DRAFT, March 18 2002 Gillian Sankoff

(to appear in Norbert Dittmar et al, Handbook of Sociolinguistics)

With the exception of two distinguished avatars (Hermann’s 1929 restudy of Charmey, following on Gauchat 1905; and Brink and Lund’s (1975; 1979) monumental study of Copenhagen), cross-sectional and longitudinal research was almost unknown in sociolinguistics prior to the 1990s. In the past decade, however, there has been a great upsurge in research of this type, mainly due to the fact that sociolinguists have exploited the possibility of carrying out follow-up studies of communities originally studied in the decade between about 1965 and 1975.

Such studies are of two major types. In cross-sectional research (also referred to as trend studies), investigators re-study the same community that has been the subject of earlier research, often attempting insofar as is possible to replicate the methods (sampling and data collection methods) of the initial study. Longitudinal studies (also known as panel studies) follow particular individuals or groups of subjects over a period of time.

It might initially seem that the two methods are quite different, and appropriate for very different ends. Whereas cross sectional studies seem ideally suited to discerning trends at the level of the community, longitudinal studies are a preferred method of study in language acquisition (both L1 and L2) in which the goal is to trace the progress or chart the path of learners as they progress toward the target. Within this research tradition, even the more socially-oriented perspectives on these topics (language socialization studies in the case of L1 acquisition; language shift studies in the case of L2) chart the trajectories of their subjects against the backdrop of a target community which is taken (often tacitly) to be stable. The present article will not review either L1 or L2 acquisition studies, nor studies of language attrition or language shift, many of which use longitudinal or cross sectional methodology. Instead, I will consider research that uses longitudinal methodology, as well as cross-sectional research methodology, in the study of language change and variation within speech communities characterized by overall social stability, i.e. that have not, during the course of these longitudinal studies, experienced significant population mobility and change due to language contact.

Before considering longitudinal and cross sectional research as such, we will begin by a brief review of the major research concept that has been used in sociolinguistics to infer change in progress from synchronic observations: the concept of "apparent time". Labov introduced the paired concepts of real time and apparent time in his 1964 study of New York City, the first major research project to implement the reconciliation of synchronic and diachronic linguistics (Weinreich, Labov & Herzog 1968). In attempting to understand how contemporary, synchronic variation was a reflex of long-term trends in the New York City speech community, he showed, using earlier observations as benchmarks, that the speech of older people tended to reflect the state of the language when they had acquired it. Such an apparent time interpretation of systematic age differences depended on individuals remaining fairly stable after initial language acquisition, a hypothesis that still, justifiably, serves as the default. However, such an interpretation will clearly require some modification insofar as speakers alter their linguistic systems over the course of their lives.

As Labov (1994) points out, in synchronic studies that plot the distribution of a particular sociolinguistic variable against speaker age, there are four possibilities, as

displayed in Table 1. If a study in, say, 1970, shows no age differentiation among speakers (the "flat" pattern), it is possible that there is no change going on, that is, that both the individual speakers and the community as a whole are stable with respect to that feature (Interpretation #1). But another interpretation of a flat pattern is also possible: perhaps all the speakers in the community are changing together. In such a case, no age differences will appear because older and younger speakers are all at the same stage in a change that is affecting all of them equally (Interpretation #4). The other possible pattern of interest is a monotonic slope with age, and once again there are two possible interpretations. One is that the community remains stable over time, but that, generation after generation, individuals change in that particular feature as they get older. This interpretation (#2, referred to as "age grading") means that as individual cohorts of speakers get older, it is typical for them to show a steadily increasing (or decreasing) use of one value or variant of the variable. Alternately, it is possible that individuals retain their childhood patterns but that each individual age cohort of speakers coming into the speech community is increasing (or decreasing) their use of the variant. This is the classic "apparent time" interpretation (#3), according to which a monotonic slope according to age, measured at one point in time, is taken as a reflection of instability in the community, or change in progress.

Table 1. Patterns of change in the individual and the community [adapted from Labov 1994:83].

Trend and panel studies can help to disentangle "age grading" vs. "generational change" interpretations in two major ways. First, a comparable re-study carried out some years later will certainly disambiguate the two interpretations for the particular community in which such a study is done. As for longitudinal or panel studies, insofar as such research does identify change in individuals, there are still two interpretations: either the individuals are changing against a backdrop of relative stability in the community, or the individual change is part of a trend within the community that is changing as a whole. Given that a trend study can resolve the question, one might well reject altogether the idea of doing panel studies. However, panel studies, can be useful in resolving questions of whether and to what extent individuals, either as a cohort or individually, are changing particular linguistic features and patterning across their lifetimes. As such, they can help us trace the limits of the possible and determine what is expectable and normal in terms of the likelihood that apparent time is an appropriate interpretation. Similarly, only panel studies can inform us about the appropriateness of the critical age as it relates to a range of aspects of linguistic structure.

Although relatively few studies have so far been directed to this issue, it seems likely that different subsystems of language behave differently across individual lifespans, with phonology being the most stable component. Insofar as the studies carried out permit me to do so, in reviewing the research on panel studies I will pay particular attention to the relative stability of the different subsystems of language. The four situations described in Table 1 as a convenient rubric in organizing the relations between individual speakers and speech communities.

