Cherciov to advocate a mixed-methods approach in studies on L1 attrition and attitudes.

4. Polish L1 attrition in the study

L1 attrition here is empirically established based on a comparison between a group of Polish speakers in Sweden and a group of Polish speakers in Poland. The metalinguistic and linguistic data reported on here, and interpreted as indications of L1 Polish attrition, encompass meta-textual indicators of uncertainty and the use of first-person pronouns as subjects respectively.

As a starting point for studying meta-textual indicators of uncertainty serves a hypothesis proposed by Andersen (1982, 112): “An LA will exhibit insecurity in his competence in language X. This will be evident in his own unsolicited comments, in an interview or questionnaire, or through ethnographic participant-observation techniques.” I have assumed that the word competence in the quotation is used in a non-Chomskyan sense, given Andersen’s description of what type of evidence should be collected in order to corroborate the hypothesis (cf. Sharwood Smith and Van Buren 1991). As there was a lack of previous studies that tested this hypothesis, I conducted a pilot study; this study exhibited promising results with respect to meta-textual indicators of uncertainty (Lubińska 2003).

Finally, the choice of the name - meta-textual indicators of uncertainty - is motivated by an ambition to reach a terminological consistency in research on Polish in Sweden; this term is used by Laskowski (2014) in his work on Polish language development in children and adolescents in Sweden.

With respect to the interpretation of meta-textual indicators of uncertainty as L1 attrition, it can, indeed, be discussed if the self-perceived experience of word-finding difficulties should count as L1 attrition. Here, it is argued that it is epistemologically tenable to use subjective experience as a window into language development (cf. Kouritzin 1999; Prescher 2007; Ecke and Hall 2013). However, it is at the same time admitted that this phenomenon deserves to be investigated further in more depth, and with a more tuned methodology.

With respect to the use of first-person pronouns, previous research about Polish in immigrant contexts shows the vulnerability of this structure in L2 settings. Previous studies (e.g. Dubisz 1997, 2001; Laskowski 2014) show that the use of first-person pronouns as overt subjects, where standard Polish uses null subjects, is a common feature in spoken Polish abroad. These studies, however, do not provide any comparison data, or the baseline for comparison seems to be standard written Polish. In addition, the results in Laskowski (2014), and to some extend in Dubisz (1997, 2001), seem to regard Polish as a heritage language, i.e. Polish acquired mostly in the family domain in a L2 environment. In the present study, on the contrary, the reference baseline is a comparison group of speakers in Poland. In the following subsection, some aspects of the use of overt pronominal subjects in spoken Polish are discussed.

4.1. Pronominal subjects in Polish

In principle, Polish drops all unstressed pronominal subjects in all persons in singular and in plural, and in all tenses and modes (cf. Bondaruk 2001; Lindseth and Franks 1996). In the first language environment, the normative guideline is then to avoid unjustified use of explicit personal pronouns in subject position (e.g. Pisarkowa 1969; Swan 2002; Szczepankowska 2012).

However, spoken and written Polish differ in terms of the use of pronominal subjects. Overt subject pronouns are more common in speech, especially in conversations (Pisarkowa 1969). In contrast, written language uses overt subject pronouns to achieve a stylistic effect (Nilsson 1982). Generally, the expected percentage of use of overt subject pronouns in written language is 3-10%, while in spoken language the use is about 45% (Nilsson 1982).

There are, however, both in Polish and in other pro-drop languages, some pragmatic mechanisms that make the use of overt subject pronouns justified. For instance, 1^st and 2^nd person’s pronouns might be used in order to express emphasis or contrast (e.g., Pisarkowa 1969; Swan 2002; Szczepankowska 2012). The 3^rd person’s subject pronouns are often used in the first part of a longer utterance, but the subject might be omitted later, if the reference stays clear. In addition to these pragmatic mechanisms, there are also, according to Pisarkowa (1969), some formal mechanisms that may push the use of overt subject pronouns in Polish. Pisarkowa (1969) proposes, on the basis of her material from phone conversations, that it is not possible to remove some overt subject pronouns in certain sentences without changing the word order, and thereby the meaning of the whole sentence. This applies to constructions with enclitic use of –by and –bym.

Another important phenomenon is that of mobile verb endings in 1^st and 2^nd person singular and plural in the past tense. This characteristic of spoken Polish means that the verb endings -m, -ś, -my, -ście move to another part of the sentence. This part can be, amongst others, a personal pronoun (Zawiślakowa 1974) that then has to be overt. Today, some researchers claim, however, that the phenomenon is about to disappear, and that it is mainly found in speech of the elderly generations (e.g., Bańko 2006; Sujkowska-Sobisz 2008).

