Irregular use of lexical bundles, short recurring sequences of two or more words, which are crucial to coherent writing in academic contexts is a common problem for Iraqi EFL university students in producing natural and fluent academic writing. The aim of this corpus-based study was to determine the kind, frequency and origin of errors in lexical bundles (LBs) in academic paragraphs produced by Iraqi EFL learners and to investigate the differences in error patterns between LBs of different functions and learners' L1, Arabic vs. Kurdish. The argumentative paragraphs were gathered from 60 Arabic and Kurdish speaking undergraduate students from a public Iraqi university to form a learner corpus of 60 argumentative paragraphs with a total of 12,400 words. Frequency and range thresholds were set and 87 three- to five-word lexical bundle types (1,234 tokens) were extracted from the text using AntConc software, followed by errors being hand-coded according to an adapted taxonomy. It was found that 39.1% of the bundle tokens had errors, the highest error rate being in prepositional phrase-based bundles (48.1%), while the most common sub-type of bundle error was mis selection (33.7%). The stance bundles were significantly less accurate (53.2%) than were referential (38.8%) and discourse-organizing (29.4%; χ² = 48.31, p < .001) bundles. There was no significant difference overall between the error rates of L1 groups (p = .185), but there was between the error sub types (χ² = 13.82, p = .017), with Arabic speakers making more mis selection errors, and Kurdish speakers more omission and morphological errors. The results support the usage-based accounts of L2 acquisition in low-input EFL environments and highlight the importance of explicit instruction on stance bundles, error-specific feedback and differentiated instruction to consider L1 transfer effects.
