1 Overview

Beyond markedness in formal phonology is a collection of chapters, edited by Bridget D. Samuels, which investigate the concept of markedness and its role in phonological theory.* All the chapters in the volume attempt to go beyond the concept of markedness relations developed in SPE, although they do so in radically different ways. Some chapters reject markedness entirely, while others argue for encoding markedness in representations. The notion of markedness has played a central role in the development of theories of phonological representations and underspecification (e.g. Kiparsky Reference Kiparsky1985, Mohanan Reference Mohanan1991, Rice Reference Rice1996). Although recent work has seen less emphasis on phonological representations, the significance of markedness has, if anything, only grown with the rise of Optimality Theory, which models all phonological processes as motivated by markedness constraints. In addition to formal definitions of markedness developed in conjunction with representational theories and OT, the term is often used without explicit definition to refer to unusual sounds or processes, particularly in reference to typology or acquisition. It is this array of distinct uses of the term, often leading to ambiguity between formal and functional notions, that motivates the critical perspective on markedness which characterises this volume.

Although the contributions are not explicitly divided by topic, it is useful to think of the collection as consisting of three sections. The first four chapters present critiques of markedness, in some cases proposing alternative analyses of data which have been given previous accounts relying on the concept. The next three chapters all propose definitions of markedness, or replace reference to markedness with other formal, representational mechanisms, within the framework of Government Phonology. Whereas the initial set of chapters in the volume largely argue for dispensing with markedness, these chapters advocate formal models of markedness based on representational complexity. The final two chapters consider markedness from a biological perspective, one assessing investigations into markedness using brain-imaging methods and the other considering conceptual issues that arise when markedness is considered within the context of biolinguistics and language evolution. In the following review, I consider each section in turn, providing a brief summary of each chapter before discussing the section as a whole.

2 Critiques of markedness and substance in phonological grammars

In the first chapter, ‘Markedness in substance-free and substance-dependent phonology’, David Odden reviews influential developments in the concept of markedness throughout its history in phonological theory. Odden begins by summarising Trubetzkoy's characterisation of privative contrasts in which one member of an opposition is ‘marked’ as having some property that the other member lacks. Odden assesses this as a formal and logical definition of markedness which is muddled by reference to substantive factors in later developments in Trubetzkoy's own work, as well as in work by Jakobson, in SPE and in Optimality Theory. In Odden's view, any notion of markedness which relates to substantive factors, such as articulation and perception, has no role in the phonological computation. Nonetheless, he concludes that questions regarding the nature of phonological features, which motivated early work on markedness, remain worthy of serious inquiry.

Charles Reiss also argues against reference to substantive factors in phonological models in his chapter ‘Contrast is irrelevant in phonology: a simple account of Russian /v/ as /V/’. The chapter argues against the significance of phonemic contrasts in shaping phonological processes, and addresses the issue of markedness only indirectly. Reiss provides an analysis of Russian voicing alternations which focuses on the idiosyncratic behaviour of /v/. Like obstruents, /v/ undergoes final devoicing and voicing assimilation. Like sonorants, it fails to trigger voicing assimilation. While previous analyses have accounted for this behaviour by arguing that /v/ is underlyingly a sonorant (e.g. Hayes Reference Hayes, Aronoff and Oehrle1984, Kiparsky Reference Kiparsky1985, Padgett Reference Padgett2002), Reiss instead argues for a solution which ‘fiddles around with the feature Voiced’ (p. 38). He proposes a derivational account in which assimilation is analysed as a sequence of feature deletion and feature insertion, formalised using set theory as set subtraction and set unification respectively. In combination with an underlying representation in which /v/ is unspecified for [voice], this analysis is able to account for its exceptional behaviour. Although Reiss provides minimal discussion of markedness, he states that markedness and contrast are linked, making an argument against contrast also an argument against markedness.

In ‘What are grammars made of?’, Juliette Blevins argues that markedness universals in the domains of distinctive features, sonority and the prosodic hierarchy, are ‘nothing more than universal tendencies’ (p. 55). Blevins summarises proposed universals in each area before reviewing documented counterexamples. With respect to features, for example, counterexamples consist of unnatural classes referred to in ‘crazy rules’ (see e.g. Mielke Reference Mielke2008). Blevins argues that features, sonority scales and prosodic domains are language-specific, learned properties. These properties tend to be similar across languages because of the influence of phonetic factors such as ease of articulation and perceptual salience. Blevins argues against incorporating these phonetic properties into formal grammars, however, noting that proposals which do so (for example the P-map of Steriade Reference Steriade, Hanson and Inkelas2009) result in substantial increases in the complexity of the phonological computation, as well as duplication between the phonetic and phonological components. Instead, Blevins proposes that formal grammars consist of language-specific, learned rules and constituents which closely resemble descriptive grammars.

