The Centrality of Time in SLA

Why is longitudinal research essential to advancement of knowledge in the field of SLA? The simple but uncontestable answer is that many questions concerning second language learning are fundamentally questions of time and timing. A cursory look at some central SLA research areas can easily show this.

For one, fundamental questions related to time concern biological time at the onset of learning, including not only “at what age” but also “for how long” L2 learning should occur, in contexts where near-native success is the desired outcome. These time-related questions have inspired basic research about maturational constraints for L2 learning and the critical periods hypothesis (Hyltenstam & Abrahamsson, 2003; Scovel, 1988). They have also been central in evaluating the research base for educational policy decisions regarding early or late exit from transitional bilingual education programs in the United States (Collier, 1989), and in evaluations of early versus late start of French immersion school education in Canada (Harley & Hart, 1997).

Even in SLA domains where the preponderance of studies favors cross-sectional practices, longitudinal claims are often implicitly posited. Task-based language learning proposals, for example (see Ellis, 2003; Robinson, 2001; Skehan, 2002), make theoretical inferences about competence-related changes that are posited to deploy over time, but they do so on the empirical foundation of static snapshots of learner's capacity for action in the L2 at a given point in time. The same is true of many other areas of instructed SLA research. For example, time issues are central in the empirical determination of ideal lengths and intensities of instruction in a given curriculum, once again, if the goal is to foster advanced L2 capacities: How much instructional time is optimal in a given school curriculum, overall, and how intensively packed into the institutional schedule should it be (Collins, Halter, Lightbown, & Spada, 1999)? Questions about the timing of instruction and how instructional intensity moderates L2 learning outcomes are implicated as well, if perhaps at a more abstract level, in the oft-lamented problem of articulation from high school to college in foreign language education in the United States (Lally, 2001). They also underlie characterizations of the crisis of FL collegiate education in North American universities, which many blame at least in part on the dichotomization of programs into lower-division language courses, where time is devoted to the basics of language learning alone, and upper-division content courses, where students experience discontinuity at a turning point when instructional time is suddenly devoted to the exclusive learning of literature and cultural studies (Byrnes, 2002).

Many broader SLA questions could be posed that require answers mustered in longitudinal evidence: What do we know about the pace and pattern of development in second language and literacy, throughout the lifetime of L2 learners? What critical transition points in L2 development need to be taken into account when planning and evaluating educational policy and practice for specific L2 learner populations? Ultimately, it is through cumulative longitudinal findings that the SLA research community would be able to contribute meaningful characterizations of the gradual process of attaining advanced second language and literacy competencies across various contexts. Given the centrality of time in SLA research programs, then, more attention to longitudinal research practices and to findings gleaned from longitudinal studies is desirable.

Methodology for the Present Review

Although we make no claim to exhaustion or completeness in this review, the themes we present emerged from strategies for systematic synthesis of primary research domains (Cooper & Hedges, 1994; Light & Pillemer, 1984). We conducted electronic searches to uncover longitudinal SLA studies published during 2002 and 2003 across 20 major applied linguistics journals. This was supplemented with hand searches of 2004 journal issues that were available at the time of this writing. The search yielded 38 longitudinal studies that focused on L2 learning (including second, foreign, and heritage language learning). A little fewer than half of these studies investigated some aspect of L2 learning by young adult students enrolled in institutions of higher education, whereas about 20% investigated high school or middle school contexts. Even fewer studies addressed adult learners in nonuniversity contexts or preschool or elementary-school-aged children. Broadly speaking, then, recent longitudinal research in SLA is consistent with the focus on college-level populations that is typical of applied linguistics research in general.

We located and read these studies and discussed main methodological features. We also recorded the extent to which studies defined and documented “change over time” (quantitatively, qualitatively, or both), and we evaluated the strengths and weaknesses of each study, noting any longitudinal challenges explicitly discussed in the primary reports. This process compelled us to refine and broaden our working definition of “longitudinal” SLA research, as we realized the study of second language and literacy development over time has been undertaken from a number of distinct epistemological traditions, including quantitative descriptive studies of linguistic features and qualitative interpretive studies of sociocognitive and sociocultural dimensions of multilingual development. In the remainder of this chapter, we characterize four broad trends in recent longitudinal SLA research. We conclude with a methodological discussion that we hope is useful for SLA researchers and graduate students who plan to conduct, or are in the process of reporting on, longitudinal studies.

Descriptive-Quantitative Longitudinal Studies of L2 Development

Many longitudinal SLA studies published in the last three years have adopted a descriptive (that is, nonexperimental), quantitative design. Such studies focus on quantifiable variables, but without researcher's manipulation and without random sampling or random assignment of participants to conditions or to contexts for learning. They are characterized by multiwave data collection from the same individuals over a relatively long period of time, which typically spans anywhere between four months and four years. Although the longitudinal data in these studies are quantified, the number of participants is far too small to warrant the use of inferential statistics. For this reason, descriptive statistics displayed in the form of frequencies, percentages, and proportions, and other analytical tools, such as visual displays and implicational scaling, are favored.

The research focus is often strictly linguistic, following up on the tradition of early SLA longitudinal studies that have become classics (e.g., Huebner, 1983; Sato, 1990; Schumann, 1978; but see Schmidt, 1983, for a broader linguistic focus). Although the early interlanguage work involved intense case studies of usually one or two participants only, more recent studies increasingly tend to involve multiple participants, although still typically small groups. For example, Perdue, Benazzo, and Giuliano (2002) charted the development of finiteness in the spoken production of five adult Italian and Spanish immigrant workers in Europe during 30 months of naturalistic exposure to English or French, and Bardovi-Harlig (2002) investigated the development of future tense as evinced in the written production by 16 college ESL learners during 7 to 17 months of study in a university's English Language Program (this is the same sample and longitudinal corpus previously studied in Bardovi-Harlig, 1994). Well established descriptive-quantitative longitudinal lines of SLA research to the present day concentrate on L2 morphology, as the studies just mentioned illustrate, and on L2 phonology (e.g., Abrahamsson, 2003; Hansen, 2004).

Descriptive-quantitative longitudinal SLA research draws on linguistic theories, often of a functional bent, and seldom includes nonlinguistic variables. These stable trends notwithstanding, it is interesting to note that in other recent SLA studies employing this type of design there has been a broadening not only of the linguistic focus but also of the epistemological approach to language development. This is the case of a longitudinal study conducted within a Vygotskian sociocultural framework by Belz and Kinginger (2002). The researchers documented critical incidents that contributed to the learning of indexical politeness (specifically, the use of tu/vous and Du/Sein) by two fourth-semester foreign language students at a U.S. university. The participants were involved in a telecollaboration project that afforded them the opportunity to establish a personal relationship with a native-speaking peer in France and Germany, respectively. The critical incidents were defined by the researchers a priori as instances in which a student chose the formal you during an e-mail exchange in the L2, causing the peer interlocutor to focalize overtly on the inappropriateness of this choice. The researchers argue that solidarity (for the French student) and even intimacy (for the German student) developed between cross-site partners, and that this personal investment infused the choice of term of address with special significance. In sum, the personal communicative investment enabled by the telecollaboration assignment outside of the classroom made choice of address term functionally salient for these students, whereas it also afforded them fine-tuned and timely assistance from their native-speaking peer.

The descriptive-quantitative type of design discussed here corresponds closely with traditional definitions of “longitudinal” research in a number of fields. Its main strength, according to the canon of statisticians (Johnson, 2001), is that it provides evidence for time relationships among variables, thus helping not only to describe phenomena and association among variables, but also to begin to explain causes and effects by looking at antecedent-consequent relationships (e.g., the cooccurrence of critical incidents and systematic shifts in the use of indexical politeness documented by Belz and Kinginger, 2002). In the social sciences, the longitudinal tradition of life-course studies is based on this descriptive, quantitative approach, and has contributed massive panel studies as well as more developmentally oriented studies, in both cases involving hundreds or thousands of participants followed over what is some times several decades of their lives (Phelps, Furstenberg, & Colby, 2002). However, the longitudinal SLA tradition, with its small samples and short-span lengths of investigation, is much closer to the spirit of longitudinal research in the feeder field of first language acquisition (e.g. Karmiloff-Smith & Karmiloff-Smith, 2001). Here, detailed case studies are the norm, and the chronological frame of several months up to five or six years is well established as sufficient for any given study, because this is the biological time scale naturally needed to capture the object of inquiry in child first language acquisition.