1. Stability.

Stable sociolinguistic variables have traditionally been characterized as those that show no age differential in the way they pattern in the community. To my knowledge, no re-study of any community has specifically examined the stability over time of any sociolinguistic variables that appear to be stable on the basis of extrapolation from synchronic data. However, studies of English (t.d) deletion and (ing) have shown that children as young as 5 reproduce the probabilities of the stylistic constraints found in the grammars of their individual parents (Labov 1989, Roberts 1993). Such evidence, though it points in the direction of stable variation for individual speakers over time, has yet to be verified with either panel or trend studies. Fischer’s (1958) original study of (ing) located an important stylistic component in the variation, in that learnèd vocabulary was more likely to be associated with the (ing) variant. It would seem that, in comparing sociolinguistic indicators (showing inter-individual or social variation but not stylistic variation) with markers (showing both social and stylistic variation) that those variants within speakers’ ability to vary according to style might well be more likely to be varied also across speakers’ lifetimes, however, no research has so far been directed to this question.

Both (t/d) and (ing) involve phonological variation with morphological constraints. Within the realm of phonology per se, several panel studies have shown great individual stability in vowel systems. In Copenhagen, Brink & Lund (1975) examined recordings of several Danish speakers over many decades, in one case with a 50 year interval, and found all speakers’ phonologies to be extremely stable. Labov & Auger (1998) traced a group of middle-aged Philadelphia speakers in a real-time study that showed no changes in their vowel systems over a 17-year time span. Prince (1987), tracking one Yiddish speaker (the singer Sarah Gorby) across many decades of her life, found that vowels in close-class lexical items were resistant to the phonological influences from the prestige Yiddish dialect that Gorby had attempted to emulate in open-class words.

As far as morphology is concerned, two studies based on real-time research in Montreal French have reported stability. As described in more detail in 3.2 below, data from 1971 and from 1984 has been used to assess real-time change, in several cases by using the "date of recording" — 1971 or 1984 — as one of the factor groups in a Varbrul multivariate analysis. Of the two M.A. theses specifically devoted to the topic, neither located real-time effects. Daveluy’s analysis (1987) of the variant forms of the feminine demonstrative cette ([set], [ste], and [se] before consonant) showed no change between 1971 (60 speakers studied) and 1984 (the same 60 speakers and 12 younger speakers studied). Lessard (1989), using the same sample as Daveluy, investigated the variable presence of the preposition de in infinitival constructions, and once again found no change. Such results are not unexpected, since the default assumption in examining any speech community is that at any given period, most aspects of its language will not be involved in ongoing change. In a study of AAVE negation, however, Baugh (1996) found only one of the four speakers he followed over time to be stable: the three others reduced their usage of nonstandard variants as they grew older.

 

2. Age Grading: individuals change while the community remains stable.

Lexicon may well be the least stable linguistic system in terms of change over individual lifetimes. D. Sankoff & Lessard (1975) show a steady incrementation of lexicon over successive age cohorts for the 120 speakers in the 1971 Montreal French corpus, an effect that was independent of social class and educational achievement. Their interpretation of this synchronic data was that individual speakers expand their vocabularies throughout life. Also on the basis of synchronic data, Labov (1966) suggests that in the case of the stigmatized pronunciations of (dh) and (th) as stops in New York City, the 16-year old spike in the distributions is evidence of age grading. His interpretation is that one would expect speakers to use more of a stigmatized feature in their youth, subsequently pulling back from this high level of usage to some degree. Sankoff (n.d.) offers an "age grading" interpretation of a similar pattern in Macaulay’s data on glottal stop in Glasgow (1977). Macaulay’s data were drawn from an elegantly constructed, balanced sample in which children 10 and 15 years old from four social class backgrounds were selected from Glasgow schools representing upper class, upper middle class, lower middle class and working class populations. Each was represented by 12 speakers (2 male and 2 female speakers in each age bracket). Whereas the 10 year olds as a whole use glottal for /t/ at a rate of more than 65%, the upper class 15 year olds withdraw considerably from this pattern, as do the upper middle class girls. Among adults, both upper and upper middle class groups are reduced to less than 60%, whereas lower middle and lower class adults maintain a usage of 80% and higher. My interpretation of these data is that the middle and upper class adolescents withdraw from glottal stop as they get ready to enter the labor force, where the evaluation of this feature on the linguistic market is that the standard pronunciation is more appropriate to their position in society. Such a statement can of course only be made with caution: it would require a panel or trend study to establish whether this interpretation is appropriate. However, the fact that the Glasgow data has received the same, age-grading interpretation from Chambers (1995), also receives support from another source. Collins and Mees (1996) confirm that glottal stop for /t/ was first cited in Scots English and, based on a study of early audio recordings that include speakers born as early as 1848 (the actress Ellen Terry) and 1888 (broadcaster Freddy Greenwood), date the appearance of glottal stop in British RP as considerably earlier than has generally been appreciated.

