A demographic analysis of the dynamics of language behaviour

2Many western countries are currently facing the foreseeable consequences of demographic ageing : decreasing natural growth, a shrinking workforce, increases in dependency, and rising healthcare and pension costs. In a number of these countries, international immigration has been chosen as an obvious response to contemporary demographic challenges. Yet low fertility and high immigration rates are leading to profound changes in the characteristics of their populations. These transformations are so extensive that some have labelled this phenomenon a “third demographic transition” (Coleman, 2006).

3Canada possesses all of the characteristics associated with this third demographic transition (Dion et al., 2013). According to the 2006 census, 20% of the Canadian population was born abroad, and this percentage is set to increase to between 25% and 28% in 2031 (Caron Malenfant et al., 2010). Moreover, new immigrants show increasingly diverse linguistic characteristics and behaviours. In Canada, the proportion of persons whose native language is neither French nor English (Canada’s two official languages) rose from 10% in 1981 to 20% in 2006, and should reach between 29% and 32% by 2031 (Caron Malenfant et al., 2010). Given the importance of language skills on the labour market (Chiswick et Miller, 2002) and the lower literacy [1] levels of allophone [2] immigrants (OECD, 2013), change in the linguistic composition of destination countries such as Canada is of great economic importance.

4Most linguistic behaviours are not static, but vary over time, particularly among immigrants and their descendants. The language spoken at home, the language used at work, and knowledge of these or other languages can change over the course of an individual’s life. Such changes can offer an index of the cultural and linguistic integration of first- and second-generation immigrants.

5There is a large multidisciplinary research literature on patterns of change in the linguistic behaviours and characteristics of immigrants and linguistic minorities (Abrams and Strogatz, 2003 ; Castonguay, 1994 ; Crystal, 2000 ; Fishman, 1991 ; Rumbaut et al., 2006 ; Wickstrom, 2005). In this article, we will specifically examine language shifts, wherein individuals abandon their native language as the principal language spoken at home and adopt another. In the native-born population, the language most often spoken at home and in the public sphere is ordinarily the first language spoken in childhood. The pattern can be different among allophone immigrants and members of linguistic minorities, who may later come into contact with speakers of the majority language both in intimate relationships (exogamous couples, friendships) and in public ones (at work, in the civic sphere). To understand one another, individuals who speak different languages must adopt a common language of communication. In this situation, it is usually the most useful or prestigious language that wins out : either the language that is most widespread in the population, or the language of institutions and business (Laponce, 2006). Change in the language used at home is known as “language shift”. Such a shift may occur, for example, when a brother and sister who are native speakers of Chinese attend English-language schools in Canada, and start conversing in English at home ; [3] or when a young man whose native language is Arabic leaves his parents’ home where Arabic is the dominant language to move in with his French-speaking spouse.

6In the absence of factors that slow down language shifts (strong linguistic endogamy, social or geographic isolation), the native language of a cohort of immigrants – due to its lack of use in institutions and its limited usefulness in public contexts in comparison to the official language of the host country – disappears as the main language used at home within three generations. The extinction of allochthonous languages in three generations is an empirical theory supported by many scientific studies (Rumbaut et al., 2006 ; Bélanger et al., 2011a ; Alba et al., 2002), although obviously there are some exceptions. [4]

7With regard to language shifts, several factors distinguish longstanding linguistic minorities from immigrants. The former are typically more numerous and geographically concentrated (Mougeon and Beniak, 1994), benefit from the structuring contributions of ethnolinguistic institutions that have developed over time (Breton, 1964 ; Chiswick and Miller, 1996), and often enjoy some legal and political recognition (Leclerc, 1986), enabling them to preserve their language more effectively.

8The dynamics of language shift over time – that is, the proportion of language shifts occurring at different ages – has rarely been studied due to a lack of appropriate data. Canadian census data include information on native language and the language spoken at home, but it is impossible to establish the exact moment when a given individual makes a language shift : either it has not yet happened, or it happened some time between birth and the time of the census. Measuring language shifts using these data is nonetheless a useful way to assess the “vitality” of a language (Castonguay, 2005a, 2005b) or to study the determinants of language choices (Bélanger et al., 2011b), but they cannot be used to observe the dynamics of shifts. Rigorous estimates of language shift rates by age or region of residence are important both in theoretical terms (understanding the phenomenon) and in practical terms (for demolinguistic projections).

