Possible and impossible segments

In this article we consider the relationship between phonetic possibility and phonological permissibility of segment types. We ask (i) are any phonetically impossible segments phonologically permissible? and (ii) are any phonetically possible segments phonologically impermissible? Our main focus is on answering (ii). We analyze the implications of the only relevant case we can find, which is in Cohn's (1990, 1993a) examination of nasality spreading in Sundanese, and relates to the description of glottal nasals (produced with glottal place of articulation and lowered velum). Cohn tentatively proposes that nasalized [h] and [?] occur phonetically but not phonologically. We show that a persuasive theory of nasality spreading suggests otherwise, and it is supported by evidence from several languages. Our conclusion is that no sound argument exists for excluding any pronounceable segment from phonology on theoretical grounds. The relation between the phonetically possible and the phonologically possible accordingly becomes somewhat more straightforward.* What is the relationship between pronounceability of segments on the phonetic level and licitness (or illicitness) of segments in terms of phonological theory? The question can be separated into two more specific questions about the relations between phonology and phonetics. Let us call a segment type PHONOLOGICALLY PERMISSIBLE if and only if no principle of phonological theory is infringed by the appearance of a segment of such type in an underlying representation or in the input/output of a phonological rule or level, and let us call a segment type PHONETICALLY POSSIBLE if and only if segments of such type can actually be pronounced. The cross-cutting of the two distinctions here (phonetic versus phonological and possible versus impermissible) defines four logical possibilities, but we will dismiss two of them immediately. We will assume that under every theory of phonology at least some phonetically possible segment types are phonologically permissible. We will also assume that every phonological theory bans some segment types that are phonetically impossible (both phonology and phonetics agree in the complete absence of apico-uvular stops, velar trills, and so on). This leaves two issues: