What does it take to run a full substance-free phonology (SFP)? Because in classical approaches only items below the skeleton have phonetic properties that, according to SFP, need to be expunged, current work in SFP only ever concerns segmental aspects. If substance is removed from segmental representation and primes and computation are therefore arbitrary, the non-trivial question arises: how can such a system communicate with a system where primes and computation are not arbitrary (at and above the skeleton)? The two phonologies below and at / above the skeleton that exist in production are complemented with a third phonology that occurs upon lexicalization, that is, when L1 learners or adults transform the acoustic signal into a stored representation. The core of this article argues that this broad architecture is inhabited by three distinct computational systems along the classical feature geometric divisions: Son(ority) is located at and above the skeleton, while Place and Lar(yngeal) live below the skeleton. The question then is how a multiple-module spell-out works, that is, how ingredients from three distinct vocabularies can be mapped onto a single phonetic item. It is argued that the skeleton plays a central role in this conversion.

This article argues that phonological features have no substantive properties, instead, segments are assigned features by learning strategies set to the task of devising a computational system for a phonology that is consistent with the requirements of UG. I address two problems for such a substance-free model. The first is the Card-Grammar problem, which has been suggested to argue for universal substantive features, on the premise that, otherwise, language data cannot be stored in a fashion necessary to correct learning errors. The Card Grammar problem disappears, in a suitably modular theory of mind with learned interfaces, where the mind still can retain information not parsed in a particular grammar. The second problem is the need for a demonstration, not just an assertion, that a reasonable theory of grammar and learning which has no access to phonetic substance can yield a coherent system of feature assignments. This is accomplished by modeling the learning of features necessary for the phonology of Kerewe.

Substance-free phonology (SFP) is based on the hypothesis that phonological computation makes no reference to phonetic substance, and that phonological features are treated as arbitrary symbols for the purposes of computation. However, phonologists within the SFP tradition disagree about whether the content of phonological features is innate or learned (“emergent”), and if learned, whether the acquisition process is based on phonological patterning alone or refers to phonetic substance. In the present article we identify predictive differences between these accounts. We conclude that there is an innate basis to phonological features, but that featural content is not innate. We suggest that a hybrid phonetic-phonological approach to feature content acquisition may ultimately be the most successful.

We argue that the representational primes of the human phonological faculty, the so-called distinctive features, are innate and substance-free. Our arguments for the innateness of features are built on traditional and novel logical arguments, experimental evidence accumulating over recent decades, and somewhat detailed proposals about their neurobiological reality. As symbols in the brain, features are substance-free, that is, they are devoid of articulatory and acoustic content, or even any direct reference to such phenomena. This is consistent with our substance-free conception of phonological computation, an approach that eschews functionalist notions like markedness, ease of articulation, and so on. We also outline a neural model of the phonetics-phonology interface called Cognitive Phonetics, which transduces innate features to speech articulation and from speech acoustics. These extra-grammatical transduction procedures are also part of the human biological endowment, which leaves no room for language-specific phonetics in our theory of the externalization of language. We show how Cognitive Phonetics can account for traditionally recognized intersegmental coarticulation, as well as the previously unexplored intrasegmental coarticulation, strongly suggesting that the basic units of speech production are transduced features.

Theories of phonology claim variously that phonological elements are either innate or emergent, and either substance-full or substance-free. A hitherto underdeveloped source of evidence for choosing between the four possible combinations of these claims lies in showing precisely how a child can acquire phonological elements. This article presents computer simulations that showcase a learning algorithm with which the learner creates phonological elements from a large number of sound–meaning pairs. In the course of language acquisition, phonological features gradually emerge both bottom-up and top-down, that is, both from the phonetic input (i.e., sound) and from the semantic or morphological input (i.e., structured meaning). In our computer simulations, the child's phonological features end up with emerged links to sounds (phonetic substance) as well as with emerged links to meanings (semantic substance), without containing either phonetic or semantic substance. These simulations therefore show that emergent substance-free phonological features are learnable. In the absence of learning algorithms for linking innate features to the language-specific variable phonetic reality, as well as the absence of learning algorithms for substance-full emergence, these results provide a new type of support for theories of phonology in which features are emergent and substance-free.

Substance-free phonology or SFP (Reiss 2017) has renewed interest in the question of abstraction in phonology. Perhaps the most common form of abstraction through the absence of substance is underspecification, where some aspects of speech lack representation in memorized representations, within the phonology or in the phonetic implementation (Archangeli 1988, Keating 1988, Lahiri and Reetz 2010 among many others). The fundamental basis for phonology is argued to be a mental model of speech events in time, following Raimy (2000) and Papillon (2020). Each event can have properties (one-place predicates that are true of the event), which include the usual phonological features, and also structural entities for extended events like moras and syllables. Features can be bound together in an event, yielding segment-like properties. Pairs of events can be ordered in time by the temporal logic precedence relation represented by ‘<’. Events, features and precedence form a directed multigraph structure with edges in the graph interpreted as “maybe next”. Some infant bimodal speech perception results are examined using this framework, arguing for underspecification in time in the developing phonological representations.