The Differentiation of Allophonic and Nonallophonic Segments in Phonemic Analysis

Abstract

TtHE NOTION IS WIDELY HELD that phonemic analysis of complex distributional data may be effectively carried out solely on the basis of complementary distribution, phonetic similarity, and pattern congruency.1 It can be shown that these criteria are not so rigid or completely objective as is commonly supposed. Too often complications arise when more than one grouping is possible with phonetically similar phones in free variation or complementary distribution.2 Where the distributional data make it difficult to decide phonemic grouping, or where underor overdifferentiation is suspected in an analysis, phonemicization may be better effected by employing the test of segment substitutability. The test involves the commutation of phonetically similar segments in common environments of paired or phonetically similar utterances to determine whether the original utterances undergo a change in meaning.3 If a change in meaning 'does occur' in one of two compared utterances, then the phones are contrastive. If no change in meaning occurs, the phones are non-