Over the last few years, much work in phonology has been devoted to exploring the way features are specified for segments; in particular, to what extent feature specification may be underlyingly present and/or acquired by rule or default in the course of a derivation. While a number of proposals have been made attributing various degrees of underspecification to abstract levels of the phonology (Kiparsky 1985; Steriade 1987; Archangeli 1988), it has been generally assumed that where phonetic implementation comes into play, i.e. at the end of the derivation, segments are exhaustively specified.