Though the "age-grading" interpretation of the synchronic data discussed above has not been subject to longitudinal verification, longitudinal research on other aspects of Montreal French has revealed that age grading is the appropriate interpretation of some of the results. Based on the 1971 data, Sankoff & Laberge (1980) had identified the replacement of generic on by generic tu as a change from below, the second stage in a morphological chain-shift in which the first stage was the shift to on as a replacement for nous ‘we’. Thibault (1991) confirmed this analysis, using a sample of 25 of the 1984 speakers who showed dramatic change over the 13-year interval. Dubois studied "extension particles", i.e. the discourse particles and short phrases "serving to extrapolate from what has previously been said" (1992:179-80) in the 1971 and 1984 data. Studying the same 60 speakers who were interviewed in both 1971 and 1984, as well as the 12 younger speakers added in 1984, she interpreted age effects as an age-grading phenomenon, rather than as evidence of real time change. She reports that "the youngest speakers used more extension particles, but this tendency is lost after the teen years" (p.199).

Another longitudinal study of Montreal French discourse markers (Thibault and Daveluy 1989) proposes a combination of real-time change with age grading. This paper will be reviewed in 3.2 below, along with several other studies in which the re-examination of a community seems to show a combination of change among individuals and change within the speech community as a whole.

In sociolinguistic research in Brazil, Callou et al (1998) report on a longitudinal study of the weakening of syllable- and word-final (r) in Carioca Portuguese. A balanced sample of 66 male and female subjects recorded in the early 1970s, was also stratified into three age groups: speakers from 25-35; those 36-55, and those 56 and older (note that all were middle class speakers because of the project’s goal of characterizing the standard spoken language or norma culta). Five male and five female speakers were re-recorded in the 1990s, along with a new sample of younger speakers (9 male and 9 female). The original study had revealed that younger speakers deleted final (r) at a rate much higher than the middle-aged and older speakers. However, not only had the 10 re-interviewed speakers reduced their frequency of (r)-deletion as they got older, but the new sample of young people almost perfectly matched their peers of a generation earlier. This demonstration of the parallel results of longitudinal and cross sectional research argues convincingly for age grading as the appropriate interpretation in this case.

3. Generational change.

Attempting to review the extensive literature on corpus-based studies of language change within historical linguistics within historical linguistics would be well outside the scope of the current discussion. However, it is important to note that an increasing number of studies in this field have been explicitly related to the theoretical concerns and the methodology of quantitative sociolinguistics

The corpus-based historical work of Kroch and his associates has pioneered the use of quantitative methodology and quantitative reasoning in studying the real-time diffusion of linguistic innovations according to a model that has come to be known as the "competing grammars" approach. As in synchronic sociolinguistic studies that use the variable rule methodology, a major part of such work is in defining the envelope of variation within which competition occurs. The first major study in this tradition (Kroch, Myhill and Pintzuk 1982) traced the evolution of do-support in English beginning in the 15^th century. Kroch (1989) enunciated the concept of the constant rate hypothesis, a logical extension of the independence of constraints in the variable rule model to the analysis of language change. There followed, among others, Santorini (1993) on phrase structure change in the history of Yiddish; Taylor (1994) on word order change in Ancient Greek; Pintzuk (1995) on Old English clause structure, and Kroch and Taylor (1997) on verb movement in Old and Middle English. Based on a sequence of historical corpora, all of this work falls squarely within the rubric of cross-sectional, or trend studies.

Other work that takes up issues raised by sociolinguists includes Raumolin-Brunberg’s (1996) research on the history of English. This paper addresses Labov’s suggestion (1994:84) that morphological change is typically generational. Raumolin-Brunberg’s trend study is based on the correspondence of younger and older generations in two English wool merchant families: the Cely family between 1472 and 1488, and the Johnson family between 1542 and 1553. She reports on four morphological variables: the replacement of older subject ye with you; relative the which replaced by which; the introduction of subject relative who (a change from above); and the replacement of third singular —th with —s. Of these, all but the last appear to be generational in character, in keeping with Labov’s characterization of the situation. Based on the fourth case, Raumolin-Brunberg states that "some morphological changes may be diffused in a communal manner . . . the variation between —th and —s around 1600" (p.94) being such a case.

In another important longitudinal study, Arnaud (1998) used a combined trend and panel design to track the rise of the progressive in English in the 19th century via the informal letters written by a sample of male and female writers, including the most well known authors of the day. Arnaud’s 'density' measure is calculated as the number of progressives per 100,000 words of text. As with some of the studies we will review in 3.2 below, Arnaud showed both types of change going on in tandem. For the period between 1830 and 1845, he found that authors born prior to 1780 used the progressive with a mean density of 170; whereas those born after 1790 used it with a mean density of 241. Further, for two of the three authors whose usage he tracked across the decades of their writing, Charles Dickens increased from 270 in the 1830s to 540 in the 1870s, and Elizabeth Gaskell increased from 210 in the 1820s to 495 in the 1870s (Thackeray remained stable at c. 500 across five decades). Thus, along with higher rates among those born later in the century, there a tendency among some authors towards a gradual increase in progressive usage across their lifespans.

An historical corpus of a different sort is the corpus of recordings made in the 1940s of 250 New Zealanders born in the 19^th century, some as far back as the 1850s. Gordon and Trudgill (1999) have traced the "embryonic" presence of some well-known New Zealand phonological variants (notably the centralized, stressed short /i/ as in [p´g] ‘pig’), raising the question of "why some early variants should later develop into present-day features of New Zealand English and others disappear completely" (p.111).