Longitudinal Research on L2 Program Outcomes

A very different kind of longitudinal design has been adopted more rarely in recent SLA studies, which we would like to call, in lieu of a better term, “programmatic” longitudinal research. By comparison to the descriptive-quantitative longitudinal design discussed in the previous section, the programmatic longitudinal design is used for evaluation of L2 curricular options, employs relatively large samples, and focuses on longer periods of observation scaled on institutional time (typically four to six years), coupled with fewer data collection waves spaced along wider time gaps (there are often one or two collection points per year). Although this longitudinal design has been infrequently employed in SLA, it addresses questions of utmost importance for policies and practices in L2 programs, such that an increase of longitudinal investigations of this kind would be a positive development in future instructed SLA research. In this section we discuss two representative studies: Lightbown, Halter, White, and Horst (2002) and Klapper and Rees (2003).

The participants in Lightbown et al. (2002) were a cohort of 73 French-speaking school students in the Canadian province of Nouveau-Brunswick who had received all their English L2 instruction via a self-access, comprehension-only experimental curriculum (that is, the program was entirely based on self-paced reading and listening activities). This cohort's overall proficiency in English was mapped during six years of comprehension-based instruction, with data points on the students' first year of English study (in grade 3), their last year of primary school (in grade 6), their first year of secondary school (in grade 7), and a year later (in grade 8). In addition, the cohort's L2 competencies were also compared to a group of 27 students from the same province who studied English under the traditional (audiolingual) curriculum and—for grades 6, 7, and 8 only—with another group of 14 traditional program students. The fact that these learners were followed for six years makes this one of the longest periods of observation investigated in any longitudinal SLA study focusing on group, rather than individual, performance. The analyses based on second-wave observations in grade 6 (the last year of primary school) yielded data that supported a no-difference interpretation. An inspection of the evidence over the longer period of six years, however, revealed a similar-sized improvement in reading and listening abilities for the experimental and the traditional curriculum samples, but a clear advantage in productive skills (particularly L2 writing, including the use of verbal morphology) for the traditional curriculum groups.

Klapper and Rees (2003) undertook a four-year comparison between two groups of university students (a total of 57 learners) in the same German department in the United Kingdom, who experienced a traditional grammar instruction curriculum (mostly grammar and grammar-translation courses for 2 to 3 hours a week in a major degree concentration) or a focus-on-form curriculum (mostly content- and task-based courses for 2 to 3 hours a week in a minor degree concentration). The two groups began their programs with comparable scores and received near identical amounts of instructional time during the first two years of study. Both groups' programs were enriched with a one-year study abroad component on the third year of the four-year college curriculum. Klapper and Rees examined changes on a grammar test and a test of general academic language proficiency, both administered in three waves, yearly, for years 1 through 3 of the program. The comparisons also included one-time indicators of academic ability and success (an IQ test at the beginning of the program; final exams and degree marks at completion of the program in year 4). Klapper and Rees concluded that the pace of development of the focus-on-forms and focus-on-form cohorts was different, but the final outcomes were comparable. Specifically, the focus-on-forms group showed large increases in grammar and proficiency test scores during the first two years of the program, whereas the focus-on-form group showed stagnation during the same period but similarly large progress after the one-year study abroad experience on the third year. In essence, then, the two groups had started equal and finished equal as well, but they exhibited interesting differences in the pace of development within the four-year observation period. Notably, these differences were eventually equalized by the turning-point experience of study abroad.

