Segmental and prosodic effects on intervocalic voiced stop reduction in connected speech.

De scriptions of lenition have often assumed that connected speech reductions are the phonetic precursors of phonological lenition processes. In this article, production of intervocalic voiced stops during reading in American English is examined to determine whether connected speech reduction processes mirror the stages of lenition that have been posited in the phonological literature. The first result shows that American English speakers never lenite to fricatives or debuccalize to [h] or glottal stop, but rather produce approximants whenever reduction occurs. Second, stress plays an essential role: 51% of stops are produced as approximants when stress is on the preceding syllable (e.g. 'yoga'), but only 7% of stops weaken when stress is on the following syllable (e.g. 'lagoon'). Approximant productions are longer and higher in intensity than stop productions when stress precedes the target consonant, but when stress follows the target consonant, the stop cues are enhanced. These acoustic findings suggest that English speakers prioritize the realization of acoustic cues to stress, including the robust production of stop consonants, over pressures to reduce or weaken consonants in intervocalic position.