This book on the assimilation and dissimilation of consonants presents the Surface Correspondence Theory of Dissimilation (SCTD), which aims to show that assimilation and dissimilation can be represented by means of the same unified framework, though there is a systematic mismatch in the way the two processes apply. The theory is based on previous work that showed how assimilation could be explained in terms of correspondence between consonants that share certain features combined with agreement for certain other features for consonants that correspond. The innovative aspect of the SCTD is to extend this idea to providing a model for long-distance dissimilation as well as assimilation. It uses Optimality Theory (OT) to show how the SCTD can represent patterns of assimilation and dissimilation in a wide range of different languages spoken in Africa, such as Kinyarwanda, Obolo and Zulu, in Central and South America, including Quechua and Chol, the Austronesian language Sundanese, the Polynesian language Ponapean, the Australian aboriginal language Yidiny, the Kartvelian language Georgian, and also Latin.

There are ten chapters in this book. Chapters 1 and 2 present an outline of the theory. Chapter 3 shows how the SCTD models both the assimilation of sibilants within the stem and the dissimilation of voicing in prefixes in Kinyarwanda. In Chapter 4, assimilation and dissimilation are analysed for the plural /ar/ infix in Sundanese which sometimes gets pronounced as [al]. Chapter 5 applies the SCTD to the selection of [h] or [ʔ] as an epenthetic consonant at the start of words in the Cuzco variety of Quechua and also to nasal assimilation in Obolo. In Chapter 6, the complete identity effect is analysed for Chol, in which ejectives in the root must be identical for both place and manner, and sufficient identity effects are discussed for Ponapean, in which plain labials and labio-velarised labials do not occur in the same root. In Chapter 7, it is shown how the SCTD can model patterns of dissimilation in Zulu, in which labial consonants in the root become palatals when there is a passive /-w/ suffix but root-initial consonants are exempted from this dissimilation, and it is argued that an alternative approach, based on the Obligatory Contour Principle (OCP), cannot successfully model this phenomenon. Chapter 8 shows how the SCTD models /l/ to [r] dissimilation in Yidiny, in which an earlier /r/ in the root of the word prevents the dissimilation occurring, and /l/ dissimilation in Latin, in which the /-alis/ suffix dissimilates to [-aris] under some circumstances. However, it is admitted that the SCTD fails to model dissimilation properly in Georgian, in which the dissimilation of /l/ to [r] after a previous /l/ is blocked by an intervening /r/ but not if the /r/ occurs earlier in the word. In Chapter 9, the typology of dissimilation and predictions of the SCTD are considered in a survey of over 130 languages. Finally, Chapter 10 provides a brief summary.

In one respect, the representation is unnecessarily complex and also inconsistent. In the first two chapters, correspondence is shown using numbered subscripts; but in fact in no case is more than a single correspondence chain considered. In other words, in the analyses presented, consonants either correspond or they do not, so there seems to be no need to number them. For example, on page 67, the surface correspondence for the Latin word plumalis ‘feathered’ is shown as [p₁l₂um₂al₂is₃] to indicate that [l m l] are in correspondence while the initial [p] and final [s] are not, something that is also redundantly shown by means of R: {p}{l m l}{s}. However, the subscript notation is abandoned from Chapter 3 onwards. For example, when dissimilation for Latin is reconsidered in more detail in Chapter 8, the correspondence for legalis ‘legal’ is shown on page 295 quite adequately as R: {l g l}{s}, and no numerical subscripts are used. It is not clear why the superfluous subscripts are adopted in Chapters 1 and 2 but not elsewhere, especially as the R notation is perfectly clear. And, if subscripts are to be used, it is not clear why multiple numbered correspondence chains are shown when simple use of 1 and 0 would suffice.

While the representation of assimilation and dissimilation in terms of the same framework is certainly an attractive goal for phonology, it results in some quite contorted derivations. For example, let us consider the alternation (discussed on pages 66–68) between the Latin words lunaris ‘lunar’ and plumalis ‘feathered’. Both these words have the -alis suffix, but dissimilation changes the second /l/ to [r] in lunaris while there is no such dissimilation in plumalis, because the intervening /m/ blocks dissimilation from occurring. This analysis, in which coronals such as /n/ allow dissimilation but non-coronals such as /m/ block it, is reasonably straightforward. Now let us see how it is represented in the book. The SCTD suggests that the two tokens of /l/ must be in correspondence because they are both laterals in the stem, but correspondence can only occur in neighbouring syllables, with the result that the intervening /n/ in lunaris has to be recruited as part of the correspondence chain. However, the /n/ cannot be in correspondence with /l/ because it is [–lateral], so the optimal candidate is the one in which the second /l/ changes to [r]. Once the second /l/ is changed to [r] in the output, there is only a single lateral and therefore no need for correspondence between the consonants, so only the lower ranking faithfulness constraint is violated. What about plumalis? In this case, because /m/ is [–coronal], it is unspecified for laterality, so it is not prevented from participating in the correspondence between the two tokens of /l/. Given that the two /l/s can recruit the help of /m/, there is no problem with them being in non-neighbouring syllables, so they can both surface faithfully as /l/. This seems to be a somewhat convoluted explanation for what initially appears quite a simple alternation. Furthermore, as is admitted on page 300, the SCTD fails to explain why solaris ‘solar’ does not undergo dissimilation, as the model predicts that *[solalis] should be the optimal output.