Despite rather large differences in L2 learning context and in instructional focus, these two programmatic longitudinal studies share strong design similarities. They are similar in the quantitative orientation of the analyses, the sampling of rather large groups, the comparison of broad curricular programmatic options, and the unusual length of the observation span, which was scaled on institutional time, coupled with infrequent, year-spaced data collection waves. Perhaps these shared features are responsible in part for the fact that in both studies the findings changed radically when the length of observation was doubled, from three to six years in Lightbown et al. and from two to four years in Klapper and Rees. Had the research teams in either study decided that three or two years was “long enough,” this choice would have led them to the conclusion, erroneous in a fuller longitudinal perspective but consistent with the first available waves of data, that there was no difference between the two programs compared, in one case (Lightbown et al.), or that there was a difference, in the other (Klapper & Rees). Thus, these two programmatic SLA studies attest to the tremendous value of taking a very long view on L2 development in SLA research.

The Longitudinal Investigation of L2 Instructional Effectiveness

A third recent, noteworthy trend is the gradual emergence of what could eventually give rise to a longitudinal design tradition in effectiveness of L2 instruction research. For one, there is a pronounced tendency in recent quasi-experimental studies of instruction to feature longer-lasting interventions (eight weeks seems to be a favored choice) and to include delayed posttests (often only a month after instruction, some times up to three months later). This, for example, can be found in the designs adopted by Han (2002) and Lyster (2004). This is a valuable improvement over the prevalent practices in this research domain up until 1998 reported by Norris and Ortega (2000). Of the 49 studies included in that meta-analysis, the majority (30 or 61%) featured short duration treatments of a maximum of two hours, and only 22 (or 45%) included a delayed posttest. Although the presence of long-lasting interventions and delayed post-tests does not amount to longitudinal research per se, when data are collected and analyzed across multiple waves to document gains on several posttests, some times even repeatedly during the treatment period as well, the process of learning is presented in longitudinal perspective to some extent.

Moreover, some recent studies have implemented true longitudinal intervention designs, although they cannot be characterized as fully quasi-experimental because no control or comparison group was employed. A representative case in this category is an investigation of instruction by Kim and Hall (2002). These researchers, like Belz and Kinginger (2002), sought to document L2 pragmatic learning through the study of Vygotskian microgenetic change, taking a close look at the process of change throughout the history of a learning event. More specifically, their purpose was to investigate the benefits of interactive book sessions that took place between the teacher-researcher and four third-grade Korean students who had been learning English for a year at the time of the study. The interactive book reading sessions (the instructional ‘treatment’) took place over four months in 30-minute biweekly meetings, during which seven age-appropriate English books were read. One book was read per session during the first month, and then the same seven-book cycle was repeated each successive month. After 20 minutes of interactive reading, the last 10 minutes in each session were typically devoted to the children's role playing of some aspect of the story. This role play was identified as the ideal “testing” context, from the viewpoint of microgenesis. Pragmatic L2 development was operationalized as increases during these role play events in mean number of words, mean number of utterances, average use of moves that served to manage talk (that is, initiations, elaborations, conclusions, and formulaic expressions), and average use of moves that served to manage meaning (that is, self- and other-corrections). The researchers found clear improvement in the four children's EFL pragmatic competencies in terms of all dependent variables but one (the meaning management moves) over the four-month period.

Although the research trends in effectiveness of instruction discussed so far are clearly encouraging and take a longer view of instructed L2 development, little progress seems to have occurred regarding the use of time-series designs in SLA research. Time-series design constitutes the single best formal strategy for investigating effects of instruction longitudinally. The principles for this design were clearly and forcefully laid out as a promising longitudinal experimental research option in SLA by Mellow, Reeder, and Forster (1996). After nearly a decade since the publication of that article, we were able to uncover only one study of L2 instruction adopting a time-series longitudinal experimental design, Ishida (2004).