9This article provides an empirical description of the timing and intensity of language shifts, without seeking to identify the full range of sociodemographic determinants of the phenomenon (Bélanger et al., 2011b). We propose a new method for estimating rates of language shift over the life course on the basis of a single Canadian census, which avoids some of the limitations inherent to previous methods, linked either to the methodology itself (Termote et al., 2011) or to the sample (Corbeil and Houle, 2014).

10While in Canada as a whole almost all language shifts are towards English, in Quebec both French and English are languages of convergence. In the metropolitan region of Montreal where more than 80% of immigrants to Quebec are concentrated, around half of language shifts are towards French. We thus performed a complementary analysis for Quebec, examining the proportion of shifts towards French by age and time since immigration.

11With certain hypothetical assumptions (stationary rates of language shift, negligible impact of demographic behaviours that differ with language, no more than one language shift per individual), this method can be used to make comparisons across time (using the four censuses from 1991 to 2006), space (different Canadian regions), sociodemographic traits (immigrant status and level of education) and linguistic characteristics (allophones and francophones). These comparisons suggest that the process of language shift is quite stationary, as the estimates remain stable between censuses.

Data used

12The data used here were drawn from the files of individual and confidential data (20% sample) from the Canadian censuses of 1991, 1996, 2001, and 2006.

13The language variables analysed were mother tongue and language most often spoken at home. The following question was used to establish mother tongue : “What is the language that this person first learned at home in childhood and still understands ? If this person no longer understands the first language learned, indicate the second language learned”. The concept of mother tongue is analogous (but not identical) to the concepts of native language or first language (lengua inicial) (Jiménez-Salcedo, 2011). Francophones, anglophones, and allophones are defined as individuals whose mother tongue is French, English, or another language, respectively. The expression “linguistic minorities” is reserved for official language communities that are in a minority with respect to the language most widespread in the region : namely, anglophone communities in Quebec and francophone communities in the rest of Canada. The language spoken at home is derived from a simple and unambiguous question : “What language does this person speak most often at home ?”. We refer to an individual who speaks a given language at home as a speaker of that language.

14Although the formulation of these questions suggests that a single answer is expected, Statistics Canada accepts multiple answers to language questions. In our data, multiple answers were given in 1.3% of cases for the mother tongue question and in 1.8% of cases for the question on language spoken at home. In order to simplify the analysis, we reduced the number of possible categories from seven [5] to three (French, English, other). Multiple answers were thus simplified by randomly choosing a single one of the reported languages. For example, an individual who reported speaking French and another language is randomly classified as speaking either French (with probability 0.5) or another language (probability 0.5). The issue of multiple answers has been the subject of numerous analyses, which we will not review here. [6] A sensitivity analysis shows that random allocation of multiple answers offers a good intermediate solution (data not shown, see Appendix 2).

15A language shift is defined as most often speaking a language other than one’s mother tongue at home. Language persistence is the proportion of a cohort that has not gone through a language shift.

Earlier methods

16Termote and collaborators (Termote et al., 2011 ; Termote and Thibault, 2008) have previously calculated rates of language shift by age and immigrant status for purposes of demolinguistic projection. The rates were obtained by comparing the linguistic characteristics of five-year cohorts in two successive censuses. As language shifts are relatively rare events, at least in the first generation of immigrants, these comparisons linking cohorts across two censuses are subject to sampling errors, differential undercounting by age, and other interfering phenomena such as migration. And as these errors are generally larger among allophones and in the age groups where language shifts are more frequent, it is difficult to obtain estimates that are stable across censuses. The results obtained using this estimation technique are highly variable and present certain anomalies. Levels of final language persistence (after age 50) also vary significantly, depending on the censuses used for the comparison.

17In theory, analysis of data from event history or longitudinal surveys is still the best method for deriving rates of language shift. Corbeil and Houle (2014) used such a survey to derive rates by age among allophones in the Montreal region. Unfortunately, the small sample size and the nature of the study population make interregional comparisons problematic. In addition, because this survey will most likely not be repeated, behaviours cannot be compared over time.

18Finally, data from two successive censuses can be linked [7] to permit true longitudinal measurement of linguistic behaviours, but this method requires a linkage key (not available to researchers outside Statistics Canada at the time of analysis) or a linkage algorithm, also subject to various types of errors. Statistics Canada has already used census record linkage to study transitions in Aboriginal identity among Canadians (Caron Malenfant et al., 2014).

