This book presents a new dialect of Government Phonology (GP), which the author dubs the model of Complexity Scales and Licensing (CSL). This particular version of GP was put forward earlier as Cyran (Reference Cyran2003), and is updated and slightly refined in the present book. The central idea extends Charette's (Reference Charette1991) Government Licensing, Kaye's (Reference Kaye, Durand and Katamba1995) Licensing Principle and Harris's (Reference Harris1997) notion of Licensing Inheritance: that nuclei are the ultimate source of licensing in a phonological string and that certain nuclei, due to prosodic position or melodic content, are better licensors than others. At the same time, certain onsets or clusters are easier to license than others. Phonotactic restrictions are (to a large part) the result of the interaction of these two factors.

The book consists of three chapters. In the first two, the basics of GP and CSL are presented; the last chapter discusses data problematic for CSL. Chapter 1 discusses element theory, illustrated with a wide variety of vowel alternations and consonant mutations in Celtic and examples from Polish. The notion of ‘substantive complexity’ is introduced: the more elements a sound contains, the more complex it is. This concept plays an important role later on in the book. Chapter 1 also discusses the idea that certain elements, such as the ‘noise element’ h, are systematically absent (by parameter) in certain languages, although this idea plays only a subordinate role in the remainder of the book. Cyran argues that a number of differences in the consonant systems of Welsh and Polish, e.g. the presence or absence of affricates and the nature of the set of fricatives, can be derived from such parametric differences.

Chapter 2 turns to formal complexity, i.e. constituent structure. After an exposition of the classic GP view, Cyran takes up an observation made in Kaye & Lowenstamm (Reference Kaye, Lowenstamm, Belletti, Brandi and Rizzi1981), viz. that a language with branching onsets will necessarily have branching rhymes, i.e. coda–onset clusters (but not vice versa). He argues that a scale is involved, one that ranges along clusters of different complexity. A branching onset (e.g. pr in apron) is more difficult (‘costly’) to license than a coda–onset cluster (e.g. rp in a rhotic pronunciation of carpet), which in turn is more difficult to license than a single onset (e.g. p in leper). Thus, the observation made in Kaye & Lowenstamm (Reference Kaye, Lowenstamm, Belletti, Brandi and Rizzi1981) can be derived: if in a certain language a particular type of nucleus, say, a full vowel, is able to license branching onsets, the most difficult cluster type of all, then in that language that same type of nucleus will also be able to license a coda–onset cluster, which is less costly. And if a coda–onset cluster can be licensed, then so can a single onset. The reverse implications, however, do not necessarily hold.

Cyran adds to this the idea of licensing strength, which again is argued to be a scalar property: full vowels are better licensors than schwa, and schwa in turn is a better licensor than an empty (unrealised) nucleus. What this means is that if in a given language an empty nucleus is able to license a certain onset/cluster, then so will schwa. If schwa is able to license a certain onset/cluster, then so will a full vowel. Again, the reverse implications do not hold: for example, if a full vowel can license a certain onset/cluster, there is no guarantee that schwa can license that onset.

In summary, there are two different three-level complexity scales, each of which runs on its own axis. Together, they open up a two-dimensional grid of syllable structure typology with nine cells, where one axis represents the licensing strength of nuclei and the other one the complexity of clusters. A language may choose any point in that grid as an arbitrary cut-off point: for example, language L might allow schwa to license coda–onset clusters, but not the more complex branching onsets. From that choice, several other properties follow: firstly, schwa will also license single onsets (which are less complex than coda–onset clusters). Secondly, if schwa can license single onsets and coda–onset clusters, then so can a full vowel. Thirdly, if schwa cannot license branching onsets, then neither can an empty nucleus. Note, however, that one chosen cut-off point does not necessarily determine all properties of the system. In our example, it is left open whether full vowels can license branching onsets. (We only know that they can license single onsets and coda–onset clusters.) It is also undetermined what exactly empty nuclei can license. (We only know that they cannot license branching onsets.)