Ishida investigated whether a regime of intensive recasts on the aspectual Japanese form —te i-(ru) delivered over four 30-minute, biweekly tutorial sessions would have lasting effects on the oral accuracy exhibited on this form by four second- and third-year semester students. The time-series design was operationalized as two waves of data collection that established a baseline for all four participants (on week 1), followed by four more waves in which the treatment was delivered and immediate effects were documented (on weeks 2 and 3), and two final waves that served as posttest (on week 4) plus a final wave of data collection seven weeks later (on week 11 of the study) that provided delayed posttest observations. Ishida found that the intensive recasting treatment exerted an immediate effect, as shown in learners' more accurate use of —te i-(ru) immediately after the onset of the treatment. The accuracy gains were sustained on each successive wave of observation up until week 11. These results, together with those reported by Han (2002), offer longitudinal evidence that intensive, focused recasting is likely to be causally associated with higher levels of accuracy. When these findings are compared to those reported by Lyster (2004), the converging evidence suggests that recasts under tutoring conditions, but not in larger classroom settings, can be an effective means to promote long-term formal accuracy in a second language.

A longitudinal view on L2 instructional effectiveness is not only useful to establish the durability of effects, but it also helps elucidate effects that deploy unevenly and nonlinearly over time (cf. Mellow et al., 1996). A recent meta-analysis by Keck, Iberri-Shea, Tracy, and Wa-Mbaleka (forthcoming) yielded findings that underscore this point. These researchers synthesized 14 quasi-experimental studies (published between 1994 and 2003) that tested the direct link between interaction and acquisition of L2 forms. When they aggregated and compared effect sizes on immediate versus delayed posttests, they uncovered a systematic association between the timing of the posttests and the magnitude of effects for the two design features of task-essentialness and task usefulness (see Loschky & Bley-Vroman, 1993). Namely, aggregated effect sizes yielded by immediate posttests suggested no difference between task essential and task useful interaction treatments, as both were equally effective; but when effect sizes from delayed posttests were inspected, a trustworthy (that is, statistically significant) and large difference was noted, as groups who had experienced L2 forms through task essential treatments, rather than task useful treatments, were able to demonstrate more knowledge of the given forms on delayed post-tests (d=1.66 versus d=.76, respectively). Keck et al. were able to capture this important effect for task-essentialness, which was theoretically postulated first by Loschky and Bley-Vroman (1993) but had not been investigated in any single primary study, thanks to the meta-analysis methodology, but also because 10 of the 14 interaction studies synthesized featured delayed post-tests, scheduled as soon as seven days and as late as sixty days after the treatment.

In sum, we wish to reiterate the importance of this overall move towards a more longitudinal view on instructional effectiveness in SLA research. Even though quasi-experimental, cross-sectional designs have been predominant in this research domain, issues of intensity, timeliness, duration, and durability are hotly debated (see Doughty & Williams, 1998; Mellow et al.,1996; Norris & Ortega, 2000; Spada, 1997). Because the theoretical and educational claims implied in L2 instructional studies are tantamount with claims about change that unfolds gradually (but not linearly) over time and that is long-lasting, they will ultimately demand longitudinal evidence from researchers working in this SLA domain.

Recent Qualitative Longitudinal SLA Research

A fourth noticeable trend is the gradual accumulation of longitudinal research that employs primarily qualitative methodologies. These qualitative longitudinal studies are usually framed within Vygotskian sociocultural theory (Lantolf & Thorne, forthcoming), or within a socialization theory perspective (Kramsch, 2002; Watson-Gegeo, 2004), and they some times draw from both. Accumulation is only gradual, however, because the focus in these longitudinal studies is far from homogeneous.

Future Directions

The studies reviewed in this chapter, together with many other longitudinal studies of L2 learning and learners that have accumulated over the years since the inception of SLA, point at many implicit observations and questions regarding the unique benefits and the special challenges of investigating second language and literacy development over time, from a truly longitudinal perspective. One generalization that should serve SLA researchers well is that longitudinal research can be as diverse as the epistemological approaches that inspire it. This diversity is likely to continue in the future, and we hope it will also encourage longitudinal research that capitalizes on the strengths of mixed methods designs (see Johnson & Onwuegbuzie, 2004; Tashakkori & Teddlie, 2002). The following are some of the more persistent issues that we saw emerging across studies and authors.