Supplementary Materials for "Learning Contexts, Not First Language: Language of Schooling Confounds Native Language Identification of Learners of Arabic": Anonymized Instruments, Participant-Level Data, Qualitative Sub-Dataset, and Recalculated Analysis

This record contains the complete supplementary materials (File S1) accompanying the study “Learning Contexts, Not First Language: Language of Schooling Confounds Native Language Identification of Learners of Arabic.” The study examines whether machine-learning identification of a learner’s first language (L1) from Arabic performance data reflects cross-linguistic transfer or the collinear influence of the learner’s language of schooling. Data were collected from 65 learners of Arabic at Prince Sattam bin Abdulaziz University, Al-Kharj, Saudi Arabia; 64 were retained for five-group classification after one heritage/contact case (P038) was flagged and excluded. The materials comprise three appendices: Appendix A — Research instruments: an adapted Language Experience and Proficiency Questionnaire (LEAP-Q) targeting L1 macro-group, language of schooling and heritage/contact status; two controlled writing prompts (descriptive and argumentative); a read-aloud passage eliciting pharyngeal, emphatic and uvular phonemes and vowel-length contrasts, with annotated target loci; the full 40-item Arabic proficiency battery (20 grammar/morphosyntax items, 20 vocabulary/semantics items) with alternatives, master answer key and partial-credit scoring specification; independent two-rater agreement statistics for writing and speaking (ICC(2,1), absolute agreement); and the sequential-explanatory qualitative interview guide. Appendix B — Participant-level data: a variable dictionary; the full anonymized background and proficiency table for all 65 participants (L1 macro-group, language of schooling, nature of prior Arabic exposure, institutional level, grammar and vocabulary scores, blinded writing and speaking ratings, standardized proficiency composite); manual cross-linguistic transfer coding (16 records from 15 participants, writing and speech, with pharyngeal/emphatic error counts); and a reflexive thematic sub-dataset of one primary theme and one representative excerpt for each of 21 interviewed participants. Appendix C — Recalculated statistical analysis: the reproducibility protocol (class-weighted linear SVM, nested stratified cross-validation, bootstrap confidence intervals, permutation diagnostics); chi-square and Cramér’s V tests of the L1-by-schooling association; overall and class-level classification performance with confusion matrix; a feature-block analysis isolating the contribution of language of schooling; and a robustness re-analysis across ten outer-fold seeds in an independent software implementation. Participants are identified only by sequential pseudonymous codes (P001–P065). Direct identifiers were absent from the source data, and location-bearing wording in qualitative excerpts was generalized. No participant records were generated, replaced or imputed. Audio recordings, full interview transcripts and raw learner texts are not included. Files: a compiled PDF of the supplementary materials and the XeLaTeX source used to produce it.