Recursion is a key concept in both the organisation and the origin of language, claims Kuniya Nasukawa, the editor of this collection of papers (p. 1). If this is indeed the case, recursion ought to be found not only in syntax, but also in phonology. Phonologists are divided on whether phonology ‘proper’, i.e. the structure of syllables and segments, in fact involves recursion. The papers presented in this volume all address this question in greater or lesser detail. With one exception, Clemens Poppe, all authors conclude that phonology does involve recursion.

Half of the papers were first presented at a workshop entitled ‘Recursion in phonology’, which was held in 2016 at Tohoku Gakuin University, Sendai; the others were written for this volume. The ‘morpheme-internal’ in the title of the volume is revealing and relevant. Recursive structures have been applied to rhythmic patters in phonology before, but Scheer (Reference Scheer2011) and Nasukawa (Reference Nasukawa, van Oostendorp and van Riemsdijk2015) argue that these structures reflect (morpho-)syntactic rather than phonological structure. Most of the discussion in the present book is restricted to recursion within morphemes, where morphology and syntax have no role to play. The aim of the volume is to initiate a debate by taking a firm stand on the side of those arguing for the existence of morpheme-internal recursion in phonological representations.

The book comprises an introduction by the editor and twelve papers by sixteen authors. Most of the papers discuss phonotactics and the representation of vowels and consonants, and of syllable structure. Three of the papers consider stress at some length, and one discusses the influence of adjacent consonants on tone. However, in most papers these topics are intertwined: tone and the laryngeal specification of consonants are intimately related issues, as are vocalic elements and the place specification of consonants, and vowel complexity and stress. It is hard, if not impossible, to set up any thematic classification. The papers of the volume are arranged in alphabetical order; I will follow this order below.

Phillip Backley & Kuniya Nasukawa (pp. 11–36) discuss the representation of both vowels and consonants. They assert that prosodic units like the nucleus or the syllable are projections of melodic elements, so that a word is ultimately a projection of one of its vowels. A polysyllabic word thus involves several layers of recursive structures: one vowel may be a dependent of another vowel. Furthermore, several instances of the same phonological element are claimed to participate in the make-up of a single vowel. Another syntax-like feature of their framework is the assumption that it is the dependents that contribute linguistic information; heads are mostly just structural elements of the representation. A consonant is therefore a dependent of a following vowel, and, somewhat unexpectedly, a stressed vowel or syllable is a dependent of an unstressed one.

Edoardo Cavirani & Marc van Oostendorp (pp. 37–56) propose a maximally radical reduction of the number of vocalic elements from three (A, I, U) to zero. Similarly to the replacement of the element A by a specific structural configuration advocated by Pöchtrager (Reference Pöchtrager2006), the other two elements, I and U, are here also replaced by structure. In earlier models in which elements were replaced with structure, the aim was not only to reduce the number of elements (thereby also cutting back their combinatorial possibilities), but also to explain why certain assimilations are not attested. In the current proposal, the idea that structure cannot spread must be abandoned. Cavirani & van Oostendorp argue that elements are not atomic, but can be broken down into even smaller components. After all, modern physics has taught us that atoms themselves are not atomic.

Marcel den Dikken & Harry van der Hulst (pp. 57–115) contribute the longest and perhaps most theoretically ambitious paper of the volume. In it they contend that phonology and syntax are basically the same linguistic module, differing only in that they operate on different categories: the former on vowel (V), the latter on verb (V), for example. Den Dikken & van der Hulst assign a phonological role to the notion of ‘light v’, claiming that the so-called ‘lax’ vowels of some Germanic languages (Dutch and English are discussed in the paper) are spelled out in this node, while ‘tense’ vowels are spelled out in the ‘traditional’ V node. It remains to be seen how the distinction between the two vocalic nodes translates to languages that do not make use of a tense–lax contrast in their vowel system.

Chihkai Lin (pp. 117–142) discusses glide formation in Japanese, Korean and Tsou, a Formosan language. The analysis uses recursively structured representations of segments; however, it does not become clear why this offers advantages in comparison to non-recursive accounts of the phenomena discussed. Various parts of the analysis remain unconvincing. To cite just one case, in the change from [oi] to [we] in Korean, the dependent A and the head U must reverse dependency. However, no explanation is given for why this should be the case, apart from the fact that this is how the analysis works out (p. 137f).

