“…Sequences of linguistic units that are of high frequency or resemble sequences of high frequency will be judged more acceptable than those that are of low frequency or do not resemble frequently used structures” (Bybee & Eddington, : 349). A number of studies in both the generative and usage‐based traditions have, however, confirmed the existence of a grammaticality‐frequency discrepancy, if not a gap (Kempen & Harbusch, , ), for acceptability ratings: Corpus frequencies are poor predictors for off‐line acceptability ratings, in particular at the lower end of the frequency spectrum, in morphology and syntax (Keller, ; Kempen & Harbusch, , ; Arppe & Järvikivi, ; Divjak, ; Bader & Häussler, ; Bermel & Knittl, ,b; but see the opposite tendency in the results of Lapata, McDonald, & Keller, , for adjective‐noun combinations). This has strengthened generativists in their belief that “simple frequency data” could and should be ignored in theoretical linguistic analyses.…”