There is extensive evidence that structural regularities affect the processing of visually-presented words. However,it is not known whether the processing consequences of an orthographic violation are limited to the offending subpart (e.g.,an unattested onset cluster) or apply more globally (e.g., to the entire word). We provide evidence of global disruption fromthe recall of briefly-presented nonwords that were manipulated for degree of orthographic markedness and length. Error rateswere higher for both the initial and final portions of nonwords beginning with more marked onsets; symmetrically, report ofmarked onsets was degraded in words with longer endings. These effects suggest that, as in other visual tasks, the fidelity withwhich one element can be represented depends on the overall stimulus complexity. We present a modified version of rationalmodels of visual word perception in which global effects result from the distribution of a limited processing resource over letterpositions.

Perception and memory of linguistic information is biased in favor of stimuli that conform to structural regularities.At the level of word form, there is evidence that initial consonant clusters varying in grammatical status (e.g., br vs. *bn,*bd, *rb) differentially affect response times in same-different matching (e.g., slower responses to rbif - REBIF than to brif- BERIF; Berent & Lennertz, 2010). Previous results are consistent with two hypotheses: non-conforming clusters could bemodified by a specific ’repair’ (e.g., rbif recoded as rebif), or the encoding of such clusters could be more uncertain and theirrecall more variable. A series of matching and full-recall experiments support the second hypothesis: the response time effectfor non-conforming clusters is observed for both *rbif - REBIF and *rbif - RBIFE, but only the former ’repairs’ the cluster;furthermore, errors made in recall exhibit high variability and do not systematically improve structural well-formedness.