Theoretical precursor: Feature Classes

The rejection of major class nodes/features and the adoption of an explicit system of Feature Co-occurrence Constraints is reminiscent of earlier work on Feature Classes by Padgett (2002) and Yip (2011). Padgett rejects feature ge-ometry and develops a system of Feature Classes motivated by partial class behaviour of segments. Padgett (2002) notes that Feature Geometry is right in attempting to capture class behaviour, but at the same time, it is too rigid in forcing features to fit into classes defined by single class nodes. Take, for exam-ple, the feature [back], which contrastively denotes back vowels. There are

con-18This is, of course, reminiscent of the observation that major class features are never contrastive.

vincing phonetic (e.g., second formant enhancement effects) and phonological reasons to assume that it forms a class with [round]. Much of the phonological evidence comes from vowel harmony in Turkic, Altaic and Uralic languages, many of which display both backness and roundness harmony, which occurs jointly (in the sense that suffixes that are exceptions to backness harmony are also exceptions to roundness harmony, and vice versa). Padgett takes this, in standard autosegmental fashion, to be evidence for a class consisting of [back]

and [round], which he dubs color.

Having established the class, Padgett proposes that rather than two dif-ferent rules or constraints driving roundness and backness harmony, there is a single mechanism referring to color. So far, the account is compatible with Feature Geometry, but the picture becomes complicated when we take into account that roundness harmony in Turkish, for example, is limited to high vowels. Rather then retreating from the position that color is a class, Padgett proposes that rules and/or constraints in the theory he coins Feature Classes refer to individual features, rather than to class nodes. Hence, class nodes cease to exist, and are replaced by set-theoretic classes: color = {[back], [round]}, place = {[labial], [dorsal]}, et cetera. The fact that non-high vowels fail to harmonise on roundness is captured in an OT grammar by a high-ranking feature co-occurrence constraint *[-high][+round]. Summarising, although har-mony drives both [back] and [round] by referring directly to them jointly, via the color class, it fails to apply because non-high round vowels are ruled out by an FCC.¹⁹More generally, Padgett shows that Feature Class Theory is able to account for partial class behaviour.

Other evidence that features are targeted directly even when behaving as a class exists, too. Padgett mentions dissimilation, which, when seen as an OCP-effect, is problematic for Feature Geometry without extra stipulations.

Although it is entirely possible to posit a rule or constraint OCP(place), such a mechanism, when taken seriously, would ban sequences such as in (36):

(36) Unwanted OCP(place)-effects

*root root

place place

In other words, adjacent roots (either absolutely adjacent or on some tier) specified for place are banned. This is obviously not the desired result. The goal of OCP(place) is to rule out identical features to the degree that both belong to the class of place features – not identical classes.²⁰

19It should be noted that Turkish allows for non-high round vowels, but only in roots;

that is to say, not as the result of vowel harmony. Hence, the analysis in Padgett (2002) also involves a faithfulness constraint.

20Yip (2011) applies this framework to the question of the representation of laterals, which do not show typologically uniform behaviour with respect to continuance, but also with

Partial class behaviour is not something we are particularly concerned with in the current thesis, but there are at least two aspects of Feature Class Theory that make it an appealing companion for the Feature Co-occurrence Constraint Theory. First, it rejects the notion of overarching class nodes and/or features.

All sub-segmental entities are created equal and all are targeted by rules or constraints individually. This resonates well with the finding that major class features and grouping features are incompatible with our Feature Co-occurrence Constraint Theory. The other striking point is that Feature Class Theory, in order to express the things that set it apart from traditional Feature Geometry, relies heavily on feature co-occurrence constraints.

This raises the question of how Feature Classes are to be formalised. Padgett (2002) proposes a set-theoretic set of definitions, such as in (37) (adapted from Padgett (2002, exx. 24, 25)).

(37) a. Simplex classes Nasal =def {[nasal]}

Voice =def {[voice]}

Dorsal =def {[dorsal]}

continuant =def {[continuant]}

b. Complex classes

Laryngeal =def {[voice] ∪ [constricted glottis] ∪ [spread glottis] } Color =def {[back] ∪ [round] }

Place =def {[labial] ∪ [distributed] ∪ [dorsal] }

Although this formalisation is logically adequate, it opens the question as to the ontological status of the list of definitions. That is to say, where in the mental representation of speaker and hearer does it reside? Therefore, I propose to go one step further and rather than assigning features to classes, to assign classes to features: features are ‘labeled’ or ‘indexed’ for the classes to which they belong. This results in a (non-exhaustive) list as in (38):

(38) [dorsal]place

[voice]laryngeal

[back]colour

Although formally equivalent to Padgett’s proposal, this fashion of representa-tion immediately displays the classes to which a feature belongs.²¹

Earlier a concern was expressed regarding the indexing of features that is implicit in Nevins (2010). It should be noted that the indexes proposed here

respect to sonorancy and coronality. It is possible, technically, that the constraint in 36 is present in the grammar, but must be dominated. This is problematic in two ways: first, it runs counter to the notion of Factorial Typologies, and second, unless constraints are innate, it is hard to see how such a constraint can be learned.

21One open question is whether the equivalence between the two definitions extends to the simplex classes. That is to say, I will remain agnostic for the time with respect to the question whether [dorsal]placeshould really be represented as [dorsal]dorsal,place.

refer to inherent properties of features (laryngeality is an important aspect of [voice], for example), rather than a system-dependent and non-substantive label such as ‘contrast’. The degree to which this difference is important must be explored further.

A full exploration of this version of Feature Class Theory is far beyond the scope of the current thesis, for which it bears no direct relevance. For example, the precise status of class indices must be investigated; they are not features and hence do not behave like features. The question then arises whether they are mere shorthands for groups of features. Remember that major class features must be discarded; this hints to a rejection of Feature Geometry. Feature Class Theory has the obvious advantage of retaining crucial insights from Feature Geometry, namely class behaviour. Finally, its reliance on feature co-occurrence constraints resonates well with the current proposal.