Chapter 4, ‘Consonant epenthesis and markedness’, argues against epenthesis as a markedness diagnostic. On the basis of both cross-linguistic variation and an analysis of English intrusive /r/, Bert Vaux & Bridget D. Samuels demonstrate that historical factors must be taken into account in analyses of epenthetic consonants, even if those factors play no role in synchronic grammars. In the case of English intrusive /r/, the authors provide a rule-based analysis that includes an /r/-deletion rule ordered before /r/-insertion. The fact that /r/ is the epenthetic segment does not follow from proposals that epenthetic segments are unmarked defaults (e.g. Lombardi Reference Lombardi2002) or that epenthesis minimises perceptual distance between epenthesised forms and their inputs (Steriade's P-map). Rather, the fact that /r/ is the epenthetic segment arises from historical factors connected to /r/-deletion and hypercorrection.

Considering the first set of chapters as a group, they provide strong critiques of markedness on conceptual (Odden, Reiss) and empirical grounds (Vaux & Samuels, Blevins). Though I found the issues raised here to be clear and insightful, it is worth noting that that a number of previous reviews (e.g. Rice Reference Rice1999, Reference Rice and de Lacy2007, Hume Reference Hume, van Oostendorp, Ewen, Hume and Rice2011) and critiques (e.g. Gurevich Reference Gurevich2001, Haspelmath Reference Haspelmath2006) of the concept have been published and widely cited, including in chapters in this volume. For this reason, I was sometimes disappointed by the relative lack of serious engagement with previous work which has attempted to distinguish between the validity of different markedness diagnostics (e.g. de Lacy Reference de Lacy2006, Rice Reference Rice and de Lacy2007), and provide formal models of markedness that make specific predictions (e.g. Rice Reference Rice1996, de Lacy Reference de Lacy2006). Instead, the chapters tend to dismiss markedness as a whole, without acknowledging the range of specific proposals in previous work.

The chapter by Vaux & Samuels is an exception, in that it provides a cogent discussion of cross-linguistic variation in epenthesis patterns, a detailed analysis of English intrusive /r/ and a clear critique of specific proposals about the representation of markedness in the grammar and their potential application to the analysis of epenthetic /r/ in English. Their argument that epenthesis is not a valid markedness diagnostic is well supported, and echoes previous work which has argued against vowel epenthesis as a diagnostic of markedness (e.g. de Lacy Reference de Lacy2006).

Reiss's contribution is the other chapter in this section which provides analysis of a specific dataset. Reiss argues that contrast has no role in phonology. Contrast, like markedness, has been central to a number of theories of phonological representations (e.g. Steriade Reference Steriade1987, Dresher Reference Dresher2009) and operations (Łubowicz Reference Łubowicz2003, Flemming Reference Flemming, Hayes, Kirchner and Steriade2004). Contrast also plays an important role in work arguing that markedness is tied to representational complexity and inventory structure (e.g. Rice Reference Rice1996, Reference Rice and de Lacy2007). Reiss does not attempt to distinguish between different approaches to contrast or consider predictions of specific definitions in the literature. He seems to assume a minimal pair definition of contrast (as proposed by Archangeli Reference Archangeli1988) in which a feature is contrastive for a given segment only if there is another segment in the inventory with which it is identical in all other feature specifications. Although he argues against contrast, Reiss's analysis crucially depends on featural underspecification, a tool widely used by proponents of a significant role for contrast and markedness. In fact, the proposed analysis in which /v/ is underlyingly unspecified for [voice] is similar to previous analyses which argue for contrastive specifications (e.g. Avery Reference Avery1996, Hall Reference Hall, Arnaudova, Browne, Rivera and Stojanović2004). What distinguishes Reiss's analysis from other proposals that make use of underspecification is the use of underspecification without reference to markedness, contrast or phonetic factors. Although the chapter claims to successfully analyse a ‘longstanding puzzle’ (p. 44) concerning the behaviour of /v/ in Russian, idiosyncratic behaviour of a given segment does not seem to be a puzzle in a framework in which underspecification can be attributed to any segment arbitrarily. In Reiss's view, the combinatorial possibilities of binary features allow for segments to be specified with positive or negative values for a given feature or to be unspecified altogether, and no appeal to general principles is required to motivate the specification of any particular segment.

