1. Overview

As linguists, whether generativist or not, we cannot ignore the influence of generative grammar in contemporary linguistics. Thus, continuing to think about its core thesis – the claim that the human species is equipped with Universal Grammar (UG) – is essential for any practising linguist. In the current book, Archangeli and Pulleyblank do precisely this, from the perspective of phonological theory. These authors have published many influential articles and books using UG-based analyses (e.g., Archangeli & Pulleyblank Reference Archangeli and Pulleyblank1994, among many others), but starting around 2006 (e.g., Pulleyblank Reference Pulleyblank2006), they started addressing to what extent we can do phonology without recourse to UG. Based on the accumulation of their work since then, this book presents their (interim) conclusion – we can, and in fact should, analyse phonological patterns without UG. Instead, they propose that phonological regularities emerge as a consequence of general cognitive capacities, hence the title of the book Emergent Phonology.

The claim about UG in my view can be decomposed into three sub-claims: (1) human beings are born with some sort of ability to acquire language, (2) that ability is specific to the human species, and (3) that ability is specific to language, being to a large degree independent of other cognitive modules. In this book, the authors clearly reject the third claim, and they are sceptical about the second claim as well. Of course, the authors are not claiming that non-human animals can potentially acquire human languages; they instead refer to studies which show that abilities that have been considered as prerequisite for language acquisition are not specific to the human species (e.g., pp. 11, 55). My reading of the book is that they are still willing to accept the first claim.

Although it is not mentioned in the book, this attempt appears to be in the same spirit as the Minimalist Program in syntax (Chomsky Reference Chomsky1995 and subsequent work), which attempts to minimise the content of UG as much as possible. The conclusions are rather different, however. Archangeli and Pulleyblank, albeit tentatively, conclude that no phonology-specific UG is necessary; the Minimalist Program still posits Merge as an – and perhaps the only – operation made available by UG.

The book is not very long, consisting of 148 pages without the references and the indexes. It comes with a rich list of references, which is 26 pages long, along with useful indexes by name, language and subject (although readers can also get a free, searchable PDF copy of the book). The book is overall clearly written, but readers should be aware, as is stated by the authors, that the meanings of some technical terms are used to mean something different from their traditional uses. For example, the term ‘optimising’, which has a specific meaning in Optimality Theory (OT), is used to refer to cases in which the choice of a particular form, among other possibilities, ‘is determined by phonological properties’ (p. 73). The book has many ‘boxes’ that highlight important points and lay out technical details and comparisons with previous theories such as OT.

The book is probably too advanced for undergraduate students, as it assumes basic familiarity with previous influential theories such as OT and Exemplar Theory (Pierrehumbert Reference Pierrehumbert, Bybee and Hopper2001). However, graduate students who have completed their introductory phonology course would benefit hugely from reading this book, as it will give them an opportunity to rethink if – and more importantly, why – we need to posit the mechanisms that are introduced in such introductory courses. Grown-up researchers would likewise learn a lot from reading the book. It is unlikely that a reader will agree with every aspect of the book. But it is precisely this characteristic of the book that makes it readable: it forces us to pause for a moment and rethink whether and why theoretical postulates that we posit in phonological analyses are necessary. I would particularly like to invite practising phonologists to take a look at Box 3.6 (p. 40), which presents a comparison between the authors’ proposal and OT, a point that I will come back to below.

2. Summary of chapters

Let me now proceed to a brief summary of each chapter. Chapter 1 lays out the authors’ conceptual foundation, addressing the question of whether we need a language-specific acquisition device that is separate from other cognitive modules in order to learn phonological regularities, a question the authors answer in the negative. They argue that abilities to (1) attend to language, (2) remember, (3) identify similarities, (4) process sequences, (5) attend to frequency, (6) form categories, (7) generalise and (8) generalise over generalisations (and possibly other domain-general mechanisms) suffice to acquire phonology. Chapter 2, using data from Yangben, illustrates how learners equipped with these abilities are able to identify phonologically relevant ‘partitions’ (roughly corresponding to ‘natural classes’ in more traditional phonological analyses). The authors then present how learners may acquire the phonotactic restriction in terms of tongue root position observed in this language. Word-final devoicing of low-toned high vowels is also discussed in this chapter.