Another case for which the derivation seems somewhat convoluted is the dissimilation proposed in Chapter 5 for the selection of an epenthetic consonant at the start of words in the Cuzco variety of Quechua. Again, the pattern is fairly straightforward, with [ʔ] being inserted at the start of words with an initial vowel (e.g. /asikuj/ ‘to laugh) → [ʔasikuj]) unless the word contains an ejective, in which case [h] is inserted (e.g. /ajk’a/ ‘how many’ → [hajk’a]). The SCTD suggests that this occurs because both [ʔ] and [k’] are [+constricted glottis] ([+c.g.]) sounds, and two [+c.g.] sounds in the root of a word are required by Corr-Root·[+c.g.] to correspond but prevented by CC·Edge-(σ) from corresponding if they occur in different syllables, so the optimal solution is for the epenthetic [ʔ] to dissimilate to [h] to avoid any need for correspondence. The problem with this is that correspondence never occurs for any word in Quechua, as is confirmed on page 178, where it is stated that ‘the effect is a blanket prohibition of correspondence’. In other words, Corr-Root·[+c.g.] – the constraint that enforces correspondence – only exists to prevent itself ever occurring, like some phonological Cheshire Cat whose sole purpose is to leave behind a ghostly impression of itself. It seems a bit bizarre to propose something which never actually occurs, though of course that does not make it wrong, and it is certainly true that a uniform model for assimilation and dissimilation is attractive.

The SCTD analysis seems to make more sense when patterns of assimilation and dissimilation are shown to occur in the same language, especially when the correspondence requirements for assimilation are complementary to those for dissimilation, as is shown for Kinyarwanda in Chapter 3 and Sundanese in Chapter 4. However, while the unification of assimilation and dissimilation proposed by the SCTD is quite elegant, at least for these two languages, the analysis is always highly complex, with multiple OT tableaux to show how the various constraints for correspondence and faithfulness interact to derive the expected phonetic output, and one problem here is that sometimes the contents of the tableaux and also the text are flawed. For example, right at the start of the book, in the explanatory tableau (4) on page 3 to show how a hypothetical input /bap/ might be dealt with, it is suggested that the output candidate [b₁ab₂] is unfaithful for labiality as well as voicing, but it is hard to see how this candidate can be unfaithful for labiality when both the input and the output have initial and final labial consonants; and similarly the final row of the same tableau suggests that the candidate [b₁ak₁] is unfaithful for voicing, which again is hard to understand as both the input and the output have a voiced initial and voiceless final consonant. (Or perhaps I misunderstood something.) In fact, in the first 100 pages of the book, I found the following problems: on page 36, mention is made of candidate (8a) [t₁. . .t₁] and (8b) [d₁. . .d₁], but candidates (8a) and (8b) are actually [t₁. . .t₂] and [d₁. . .d₂] (i.e. with no correspondence between the consonants); on page 42, in discussing the tableau in (18), it is stated that the candidate in (18b) has a sibilant that does not correspond with the other sibilants [ʂ ʐ] and does not incur any violations of Corr-Stem·[+sibilant], but in fact in (18b) [z] does correspond with [ʂ] and furthermore it has a single violation of Corr-Stem·[+sibilant], and one suspects that the text should refer to (18d) rather than (18b); on page 79, in discussing three previous approaches to linking dissimilation with assimilation, it is stated that ‘all three pair dissimilation of a feature to dissimilation for the same feature’, and one assumes that one of these tokens of ‘dissimilation’ should be ‘assimilation’; on page 85, reference is made to two alternatives for the structure of Kinyarwanda words, ‘the simpler structure in (8)’ and ‘the full structure in (9)’, and this should be (7) and (8); and on page 89, mention is made of optional harmony in example (14d), and this should be (15f). Although all these might be regarded as simple typos, they do not facilitate understanding of the rather complicated assimilation and dissimilation processes and the interactions between the various OT constraints that are presented in a large number of tableaux analyzing a wide range of different languages. And I have only listed some of the problems I found in the first 100 pages.