Studies that have suggested an apparent time interpretation of monotonic age distributions in synchronic studies are legion in sociolinguistics, but although relatively few trend studies have so far been conducted with the goal of verifying these interpretations, sociolinguistic work of the late 1960s and early 1970s has been followed up in a growing number of cases. In addition to the follow-up by Callou et al (1998) on early work in Rio de Janeiro and cited in 2. above, these include Cedergren’s (1984; 1987) restudy of Panama City; Trudgill’s (1988) restudy of Norwich; Fowler’s restudy of (r) in New York City; Hansen’s (2001) trend study of Parisian nasal vowels, based on recordings from 1972-74 and 1989-93; and the cross-sectional studies of Texas dialect features (Bailey et al 1991; Bailey 2001). Paunonen (1996) reports on a restudy of the 1971 Helsinki survey that was carried out 20 years later. Lastly, a restudy of the original 1971 research on Montreal French (Sankoff & Sankoff 1973) was carried out in 1984 (Thibault and Vincent 1990) with another follow-up in 1995 (Vincent, Laforest & Martel 1995).

Labov (1994) summarizes the restudy by Hermann (1929) of the original research on the Suisse Romande village of Charmey carried out by Gauchat (1905) between 1899 and 1905. The three vowel changes identified by Gauchat on the basis of apparent time calculations were all advanced, but of the two consonantal changes Gauchat had postulated, "the aspiration of [Q] had not advanced, but showed the same type of variation that Gauchat had found in 1904" (Labov 1994:85). Labov concludes that "Gauchat had indeed succeeded in locating linguistic change in progress; but . . . real-time information was needed to resolve the ambiguity of data drawn from apparent time" (ibid.)

The second study reported on by Labov is the replication by Fowler (in 1986) of his own "rapid and anonymous" study (r)-pronunciation in three New York City department stores, carried out in 1962. The constricted pronunciation of post-vocalic (r) is a change from above, and Fowler’s trend study (an exact replication of Labov’s methods) allowed her to uncover "clear evidence of generational change in progress in real time" (Labov 1994:90). But as Labov goes on to remark, this change had advanced only a small amount between 1962 and 1986, and " is quite slow compared with the evolution of the New York City vowel system or the palatalization of /l/ that Gauchat observed in Charmey’ (ibid, pp.90-91). Hermann’s restudy was not designed to examine age grading, but in the case of NYC (r), Fowler’s restudy discovered that age grading accompanied the generational change,

In the early 1970s, Helsinki was the site of one of the first urban sociolinguistic studies, with 96 subjects recorded according to a carefully constructed sample. Half of the speakers were male, half female; the sample was stratified according to upper middle, lower middle and working classes; and three age groups were represented: elderly people born just after 1900; middle-aged people born in the 1920s, and young people born in the 1950s. In the restudy carried out in the early 1990s (Paunonen 1996), 29 of the original speakers were re-interviewed: 15 of the original 32 born in the 1920s, and 14 of the original 32 born in the 1950s. To these were added 8 male and 8 female speakers born in the 1970s. Paunonen’s report on a morphological change in progress in Finnish thus offers both panel and trend data, and allows for comparison with Raumolin-Brunberg’s historical study of morphological change reported in 3.1 above.

The change involves two types of possessive constructions in the first person singular. In the first, "reflexive", the subject of the sentence is co-referential with the possessor, e.g. ‘I found my book’. In the second, "non-reflexive" type, the subject of the sentence is not co-referential with the possessor, e.g. ‘Mother found my book’. There is variation in the use of the independent pronoun, which precedes the possessed object and may be either the standard minun or the non-standard mun. In addition, speakers vary in their use of the suffix —ni. These two dimensions of variation thus allow for the "synthetic" kirja-ni (‘book’-my’), the "double-marked minun/mun kirja-ni" (‘my’-‘book’-my’), and the "analytic" minun/mun kirja (‘my’-‘book’). Results revealed real-time generational change for both reflexive and non-reflexive possessives. Speakers born around 1900 used the synthetic form almost categorically in reflexives, and about half of the time in non-reflexives; whereas those born in the 1970s used the analytic form accompanied by the non-standard independent pronoun mun 60% of the time in reflexives and almost categorically in non-reflexives. Intermediate generations showed values between these extremes. By and large, speakers retained their original patterns throughout their lives, but there was a small age-grading component among speakers who were middle-aged in the 1970s (those born in the 1920s), and who were restudied. Four of the 5 women and 3 of the 4 men who showed no use of the popular incoming form (mun kirja) in the 1970s registered some degree of its use (between 10% and 50%) in the 1990s. In keeping with previous findings in morphology, the general trend is one of generational change, here with a small additional component of change over individual lifespans for some speakers.

