Patterns of Segmental Modification in Consonant Inventories: Contrastive vs. Redundant Systems Jeroen van de Weijer and Frans Hinskens Leiden University and Meertens Institute Abstract Segmental modification types, such as labialization, aspiration and prenasalization, usually appear on natural classes in consonantal inventories. There appear to be two typical situations in which segmental modifications pattern, which can be referred to as ‘redundant’ and ‘con- trastive’: in redundant systems the segmental modification is superimposed on another, pri- mary distinction, and therefore acts like an additional way of making a segmental contrast, while in contrastive systems there is no distinction between two (sets of) segments other than the segmental modification. In this paper we investigate the cross-linguistic occurrence of both types of systems, and offer a formalization couched in Optimality Theory, involving faithfulness constraints relativized for natural class. 1. Introduction: The symmetry of segment inventories* It has long been recognized that the segments of a language are usually neatly organized into groups, or natural classes, contributing to overall symmetry or similar types of orderliness within segmental systems. For instance, many lan- guages will have voiced counterparts of every voiceless obstruent in their seg- ment system. In other languages every dorsal consonant has a labialized coun- terpart. Of course in many languages there are also exceptions to such gener- alizations. In Dutch, for instance, there is no voiced velar stop phonemically (except in a limited number of loanwords), and other languages lack a voiceless labial stop, to mention two common patterns. Such gaps need to be explained either from a phonetic, i.e. articulatory or perceptual, viewpoint or from a pho- nological perspective. A phonetic account is readily available in these cases. As is well-known, a voiced velar stop is relatively hard to produce, and a voiceless labial stop relatively hard to hear, which can be used to account for gaps such as these. However, the basic symmetry of segmental inventories has not re- ceived the recognition it deserves. In many grammars and analyses, the seg- ments of a language are just listed as a matter of course. However, lists are ar- bitrary and this approach should be avoided, because it presents no hypothesis about what is a possible phoneme system, and what is not a possible phoneme system, and what are possible gaps. Therefore it is necessary to explore the