Romanized transliteration is widely used in internet communication and global commerce, yet we know little about its behavioural and neural processing. Here, we show that Romanized text imposes a significant neurocognitive load. Readers faced greater difficulty in identifying concrete words written in Romanized transliteration (Romanagari) compared to L1 and L2. Functional neuroimaging revealed that the neural cost of processing transliterations arose from significantly greater recruitment of language (left precentral gyrus, left inferior parietal lobule) and attention networks (left mid-cingulum). Additionally, transliterated text uniquely activated attention and control areas compared to both L1 (cerebellar vermis) and L2 (pre-supplementary motor area/pre-SMA). We attribute the neural effort of reading Romanized transliteration to (i) effortful phonological retrieval from unfamiliar orthographic forms and (ii) conflicting attentional demands imposed by mapping orthographic forms of one language to phonological-semantic representations in another. Finally, significant brain-behaviour correlation suggests that the left mid-cingulum modulates cognitive-linguistic conflict. (Contains 1 table and 3 figures.)