In sum, it can be said that the unmarked case in spoken Polish is that pronominal subjects are covert. Overt pronominal subjects in spoken Polish occur according to certain contextual and formal mechanisms: (i) for emphasis or contrast (1^st and 2^nd person’s pronouns); (ii) referentially at the beginning of a longer utterance (3^rd person’s pronouns); (iii) when the mobile verb endings in 1^st and 2^nd person singular and plural in the past tense move from the verb to the personal pronoun; and (iv) in constructions with the enclitic use of –by and –bym.

5. The concept of language attitudes

The concept of “language attitude” is not only important in language studies but is also a subject for different understandings and conceptualizations, particularly with respect to the following issues: (i) the definition of attitudes; (ii) the nature and number of attitude components; (iii) the uni- vs. multi-dimensionality of attitudes; and (iv) the stability of attitudes. Accordingly, there is a need for an outright clarification of how the “language attitudes” concept is viewed in this paper, along with the underpinnings of this view.

As mentioned in the introduction, for the purposes of this study, a working definition of an attitude originating from social psychology (Giles et al 1987; Baker 1992; see also O’Rourke 2011) is employed: “an attitude [is] a summary evaluation of an object [ … ]” (Bohner and Wänke 2002, 5; cf. also Aiken 2002; Fiske 2010; Garrett 2010; McKenzie 2010). In the case of language attitudes, the term may cover evaluations of both languages and language varieties (cf. Giles and Billings 2004; Lasagabaster 2004), along with evaluations of language-related phenomena, e.g., language maintenance or speakers of a particular language (Agheyisi and Fishman 1970; see also Baker 1992; O’Rourke 2011), alongside other language-related issues, e.g., language preference (Giles, Hewstone, and Ball 1983).

Moreover, attitudes are often considered to include three components (Baker 1992; Bohner and Wänke 2002; Fiske 2010): affective, cognitive and behavioural. The affective component concerns aspects related to feelings about the attitudinal object. The cognitive component relates to individual thoughts and beliefs about the object. Finally, the behavioural component refers to the predisposition and intention to act in a certain way. As to the relationship between the components, a common perspective within the attitude research in social psychology seems to be that the components “are not necessarily separable from each other and do not necessarily represent three independent factors” (Bohner and Wänke 2002, 5). In addition, some attitudes may comprise only one of the components; consequently, not all three of these components need to be represented in an attitude (Bohner and Wänke 2002; Oakes 2001; Fiske, 2010; Edwards 2011).

With respect to the uni- vs. multi-dimensionality of attitudes, the working definition proposed in this study suggests a relatively unidimensional conceptualization, as it is “a summary evaluation” (cf. also McKenzie 2010, my emphasis). In contrast, the current view of language attitudes seems to be that attitudes are multidimensional in nature (Baker 1992; O’Rourke 2011). This view, however, presents a few challenges. First, the view calls for determining which dimensions should be included in the concept of language attitude (Baker 1992). Second, the dimensions most likely vary across time, contexts and samples (O’Rourke 2011). Third, it is necessary to consider whether the conceptual dimensions truly reflect the personal constructions of individuals (Baker 1992). Finally, the multidimensional view runs the risk of overlooking some dimensions while constructing a measurement tool and/or of overlapping some dimensions (Baker 1992). Interestingly, some researchers (Santello 2015) seem to see components and dimensions as mutual substitutes. This is an interesting view, a possible consequence of which might be that a separation of dimensions/components is not possible or necessary and that they do not necessarily represent separate factors, or that not all components/dimensions are represented in an attitude (Bohner and Wänke 2002; Oakes 2001; Fiske, 2010; Edwards 2011). However, language attitude researchers have aimed at a multicomponential/ multidimensional construct. For instance, Baker (1992) presents an attitude concept comprising six dimensions for the context that he investigates, although he simultaneously underlines that all dimensions relevant to one context are not necessarily relevant to another context (see also Dörney and Ushioda 2009). To counteract these unsolicited effects, Baker (1992) suggests that both the type and the number of dimensions should be based on the context in which attitudes are studied (cf. Köpke 2007; Köpke and Schmid 2004). However, the challenge of the multicomponential/multidimensional construct also includes the fact that the type and number of dimensions can best be verified in large empirical studies. In such studies, the data can first be assessed for their suitability for a factor analysis, and next a factor analysis (principal component analysis) can be performed, if appropriate. Because this type of analysis can group together items that have something in common, with suitable data the multicomponentiality/multidimensionality can be validated. In this study, however, this type of statistical calculation was not feasible because of the size of the studied group. Nevertheless, the idea of possible multicomponentiality/multi-dimensionality has been adopted to some extent. This is described in more detail in the methodological section.