The notions of cluster complexity and licensing strength are themselves derivative: a branching onset is more complex because it involves a non-local relationship. In cobra the final vowel needs to license the head of the onset b, which it is not strictly adjacent to, and b in turn licenses r. In a coda–onset cluster, on the other hand, the head is adjacent to its licensor: in vulgar the second nucleus licenses the g, which in turn licenses the l. This distinction corresponds to what Charette (Reference Charette1991) calls Indirect and Direct Government Licensing. In this context, substantive complexity from Chapter 1 also makes its appearance: the bigger the difference in the number of elements of the two members of a cluster (Harris's Reference Harris1990 Complexity Condition), the easier it is to license. In other words, the syllabic typology is more fine-grained than the nine-cell typology suggests: certain branching onsets are still ‘easier’ than others, because of their elemental make-up. It is unfortunate that despite the very clear presentation of substantive complexity in Chapter 1, the internal structure of sounds is dealt with only rather cursorily in Chapters 2 and 3. It is therefore hard to assess what counts as easy, an issue which we will come back to later. In the same vein, the licensing strength of nuclei is also claimed to result from their internal structure (elements) and their prosodic position (e.g. within a foot), but again this point is not made entirely clear: for example, schwa, which plays an important role in the establishment of the syllabic typology, can or cannot contain melody (pp. 7ff, 210 et passim), and is sometimes, but maybe not always, defined solely by its location in the non-head of a foot (pp. 215, 217, 271). No such qualification is made for non-schwa vowels (p. 115).

Chapter 2 closes with a long discussion of the status of empty positions in CSL, and raises the issue of whether a CV model, where constituency consists only of a series of strictly alternating consonant (C) and vowel (V) positions, wouldn't be more adequate. Both issues will be discussed at greater length below.

Finally, Chapter 3 deals with two problematic sets of data. The first one is (reconstructed) Late Common Slavic, where various diachronic changes had led to a phonological system that had lost all its coda–onset clusters, but had retained (or even added to) its branching onsets; compare reconstructed Proto-Slavic *párgu to (metathesised) modern Polish próg ‘threshold’. This synchronic state of affairs in Late Common Slavic seems to go against the implicational scale proposed by CSL, viz. that a system possessing more complex structures (branching onsets) must also contain the simpler ones (coda–onset clusters), because Late Common Slavic seems to have the opposite. Cyran argues that the system came about through a change in prosodic strength, in particular the establishment of a trochaic foot. As a result, the complement of the foot (in *párgu) was no longer able to license (any) preceding cluster (and was later lost itself), while the head of the foot (*párgu) was. Metathesis ensued. Given that Late Common Slavic is reconstructed, however, any claims based on it have to be taken with great care, and indeed are speculative. In order to circumvent these issues, Cyran adds data from Irish and Bulgarian which display synchronic variation between forms, mirroring the diachronic stages of Slavic. (One might have wished that the discussion of Irish and Bulgarian was as detailed as that of Late Common Slavic, however.) As the author argues in conclusion (p. 251), the problem for CSL posed by the synchronic system of Late Common Slavic is only apparent: it stems from the methodologically incorrect comparison of the existence of branching onsets licensed by prosodically strong nuclei to the lack of coda–onset clusters licensed by prosodically weak nuclei. In other words, the prosodic position within the word has to be acknowledged in a complete analysis. Throughout the book, Cyran only hints at how the foot could play a role (also for other languages, e.g. Dutch), without going into too much detail. (Prosodic levels higher than the foot are left unmentioned.)

The second set of data taken up in Chapter 3 concerns the notion of binarity. The constituents of classical GP could dominate maximally two positions. This (rightly) predicted that words like clinker should be possible (branching onset cl- plus branching rhyme -in-) while it (wrongly) excluded words like mountain, where the rhyme dominates a long vowel/diphthong (ou) and the coda n, i.e. three positions. Cyran tries to find a solution to that problem in his own model, which does away with branching constituents altogether, a topic to which we will return in more detail below.

Cyran's book presents an interesting development of GP. The style of writing is generally very clear and accessible. Besides presenting the theory of CSL, the author also discusses the major assumptions of GP; thus the book will also be of interest to an audience less familiar with GP. However, there are certain areas where the argumentation could have been more to the point. Very often, the author shows where his model makes the same predictions as GP, instead of investigating the different predictions made by the two models – it remains to be seen if we are really dealing with radically different proposals or rather with notational variants. Let me address two such cases: the claim that all languages are CV, and the treatment of empty positions. (The two are of course closely connected.)