Chapter 3 turns to what we would call ‘morphophonological alternations’ in more traditional frameworks, specifically tongue root position harmony and rounding harmony patterns observed in Yangben. In this chapter, the authors define their critical notions such as morphs (roughly corresponding to ‘morphemes’), morph sets (roughly corresponding to ‘morphologically related paradigms’) and morph compilation (roughly corresponding to ‘morpheme concatenation’). The authors are very careful not to use the traditional terms, to highlight the fact that they are not relying on the traditional analytical tools. This chapter also introduces wellformedness conditions, which are similar to markedness constraints used in the OT literature, and assessment tables, which resemble OT tableaux – a visual illustration of how conflicting, violable and ranked wellformedness conditions work to select actual, observed forms. Two additional principles – Morph Set Relations (MSR), a statement of correspondence among morphs in a single morph set, and Morph Set Conditions (MSC), a statement of the productivity, if any, of the related MSR – are introduced in this chapter.

In Chapter 4, the authors challenge one of the fundamental concepts that has been shared by most modern phonological theories, namely the postulation of underlying representations (or inputs in OT), a principle that states that whenever possible, analysts (linguists and language learners) should posit a single form per morpheme. (This book is not the first proposal to eliminate underlying representations from phonological theory. See Hyman (Reference Hyman2018) for a review and a response to such proposals.) In this traditional view, when there are two realisations of a single morpheme, [X] and [Y], the relationship between [X] and [Y] is captured indirectly by virtue of these two forms sharing the same underlying representation /Z/. The authors instead argue that systematic relationships between [X] and [Y] should be more directly captured at the surface level by MSR. A simple but compelling argument is that in traditional analyses, it is natural and straightforward to posit /-d/ as the underlying representation of the past tense morpheme in English and derive three distinct surface forms ([-t], [-d] and [-əd]) by means of phonological rules or constraint interactions, but it is not possible to posit a single underlying form for the verb go, which can be realised as /goʊ/ or /wɛnt/; why learners should posit a single underlying form for the former, but not the latter, appears to be rather arbitrary. The book instead proposes that [goʊ] and [wɛnt] constitute a morph set, with their relationship being regulated at the surface level, and the same mechanism holds for [-t], [-d] and [-əd]. In this chapter, the authors also present a historical review of how the notion of underlying representations has arisen in the history of phonological thinking, which itself is very readable.

Chapter 5 discusses the consequences of their main proposal, especially that of eliminating abstract underlying representations. It presents five case studies, starting with the complementary distribution of stops and their continuant counterparts in Warembori. The second case study is a pattern that is probably familiar to all phonologists: nasal place assimilation in English, as observed in the behaviour of the prefix in-. The focus is the question of why other prefixes such as un- and non- do not undergo the same assimilation. The answer provided in the book, in short, is that for these prefixes, forms with nasal place assimilation are simply unavailable, that is, not a part of their morph set. This analysis highlights a clear difference between the current proposal and OT, which claims that all potential candidates should be considered in phonological analyses. Three other cases discussed in this chapter are ternary suffixal distinctions in Mayak, tonal patterns in Kinande and Polish yers.

Chapter 6 provides a summary of each chapter, and touches upon various topics such as directionality, non-iterativity, saltation, opacity (§6.1), syllables, stress (§6.2), language change, multilingualism, reverse engineering and frequency (§6.3). Discussions in this chapter tend to be brief, and readers are often directed to other papers written by the authors and their students. Many questions are raised but are left for future research.

3. Commentary

Now I would like to move on to more subjective experiences that I had while reading through the book. As somebody who has engaged in OT-based research, I took particular interest in Box 3.6, which summarises the major differences between Emergent Phonology and OT, and I imagine that this will be so for many readers of this journal. The four major differences are: Emergent Phonology does not posit (1) Richness of the Base, (2) Gen, which generates an infinite number of candidates, (3) faithfulness constraints or (4) Con, a set of universal constraints. While the consequences of these differences are not fully explored in the book, it is a great exercise to think carefully about what these differences actually mean.