Furthermore, there is a debate within social psychology about attitude conceptualizations with respect to “whether evaluations have to be stable over a longer time period and have to be stored in long-term memory to qualify as an attitude” (Bohner and Wänke 2002) or whether instead, attitudes are temporary constructions that people construct at the moment of evaluation and do not retrieve from their memory (Bohner and Wänke 2002). This issue is also raised in L1 attrition research (e.g., Schmid 2011). As Bohner and Wänke (2002) note, there is evidence for both perspectives, i.e., attitude as a mental file, also called the mentalist perspective (O’Rourke 2011), vs. attitude as an on-the-spot construction susceptible to situational influences, also called the behaviourist perspective (O’Rourke 2011).

To summarize, in this study a language attitude is conceptualized as a summary evaluation of a language and language-related issues, phenomena and objects, i.e., as a concept, in which dimensions and components might have merged. Finally, the stability vs. shifting vs. dependency of language attitudes is recognized as a complex issue with inconclusive evidence.

6. Methodology

6.1. Participants

Two groups of 25 individuals each participated in the study.² The experimental group (the SE group)³ consisted of adult Polish speakers living in Sweden in the Stockholm-Mälardalen Region. They learned Swedish in adulthood after immigrating to Sweden. The control group (the PL group) consisted of individuals from two counties, pomorskie and kujawsko-pomorskie;⁴ they have spent their entire live in Poland. They did not use any language other than Polish in everyday communication to any notable extent and matched members of the experimental group in age and education. Both groups had studied at least one foreign language at school, mainly Russian, English, German and French. None of the participants worked with Polish on a daily basis, for instance as a Polish-language teacher. To minimize the age effects, the age limit was set at 65, because it is known that all language skills may be affected by ageing (Baddeley 1999), and that some language skills are more sensitive than others to the natural effects of ageing (e.g., Baddeley 1999; Goral 2004). The participants were recruited through my personal networks and through various Polish organizations (the SE group). Additionally, a snowball method was used (Denscombe 2014). The background information is summarized in Table 1.

Table 1. Participants of the study: Background data

Background factor	Experimental group (SE-group) (n = 25)	Comparison group (PL-group) (n = 25)
Age of immigration	18-34 yrs. (mean: 25)
Age at the point of the study	45-64 yrs. (mean: 57)	47-65 (mean: 55)
Length of residence	20-45 yrs. (mean: 32)
Education level
High school	5	5
Post-secondary	3
University	17	20

6.2. Instruments and procedures

6.2.1. Background questionnaire

The background data on the participants (Table 1) were collected with a background questionnaire during the first phone contact with a potential participant. The metalinguistic and linguistic data and the data on attitudes were collected later, during an individual meeting with each participant.

6.2.2. Meta-linguistic and linguistic data

The instruments for examining the attritional effects consisted of (1) a Charlie Chaplin film retelling task, and (2) an odour cue task. In the retelling task, the participants were asked to recall ten minutes of Charlie Chaplin’s silent movie (“Modern Times”). In the odour task, a smell of cinnamon was used as a stimulus. Participants were asked to talk about associations and / or memories the odour provoked. The idea of using an odour association was derived from psychological research on autobiographical odour memories (Willander and Larsson 2006). However, transferring an instrument from one context to another may be problematic. There was no guarantee that associations and / or memories should appear or how long the story would last. Therefore, in order to collect a relatively spontaneous speech of about 30 minutes a set of questions about other topics was prepared. When the participants’ memories and associations seemed to be exhausted, a topic from the set was introduced and follow-up questions were asked if something seemed to engage the participant. For example, some topics like school memories were almost inexhaustible for many of the participants, while others preferred talking about travel memories (see Lubińska 2011 for further details). The two tasks served to elicit relatively free speech. The speech was then transcribed and coded in the CHAT-programme and analysed, with a focus on (i) meta-textual indicators of uncertainty, and (ii) the use of first-person pronouns.

The meta-textual indicators of uncertainty were then operationalized as the amount of parenthetical comments on word finding difficulties or difficulties with the formulation of utterances in relation to the length of speech in both elicitation tasks. Example 1 (Lubińska 2011, 103, modified) illustrates what was counted as parenthetical comments (in braces):

(1)

PL: Starałam się skontaktować i skontaktowałam się z tym {jak się mówi} // który był później moim {nie wiem jak to powiedzieć} handledare .

