In studies of second language (L2) learning aptitude, research using the LLAMA aptitude tests (Meara,2005) has grown in recent years to over 40 published studies so far. Although LLAMA was not validated priorto its release, later research (e.g. Grañena & Long, 2013) has contributed validity evidence mainly from samplesof intermediate or advanced L2 learners. Often, however, L2 is narrowly represented by, for example, agrammaticality judgement task, and there is also a lack of consistency in the correlational evidence betweenLLAMA and L2 tasks. Some studies rely on small, outlier sensitive, samples. All together, no clear picture hasemerged regarding the predictive validity of the four subtests of LLAMA, the facets of L2 proficiency they relateto, and covariation with learner level. A continuous investigation of validity evidence is therefore necessary.

Mixed L1 adult learner groups at beginner level is a common situation in many L2 classrooms today. Skehan(1998) suggested that for L2 beginners, auditory processing ability is of particular importance. This has gainedsome LLAMA-derived evidence recently (Artieda & Muñoz, 2016) but otherwise, general proficiency amongL2 beginners is not well represented in existing LLAMA research. The aim of the present study was to explorethe predictive validity of LLAMA in a sample of mixed L1 adult beginners of Swedish as a L2, and possiblybeing able to confirm the role of auditory processing among beginners.

International newly arrived university students (n = 93) with a range of L1 backgrounds, participating in aSwedish (L2) language course for beginners, took part in the study. Half way into their six weeks’ training, thestudents were administered the LLAMA battery, consisting of four subtests: B (vocabulary learning), D (soundrecognition) , E (sound-symbol correspondence) and F (grammatical inferencing). Upon finishing the course,they completed a C-test (cf. Klein-Braley, 1997), measuring general L2 proficiency.

LLAMA scores were correlated with scores from the C-test, reaching significance only for LLAMA D (r = 0.30).However, when controlling for participants’ L1, by grouping them into European (mainly Germanic) and non-European (mainly East-Asian) L1s, the correlation with LLAMA D in the European group was 0.50 whenapplying a robust (less outlier sensitive) regression method (cf. Figure 1). A weaker significant correlation wasobserved for LLAMA E (sound-symbol correspondence). In the non-European group, no correlations reachedsignificance. The differential impact of aptitude observed in the two L1 groups was likely due to speakers ofnon-European, typologically distant, L1s having difficulties in acquiring even basic Swedish skills in a shorttime, resulting in low C-test scores also for high-aptitude learners.

The results confirm previous findings that auditory processing ability (LLAMA D) seems to be a valid predictorof initial L2 learning. They also highlight methodological considerations, such as the effect of applying robuststatistics, as well as using tasks of appropriate difficulty when subsets of participants may be expected to performat different proficiency levels.

The present study makes an important contribution to L2 learning aptitude research by adding knowledge aboutthe predictive validity of LLAMA.