Method of analysis

19Here we describe a simple method for estimating the probability of language shift, designed to avoid the problems described above. It is based on the principle of fictitious cohorts, and can be used to determine the timing of language shifts by year of age (natives), or by years since immigration to Canada (immigrants) in various regions of Canada and different mother tongue groups (francophones, anglophones, allophones).

20Annual language shift probabilities were drawn from a “survival curve” established by calculating and juxtaposing the linguistic persistence of successive cohorts. Cohorts were defined by age for natives and by time since immigration for immigrants. We assumed that an individual’s mother tongue is fixed at birth and does not vary over time, which is a reasonable approximation (Lepage, 2011). We studied only language shifts that occur before age 50, as they are rare in older groups. Only immigrants who arrived after 1970 were included in the analysis because numbers of immigrants per age at arrival are too low for earlier cohorts. Moreover, the composition of immigration and language policies in Canada in general, and in Quebec in particular, have greatly changed since the 1970s, and the analysis of previous language behaviours is now a matter of mainly historical interest. Non-permanent residents were excluded because, by definition, they will either leave the country when their temporary visa expires or, if they become permanent residents, join another cohort of immigrants.

21The principle of fictitious cohorts assumes that rates of language shift are stationary, i.e. they do not vary over time. This possible limitation on the method is compensated by the fact that we were able to compare curves of language persistence over a period of 15 years (four censuses). The similarity of observed persistence curves suggests that the rates are indeed relatively stationary, at least in the medium term (Figure 1, Figure 2E). This method is also subject to errors in rates by age or mother tongue due to undercoverage, insofar as the characteristics of the “missing” individuals may differ from the population covered. However, undercoverage mainly affects young adults, [8] and its overall impact on the estimation of the survival curve was likely negligible. Finally, differential mobility between groups with different language characteristics could lead to under- or over-estimation of language persistence curves at sub-national level. We examine these potential biases in detail in the discussion.

Figure 1

Language persistence in the allophone population by age and time since immigration

Language persistence in the allophone population by age and time since immigration

Notes : A. All allophones ; B. Allophones born in Canada, non-Aboriginal allophones born in Canada, and allophones born in Canada to at least one immigrant parent (2006 census) ; C. Non-Aboriginal allophones born in Canada ; D. Immigrant allophones ; E. Immigrant allophones by age at arrival in Canada.

Figure 2

Modelling of the curves of language persistence in the allophone population using polynomial linear regression

Modelling of the curves of language persistence in the allophone population using polynomial linear regression

Note : A. Raw data and model, population of non-Aboriginals born in Canada ; B. Modelled curves of language persistence, non-Aboriginals born in Canada. C. Annual probability of language shift among immigrants arriving in Canada, by age at immigration ; D. Language persistence at arrival in Canada, by age at immigration ; E. Value of the regression coefficient for each of the census variables (reference 1991).

22The parameters of the survival curves were obtained using a linear regression on census microdata. The regressions, whose dependent variable, language shift, takes a value of either 0 or 1, include effects of time (age or time since immigration), time squared, and census year to control for period effects. Contrary to the parameters of the functions associated with risk and duration models, the parameters of the linear regression do not necessarily lead to a strictly decreasing function (here, a quadratic function) ; this is a problem insofar as a survival curve can never be an increasing curve. [9] In order to avoid this issue, any positive slope in the observed survival curve was transformed into a zero slope. Note that this situation only occurred in cases where the probability of language shifts was very low (generally after age 40 ; Figure 2C). This adjustment did not have a significant impact on the results, and mainly served the objective of theoretical coherence.

Language persistence in native-born allophones and immigrants

23Figure 1 shows allophones’ language persistence by age and time since immigration. Figure 1A presents persistence for the whole Canadian allophone population from the 1991, 1996, 2001, and 2006 censuses. It stands at around 0.8 at age zero, suggesting that 20% of allophones made a language shift “at birth”. Such early language shifts may seem surprising, given that newborns are not yet able to speak. The instructions in the Census Guide ask parents to “report the language spoken most often to this child at home”, [10] both for the mother tongue and the language most spoken at home. A fifth of parents thus give two different answers to the same question. It is possible that allophone respondents who have themselves made a language shift give their own mother tongue as a marker of identity for the child, despite the fact that one of the official languages is spoken more often at home. Whatever the explanation, analysis of these early language shifts is beyond the scope of this article.

