Abstract
Greenberg’s paper on universals (1963) contains an interesting set of generalizations relating to features. It is a good time to review the issues involved in establishing universals of features. These verge on the philosophical at one extreme, while at the other they concern the practical question of how we present and gloss examples. Various initiatives concerned with standardization, taken broadly, are under way, and it is important that they should be fully informed by the linguistic issues. There are two main areas to discuss: the Analysis problem and the Correspondence problem. The Analysis problem: for a given language, we need to be able to justify the postulation of any feature (such as number or case). Equally, for each feature in the language we need to be able to justify the set of values postulated (for example: singular, dual, paucal and plural; nominative, accusative and genitive). For some languages the analysis is trivially simple, in others it is exceptionally complex (for some there have been long-running debates). In this context, it is worth reviewing the work of the Set-theoretical School, given its undoubted relevance for typology. The difficulties posed by hybrids will be discussed; this leads naturally to typological hierarchies and the ‘Canonical’ approach in modern typology. The Correspondence problem: as typologists we need to be able to justify treating features and their values as comparable across languages. This is not straightforward, and yet a good deal of typology, including enterprises such as the World Atlas of Language Structures, depends upon it. The problem has a second, more subtle version. Even within a single language, features and their values do not necessarily line up consistently. In Bayso, the number system of nouns and verbs interact in a complex way. In Romanian, the genders of nouns and adjectives differ, and there are many more such examples. Here a typological perspective can inform the analysis of a single language and, of course, a typology which ignored these languages would be considerably impoverished. Features are an area where the concerns of the typologist meet those of computational linguists, formal linguists, fieldworkers, in fact linguists in many different guises. As we put increasing theoretical weight on features, it is important to review our assumptions and check our progress in understanding them.