‘I tried to contact and I contacted this {how to say}// this one who later was my {I do not know how to say this} handledare-CODE SWITCHING.’ = supervisor

It has also to be pointed out that code-switching was not counted as meta-textual indicators of uncertainty. Such an approach would be in conflict with the evidence from research on multilingual communication; as both the interviewer and the participants are bilingual, code-switching has to be seen as a natural way of communication.

The use of a first-person pronoun was calculated based on the odour cue task, as the first-person pronouns were barely used in the retelling task. In the present material all prodrops and overt pronominal subjects were coded. From the coding of the 1^st person's prodrop, the expression że tak powiem, “so to say”, was excluded. The reason for the exclusion was that this expression might be treated as a fixed expression.

The instances of prodrop in coordinated sentences like On się zdenerwował i poszedł do domu, “He got angry and went home”, were included in the prodrop counts. There is evidence, even if it comes from speakers of Polish as a foreign language, that personal pronouns are used overtly in such constructions (Pisarkowa 1969). Thus, coordinated sentences of this type are possible candidates for unjustified use of personal pronouns.

6.2.3. Attitude questionnaire

There are a variety of techniques for measuring language attitudes. They can be generally divided into direct and indirect types. The direct methods ask direct questions about people’s attitudes, i.e., about their beliefs, feelings and knowledge of the given object, usually through interviews and/or questionnaires. The indirect methods try to make the purpose of the investigation less obvious to the participants; the most popular type of indirect method is a matched-guise technique (cf. Garret 2003). As Baker (1992) and Edwards (2011) note, there are problems with all techniques, but some researchers (e.g., Oppenheim 1992; Payne and Payne 2004; see also Garrett 2010) assume that attitudes can be deduced through direct questions, using questionnaires based on the so-called Likert scale. Baker (1992) lists also some advantages of this type of questionnaire: (i) responses can easily be categorized; (ii) comparisons and contrasts can be drawn; and (iii) statistical analyses are relatively easy to conduct. On the one hand, in this study, the choice of this technique was dictated by the advantages listed by Baker, especially the future opportunity to compare and contrast results from other Polish migrant contexts. On the other hand, the nature of the doctoral project, of which the questionnaire was a part, played a role because statistical comparisons were involved.

Consequently, an attitude questionnaire (see online Appendix) was delivered to the participants at the end of the personal meeting. The questionnaire contained 20 items. The participants were asked to spontaneously express the extent to which they agreed with a given item. There were five options to choose from on a Likert-like scale⁵: (i) strongly agree; (ii) agree with hesitation; (iii) have no opinion; (iv) disagree with hesitation; and (v) completely disagree. The process of creating the attitude questionnaire is described in detail in section 6.2.4.

The answers from the questionnaire were then fed into a statistical program. The alternatives in the attitude questionnaire were pre-coded 5–1, where 5 meant a positive attitude and 1 meant a negative attitude. One of the items (sentence 4) was negative; hence, 5 meant a negative attitude and 1 meant a positive. This item was reversed in accordance with applicable conventions to ensure that the high number meant a positive attitude. The reliability of the questionnaire was tested by calculating Cronbach’s alpha, which is a measure of how the questionnaire components are correlated to one another; that is, if they are measuring the same construct (Pallant 2010). In this case, the Cronbach’s alpha coefficient was 0.9. A value of approximately 0.7 or higher is interpreted as a very good correlation (e.g., Ejlertsson 2014; Pallant 2010), indicating the questionnaire has good internal consistency overall.

6.2.4. The construction of the attitude questionnaire

As mentioned, language attitudes are conceptualized here as a summary evaluation of language and language-related issues, phenomena and objects. Because the concept is a summary evaluation, it has consequences for the construction of the attitude questionnaire; the questionnaire should include a range of relevant issues which, scored together, gives one score (a summary) that represents an overall attitude.

The complex matter of which issues are relevant and should be converted to questionnaire items was addressed from two angles: theoretical, based on previous research on language attitudes and observational, based on material collected in the studied context. To put it in another way, the central questions were (i) which issues should be included into the questionnaire and (ii) how it could be assured that they applied to the studied context. As it will be shown in the following paragraphs, the angles had slightly different functions in the construction of the attitude questionnaire.

The question of what issues should be included into the attitude questionnaire was answered based on the theoretical considerations about the components (affective, cognitive and behavioural) and dimensions presented in the literature on language attitudes.

For instance, Baker (1992) proposed a six-fold conceptual distinction, with the caveat that these dimensions may not be relevant in all contexts (Baker 1992). O’Rourke (2011) stated that two dimensions seemed to be particularly used by researchers across boundaries of time, sample and nation, and they appeared frequently as a dichotomy in research on language attitudes. These are the integrative and the instrumental dimensions (see also Baker 1992). The integrative dimension is rooted in the idea that language has a function of binding people into a group of shared understandings and common identity, giving them a feeling of affiliation (O’Rourke 2011), and probably could be merged with Baker’s dimensions of ethnic traditions and personal ideologies (Baker 1992). Furthermore, O’Rourke notes that such a view transforms language into a symbol of group solidarity and a distinguishing factor for members and non-members of a specific group. Baker (1992) views this dimension as social and interpersonal in orientation and defines it as attachment or identification with a language group and its cultural activities.

The instrumental dimension is viewed as a utilitarian component (Baker 1992). As opposed to the integrative dimension, the instrumental component is relatively individualistic and self-oriented in nature (Baker 1992). In the context of minority language development and use, a language might be valuable, for instance, for vocational reasons or the achievement of status and personal development (Baker 1992). Baker also describes this as economic and social dimensions.

To summarize, these two dimensions – integrative and instrumental – seem to be traditionally used when studying language attitudes within the context of bilingualism; it has therefore been justified to use them as the underlying concept for the construction of the language attitude questionnaire in this study. However, in the present study the components and dimensions are perceived not necessarily as separable and/or probably overlapping (see section 5). Therefore, they do not form any separate scales in the attitude questionnaire.

Further, as has been stressed (e.g., Baker 1992), the integrative and instrumental dimensions alone are probably not sufficient for an attitude concept. In studies on the role of language attitudes in second language acquisition (Gardner 1985a, 1985b), the offered language attitude concept includes several other dimensions besides the integrative and the instrumental. However, the other dimensions used in Gardner’s model are not useful in this study, as they refer to issues relevant to second language acquisition. Accordingly, transferring dimensions directly from one setting to another is hardly possible; the dimensions must either be adjusted to the studied context or abandoned, or possibly even replaced by more appropriate alternatives for the context (cf. Baker 1992; Oppenheim 1992; see also Köpke 2007). Thus, in this study, two further dimensions are used – language as a core value and language maintenance-oriented – but they are partly treated as an extension of the integrative component. The reason for including these two components is subsequently explained.

Language as a core value dimension is rooted in the idea that a language might be a core value of culture (see Smolicz 1980; Smolicz 1992; Smolicz and Harris 1977; see also Hofman and Cais 1984; Oakes 2001)⁶ and as such belongs to a precious cultural heritage. What can be included in a group's cultural core value is easiest to determine when the group is forced to defend itself against an external threat (Smolicz 1980). According to Smolicz (1980), ethnic groups differ in terms of their core cultural values. For instance, Poles⁷ are considered a group for which language is such a core value because of historical circumstances during Russian, Prussian and Austro-Hungarian colonisation (1772–1918). Many of the systematic efforts employed by the colonial powers were directed against language and culture. Therefore, the Polish language became a crucial value, a symbol of national survival and cultural preservation (e.g., Davies 2005). The language maintenance dimension is indeed somewhat connected to language as a core value construct (cf. Hofman and Cais 1984), because an ethnic group that experienced persecution related to its mother tongue might have a strong need to protect and maintain it, even if the persecution has ceased. In connection with the idea that the members of the group wish to transfer the L1 to future generations (cf. e.g., Schmid 2011), it seemed reasonable to include this dimension in the questionnaire.

Finally, a slightly blurred dimension of so-called general approval seems to be quite common across ethnic groups. A review of studies on language attitudes (e.g., Lewis 1975; Oakes 2001; cf. also Baker 1992) confirms that there is a tendency to embrace this dimension in the concept of a language attitude. Baker (1992) exemplifies this dimension with “I like speaking language X”. Therefore, the dimension of general approval is also included in this study.

In order to assure that the theoretical constructs described above applied to the studied context, observational material was used. The material included handwritten notes with utterances about languages and language-related issues expressed by Polish people in formal and less formal social settings in which Polish people in Sweden meet. The utterances were collected over several years in metropolitan Stockholm by me while I was familiarizing myself with a new life and work context. The utterances were usually written down verbatim. It has to be acknowledged that the observations were not guided by a specific research interest or question, and there was no immediate plan that the notes would prove useful to a research project. Consequently, there was no stringency, which usually would characterize observation as a data collection instrument (Cohen, Manion and Morrison 2011). For instance, I did not write down when, where and who said what. During the period the notes were taken, I never made any further reflections on them, which normally would be the case with field notes in a research project. Undoubtedly, the observations and following notes were a subject for bias, just as a stricter collection can be (Cohen, Manion and Morrison 2011). Another possible bias was selective memory bias (Cohen, Manion and Morrison 2011). If we write after an event, our memory neglects and selects data, and sometimes we record the interpretation of what has been said, not of what has actually been said. Despite these disadvantages, the utterances proved to be of value for the language attitude questionnaire, primarily in the naturalistic formulations, which were used in the questionnaire almost directly (Oppenheim 1992). To some extent, the utterances also supported the choice of components/dimensions that were included in the questionnaire, for instance in the case of the general approval dimension. The reliance on this naturalistic material also explains why 19 out of 20 items in the questionnaire were positive, even though it is recommended to use a more balanced mix of positive and negative items in attitude questionnaires (Cohen, Manion and Morrison 2011). Consequently, the items in the questionnaire became a blend of 20 utterances. The utterances covered totally eight components/dimensions as summarized in table 2.