24Persistence increases slightly up to the age of two years, the time when children generally learn to speak, and then begins an uninterrupted decrease. Note also that most language shifts take place early in life, at ages 2-10, and that persistence reaches a level of around 0.5 between ages 10 and 20. Finally, the persistence curves derived from the four censuses are relatively similar, suggesting that the distribution of language shifts remains stable over time.

25This result conceals an important distinction, however : allophone immigrants do not have the same probability of making a language shift as allophones born in Canada. Figure 1B shows the persistence curve for individuals born in Canada, based on the 2006 census. Here it can be seen that language persistence is lower in native-born allophones than in the total allophone population. In the former group, after age 50, it reached a steady level of around 0.2, versus 0.5 in the latter group. Moreover, among allophones born in Canada, two groups can be distinguished : those born to immigrant parents (generation 2) and those born to parents who were themselves born in Canada (generation 3), whether or not they are of Aboriginal [11] origin. Language persistence is lower among non-Aboriginal allophones, and particularly among the children of immigrants. This result is unsurprising : Aboriginals form longstanding linguistic communities, and the language persistence of allophones born in Canada is increased by individuals who are third-generation or more, who form a residual category. The strong language persistence of this third generation may be explained by a range of factors : high levels of religious endogamy, spatially concentrated communities, the presence of strong ethnocultural institutions, etc.

26Figure 1C shows that the distribution of language shifts among non-Aboriginal native-born allophones is relatively robust over time, and that shifts are highly prevalent, with language persistence dropping to around 0.1 at age 50.

27Figure 1D shows language persistence among immigrants by time since immigration. Around 20% of immigrants have already made a language shift before their arrival in Canada, and language persistence decreases in a linear fashion in the years after their arrival. Note that only immigrants who arrived after 1970 were included in the analysis, which explains why the curves are truncated. Because age at arrival is an important determinant of integration into the host society, immigrants are further categorized according to this variable. Figure 1E shows persistence among immigrants who arrived at ages 15-29, compared to those who arrived before age 15. Persistence is markedly lower among immigrants who are younger on arrival. Language shifts also seem to have slowed in more recent censuses in both age groups.

28One way to facilitate the analysis of the information contained in the language persistence curve is to fit a mathematical function to it, and calculate the associated parameters. A linear regression establishing a quadratic relationship (second-degree polynomial) between language persistence and time (age or time since immigration) is used (see Appendix 1 for other examples of possible models). As Figure 2A shows, the obtained fit adequately describes the persistence curve, although the quadratic function is not able to track the slight increase in rates of language shift at preschool ages and in early adulthood. The two coefficients and the constant of the regression (the census coefficients are excluded) thus provide parameters that summarize the curve of language persistence.

29Figure 2B shows results for allophone immigrants by age at arrival in Canada. First, it can be seen that the younger immigrants are when they arrive, the lower their language persistence. For an allophone immigrant who arrived at age 40, the probability of shifting languages is almost nil, whereas for a child who arrived in Canada before age 10 it is practically identical to that of an allophone who was born in Canada. A qualitative shift can also be seen for immigrants who arrived at age 15 or later : persistence is significantly higher in these age groups than in younger groups. This is likely explained by the normative impact of school socialization. It is at school that codes as well as cultural and linguistic references of the host society are transmitted ; it is also here that durable bonds are formed between allophones and speakers of an official language. Figure 2C presents the annual probability of a language shift among immigrants at the time of their arrival in Canada. It can be seen that immigrants who arrived at a very young age are highly likely to make a language shift. This initial probability quickly decreases over the time of schooling, falling from 0.038 at age two to 0.022 at age 16 (– 0.016). It continues to decrease after this age, but more slowly, reaching 0.011 at age 30 and stabilizing around zero after age 40.

30Figure 2D shows the curve of persistence at arrival in Canada, by age at immigration : this is an estimate of the proportion of allophone immigrants who had already gone through a language shift abroad. Language persistence at arrival in Canada varies with age at arrival. The least persistent immigrants are those who arrive at ages 25-30 or thereabouts. Moreover, it can be seen that the individuals who are around the right ages to be this group’s children (ages 2-10) are also among the least persistent. The level of persistence also increases with age, beginning at age 30 : the greater an immigrant’s age at arrival in Canada, the less likely he/she is to have already made a language shift abroad. Overall, however, the proportion of immigrants who had already gone through a language shift before arriving in Canada is relatively constant, and persistence ranges between 75% and 85% depending on age at arrival. These results are similar to those obtained elsewhere (Castonguay, 2005b).

