A commonly made, but rarely defended, assumption is that phonetic reduction processes apply to hyperarticulated phonetic targets. Results from experiments reported in this paper support this assumption. In various experimental conditions, listeners adjusted the input parameters of a speech synthesizer until the vowels it produced sounded like the vowels found in a set of example words. A preliminary study indicated that the method of adjustment is a feasible tool for studying vowel systems. Interestingly, listeners in the study chose vowels that were systematically different from those measured in productions of the set of example words: high vowels were higher, low vowels were lower, front vowels were farther front, and back vowels were farther back. We hypothesized that this extreme vowel space corresponds to phonetic targets that are hyperarticulated: HYPERSPACE. This hypothesis was tested in the two main experiments. The first experiment controlled for possible effects of instructions and phonetic training on the listeners' choices. In the second experiment, we improved the naturalness and distinctiveness of the synthetic vowels. The results indicate that the extreme vowels chosen by the listeners were consistent with those produced in hyperarticulated speech; moreover, the hyperspace effect is robust across experimental conditions. These results validate the hypothesis that phonetic targets are hyperarticulated, and are consistent with a two-stage model of phonetic implementation: at the first stage distinctive features are mapped to hyperarticulated phonetic targets, and at the second stage these phonetic targets are reduced.
Stored only when both papers are in the corpus — 348 citations exist upstream.