Xiaoxi Liu & Nancy C. Kula (pp. 143–180) examine consonants that act as tone depressors, i.e. consonants which interact with vowels, lowering the tone of those vowels. The authors also use Element Theory, which inherits the element L representing both obstruent voicing and vocalic low tone from classical Government Phonology (Kaye et al. Reference Kaye, Lowenstamm and Vergnaud1990). In later work (e.g. Nasukawa Reference Nasukawa2005), the scope of L was extended to nasality. In this light it is surprising that plain nasal consonants are not depressors, while in some cases voiceless, voiceless aspirated and even breathy nasals may be. Liu & Kula propose representations for laryngeal categories that explain the unexpected depressor effects.

Filiz Mutlu (pp. 181–205) proposes consonantal representations that involve recursive structures. She sets up a strength scale (similar to the standard sonority scale), and claims that a consonant may be embedded in another, stronger consonant, producing what we call a consonant cluster. Mutlu makes some interesting predictions, but is rather inaccurate both on data and on philological detail. We read that ‘nasals assimilate to following obstruents’ (p. 182), but it turns out that only plosives are intended. Nasals in English, for example, keep their place contrast before a fricative (info vs. nympho, Kansas vs. crimson). The claim that affricates do not occur before plosives (p. 184) also has to be refined: while [ʦp] does not appear to occur in some languages, [ʦk] does occur in, for example, Hungarian ([pɑlɑʦk] ‘bottle’). Similarly, Polish provides several examples of word-final fricative + affricate clusters ([deʂʦ̢] ‘rain’, [ɡoɕʨ] ‘guest’), as well as fricative + nasal clusters ([taɕm] ‘ribbon-gen.pl’, [awtizm] ‘autism’), which are claimed to be impossible. Phonotactics is far too complex to describe in such general terms. Mutlu mentions that ‘|I| and |U| both have positive charm’ (p. 183). However, in Kaye et al. (1990: 202), the source she cites for charm theory, these two elements are charmless. In an earlier paper, Kaye et al. (Reference Kaye, Lowenstamm and Vergnaud1985) assign negative charm to these elements (this paper is not referred to in the text, although it does appear in the references).

Hitomi Onuma & Kuniya Nasukawa (pp. 207–236) analyse velar softening in English. This is a rather surprising endeavour in the Government Phonology-like framework assumed by the authors, since velar softening is not a phonological process in English in a Government Phonology approach (Kaye Reference Kaye, Durand and Katamba1995). It is a sound change that took place more than fifteen centuries ago in another language, Late Latin. Even more surprisingly, the analysis appears to take the spelled forms of words to be their underlying representation: ‘two identical consonants (in the spelling) become a singleton in speech production’ (p. 208; emphasis mine), ‘su[kk]ess → su[ks]ess’ (p. 209). That velar softening in English has nothing to do with phonology is demonstrated by the fact that the authors find it rather difficult to explain why [g] is ‘softened’ to [ʤ], rather than the expected [z]. The trigger of velar softening is the unnatural class [ɪ ə iː aɪ]. The authors gloss over this important difficulty of their analysis by claiming ‘as the spelling suggests, they [the different triggers] were presumably pronounced -[ɪ]ty, -[ɪ]zm and -[ɪ]se/[ɪ]ze at some earlier stage in the history of English. … There seems to be a consensus that /ɪ/ is the lexical form of the initial segment of the suffixes’ (p. 210; emphasis mine). It is not clear what ‘lexical form’ means in the context of this paper, but in any case the present-day [aɪ]'s which cause velar softening were never short [ɪ]'s in the history of English. Note also that the example of success suggests that the trigger may also be [e]; however, it may not be [eɪ] or [eə].Footnote ¹

Markus Pöchtrager (pp. 237–266) considers the nature of recursion, arguing that it is not restricted to just the involvement of a category within another instance of the same category. Rather, recursion is present in any phonological representation with hierarchical structure. The paper is a clear introduction to the framework known as GP 2.0, first introduced in Pöchtrager (Reference Pöchtrager2006). The importance of non-arbitrariness is emphasised: for any alteration, there must be a reason for why it happens where it happens. Pöchtrager explains the advantages of replacing the element A with structure (an approach which Cavirani & Oostendorp take to its extreme, as we have seen), and how this accounts for asymmetries in English and Putonghua diphthongs. The chapter also considers why there should be a limit on the degree of phonological recursion (or syntactic recursion). I found this to be the best paper of the volume.

Clemens Poppe (pp. 267–306) also discusses the representation of vowels. Based on data from one obsolescent and one obsolete system, English Received Pronunciation (RP) and Middle Korean, he argues that both dependency relations and multiple copies of an element are needed to analyse these vowel systems. However, no more than two copies are necessary, so ultimately the framework described in the paper does not include recursion proper. For RP, Poppe adapts Backley's (Reference Backley2011) inventory, which categorises vowels based on a hybrid transcription convention, in which vowels contrast both in length and in tenseness vs. laxness. Poppe argues that the basis of the contrast is tenseness vs. laxness, since length is not contrastive in all varieties of English (and Dutch). This is a valid argument only if all these varieties of English (and Dutch) are to be derived synchronically from the same underlying inventory. It is not obvious to me that this is necessary, or for that matter, desirable. My other main objection is that the labels ‘tense’ and ‘lax’ for the two relevant sets are not justified in any way. Backley's ‘long’ and ‘short’ at least map onto the transcription symbols, even if the two sets do not form phonological natural classes. But for some of Poppe's ‘tense’ vowels (e.g. /ɔ/) there is neither a phonological justification nor a phonetic reason for assuming they are tense.

Geoffrey Schwartz (pp. 307–327) aims at providing a representation of both segments and syllable structure, and of other prosodic elements of speech, like stress. My impression is that the current state of Schwartz's theory does not warrant all of the general claims made: there are too many ‘maybes’. The explanation for why the second stress is more prominent in ˌCindeˈrella, when the word is pronounced phrase finally or in isolation, leaves open the question of why the first one is more prominent in ˈCindeˌrella's ˈgodmother, as per the English Rhythm Rule. Likewise, I feel that the explanation for why holi is more prominent than day in ˈholiˌday will necessarily be ad hoc when there are plenty of counterexamples in both directions: ˈanˌcestor, ˌpresent-ˈday, etc. A less ambitious paper might have been more convincing.

Ali Tifrit (pp. 329–382) looks at the behaviour of obstruent + liquid clusters. His main claim is that, as opposed to obstruents, liquids must find a host in the phonological representation, and this host is typically an obstruent. Thus, Tifrit argues, a number of phenomena that characterise obstruent + liquid clusters can be explained. The summary of ‘wandering’ Sardinian liquids is truly intriguing. Some parts of the analysis remain unclear though (to me at least). Why, for example, is a liquid in an unlicensed O‴ voiceless in French (p. 359), if French does not exhibit word-final devoicing elsewhere? If word-final consonants (all unlicensed) are not voiceless, this seems to be a claim the reader simply has to believe. What is the representation of a voiceless sonorant? The development of word-final aspiration in the same context is perhaps even more mysterious.

Hisao Tokizaki (pp. 383–408) discusses stress assignment in English. He argues that when a set of elements is merged with a terminal, the set will carry the label ‘Strong’. Unexpectedly, while syllable weight is defined in §4.2 exactly as in strict CV analyses, e.g. Scheer & Szigetvári (Reference Scheer and Szigetvári2005) or Ulfsbjorninn (Reference Ulfsbjorninn2014), no reference is made to this approach to phonological representations. It is rather ambitious to expect that a single example, visit Disneyland, will illustrate that all predictions of the theory about the location of word stress, compound stress and phrasal stress are correct (p. 403). The alternation of the prominence of stress (or accent) in cases like ˌJapaˈnese vs. ˈJapaˌnese ˈgarden (the Rhythm Rule, once again) would have deserved some discussion.

The typographically sensitive reader will find the book painful to read, but any reader will often wonder about the many oddities, typos and mistakes. The typesetter has often failed to set the authors’ symbols correctly. Some of these are familiar mistakes; for example, a hard hyphen is found where a soft hyphen is required (pp. 69, 107), a hyphen is used instead of a minus sign (p. 182), an opening single quote occurs instead of an IPA stress mark (p. 356) and a closing single quotation mark occurs where a prime is required (p. 357) (although the diagrams on the same page have the correct symbol). The typesetter has also changed symbols for the worse: for example, in (11) on page 65 we find the digit ‘1’ instead of letter ‘l’; furthermore, ‘w’ in this example is also mistakenly capitalised. I have seen the draft version of this paper, which did have the correct symbols. The rows in (7) on page 335 were shifted during typesetting, making this table more of a puzzle than a display of data. And there are scores of other such blunders.

It is certainly not a reviewer's task to investigate where the responsibility for such mistakes lies. However, it is clear that the text which ultimately went to press is not what the volume's editor or contributors had intended. The authors of the papers in this volume certainly deserve a better product than what they got.