31Finally, figure 2E illustrates the effect of period on language persistence (as reflected by the regression coefficients for each of the dichotomous variables representing the censuses ; the 1991 census is used as the reference). It can be seen that the effect of period is relatively modest : it generally ranges between –0.02 and 0.04, which roughly corresponds to the effect of one or two years of exposure to this risk in a young second-generation allophone. Moreover, variations between censuses seem to be random, and there is no clear trend over time.

Interregional comparisons

32The analysis thus far has focused on language persistence among allophones throughout Canada. Figure 3 presents an inter-regional comparison of persistence among native-born allophones in Quebec, Ontario, British Columbia, and the rest of Canada. Figure 3A shows that there are few disparities between the different regions among Canadian-born allophones. The dynamics of language shifts thus seem to be robust even at this sub-national scale. Quebec alone stands out as showing slightly higher persistence overall than other regions. Figure 3B shows curves of language persistence in Quebec from four censuses. The trend seems to be stable over time, although some variability can be seen, particularly in adolescence.

Figure 3

Interregional comparison of language persistence among allophones

Interregional comparison of language persistence among allophones

Note : A. Persistence in allophones born in Canada to at least one immigrant parent ; B. Persistence in non-Aboriginals born in Canada.

Persistence of official language minorities

33This model of language persistence may also apply to individuals from minority official language communities. Figure 4A presents persistence curves for francophone minorities outside Quebec, as well as those of the anglophone minority and francophone majority in Quebec (with the latter taken as a control group in this context). The language persistence of Quebec francophones is very high (a decrease of only a few percentage points, and no significant variation over the life course), but that of francophones living west of Ontario is markedly lower (“Rest of Canada” curve in the graph). Their persistence is comparable to that of second-generation allophones : in this case, around three quarters of a francophone cohort speak English at home before age 50. Contrary to the observed patterns among allophones, the persistence of linguistic minorities varies widely between regions. Persistence is highest among francophones in northern New Brunswick : from 96% at birth, it falls to around 90% at the end of the risk period. The dynamic of language shifts among Quebec anglophones is similar to that of francophones in northern New Brunswick, although the initial level of persistence is lower. The former group begins with persistence of around 91% at birth, decreasing to around 83% after age 30. This level remains relatively high given that the English mother tongue population represents around 8% of the Quebec population (and around 13% in the Montreal metropolitan region), whereas New Brunswick francophones represent nearly a third of the provincial population (and nearly 60% in the north of the province). Francophones in the regions of Ottawa, northern Ontario, and Atlantic Canada (without northern New Brunswick) show intermediate patterns of persistence.

Figure 4

Curves of language persistence in official language communities

Curves of language persistence in official language communities

Note : A. Regional comparisons ; B. Comparisons over time for three regions.

34Does the persistence of these minorities change with time ? Figure 4B shows the persistence curves of francophone minorities in three regions over four censuses. The curves did not substantially vary over the 15-year period covered by the four censuses. A slight decreasing trend is apparent, however, among francophones in northern Ontario between 1991 and 2006, particularly between at ages 10-20 (figure 4B).

The impact of education on language persistence

35Insofar as education is considered to promote language learning and linguistic integration, attaining a higher level of education might be expected to increase the probability that an allophone will make a language shift. Figure 5 shows that level of education indeed plays a role : allophones without a high-school diploma are markedly less likely to shift languages than allophones with at least a high-school diploma, whether they are immigrants (figure 5A) or native-born (figure 5B). Among immigrants, holders of a high-school diploma are slightly less likely to shift languages than those with a postsecondary qualification (university or other). Among individuals born in Canada, there are no major differences across different educational levels. The pattern in this case can be described as a threshold effect, rather than a gradient.

Figure 5

Language persistence in the allophone population by highest educational qualification obtained

Language persistence in the allophone population by highest educational qualification obtained

Analysis of the language choices of allophones in Quebec

36While almost all language shifts in Canada outside Quebec are towards English, the pattern is different in Quebec, where French and English are both languages of convergence. In principle, there is no reason to believe that the number of languages of convergence has a significant effect on language shift rates. Figure 3 shows, moreover, that the profile of language persistence in Quebec is similar to those of other Canadian regions. The sociolinguistic forces underlying language shifts (the weak position of foreign languages with respect to the official languages) are the same in Quebec as in the rest of Canada. Only the attractiveness of one or the other official languages varies across regions. The proportion of language shifts towards French thus reveals its relative status with respect to English, a fundamental aspect of language dynamics in Quebec.