Table 2. Distribution of components and dimensions in the attitude questionnaire.

Component/dimension	Dimension/component
	INT	INSTR	CORV	MAIN	GEN
AFFECTIVE	19	6, 16,	9, 11, 18		5, 12, 14
COGNITIVE	10, 15, 17	4	7		2
BEHAVIOURAL		20		1, 3, 8, 13

INT = the integrative dimension, INSTR = the instrumental dimension, CORV = the language as a core value. MAIN = the maintenance oriented dimension, GEN = the general approval dimension

As table 2 shows, the affective component was represented by nine utterances, the cognitive – by six, and the behavioural – by five. Thus, the questionnaire was created in Polish and discussed with two Polish speakers in Stockholm who matched the participants of the study in age and education, and who did not participate in this study. The aim of this procedure was to select items that were most engaging and to receive feedback on the clarity and naturalness of the language in the questionnaire and on the relevance of the content for the group being targeted (cf. Oppenheim 1992; Edwards 2011). The questionnaire is presented in the online appendix in the Polish original with English translations. However, it has to be reminded that the translations should be seen as approximations only. This means that they rather do not carry the load of connotational meaning (cf. Fawcett 1997) that the original Polish utterances most probably do.

6.3. Ethical statement

The study followed the Swedish Research Council guidelines for the Humanities and Social Sciences (Vetenskapsrådet 2011/2017).

7. Results

As mentioned, the background data on the participants was collected with a background questionnaire and is accounted for in the methodology section (Table 1). The meta-linguistic and linguistic data along with the attitudinal data are accounted for in the present section.

The material consisted of transcripts coded for meta-textual indicators of uncertainty and the use of a first-person pronoun. The size of the recordings with respect to the length and the word amount is summarized in table 3.

Table 3. Amount of words and length (in minutes) of the recordings

Group	Word count		Mean		Length (range in min.)				Mean
					CHCH		OAT
	CHCH	OAT	CHCH	OAT	min	max	min	max	CHCH	OAT
SE	22 311	102 637	892	4108	4	14	5	48	8	32
PL	21 585	95 151	863	3806	4	14	20	41	7	28

CHCH = Charlie Chaplin retelling task, OAT = odour association task

There were no significant differences in the duration and word count in the elicitation tasks, with the exception of the odour task, as there was a larger variation in duration in the SE group compared to the PL group. The reason for this was that one of the participants in SE group chose to interrupt the association task.

As mentioned, the meta-textual indicators of uncertainty were operationalized as the amount of parenthetical comments on word finding difficulties or difficulties with the formulation of utterances in relation to the length of both elicitation tasks.

The first-person pronoun variable was operationalized as a proportion of explicit uses (overt uses) of the pronoun, both in justified and unjustified contexts and the duration of the retelling task. The calculations for both variables are presented in table 4.

Table 4: Distribution of meta-textual indicators of uncertainty and the use of first-person personal pronouns

SE –group (n = 25)				PL-group (n = 25)
Meta-textual indicators of uncertainty
No.	Length in min.	No./ length	Mean	No.	Length in min.	No./ length	Mean
70	1005	70/1005	0,07	6	886	6/886	0,00
1^st person personal pronouns
Prodrop	Explicit use	Tot. no.	Explicit use/tot. no.	Prodrop	Explicit use	Tot.no.	Explicit use/tot. no
3739	1717	5456	1717/5456 (31%)	3400	883	4283	883/4283 (21%)

Statistical comparisons between the SE group and the PL group on these two variables were conducted. The Mann-Whitney U-test revealed significant results both for meta-textual indicators of uncertainty (p = 0.000***, α = 0.001) and for the use of first-person personal pronouns (p = 0.026*, α = 0.05). These results were interpreted as L1 attrition.

The attitude value was calculated as an average based on the 20 items in the attitude questionnaire. In the answers, one participant did not take a stand on one item. In addition, five participants left out the items concerning children/grandchildren because they did not apply to these participant. The mean attitude value was 3.8, with a range of 2.4 to 4.8.