3 Markedness and complexity in Government Phonology

In contrast to the first set of chapters, the contributions in the second section of the volume argue for formal definitions of markedness and complexity within the framework of Government Phonology. Chapter 5, ‘On silent markedness’, is the first contribution in this section. In this chapter, Edoardo Cavirani & Marc van Oostendorp argue not only that empty categories play a role in phonological representations, but also that there are multiple distinct empty categories, with some being more marked than others. The chapter adopts the position advocated in Government Phonology that word-final consonants are onsets to empty nuclei. Differences between different types of final empty nuclei are encoded using a distinction between projection and pronunciation proposed in Turbidity Theory. A skeletal slot projects its melodic content, whereas melodic elements pronounce their associated skeletal slots. The distinction between these relations allows the authors to propose two types of empty categories, one which consists of a skeletal slot with no melodic material and another which consists of a skeletal slot which projects a melodic element but lacks a pronunciation relation. Evidence for this distinction includes data from variation in the patterning of final devoicing across Dutch dialects. In some dialects, final devoicing fails to apply just in those cases where the relevant form is a verb in the first person, e.g. Tilligte /ɣeløːv/ → [ɣeløːf] ‘faith’ vs. /ɣeløːv/ → [ɣeløːv] ‘I believe’. Historically, these verb forms had a final schwa, which has been lost in most varieties. Those dialects which retain final voicing in these forms are geographically adjacent to dialects in which the final schwa is retained as a first person marker. These patterns are analysed by representing standard Dutch forms with a final empty nucleus with no melodic content. In dialects where devoicing fails to apply, the relevant inflected forms end in an ‘empty’ nucleus, which has melodic content but no pronunciation relation. The nucleus with unpronounced melodic content is more complex, and hence more marked, than the empty nucleus which lacks melodic content altogether. This more complex empty nucleus is able to license voicing of its onset, leading to the preservation of final voicing in the first person verb forms. In this view, markedness is defined as structural complexity. Both melodic and prosodic content contribute to structural complexity, as do the pronunciation and projection relations between them, allowing for a range of empty categories which vary in markedness and licensing capacity.

In ‘The phonetic salience of phonological head-dependent structure in a modulated-carrier model of speech’, Kuniya Nasukawa argues for a reassessment of head-dependent relationships in phonological representations. Contrary to widely adopted models, Nasukawa proposes that the head of a phonological constituent is structurally more important, but also less prominent, than its dependent. This leads to analyses in which, for example, the unstressed syllable of a foot is its head and the stressed syllable is a dependent. Arguments for this position come partly from comparisons with syntactic structure in which heads are structurally more important than dependents, but support fewer lexical contrasts, are semantically more impoverished and are phonetically less prominent. Nasukawa argues that another advantage of this approach is that it avoids markedness mismatches where, for example, the unmarked segment schwa occurs in the marked position of a foot in standard models of head-dependent structure.

‘Markedness and formalising phonological representations’, by Shanti Ulfsbjorninn, proposes that markedness can be given a formal definition in terms of representational complexity. According to the strict CV model of phonological representations adopted by Ulfsbjorninn, all syllables have a CV sequence on the skeletal tier. Word-final consonants are necessarily followed by an empty nucleus in this framework, and word-initial vowels are preceded by empty onsets. Focusing on syllable templates, Ulfsbjorninn analyses cross-linguistic variation in syllable complexity as arising from differences in parameter settings. Parameters refer to the licensing of empty categories in various contexts. If a language has more ‘yes’ settings, it will license empty categories in more positions, resulting in more complex sequences of segments in surface forms. Parameter settings are represented using a decision tree structure in which a ‘yes’ setting leads to additional parameters that must be set, whereas a ‘no’ setting renders following parameters irrelevant. If a language has more ‘yes’ settings, it will have a longer decision tree. This length of description is used as a formal definition of markedness.