3. Paris.

Hansen (2001) bases her cross-sectional study of change in French nasal vowels on 42 speakers, 16 taken from a corpus recorded in 1972-74, and 26 from her own 1989-93 corpus. In each sample, speakers were divided into older and younger, such that she has four mean dates of birth represented: 1916, 1943, 1952, and 1973. Two educational groups were selected: "well-educated informants with the baccalauréat. . . . and informants with technical/practical backgrounds, without the baccalauréat (p.213). Her study revealed a backing chain shift among the three oral nasal vowels: /E$/ à [E$~A$], /A$/ à [A$~O$]/[O$}, and /O$/ à {o$}. The change she focuses on is the one involving /A$/, represented in Table 2 as a decline in frequency of the older [A$] variant. Whereas older

	Educational groups	Mean date of birth for four age groups
		1916	1943	1952	1973
	Technical	58%	65%	55%	42%
	Baccalauréat	70%	53%	57%	55%

Table 2. Declining mean percentage of [A$] variant for Paris speakers

by educational group and mean date of birth. [Note: Percentages

extrapolated from Hansen’s Fig’s 3 and 4, pp. 217-218].

speakers with a technical background registered a lower percentage of the older form than more highly-educated speakers, they remained fairly stable until the youngest generation, who registered a significant decline. Speakers with the baccalauréat, on the other hand, registered a big decline between the 1916 and 1943 generations, and have remained fairly stable since that time. (Not represented in Table 2 is the fact that both the 1952 and 1973 age groups from the "technical" category use close to 10% of the most advanced form, {o$}, the which has not yet made any inroads among the more highly educated speakers). Hansen’s article is largely devoted to exploring lexical frequency and grammatical category effects, since in addition to phonological conditioning (preceding labials favor backing and rounding), there is evidence for lexical diffusion in her data. There is no panel component to her study, but it is clear from the cross-sectional comparison that there is real-time evidence for the chain shift in progress in the nasal vowels.

4. Texas.

Bailey et al (1991) report on a comparison of a number of Texas dialect features in two Texas surveys completed in 1989 with dialect survey work carried out 15-20 years earlier. Features examined include the /O/ ~ /A/ merger in three phonological contexts; three mergers before /l/: /i/ ~ /i/, /e/ ~ /E/, and /u/ ~ /u/; the fronting of /au/; the monophthongization of /ai/; unconstricted postvocalic /r/; the loss of /j/ after alveolars as in Tuesday; the loss of /h/ before /j/ as in Houston; intrusive /r/ in Washington; and two tense/aspect features: the use of the quasi-modal fixin to, and the use of double modals as in might could. In all except three of these, real-time trend comparisons confirmed the apparent time interpretation of the earlier data (Bailey 2001:315). In the remaining features, real-time data also confirms the original apparent time inferences in the case of the fronting of /au/: variation is stable and ethnically conditioned, with White Anglo speakers generally fronting and Hispanic and African American using the un-fronted variant (Bailey 2001, pp.317-318). Bailey notes that the distributions for /ai/-monophthongization and for the use of double modals (might could), discrepancies may be due to methodological differences between the studies, however, "when only native Texans are considered, . . . the data do in fact suggest a change in progress (i.e. monophthongal /ai/ is expanding in words like night); the apparent-time data, then, shows exactly what real-time data lead us to expect." (Bailey 2001:319). Overall, then, the results of the cross sectional research by Bailey and his collaborators indicate generational change for these phonological and morphological changes in Texas.

5. Panama City.

Two major restudies of well-known earlier sociolinguistic research are Cedergren’s restudy of Panama City, and Trudgill’s restudy of Norwich. Cedergren (1984;1987) studies the progress of CH-lenition on the basis of return trips in 1982 and 1984, a trend study comparable to her original research in 1969. Cedergren reports her own original interpretations as based on the 1969 study as follows:

Results on CH-lenition in the 1982-84 restudy are compared with those from 1969 in Figure 3. Cedergren interprets these results as a confirmation of her original "apparent time" inferences, and adds a remark about the behavior of the young adults speakers who in both studies show a "spike" in comparison to adjacent age groups:

(Cedergren 1987:53)

As the speakers with a mean age of 20 in 1969 retreat to a lower level of CH-lenition in 1982-84 when their mean age was 34, we can also see the effect of age grading on this variable, with a "young adult spike" reminiscent of the "adolescent spike" Labov discussed in both NYC and Philadelphia. Thus the restudy of CH-lenition in Panama gives evidence of a combination of real-time, generational change, as well as age grading. Labov’s conclusions from these results are worth citing in detail:

(Labov 1994:97)

The most comprehensive trend study to date is Trudgill’s follow-up study of Norwich. Carried out in 1983, it kept faithfully to the methodology of the original research in, for example, including the same reading and word list passages that were originally used. In his sample, however, Trudgill did not replicate the entire design of his 1968 research, which included 60 speakers born between 1875 and 1958. Instead, he added the next generation: 17 younger speakers born between 1958 and 1963. This new generation manifested a number of new features not at all (or only marginally) present in the 1968 study. Trudgill (1988) reports on seven variables, providing a complex picture of how different linguistic features have behaved differently in the 15-year interval. Some of these were features Trudgill had not attended to in the original study; others he had previously studied in some detail. Table 3 reproduces his own summary of the results.

