That is a function of the complex cognitive mechanisms involved in human language processing and speech, which Kormos explicitly acknowledges as possibly the most complex of all human cognitive processes (Levelt, 1995 in Kormos, 2003 p88). Given that characterization, the use of a word span test -- in which performance could quite conceivably measure other variables besides raw working memory, such as parallel recognition and various other associative or pneumonic devices -- instead of a non-word span test likely undermined the validity of the Mota study (2003) results.
Furthermore, the Kormos and Safar study (2008) employed standard objective academic tests of foreign language proficiency, whereas the Mota study (2003) employed a novel matrix of variables (fluency, accuracy, and complexity) to measure foreign language proficiency (Mota, 2003 p69). The former type of diagnostic test would appear to be an appropriately accurate and narrowly targeted measure of memory-based performance and learning in the realm of foreign language acquisition (Kormos & Safar, 2008 p265).
Meanwhile, the latter type of test would seem to introduce multiple extraneous variables in the realm of linguistic abilities much more general than those exclusively dependent on working memory. Particularly in light of the perspective...
Conversely, the Mota study (2003) yielded less concrete or meaningful results capable of multiple explanations, including one very simple explanation that may have completely compromised the relative value of that study. Namely, that individuals with better working memory may simply recall language more efficiently in general, and regardless of either proficiency in foreign languages or the precise role of memory in foreign language acquisition.
