In "The Sound Pattern of English," Chomsky and Halle maintain that the phonetic representation of most words can be generated from underlying forms and a small set of rules. Since these underlying forms are frequently close to the traditional spelling, we may hypothesize that literate native speakers share comparable internalized rules which enable them to agree on the pronunciation of new words. Twenty subjects were asked to pronounce 30 little-known English words. Because Chomsky and Halle's rules are based in part on learned Latinate vocabulary, the words were evenly divided between those entering English from Latin Greek, and Germanic sources to test for differences in performance relative to word origin. While the subjects averaged 81 percent agreement on stress assignment, the number of segmentally distinct responses ranged from one to ten per test word with an overall average of five. On the other hand, a majority of the subjects agreed in their pronunciation of 21 of the 30 words. The subjects demonstrated even more agreement in their interpretation of vowel graphemes under the same conditions of stress and syllable type. Chomsky and Halle's system reflects a fair degree of psychological reality, but it exceeds our competence on many points. (Author/CLK)