Noting that some of the changes in progress in 1968 concerned local dialect features that were already retreating, Trudgill explains that the lengthened monophthong in the lexical set of gate and face, was "vestigial in 1968". In 1983, he found that /e:/ was entirely absent from the speech of the speakers under age 25, although it could "still be heard in the speech of older people" (1988:40). Similarly, the beer/bear merger, identified in 1968 as an ongoing change from below, was complete except for Upper Middle Class speakers. The third feature concerns the collapse of a distinction (between the vowels of the moan vs. mown word classes) specific to East Anglia. In the 1968 study the merger was "confined to a small number of middle-class speakers" (p.40), and by 1983, it was "beginning to expand, and a number of speakers from other social class groups [were] now beginning variably to adopt this feature" (ibid.). The fourth feature, /r/ labialization, was barely present in the earlier data, but showed a dramatic increase among the young people of the 1983 study, with between 30% and 35% of speakers born after 1948 categorically using the innovative form. Importantly, Trudgill notes:

The fifth feature concerns the merger of /Q/ and /f/ in all positions, and of non-word-initial /D/ with /v/. "[N]ot a single speaker in the 1968 sample showed even one instance of this phenomenon, . . . [but] of people born between 1959 and 1973, . . . 41% have the merger variably; and 20% have a total merger, i.e. /Q/ has been totally lost from their consonantal inventories" (p.43). Like /r/-labialization, this dramatic change seems to be a generational one which has not affected adults.

The last two features Trudgill reports on, the replacement of [t] by glottal stop and the backing of /E/ before /l/ in words like belt, were both well advanced in 1968 but showed considerable stylistic stratification. The major change in 1983 was the rise in usage of the innovative variants even in formal styles by the younger speakers. Since the adults of the 1969 study were not re-studied in 1983, Trudgill is not certain (he in fact expresses some doubt) about whether or not there has been a rise in usage in the innovative forms over adult speaker lifetimes. Overall, his results on these seven phonological processes are a testimony to the wealth of possibilities in ongoing linguistic change, and to the necessity of careful study to disentangle the varied sources and dimensions of change.

The design of the original Montreal French study carried out by Henrietta Cedergren, David Sankoff and myself in 1971 was to record native Francophone Montrealers in a random sample of 120 speakers stratified by social class (operationalized by selecting from neighborhoods representing 6 mean income levels), sex and age (Sankoff & Sankoff 1973). Speakers’ birth dates ranged from 1885 to 1956. The 1984 follow-up study (Thibault & Vincent 1990) recontacted 60 of the original speakers and added 14 younger speakers, born between 1959 and 1969. The 1995 follow-up, designed to tap greater stylistic variation, followed only 12 of the original speakers but recorded them in many more contexts, and also followed 2 of the younger speakers added in 1984. Thus, this three-stage study covering a span of 24 years combines both panel and trend components.

In addition to the studies cited earlier in this review, in which stability or age grading was found (D. Sankoff & Lessard 1975, Daveluy 1987, Lessard 1989, Thibault 1991, Dubois 1992, Labov & Auger 1998), a number of studies have documented generational change in progress. The "apparent time" interpretation given to the age distribution of posterior (vs. alveolar) (r) by Clermont & Cedergren (1979) was confirmed by Cedergren (1987) via an examination of data from an earlier study: the 29 speakers of the Bibeau-Dugas 1963 Montreal corpus. Those born after 1930 used a preponderance of posterior (velar or uvular) [R], while those born before 1925 were virtually categorical users of alveolar [r]. Clermont & Cedergren studied. In the 1971 corpus, Clermont & Cedergren studied 57 speakers born between 1937 and 1956, and found that over half — 33 — were already categorical users of [R]; 23 were variable. This trend comparison between the 1963 and 1971 data confirmed the existence of change in progress at the community level. Sankoff, Blondeau and Charity (2001) conducted a panel study to verify whether individual speakers also changed their patterns across their lifetimes. Coding 100 tokens of (r) per speaker per year recorded, we followed 25 speakers, as shown in Table 4, which groups the speakers according to their usage. The 8 Group I speakers were virtually categorical uses of the innovative form, [R], in 1971, and were stable thereafter. Similarly, there were 8 speakers who showed stability in maintaining the older [r] variant that they acquired in childhood. The remaining third of our sample, consisting of 9 speakers, were more variable in 1971, and as a group, they increased their usage considerably between 1971 and 1984, leveling off thereafter. Of the 9, 7 individually showed statistically significant increases (<.05) in the 1971-1984 period. This change has clearly operated at the level of the community and over individual lifetimes, even in later life. Of the 7 speakers who made significant changes after 1971, one was a teenager, 3 were in their twenties, 2 were in their forties and one was in his early fifties!

Table 4. Mean [R]/([R]+[r]) for Montreal speakers, 1971 — 1995.

the 1971 total is 23 speakers. In 1995 only 14 speakers were re-recorded.