37Language choices were therefore examined in a second stage to analyse the proportion of shifts in each cohort towards French, by far the majority language in Quebec as a whole, but codominant with English in the Montreal region which attracts the great majority of allophones. Allophones were divided into two categories by “language affinity” : francotropes and non-francotropes. Francotrope allophones are those whose ethnocultural characteristics predispose them to shift languages towards French. For the purposes of the present study, individuals were classified as francotrope on the basis of their country of birth (for immigrants) or their mother tongue (for the native-born). A binomial test was performed for each group to determine whether the proportion of language shifts towards French was significantly higher than 0.5. Francotrope allophones are generally speakers of a Latin language, come from a country whose official language is a Latin language, or come from a French-speaking country. The empirical results of this approach are similar to those obtained using other, more theoretical methods (Castonguay, 1994 ; Bélanger et al., 2011a), except in the particular case of individuals of Italian origin and Italian speakers, who were classified here as non-francotropes despite their Latin origins.

38Figure 6A shows the proportion of language shifts in Quebec towards French among first- and second-generation allophones. The proportion of allophone immigrants who shift towards French rather than English has been relatively stable since the early 1980s, at around 70%. It began to rise before the 1970s, and continued to do so until the early 1980s. Among second-generation allophones, the proportion of shifts towards French increased in almost linear fashion during the entire period 1970-2006, reaching the same level as that of allophone immigrants in 2006. This discrepancy in timing is partly explained by the time between immigration to Canada and the birth of the first second-generation children. The changes observed in the first generation are transmitted to the second after a time lag that depends on age at arrival.

Figure 6

Proportion of language shifts towards French in Quebec

Proportion of language shifts towards French in Quebec

Note : A. Allophone population, both immigrant and born in Canada ; B. Allophone population born in Canada, by language affinity (francotrope or non-francotrope) ; C. Allophone immigrant population, by language affinity and age at immigration ; D. Proportion of francotropes in the total allophone population.

39Figure 6B shows the results for second-generation immigrants, both francotrope and non-francotrope. Note that the proportion of language shifts towards French has gradually increased over recent cohorts in both groups. This trend is more marked among francotropes in absolute terms (from 0.5 to 0.8), but is greater in relative terms among non-francotropes (from 0.1 to 0.3). Given the cross-sectional nature of the data, this progression might alternatively indicate that early shifts towards French have become more frequent, probably due to obligatory attendance at French-language schools imposed by Quebec language legislation since the late 1970s (Bill 101).

40Figure 6C explores trends in language shifts towards French by immigrants’ language affinity and age at arrival in Canada. We distinguished immigrants who arrived before age 15 (school age) from those who were older on arrival in order to check for the potential effect of attendance at French-language schools imposed by Quebec language legislation. Among immigrants who arrived at age 15 or later, little variation can be seen among cohorts who arrived between 1970 and 2006 ; the proportion of shifts towards French remained relatively stable, at around 0.3 among non-francotropes (slightly decreasing in recent cohorts) and around 0.8 among francotropes (slightly increasing in recent cohorts). For immigrants who arrived before age 15, a large increase in shifts towards French can be seen in the cohorts who arrived in the 1970s, followed by relative stability, at around 0.8, for those who arrived in the early 1980s. Among non-francotropes, this figure increased gradually between 1970 and 2006, from around 0.2 to around 0.5. Overall, then, the data indicate that recent cohorts of allophones, whether immigrants or born in Canada, have more often made a language shift towards French than older cohorts. Note, however, that the positive trend observed in all of the subgroups (figures 6B and 6C) is less marked than for allophones overall (figure 6A), suggesting an underlying composition effect. Figure 6D tests this hypothesis, illustrating how the proportion of francotropes within the different allophone cohorts (immigrants and native-born) changed over time between 1970 and 2006. The similarity between figures 6A and 6D is clear : the increase in the proportion of shifts towards French is strongly correlated with the change in the language affinity composition of allophone cohorts. The most recent cohorts have included a growing proportion of francotropes, which has raised the proportion of language shifts towards French.