CV models (e.g. Lowenstamm Reference Lowenstamm, Durand and Laks1996, Scheer Reference Scheer2004) assume that there are no branching constituents, and that every phonological string is a sequence of strictly alternating Cs and Vs. The author explicitly argues that that is the case in Polish, but later implicitly extends it to all languages. In his argument, Cyran applies several tests for constituency to Polish words and concludes that there is no difference in behaviour between, say, a branching onset and a sequence of two onsets that contract a left-headed onset-to-onset governing relationship. While it might be true that in the case of Polish the arguments for or against branching constituents are not decisive, it is surprising that the author still concludes that this must automatically mean that branching can be dispensed with. Why not assume that onset-to-onset governing relationships can be dispensed with in Polish? The decisive factor for Cyran seems to be an economy argument (pp. 166f): since branching onsets and sequences of onset–empty nucleus–onset can do the same job, and since empty nuclei are required in phonological theory anyway, we are better off without branching structures. But notice that Cyran still needs to distinguish between two types of onset–nucleus–onset sequence: (i) where the two onsets can see each other and contract a governing relationship, and (ii) where they do not. Sequences of the former type cannot be broken up by epenthesis; sequences of the latter type can. So there is a price to pay for getting rid of branching: the addition of another type of empty nucleus. The empty nuclei of GP were postulated for reasons having to do with alternations and distributional facts. That is, there was always an empirical motivation. Cyran's CSL model (like other CV models; cf. Ploch Reference Ploch and Ploch2003) has to distinguish between two types of empty nuclei, those which are not quite empty (alternating with an audible vowel) – type (ii) above – and those that are ‘really’ empty – type (i). Crucially, the latter ones are claimed to be invisible to the constraint *ø-ø (which excludes sequences of two unrealised nuclei) and do not have any licensing duties to fulfil. In other words, they have no empirical effect, but are posited only because the CV model requires them. Thus it is not clear that having several types of empty nuclei is more economical than having branching.

Matters only get more complex with the abolition of branching rhymes. As a consequence, coda–onset clusters also have to be treated as sequences of independent onsets, and again we have to distinguish between those that can be broken up and those that never can. As a replacement for branching onsets and coda–onset clusters, Cyran suggests that we are dealing with left-headed inter-onset government and right-headed inter-onset government. While it is true that GP also sometimes employed those notions, usually only one type (right-headed or left-headed) was found in a given language, not both. Furthermore, classical GP claimed that real clusters (adjacent consonants) and clusters involving onset-to-onset government (consonants separated by an empty nucleus) were empirically different. For example, in the first case, a sequence of nasal+obstruent has to be homorganic (English), in the second case it does not (Korean). Such a distinction is lost in CSL, and would have to be made up for by other means. Again, it is not obvious that CSL is more economical or empirically more adequate.

A final argument that is made for CV comes from structures with superheavy rhymes, e.g. English count or mountain, where a long vowel (branching nucleus) is followed by a coda–onset cluster. GP excluded such structures and, with very few exceptions (e.g. Harris 1994: 66ff), that incorrect prediction was never really addressed in the literature. Cyran's proposal is thus highly welcome and a treat to read. He presents a very insightful analysis of superheavy rhymes, where he has to reinterpret GP's branching nuclei as nucleus–onset–nucleus and GP's coda–onset clusters as onset–nucleus–onset. Cyran links superheavy rhymes to sequences of schwa and (traditional) coda–onset clusters: in both cases, the cluster following the long vowel/schwa has to be exclusively coronal or s+C (count, joust, but *coump; potent, but *potemp). A long vowel/diphthong is claimed to be a sequence of nucleus–empty onset–nucleus. As shown in the representation in (1) (building on Cyran's (63b) on page 272), the second nucleus in that sequence (dominating x₄) is the dependent, i.e. weak, just like schwa by itself.

1. (1)

   

The parallels in distribution are therefore derived. Cyran argues that in potent the final empty nucleus (x₈) licenses the preceding nt (x₅, x₇). In order to do this, it needs licensing from the next visible nucleus on the left (x₄; x₆ is contained in a cluster and does not count). Since this licensing comes from a weak nucleus (x₄ is either licensed itself, by x₂, or it is inherently weak, i.e. schwa), only an ‘easy’ cluster (exclusively coronal or s+C) can be licensed by x₈ (though the reasons for its easiness remain unexplained). In count, this licensing comes from the second half (weak member) of the diphthong ou, which, it is claimed, is as weak as schwa. While this is an intriguing proposal, it is not clear that it constitutes an argument for CV, as the author claims. The empty onset (dominating x₃) sandwiched within a long vowel/diphthong has no role to play in any of this, and so its presence is not supported by the data under discussion.