Whereas (r) is the only consonantal feature of Montreal French to have been studied longitudinally, several cross-sectional and longitudinal studies have appeared on vowels. Kemp & Yaeger-Dror (1991) is a detailed cross-sectional study of the pronunciation of (a) in the suffix -ation. Québécois French has maintained a clear distinction between front and back (a), for example in a minimal pair like patte [pat] and pâte [pAt, pOt]. However during the course of the 1940s — 1960s, the formerly back [A] of the -ation suffix underwent a change to the front category. Kemp and Yaeger-Dror document this by analyzing a corpus of early recordings of 34 speakers born between 1860 and 1912 (all but one recorded between 1927 and 1960, but mainly in the 1940s and 1950s), and comparing them with 52 speakers of the Montreal 1971 corpus divided into three age ranges: older speakers born between 1886 and 1912, middle-aged speakers born between 1918 and 1932, and younger speakers born between 1946 and 1955. This paper, exceptional in its penetrating analysis of the social changes that accompanied this change, clearly revealed this to be a generational change (from above). It is traced from middle-class women to middle class men and working class women, with working class men as the last holdouts. However, the change was virtually complete by 1960, with even for working class men born after the Second World War showing almost no instances of back [A] in -ation. They are able to include one working class male speaker, J.P., born in 1918 and recorded for a radio broadcast in 1951 when, "as the only francophone representative in the union hierarchy" (p.165), he was chosen to discuss railroad union negotiations. At that time, 19/20 of his -ation tokens were back [A], but at the time Kemp talked with him (at age 64) in 1982, "there was no evidence of back A realizations in the five items . . . that occurred during the conversation." (p.165). Despite the evidence of age grading or lifetime change for this speaker, they believe that he was an exceptional case and that, since back A was the former prestige norm, few speakers would have had any motivation to change their pronunciations during their lifetimes (indeed most of the speakers in the radio archive are public figures — highly educated and with upper class backgrounds.

Yaeger-Dror and Kemp (1992), and Yaeger-Dror (1994, 1996) also studied the three mid-low lengthened vowels, subject to raising, lengthening and diphthongization: [E:} as in père ‘father’; [O] as in porc ‘pig’; and [¼] as in peur ‘fear’. Their work was based on the analysis of speakers from both the 1971 and 1984 corpora, as well as from a sample of older speakers recorded in 1978. In all they analyzed 76 speakers for research reported in these papers, comparing three generations: speakers born prior to 1900; those born between 1900 and 1920, and those born after 1944. The overall conclusion of this work is that vowel lowering has occurred as a change from below , led by working class women. The question of the relative contribution of generational change vs. age grading is not explicitly raised in these papers. In her most recent summary, Yaeger-Dror states that their results "confirm that both etymological and exceptional lexical diffusion are occurring for these vowels along with the real-time change as shown by the significance of the year variable for most of the speakers" (1996:276). If ‘year of recording’ is significant as a factor in vowel lowering for individual speakers in a multivariate analysis, this would lead us to believe that speakers are changing over their individual lifetimes, in addition to the major trends established in the community according to speaker generations established by year of birth. Vowel diphthongization was also studied by Cedergren et al (1981) and by Santerre & Millo (1978) on the basis of the 1971 data. Cedergren et al reported that their apparent time interpretation indicated a change in progress toward increasing diphthongization of /¼:/ before /r/ (one of the vowels studied by Yaeger) and also of /o:/ as in chose ‘thing’ and gauche ‘left’. Santerre and Millo, however, found that the social class factor was predominant, while nonetheless accepting the existence of some degree of age differentiation. Sankoff (2002) examined the longitudinal data for 14 speakers, 1971 — 1995, for the same two vowels. Small token numbers (20 tokens per speaker per time period) precluded calculating significance individually, however as a group the 14 speakers showed a 32% to 43% increase in [o:] diphthongization, significant at the .05 level, and an increase in [¼:r] diphthongization from 41% to 62%, significant at the .001 level. Though these results, based on aural coding, are tentative pending the incorporation of the trend component to this study, and also pending acoustic measurement (ongoing at the time of writing), I believe they indicate some shift over individual speaker lifetimes in addition to generational change in the community.

Other real-time studies of Montreal French have focused on morphology and on discourse markers. Thibault & Daveluy (1989) report on five discourse markers and one lexical/syntactic variable: the expression of restrictives ‘only’ as seulement, rien que, and juste. Using all the data available from both 1971 (120 speakers) and 1984 (the 60 re-recorded speakers plus the 12 new, younger speakers), their multiple regression analyses were run on "the complete data sets from 1971 and 1984 (a trend study) and the combined data repeated for the 60 reinterviewed individuals (a panel study)" (p.44). Thus disambiguating trend from panel effects, they found different patterning for the different discourse markers, two of which I will review here. Alors ‘then’ (a term also associated with higher educational/ occupational status) was clearly shown to be age graded, with "speakers of bourgeois origin" in particular seen to be increasing their use of this marker as they got older. . As for tu sais ‘you know’ (a term associated with working class speakers), Thibault & Daveluy identified "a drastic increase in the number of speakers who use it exclusively, . . . much greater than the shift of speakers towards the familiar form tu from vous in other contexts" (p.144). This is clearly a trend effect. Two other results of this study, involving change across the entire community, will be reviewed in 4.0 below.

With respect to the pronominal system, the alternation between simple plural personal pronouns and their compound analogues with —autres, Blondeau (1999, 2001) identified a real-time change in which the simple forms were replacing the compound forms between 1971 and 1995. The two variants are illustrated in (1) and (2), taken from adjacent sentences in the interview of Christian B., (Speaker #25, 1995). In both excerpts, he contrasts another work place with his own.

(1) C'est pas tout' organisé comme nous-autres là.

‘It’s not all organized like we [are, in our office].’

Whereas simple forms constituted only 8% of plural personal pronouns in 1971, this had risen to 22% in 1984 and to 29% in 1995.