Linguistic persistence in allophones

42Rates of language shift among allophones born in Canada are very high. Less than 20% of a given cohort continue to speak their mother tongue at home after age 50, and most language shifts take place before age 30. This confirms that the disappearance of foreign languages “in three generations” is in fact condensed into the second generation, or even the first generation in cases where immigrants arrive in Canada in the first years of life (Figures 1 and 2).

43Differences in methodological and analytical approaches make it difficult to perform a rigorous comparison of our results with those of existing studies. Corbeil and Houle (2014), for example, derived annual rates of language shift by destination language among allophones in the Montreal region using data from a small sample, which are difficult to compare with the overall rates (independent of destination language) calculated here. It is possible, however, to compare final language persistence among second-generation allophones, calculated here at 0.1, with the higher figure of 0.34 obtained by Termote et al. (2011).

44The curves of language persistence derived in the present article are robust over time and space among allophones : little variation appeared between the censuses or between different regions of Canada. Persistence nevertheless proved to be slightly higher in Quebec than in the rest of Canada. This difference is small, but intriguing. It could be that the competition between two languages of convergence in Quebec changes the process underlying language shifts, although it is difficult to see what mechanism might be at play. Differential migration could also explain part of the difference if language shift rates are higher among departures than among arrivals, but this factor is unlikely to have a large impact (Appendix 3). Finally, some authors have suggested that a strong welfare state (Quebec has more extensive social coverage than the rest of Canada) may slow the integration of cultural communities as state financial support cushions them from the linguistic and cultural demands of the market (Koopmans, 2010).

45Among allophone immigrants, we found that language persistence in individuals who arrive before age 10 is similar to that of allophones born in Canada. This result is expected, insofar as immigrants who arrive at a young age go to Canadian schools, the main vector of linguistic integration, notably as an institution for the transmission of culture and as a place of socialization which in the long term favours linguistically exogamous unions. And indeed, the probability of making a language shift falls sharply beyond the age range of obligatory schooling. The older immigrants are at arrival, the less likely they are to shift languages. Given the median age at immigration (late twenties), it is thus logical to observe fewer language shifts in the first generation.

46Whether allophones are immigrants or born in Canada, their level of education has only a marginal effect on language persistence. Only allophones with no qualifications show substantially higher persistence.

Language persistence in official language minorities

47In contrast to our results on allophones, we found large regional variations in the persistence curves of official language minorities. These variations reflect differences in linguistic environments, such as the spatial concentration of communities, and access to public services such as schools, universities or hospitals in the minority language (Castonguay, 2005a). As in the case of allophones, the analysis showed that persistence is relatively stable over time, at least in the short term.

48Rates of language shift are high in almost all Canadian francophone communities, in some cases comparable to those of second-generation allophones. The only exception are the francophones of New Brunswick, whose final language persistence stands at around 90%. Other francophone groups outside Quebec are decreasing by 30% to 80% with every generation. In this context of linguistic transformation, it is not surprising that the notion of “francophone” has been undergoing a major redefinition in Canada outside Quebec (Guignard Noël et al., 2014).

49The linguistic persistence of the anglophone minority in Quebec is slightly lower than that of francophones in New Brunswick, but higher than that of francophones in the rest of Canada. It is important to note, however, that anglophone language shifts towards French are entirely offset by an opposite flow of shifts from French towards English (Castonguay, 2002), and by an even larger flow of shifts from other languages towards English. According to the 2011 census, there are nearly 650,000 individuals of English mother tongue in Quebec, and 835,000 people who speak English at home. Contrary to francophone communities outside Quebec, the Quebec anglophone community is not undergoing any marked overall language shift.

Orientation of language shifts in Quebec

50Our analysis of allophones’ language choices in Quebec shows that the attractiveness of French has gradually increased since the 1970s. The rate of shift has been highest among immigrants who arrived in Canada as children and among francotropes. It is interesting to note that there has been little variation in the proportion of shifts towards French among immigrants who arrived as adults, which seems to indicate that this change has mainly been due to immigration policies (immigrants are selected partly on the basis of their knowledge of French) and the provisions on schooling in Quebec language legislation (commonly called Bill 101). As can be seen in figure 6D, the increase in shifts towards French is more a result of the ethnocultural composition of immigration than of changes in allophones’ language preferences and behaviours. It can also be seen that the increase in language shifts towards French began several years before the adoption of Bill 101 in 1977. Several factors may explain this pattern. First, efforts to raise the status of French had already begun in the late 1960s. Bill 101 was not the first piece of language legislation in Quebec, and such legislation is only one of various aspects of language policy (alongside immigration policy and cultural policy). The increase in shifts before 1977 may also be an artefact of gradual change in the composition of immigration caused by linguistically differentiated migration, which was very significant in the 1970s and 1980s : after the most “anglicized” immigrants left Quebec, those who remained were more French-oriented. Nevertheless, the proportion of language shifts towards French still remains lower than the relative proportion of mother tongue speakers of French in the Quebec population, resulting in decreases in the proportion of francophones over time (Bélanger and Sabourin, 2013).

