Metaphony with unary elements

In this article, I analyze several vowel harmony systems, generally referred to as metaphony. These systems (which come in many different varieties; see below) have been described or analyzed in terms of binary features, either using the feature [±high] (Walker 2005) or [±ATR] (Calabrese 2011) or in terms of unary features. In the latter case some authors have supported the use of unary features (Maiden 1991; Canalis 2016; d’Alessandro & van Oostendorp 2016), while others have argued against their use (Kaze 1991). In this article, I adopt the use of unary elements, such as the ‘AIU’ system that has been proposed in Dependency Phonology (Anderson & Ewen 1987), Government Phonology (Kaye, Lowenstamm & Vergnaud 1985), with some modifications that have been proposed in Radical CV Phonology (van der Hulst 2005, in prep.; van der Hulst & van de Weijer, to appear). My main goal is not to motivate in general that unary features are preferred over binary features. Rather, given that we use unary elements, I investigate which set of such features is required and how metaphony is best formally represented. Kaze (1991) has argued that an ‘AIU’ system fails to provide an adequate analysis of metaphony, based on the argument that in such a system it is not immediately obvious how one can account for processes that are triggered by high vowels (such as [i] and [u]), arguing that the ‘AIU’ system is unable to treat high vowels as a natural class of vowels that share the property ‘high’. Similar objections have been raised in Clements & Hume (1995). Staun (2003), making reference to discussions of other processes that seemingly require access to a feature [high] in dependency-based analyses, remarks that “[I]n each such account the notion of a negated component, in particular negated |a| has played a central part. [...] despite the claims of both Clements and Hume and Kaze, a unique specification of high vowels is perfectly possible within the dependency-based model, viz. as |~a|.” Clearly, there is a ‘risk’ in appealing to (the spreading of) negated elements when one advocated for a unary feature system. Here I will take a different route. Whatever the merits (and dangers) of using negated elements, the adoption of a unary system does by no means imply a necessary commitment to the ‘AIU’ set, without any other features. Indeed, Anderson and Ewen (1987), as well as Kaye, Lowenstamm and Vergnaud (1985), propose additional features. The set of features in binary or unary systems can vary, depending on various considerations, both empirical and theoretical. The choice between binarism or unarism does not depend on the analysis of specific processes. As discussed in Kaye (1988) and van der Hulst (2016a), postulating a unary system is by and large the null hypothesis, since, keeping the set of feature ‘names’ constant, treating these ‘names’ as unary features leads to a more restricted theory; it only allows for half the number of natural classes and processes. Additionally, a unary system provides a head-on answer to the problem of markedness that was ‘noted’ in chapter 9 of Chomsky & Halle (1968). For these reasons, my point of departure is to explore the consequences of a unary approach which, then, requires a specific choice of unary features. While the ‘AIU’ set of features is well-founded and widely used, my own