The chapters in this section advocate abstract representations and definitions of markedness which make reference to representational complexity. Unlike the first series of chapters, these chapters as a group do not reject the concept of markedness outright, although they do address some of the specific critiques of markedness developed in previous literature. Cavirani & van Oostendorp's chapter on markedness of empty categories is perhaps the most abstract, as it argues for a robust distinction between phonetic and phonological objects and a typology of empty categories. Yet this chapter provides evidence in support of its proposals from the domain of dialect variation, and provides insight into how variation in representations can develop through the course of language change. Within an abstract representational framework, this chapter takes seriously the importance of historical change as a mechanism for markedness relations, an issue raised in some of the earlier chapters in the volume. This chapter also provides an illustration of the logical independence of markedness and phonetic substance. The chapters in the first section of this volume tie rejection of markedness to rejection of phonetic substance in the phonological grammar. The definition of markedness used in Cavirani & van Oostendorp's analysis, however, is based on representational complexity, which is not directly tied to phonetic substance. The various empty categories proposed differ in markedness, although they are phonetically alike. Ulfsbjorninn also provides an abstract representational model, one which concurs with the claims in the preceding chapters by Odden, Reiss and Blevins that markedness has no role in the phonological computation, while nonetheless providing a markedness evaluation metric in terms of length of description. Ulfsbjorninn suggests that this markedness metric is not referred to by the phonological grammar, but plays a role in constraining the hypothesis space during the course of acquisition.

4 Biological perspectives on markedness

The final section of the volume takes a different approach, looking at the investigation of markedness from a biological perspective. In ‘Are there brain bases for phonological markedness?’, Mathias Scharinger provides a useful overview of neuroimaging techniques, before summarising results of studies which have used these techniques to investigate phonological markedness. EEG studies have presented asymmetries in Mismatch Negativity (MMN) responses to various speech sounds as evidence in support of featurally underspecified representations (e.g. Eulitz & Lahiri Reference Eulitz and Lahiri2004). Scharinger summarises a number of these studies, and considers alternative interpretations of reported results. Specifically, he considers the possibility that differences in MMN responses follow from markedness, on the assumption that processing of more marked sounds requires greater neural effort. Although underspecification theory is closely tied to markedness in the phonological literature, a direct connection between underspecification and markedness is not assumed in the model of underspecification adopted in the neuroimaging work cited. The markedness hypothesis is consistent with results of some studies, but not with others, although it is not always clear which of the segments being compared is more marked or what theory of markedness such evaluation is based on (a difficulty acknowledged by Scharinger). Scharinger also considers Haspelmath's (Reference Haspelmath2006) suggestion that many phonological markedness effects can be attributed to frequency. Scharinger shows that the frequency hypothesis is not consistent with results of the summarised studies, and argues that markedness and underspecification cannot be reduced to frequency. Scharinger suggests an additional alternative hypothesis and a different conception of markedness in terms of Predictive Coding. According to the framework of Predictive Coding, perception relies on the brain's ability to make predictions which are compared with sensory evidence. Predictions are based on abstract, long-term memory information, and prediction error is determined by mismatches between these predictions and sensory evidence. This perspective suggests that more marked segments are more richly specified, and hence generate stronger, more precise predictions. Deviations from more marked segments will then generate stronger MMNs than deviations from less marked segments.

The conclusion of the volume's final chapter is evident from its title, ‘There is no place for markedness in biologically-informed phonology’. In it, Pedro Tiago Martins considers critiques of markedness from the phonological literature, presenting arguments that overlap with material from the book's early chapters. Looking at markedness from a biological perspective, Martins raises the question of whether markedness is unique to humans or whether it is part of more general cognitive structures which are not necessarily unique to the species. He argues that neither can be satisfactory. Against a human-specific development of markedness, Martins notes that little in biology is truly unique. Evolutionary biology suggests that complex cognitive mechanisms may be built up from cognitive structures that are widely shared across species. Martins is, however, equally sceptical of a concept of markedness which is not specific to humans and language. If markedness is a property shared across species, it must be defined in terms which we can look for outside of the domain of language. This will require a number of more specific senses of markedness to be spelled out. Martins argues that such a redefinition is desirable for a biologically informed study of language, but will leave little intact of the concept of markedness as phonological universals encoded in UG.

The chapters in this final section of the volume differ significantly from the previous chapters in terms of the type of evidence that they consider and the conceptual framework of their inquiry. In their conclusions, however, the chapters show an affinity with previous sections. Martins concurs with contributors to the volume's first section that markedness should be rejected as a meaningful concept in phonological theory. Scharinger's chapter on neuroimaging, on the other hand, resembles chapters from the volume's second section in including a proposal for a markedness evaluation that differs significantly from previous approaches. His suggestion that markedness effects in neuroimaging studies could be reinterpreted in terms of Predictive Coding provides a potential linking hypothesis between brain function and linguistic concepts.