Assessing the content typicality and construct of Persian language proficiency test (PLPT) for non-Persian speakers: a corpus-informed study

Abstract

AbstractDrawing on a growing body of research on the interface between corpus linguistics and second/foreign language testing and assessment, we adopted Peykare, a large-scale, annotated, Persian written language resource to evaluate the content (i.e., coverage and typicality) and construct validity of a Persian language proficiency test developed for certification of proficiency in Persian as a foreign language (PFL) of non-native speakers. Designed at the Research Center for Intelligent Signal Processing (RCISP), Peykare contains 35,058 text files over five linguistic varieties and 24 different registers of contemporary Persian. This study addresses how corpora, as rich database resources, can practically be applied to test validation purposes and insightfully inform the test life cycle. The results of content validity phase revealed evidence supporting content representativeness, relevance, and typicality of the test. The linkage between the corpus-extracted criterial features or parameters and those covered by the test was not, however, strongly evidenced by items measuring ezafeh constructions, homographs/homophones, PRO (proposition), and POST (postposition). The analysis of content typicality indicated chunks that did not closely conform to the corpus typical output. The construct validity phase, assessing the test hypothesized factor structure (i.e., hierarchical, unitary, correlated, and uncorrelated models) in two randomly split samples of PFL learners from Asian and European countries (N=121), showed that the correlated model fit the data best in both samples. The results supported the presence of distinctive factors of receptive skills, providing empirical evidence for score interpretations of the corpus-based test.