Because the sample was relatively small, the relationship between the overall attitude value and meta-textual indicators of uncertainty and the use of first-person personal pronouns was investigated using a non-parametric (Pallant 2010) Spearman’s correlation. There was a moderate negative correlation between the attitude value and meta-textual indicators of uncertainty (2-tailed, N = 25, α = 0.01, p = 0.013*, rho= - 0.491). The correlation between the attitude value and the use of first-person pronouns was not significant.

8. Discussion and conclusion

The aim of this study was to investigate the degree to which an overall language attitude was related to L1 attrition amongst educated adult L1 Polish speakers in Sweden. The underlying assumption was that the more positive the attitude, the less likely that attrition would occur.

The relationship between the overall language attitude value and one of the two attritional variables – meta-textual indicators of uncertainty – was proved significant and had a moderate strength. This means, that the more positive the attitude, the less likely that meta-textual indicators of uncertainty will occur. However, the relationship between the attitudinal variable and the second attritional variable – the use of first-person personal pronouns as overt subjects – was not significant. Thus, these results are in line with previous research: L1 attrition can to some extent be explained by language attitudes, but some instances of L1 attrition seem not to be related to language attitudes. Indeed, this result generates a new question, albeit one that cannot be answered here: why do some linguistic variables seem to be influenced by language attitudes, whereas others are not?

Even if this question is not directly and easily answerable, our understanding could possibly be enhanced within the cross-linguistic influence framework proposed by Pavlenko (cf. section 2). The observed overuse of overt pronominal subjects in the SE group should be then seen as cross-linguistic influence, more specifically as restructuring (cf. Pavlenko, 2004). In this case, Swedish is as an external variable that influences Polish because standard contemporary Swedish has a pronominal subject constraint (e.g. Hammarberg and Viberg 1977). According to Pavlenko, restructuring and attrition are not the same, even if they are induced by the same factor. Pavlenko (2004) claims that restructuring can be called L1 attrition only when (i) it is empirically proved to be permanent, and (ii) it can be also observed in a monolingual L1 context, i.e. when a person exhibiting this influence continues to retain it in her/his linguistic repertoire outside the L2 context. Then, L1 attrition, or in Pavlenko’s terms - restructuring, might appear due to cross-linguistic influences (cf. also Seliger and Vago 1991), and therefore might not be correlated with non-linguistic factors as for instance language attitudes. This would mean that overuse of first-person pronouns as overt subjects would occur regardless of an individual’s attitude, or in other words, that the positive language attitude cannot level cross-linguistic influences. This is, however, highly speculative, and probably not really empirically testable. Finally, in light of the research on the reversibility of some aspects of L1 attrition, (e.g. Hyltenstam et al 2009; Seton and Schmid 2016), another question could be asked, provided that L1 attriters would be interested in any kind of reactivation exercises: Is cross-linguistically induced L1attrition easier to reverse than L1 attrition related to attitudinal factors?

In sum, this study supports previous insights into the relationship between L1 attrition and language attitudes. Hence, it is safe to say that some aspects of L1 attrition might be influenced by the attitudinal factor, while others – not. In addition, the study shows that L1 attrition occurs in L1 Polish speakers in Sweden to some extent, and – to some extent – that it is influenced by language attitudes. Finally, most probably other mechanisms than language attitudes are also involved in the attritional process.

8.1. Limitations of this study and generalizability of the results

Clearly, this work is not without limitations, and if stronger conclusions are to be drawn, improvements should be made. For instance, the attitude questionnaire should be tested on a larger population in order to further corroborate the relevance of the items included. In addition, the entire study might be replicated in another Polish immigrant context, or enlarged in Sweden. Possibly, a replication study in a context where Polish is in contact with a typologically closer language than Swedish would be of special interest. Such a study could shed light upon to what extent the overuse of first-person pronouns is cross-linguistically induced.

This study used a pragmatic approach (Denscombe 2014) that aims at obtaining “[ … ] accuracy that is good enough for the purposes of research within the resources available for research [ … ]” (Denscombe 2014, 50). Therefore, special caution is needed with respect to the extent to which generalizations and conclusions can be made.

