In experiments reported at the last meeting of the Society, we cross‐spliced 60 sentence pairs (e.g., I say taezi, I say taeza) at the beginning of the last syllable, then compared recognition time (RT) to the final target vowel /i or a/ in cross versus intact (as spoken) sentences. The resulting slower RT to crossed targets suggested effects of vowel‐to‐vowel anticipatory coarticulation; that is, misleading information in prior context predicts the other (wrong) vowel in crossed sentences. Acoustical analysis supports this interpretation: F2 in the prior context vowel tends to be higher preceding /i/ than preceding /a/. In the present experiments, /kaezi/ and /kaeza/ were synthesized in three conditions: F2 of /ae/ was (a) “normal,” (b) raised, or (c) lowered. We predicted (compared to normal) slower RT to /i/ when F2 lowered (interference), faster RT to /i/ when F2 rose (facilitation), but vice versa for /a/. In Experiment I, using parameters derived from cepstrum and LPC analysis of real speech, the predicted interference, but not facilitation, was found. In Experiment II, using parameters from synthesis‐by‐rule, both predicted interference and facilitation were found. [Work supported by NINCDS.]