51The method used in the present article presents a number of important advantages. Conceptually, it disconnects allophones’ language shifts from their choice of destination language : in our view, these two phenomena are linked to distinct sociological forces. Shifts themselves follow from the relative status of the allochthonous language relative to the language of the host country. As allochthonous languages generally lack institutional legitimacy, they cannot be used in most public contexts : they are thus relegated to the private sphere, where they cannot persist for more than one or two generations. The dynamic underlying language shifts in immigrant populations should thus produce similar effects in comparable regions. And indeed, little variation is seen across Canadian regions.

52The choice of destination language, on the other hand, depends on the relative status of French and English. In contrast to the abandonment of allochthonous mother tongues, the choice of destination language is contingent, specific to a given period, and dependent on the singular history that has shaped the power relationship between English and French in Canada. Outside Quebec, shifts are exclusively towards English, whereas in Quebec they are divided between French and English. This power relationship is in constant movement, particularly in Quebec where, given the presence of strong institutions, English remains highly attractive despite existing language law and the low demographic weight of anglophones in the population.

Methodological limitations

53Certain limitations in the analysis presented here must be addressed. First, it is unlikely that all language groups have exactly the same propensity to shift languages : persistence curves thus offer only an approximation of allophones’ mean language persistence. Volumes of immigration to Canada have been maintained at very high levels since the late 1980s (between 260,000 and 285,000 entries expected for 2015, or two times the rate of the United States), and the current composition of immigration differs from that of the past. Dominated by British and European immigrants until the 1970s, today it is much more diverse and stems mainly from Asia, but also from Latin America and North Africa (particularly in Quebec). In the long term, then, changes in the ethnocultural composition of the immigrant population could thus change the language persistence curves presented here, most of all among first-generation immigrants, if propensities to language endogamy differ between groups. In any case, language persistence can be expected to change very slowly (immigrant stocks are large), with limited effects on the second generation (because of shared socialization at school, a lower degree of language endogamy than the first generation).

54The method presented here is also subject to biases related to differential mobility in groups with different language characteristics. As a general rule, individuals who have made a language shift towards English are more mobile. Francophones who are anglicized outside Quebec thus have a greater tendency to leave their region of origin than francophones who continue to speak their mother tongue at home. This differential mobility could thus lead to underestimation of language shifts among francophones outside Quebec. In Quebec, those who have a greater tendency to leave the province are persistent anglophones and allophones who speak English at home. This could lead to overestimation of language shifts among anglophones (since the most persistent tend to leave) and underestimation of the proportion of language shifts towards French. However, net migration between regions is relatively limited, and the impact of these factors is probably quite small (data not shown, Appendix 3).

55The method presented here focuses on a small number of explanatory variables : mother tongue, age, time since immigration and age at arrival in Canada, and, to a lesser extent, level of education. While these are determinants of language shift (Lapierre Vincent, 2004) and have been identified as such in other contexts, as in France (Jeantheau, 2007), they are clearly not exhaustive ; other, socioeconomic or cultural elements may play a role. Moreover, the variables examined here – mother tongue and language spoken at home – although indisputably central to linguistic life, are just one component of the study of language behaviours : in a multilingual context, linguistic interactions are complex and diverse (Lamarre et Dagenais, 2004). Finally, our analysis is not intended to define what linguistic integration or assimilation is, or to establish what it should be. This important but complex normative question is beyond the scope of our demographic analysis.

56Despite these limitations, the method presented here offers a useful tool for studying language persistence in Canada, and could be transposed to other national contexts where language data are available, as in Catalonia, where information on initial and habitual languages is systematically recorded. The size of the sample and the periodicity of the Canadian census allowed us to perform a detailed analysis and draw useful comparisons across time and space. The trends observed in the present article are in keeping with existing research, but are more robust and precise.