Let us now turn to the second area where the predictive power of GP and CSL seems very similar. The author shows (pp. 139ff) that classical GP had several mechanisms to keep empty nuclei silent (proper government, the parameter on final nuclei, magic licensing, licensing by inter-onset government). CSL, on the other hand, is not concerned with keeping positions silent, but rather with the circumstances under which they are interpreted; being uninterpreted is taken to be the default case. In addition, the author claims, CSL makes do with only two mechanisms: (i) an empty nucleus has to surface if it could not otherwise license a preceding consonant (cluster), and (ii) there is a constraint *ø-ø, which excludes sequences of two unrealised nuclei, forcing the first one to be pronounced. Despite claims to the contrary, GP and CSL seem to be very similar in empirical content, and it is not even clear that CSL really uses less machinery than GP. For example, in addition to the two mechanisms mentioned above, the author also assumes that empty nuclei can be ‘locked’, as he calls it, between two onsets, thus emulating, at least to some extent, GP's licensing by inter-onset government: the flanking onsets contract a governing relationship that keeps the sandwiched nucleus silent and invisible to other phonological processes. Likewise, the special status of final empty nuclei in GP is also taken into account in CSL, by a stipulation that final empty nuclei cannot have any floating melody (p. 156).

In addition, mechanism (i) will have to be refined and take into account the prosodic position in the word. As Charette (Reference Charette1991) shows for (Québec) French, empty nuclei fail to be properly governed if they have to license a preceding cluster, e.g. in fortement ‘strongly’, the empty nucleus following rt could in principle be properly governed by the following vowel, but since it has to license rt, it cannot remain silent. The situation is different in word-final position: in forte ‘strong’, the final empty nucleus does stay silent (by parameter) while at the same time licensing the preceding rt. Standard GP kept the two cases apart: one involved (the failure of) proper government; the other was due to the special status of final nuclei. Cyran's CSL will also have to distinguish these cases (as the author says himself throughout Chapter 3) if it wants to be descriptively adequate, but then the machinery dealing with empty nuclei will have to be more complex than it is claimed to be.

Furthermore, it is very unclear what different predictions are made by proper government and the constraint *ø-ø respectively. Polish pies ‘dog (nom sg)’ vs. psa (gen sg) shows a vowel–zero alternation. Both GP and CSL posit an empty nucleus between the p and the s, which in CSL needs to be marked with a floating melody that can surface when required. For GP, the final realised vowel a in pøsa (ø=empty nucleus) can properly govern the preceding empty nucleus, keeping it silent, while in pøsø the final empty nucleus is licensed to remain silent (by parameter) and thus cannot properly govern the preceding empty nucleus, hence we hear a vowel in pies. CSL derives pies by *ø-ø, which is violated in pø_esø (ø_e=empty nucleus with floating melody), so the floating melody has to be realised. (Note that if there are several empty nuclei in a row, all with floating melodies, it will remain unclear why a violation of *ø-ø always leads to the realisation of the first empty nucleus.) In the genitive pø_esa, the empty nucleus stays unrealised, because *ø-ø is not violated; we get psa. In other words, the net total of predicted patterns seems to be the same for proper government and *ø-ø.

I will close with a final note on the notion of ‘scale’: at one point in the book (p. 93), Cyran uses the term ‘quantal region’, which seems more appropriate for the characterisation of his model. It is clear that he treats branching onsets as more ‘costly’ than coda–onset clusters (or rather, the particular replacements of them that he proposes), and that there is an implicational relationship between the two types. It is not clear, though, that we are dealing with a scale, as this suggests that there are certain structures which are ‘still’ coda–onset clusters but ‘already’ quite close to branching onsets, or that there are structures that are halfway between ‘prototypical’ branching onsets and coda–onset clusters. This is not what Cyran has in mind. He does speak of ‘good’ or ‘bad’ clusters of a certain type, but (i) this does not guarantee that there is a scalar continuum, and (ii) the notions ‘good’ and ‘bad’ remain unclear, given the lack of detail after Chapter 1 on elemental make-up already alluded to above: for example, the members of tl are at least as far apart from each other as those in pl, yet play exists in English while *tlay does not. Cyran, unlike GP, treats them as the same kind of structure and invokes a ban on homorganicity as the reason for the ungrammaticality of *tl. Likewise, *kn is listed as ‘bad’ and kl as ‘good’, but the reader cannot assess this, as the internal structure of l and n (their substantive complexity) is left somewhat unclear. Note in addition that here we have a case where GP and CSL truly differ in predictions: Cyran treats kn and kl as the same kind of structure, while classic GP argued that the former is a coda–onset cluster, while the latter is a branching onset. Differences like these are left unaddressed.

CSL holds promise for tying together various aspects of phonological structure, connections that before had often only been hinted at but not developed as fully as in the present book. Cyran's contribution is therefore a more than welcome addition to current phonological research.