First, Richness of the Base was posited to avoid the so-called ‘duplication problem’, a problem that the same grammatical restriction has to be stated twice, once in the phonological component and again in the lexicon. Would the duplication problem be a genuine problem for Emergent Phonology, and for that matter, phonological theories in general? See Paster (Reference Paster2013) and Gouskova (Reference Gouskovato appear), who both argue that the duplication problem is not necessarily a problem; that is, it is necessary to state constraints both at the lexical and grammatical level. Second, given that the base is not rich, how does Emergent Phonology model how speakers produce nonce words? Japanese speakers, for example, spontaneously palatalise [s] before [i] in loanword pronunciations (Ito & Mester Reference Ito, Mester and Goldsmith1995), but how does Emergent Phonology account for this observation? Third, Emergent Phonology does away with faithfulness constraints because it eliminates underlying representations. However, Correspondence Theory (McCarthy & Prince Reference McCarthy and Prince1995) – one instantiation of faithfulness constraints in OT – was proposed not only to model input–output relationships, but also to capture similarities in, for example, base–reduplicant mappings and output–output mappings, and for that matter, relationships between rhyming elements (Holtman Reference Holtman1996). Are these observed similarities real? And if so, should a phonological theory capture the similarities? Fourth, the universality of Con makes OT an inherently typological theory – positing a constraint in the analysis of one language forces us to commit ourselves to make it a claim about all other languages. Is this property something that we are willing to let go of? The book opens up these interesting questions to be discussed in future conversation.

Another set of questions arose as I read through Chapter 3, primarily regarding whether the assessment tables – one of the central analytical devices in Emergent Phonology – can be considered to be truly domain-general. Assessment tables consist of ranked, violable wellformedness conditions, and their function is to select an observed set of forms from several possibilities; they are thus very similar to OT tableaux. One specific question that I had was where these wellformedness conditions come from. The proposal in the book is that learners can deduce these conditions by observing actual phonological regularities in the language, perhaps using domain-general mechanisms like maximum entropy, as in Hayes & Wilson (Reference Hayes and Wilson2008) (p. 24). A contrasting view would be to posit that such constraints are innate as in OT or that such constraints are deduced from learners’ phonetic experiences (Hayes Reference Hayes, Darnell, Moravscik, Noonan, Newmeyer and Wheatly1999). It is worth noting that the book deploys a specific schema for formulating wellformedness conditions (p. 44), implying that there are restrictions on what kinds of wellformedness conditions can exist (see McCarthy Reference McCarthy2003 for a similar proposal in OT). Can we, then, take this restriction as evidence for UG? (Although I note that the authors argue that the proposed schemas are ‘a formal means of representing the logical result of acquiring a phonological grammar under Emergence’ (p. 138).)

Another specific question that I had was whether all wellformedness conditions – or constraints and constraint rankings – can be learned from the ambient data or not, which has been a widely discussed topic in OT research. One argument for the latter position that came to my mind is the emergence of the unmarked effect (McCarthy & Prince Reference McCarthy and Prince1994) in L2 phonology; for example, L2 learners whose native language does not have codas sometimes show a spontaneous pattern of coda devoicing (Broselow et al. Reference Broselow, Chen and Wang1998). Similar patterns are observed in the context of loanword adaptation patterns and loanword phonology; for example, while Japanese native phonology prohibits both [si] and [ti], Japanese speakers are more prone to palatalise [si] than [ti] (producing, e.g., [ʃiti], but not [sitʃi], for city; Ito & Mester Reference Ito, Mester and Goldsmith1995). The difference between [si] and [ti], as far as I know, is something that is not learnable from phonological regularities in the native lexicon of Japanese.

The final question was where the decision-making procedure comes from. Assessment tables, like OT tableaux, instantiate a specific mechanism to decide on a ‘winner’ given a set of possibilities and a set of wellformedness conditions. It is not immediately clear whether this can be achieved via a truly domain-general cognitive mechanism, although I would like to point out that the decision mechanism proposed in the book very much resembles that of a heuristic decision-making procedure discussed in the psychology literature, described in detail by Gigerenzer & Gaissmaier (Reference Gigerenzer and Gaissmaier2011).

To conclude, even if one is not convinced by some specific aspects of the book, I believe that its most general message, quoted below, is something that every practising linguist should bear in mind:

For this reason alone, this is a book that is worth serious attention.