3. Communal change: what happens if young and old together adopt an innovation?

Labov’s characterization of this fourth possibility is as follows: "all members of the community alter their frequencies together, or acquire new forms simultaneously" (Labov 1994: 84). He further notes, "This is a common pattern of lexical change, as Payne (1976) found in her study of speakers entering the Philadelphia community." He also notes Arnaud’s analysis of the English progressive, cited above, and Sankoff and Brown (1976) on changes in relative clause syntax in Tok Pisin (a result for which, however we were not able to give quantitative evidence on individual speakers due to small sample size). A further result on Tok Pisin came from a re-examination by Labov and myself of the results originally published in Sankoff & Laberge (1973) on the de-stressing of the future marker bai in Tok Pisin. Laberge and I had originally shown that child, native speakers of Tok Pisin were carrying the de-stressing of bai further than had their parents’ generation. When Labov and I connected the dots (connecting children with their own parents) in the original diagram, we discovered that the correlation between children and parents (r=.61) was significant at the .02 level. Labov summarizes that the children’s system "is a regular projection from the language of their parents" (Labov 2001:425). A final case we can add to those in which the whole community appears to have progressed together in the adoption of an innovation is the introduction of the concatenated discourse marker tu-sais-veux-dire ("y’know I wanna say") between 1971 and 1984. Finding "no apparent time effect in either 1971 or 1984", they document "the rapid spread of this form across all sections of the population" (Thibault & Daveluy 1989:45).

4. Concluding Remarks.

There are many challenges to our understanding of the relationship between language as a mental construct in the minds of individual speakers, and language as a creation of speech communities. Cross-sectional and longitudinal studies of the future may do more to clarify these relationships, as they have to bridge the gap between synchronic and diachronic linguistics, and between real and apparent time. I believe they will also elucidate some of the aspects of the modularity of language, as we have already seen that different levels of linguistic structure are differentially susceptible to modification in later life. In reviewing the literature to date, it is clear that phonology, even though stable in most of its features across individual life spans, is nonetheless available to some speakers for some amount of modification.

Given the amount of variability we have encountered, it is probably too early to make valid generalizations. Consider, for example, the four panel and cross-sectional studies of (r) that have been reviewed in this paper: (1) Labov and follow-up studies of postvocalic (r) in New York City; (2) Trudgill’s cross-sectional studies of labialized (r) in Norwich; (3) Sankoff et al’s panel study (2001a) of the introduction of "posterior" [R] in Montreal French; and (4) Callou et al’s study of (r) weakening in Carioca Portuguese. Table 5 summarizes some of the contrasting results: in Carioca Portuguese, we find age grading alone, with no community change, whereas in Norwich we find speakers categorically using either non-labialized or labialized (r), with younger speakers adopting the change and no older speakers being influenced. Montreal and New York City each show a mixture of real-time change and age-grading. The Montreal study (the only panel study in the group) indicated that perhaps one-third of speakers who could potentially alter their (r)-pronunciation in later life in the direction of the change, actually do so. Such a mixed result is likely when other linguistic subsystems are examined in greater detail.

Table 5. Patterns of change and variation in four cross-sectional and longitudinal

We saw Labov’s generalization (1994:85) holding up quite well: lexical and discourse-level features, as well as the more lexical aspects of morphology (e.g. the French personal pronouns) seemed to be more susceptible to change over individual lifetimes than phonological features. However, to my knowledge, only one study has attempted to test such a generalization by looking at the same speakers. Blondeau et al (in press) tested three hypotheses about the relationship between phonological and morphological change across the lifespan by looking at two changes from above in Montreal French: the adoption of posterior [R] and the use of simple plural pronouns. Our hypotheses were (1) that a speaker conservative in phonology is indeterminate in terms of his/her likelihood of accepting a morphological innovation; but that (2) a phonologically variable speaker with a preponderance of innovative usage will be likely to accept the morphological innovation, and (3) that a speaker who accepts a phonological innovation will accept a morphological innovation. These hypotheses, which proved to be true in the case of these two changes from above, also appear to hold in the case of two changes from below that we are currently working on: the adoption of generic tu and vowel diphthongization (Sankoff et al 2001b; Sankoff 2002), but further work will be needed to see if they hold over a larger set of individuals.

I believe that the studies reviewed here make clear the importance of distinguishing two types of patterns that we now call "age grading" in future longitudinal and cross-sectional studies. I propose that we reserve the term "age grading" for those situations in which groups of speakers in the same community, generation after generation, find it appropriate to employ a particular pattern (say, a non-standard usage among teenagers [e.g. the Glasgow glottal stop], or higher values of a prestige variant among young adults as they enter the labor market [CH-lenition in Panama]. Such patterns may often, but not uniquely, be associated with stable sociolinguistic variables (cf. Eckert 1997). For the other type of change, in which individual speakers change over their lifespans in the direction of a change in progress in the rest of the community, I propose we dub these cases of ‘lifespan change’. From the individual perspective, the experience of living in a community where people younger than oneself are surging forward with a change that makes our own speech sound somewhat dated or old fashioned may be similar to that of adult speakers who experience dialect contact post adolescence. Trudgill’s (1983) study of the Beatles’ and the Rolling Stones’ use of post-vocalic constricted (r) and of fronted short A over the period 1963 — 1999, and my own longitudinal study of vocalic change on the part of two British speakers who moved away from their Northern dialect homes at age 16 (Sankoff n.d.) serve to illustrate the kinds of influences such speakers may be subject to, and the kinds of mixed responses they may have, even when they stay home.