Unfortunately, the flaws left me somewhat confused in places. For example, consider this text on page 95: ‘A fourth Corr constraint, defined in (27), will be considered as well, but is not crucially ranked in the analysis of Kinyarwanda facts’. Now, (27) is Corr-cvc·[–voice], which seems rather an important constraint in deriving dissimilation in Kinyarwanda prefixes. It features in all the tableaux in (37)–(41), and on page 103 it is stated that ‘the constraint that critically favors dissimilation is Corr-cvc·[–voice]’. So perhaps the text on page 95 is wrong? Maybe (26) instead of (27) is the non-crucial constraint?

Then, on page 104, in the diagram in (42), which provides a representation of the ranking of the constraints for sibilant harmony in Kinyarwanda, Ident-[sibilant] and CC·Edge-(Stem) are both shown as dominating Corr-cvc·[+sibilant], while the relationship of CC·Ident-[retroflex] with Ident-[sibilant] and CC·Edge-(Stem) seems to be unspecified. However, in the tableaux in (45)–(47) on pages 106–107, CC·Edge-(Stem), CC·Ident-[retroflex] and Ident-[sibilant] are all unranked with respect to each other but all three dominate Corr-cvc·[+sibilant]. Perhaps the structure shown in (42) was found to be inadequate? Complicating the picture even more, in the tableau in (44) on page 105, CC·Edge-(Stem) seems to dominate everything while there is no ranking between Ident-[sibilant], CC·Ident-[retroflex] and Corr-cvc·[+sibilant]. All this is rather perplexing.

There are plenty of other instances when what appears to be a flaw left me bemused. For example, in tableau (42) on page 244, the comparative score for candidate (42b) is shown as (0~1) for the faithfulness constraint Ident-[labial], which is fine except that it is shown as ‘L’, indicating the optimal candidate loses out for this constraint. Should this ‘L’ be ‘W’? And in tableau (16) on page 275, which shows blocking of dissimilation for Yidiny so that an early /r/ in the root prevents a subsequent /l/ from dissimilating to [r], the input is /burwaːliŋal/ ‘went jumping with’ and the winning candidate in (16a) is [burwaːliŋal], which seems perfectly faithful throughout. So how come it is shown with a violation of the faithfulness constraint Ident-[lateral]? And why is candidate (16b) [burwaːriŋal] shown as preferable to (16a) for the constraint Ident-[lateral] by a score of (1~0)? Shouldn’t the score be (0~1), with (16a) the winner, given that (16b) exhibits a change for one lateral but (16a) has no change? Again, maybe I misunderstood something. Finally, on page 318, in (7b), [a-n-anam] ‘one who is fond’ is given as an example of dissimilation of the agentive prefix in Tamashek Berber, but I cannot see how this is dissimilation when the [n] becomes the same as the first nasal in the root of the word. Perhaps this is correct, but I failed to understand it.

These flaws are a great pity, because there is an impressive amount of varied material collated in this book, most of which has been gleaned from a wide range of published materials describing the phonological patterns of the various languages, but wherever possible the author has checked the data via informants he had access to. Furthermore, the analysis is always detailed, profound and interesting, though one suspects that the author would have preferred to include far more extensive material and this had to be curtailed for it to be manageable in a single book. For example, mention is made on page 352 of the analysis of nasal dissimilation in clusters in Gurindji that can be found in the online supplement to Chapter 8, and in Chapter 9, the survey of over 130 languages makes regular reference to the supplementary materials in the website. One gets the impression that the original plan was to include all this material in the book.

Indeed, the work reported in the book seems to be just the start of a massive research project which may have far-reaching implications for our understanding of the phonology of consonants. However, there are clearly many issues that still need to be fleshed out. For example, the complete identity effect for Chol discussed in Chapter 6 is analysed as dissimilation, but it is acknowledged that it could just as well be regarded as assimilation, and further work is needed to determine which it is; secondly it is admitted in Chapter 8 that the analysis of the classic case of dissimilation for Georgian does not work well in the SCTD, as it does not derive the correct generalization about intervention; third, in many places throughout the book, mention is made about the failure to deal with the direction of dissimilation, something that obviously needs to be fixed; and finally, it is observed that the survey of the typology of SCTD in Chapter 9 throws up a number of problematic cases where the theory predicts no harmony but harmony is actually robustly attested, so it is a possibility that ‘the SCTD is simply wrong’ (p. 348). Clearly, all these issues need to be addressed and investigated carefully, so one hopes that the thought-provoking and important work described in this book can stimulate widespread further research into these and related issues, though it would be welcome if the outcome of future studies might be presented with fewer flaws that do not help in the interpretation of the results.