On the one hand, much research, and not just linguistic research, frequently involves relatively small numbers of participants and uses non-probability sampling (Schork and Remington 2000, Denscombe 2014). That is also the case with this study. On the other hand, a principal demand on studies that (like this study) use inferential statistics for generating results and drawing conclusions is the use of probability samples (e.g., Körner and Wahlgren 2015; Cohen, Manion, and Morrison 2011), especially samples of more than 30 individuals (e.g., Denscombe 2014). However, both the relatively small sample and the non-probability sampling can produce sufficiently accurate data for a research project (Denscombe 2014) if the limitations are considered. In this study, the way of coping with the likely impact of the small sample was the use of non-parametric statistics (Pallant 2010). Furthermore, when conclusions were drawn, the decision about how far to go in extending generalizations was made based on another procedure in which the distinction between so-called target and sample groups is essential (Colton 1974). The target group is the group about which one might want to draw general conclusions, whereas the sample group is the one from which the results come. The selective factors for the sample group are crucial and those used in this study are accounted for in the following paragraphs.

In this study, a combination of two non-probability sampling techniques was used: purposive sampling and snowball sampling (Cohen, Manion, and Morrison 2011; Denscombe 2014). Purposive sampling means that the best information can be collected by focusing on a group of individuals with certain types of attributes, such as age and education on arrival in Sweden along with age at the point of the study and profession. In this way, the present sample is exploratory and provides a way of obtaining information about attrition and attitudes among highly educated adults. This information provides valuable insights into a more general topic, i.e., adult L1 attrition (cf. Denscombe 2014), especially when compared with other attrition studies. With snowball selection, the sample was built through a process in which participants nominated others who met the selective criteria of this study (cf. Denscombe 2014). It was a valuable sampling technique that identified the majority of the participants. Of course, snowball selection increases the risk of identifying individuals from the same network who are therefore quite similar (Denscombe 2014), although without this selection method, it would have been difficult, if not impossible, to build the sample at all.

The target population of the study’s sample group was composed of highly educated Polish speakers of working age who came to Sweden as adults before 1989. The sample group was composed of individuals representing all of the immigration waves from Poland to Sweden from 1945 to 1989 (see Lubińska 2013 for details on Polish migration to Sweden). In addition, the sample group reflected the socio-economic variation that exists among this target population and the group’s vitality with respect to Polish cultural and religious activities (see Lubińska 2013).

8.2. Conclusion

Despite its limitations accounted for in section 7.1, this small-scale study contributes to Swedish research on the development of migrant languages in Sweden and to Polish research on the development of Polish outside Poland. This study adds also to a better understanding of the relationship between L1 attrition and language attitudes and confirms the selective explanatory power of the attitudinal factor in L1 attrition.

Acknowledgements

I thank my colleagues at Stockholm University for their helpful feedback on earlier versions of this paper, and the Taylor & Francis Editing Service for improving my English. I also thank two anonymous reviewers for their valuable suggestions, and the editors at Scando-Slavica for their assistance. Any remaining errors and shortcomings are mine.

Disclosure Statement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. The author has no financial interest in and derived no benefit from the direct application of this research.

ORCID

Dorota Lubińska http://orcid.org/0000-0001-9741-8670

Notes

1 The present paper is based on some data collected for a PhD project on adult L1 language attrition in highly educated, long-term Polish immigrants in Sweden. The project is reported in Lubińska (2011). For the purpose of this paper, relevant data was reanalysed and reinterpreted.

2 A group of 25 participants was not an aim in itself; the aim was to collect data from as many individuals as possible who matched the selection criteria. When it was clear that no more than 25 participants would be in the experimental group, I looked for the same number of matching participants in the control group, for the sake of balance.

3 Language codes are used according to EU standards.

4 Wojewódzwto [read as voievutsvo] is a term for an administrative division of Poland, translated into English as e.g. county. I agree that this is a rather broad description, and for instance not informative when it comes to the linguistic variation in Poland. However, the attritional variables that were studied here are not dialect-dependent. Therefore, I have chosen not to be more specific about the origins of the PL and the SE group (both with regard to the settlement in Sweden and the origins in Poland).

5 The reason why I am using the term “Likert-like scale” and not “Likert scale” is that this study did not follow the stringent procedures for attitude scaling that would justify referring to a Likert scale in the proper sense of the term (cf. Oakes 2001).

6 With respect to Smolicz’ interpretation of Polish as a core value for Poles, my opinion is that this view might require refinement in future research. Even if the Polish language is claimed to be a core value for the Polish intellectuals, the same seems not to apply to other groups (e.g., Szydłowska-Ceglowa 1992; cf. also Zaręba, 1979), which tend to see Roman Catholicism as a common value (cf. Uggla 1993). The concept is, however, useful here insofar as the data come from highly educated Polish speakers.

7 Polish has also been the only mother tongue for some Polish Jews, such as those who emigrated from Poland to Sweden after 1968 (Ilicki 1988; see also Lubińska 2013). It is an empirical question whether Polish also is a core value for this group, or how the ties to Polish might have evolved, given the historical circumstances which this particular group became caught up in.