A viable analysis of any phenomenon must achieve at least two conditions of adequacy (Chomsky Reference Chomsky1965: 27). It must be descriptively adequate, in the sense that it correctly describes the phenomenon; it captures the linguistic facts. More importantly, it must aim for explanatory adequacy, in the sense that the theory that engenders the analysis explains why the phenomenon has the properties ascribed to it and not others. There are many analyses of vowel harmony that, on the surface, achieve the level of descriptive adequacy, but do not qualify as explanatory because they do not address the question of what is a possible vowel harmony system. This question must be addressed if an analysis can claim to be explanatory. This monograph, Locality in vowel harmony, meets this test. At the level of description, Nevins provides deeply insightful analyses of an impressive range of vowel harmony systems. At the level of explanation, he proposes a theory that sharply circumscribes what is a possible vowel harmony system, ruling out certain logical possibilities in a principled manner.
A comprehensive analysis of vowel harmony is not the real goal of this monograph. The overarching and ultimate goal is to show that the locality conditions that regulate the manifestation of vowel harmony can be attributed to the same principle that regulates long-distance syntactic dependencies such as the familiar agreement relations. Hence, vowel harmony is considered to be another instance of Agree (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000). The investigation of vowel harmony as a manifestation of the Agree relation is part of an overarching goal to show that (some aspects of) phonology can be characterised in terms of the principles of the Strong Minimalist Program, the goal of which is to show that language provides optimal solutions to interface conditions. The monograph contains six chapters, including an introduction and a final summary that considers some of the implications of the theory of harmony, and speculates on the types of experiments that would help to test its validity.
The introductory chapter provides an overview of the theory of vowel harmony. Nevins rejects the conventional approach that analyses the phenomenon as a process that transmits a feature value rightwards, leftwards or bidirectionally from a donor to a recipient. His proposed alternative is a Search.and-Copy theory, in which some vowel that lacks a particular feature value seeks a source that can provide that value. Crucially, the search is always minimal, in the sense that the feature-source is always the closest possible donor. For Nevins, ‘vowel harmony emerges as a computational operation whose purpose is to provide a solution for valuing [an] ‘uninterpretable’ … vowel through the principle of minimal search for the closest goal to act as a feature-source' (p. 12). The determination of closest depends on what the needy element is looking for. Nevins argues that, cross-linguistically, locality can be relativised ‘to certain types of properties of phonological features, namely, the contrastive or marked status of the features in the segment that bears them’ (p. 16). Hence harmony may be a search for any instance of a particular feature, regardless of its status, or for just a contrastive value or for just a marked value. The exercise of this parametric choice allows a language to ignore some segments in the search for a feature value for a needy vowel, thereby accounting for the phenomenon of segment transparency. The locality principle, requiring a search to terminate with the closest relevant vowel, is universal. Nevins argues that a search may fail if it encounters a ‘defective’ feature source, namely, a segment that is disqualified as a feature donor for phonological or morphological reasons. A language may also limit the domain of a search by imposing a distance boundary, dictating how far a search can look. The final component of the theory of harmony is the role of sonority in closing off a search domain. In addition to outlining the basic features of the new theory of vowel harmony, the introductory chapter justifies the need for a new model (§1.5), explaining why the two existing models are inadequate. Nevins contends that the declarative identity-enforcement model promoted in the OT framework (e.g. Baković Reference Baković2000, Hansson Reference Hansson2001, Krämer Reference Krämer2003) is too permissive, overgenerating unattested patterns, while the sharing-by-spreading model (e.g. Goldsmith Reference Goldsmith1976, Gafos Reference Gafos1996) is too restrictive, disallowing attested patterns. Case studies demonstrating the superior explanatory power of the Search-and-Copy theory are provided in the following four chapters.
We can get a very good idea of how the Search-and-Copy model works from the analysis of Turkish vowel harmony in Chapter 2. This well-known system requires vowels to harmonise for backness with a preceding segment and in some cases for both backness and roundness. Satisfaction of these requirements results in alternation in the realisation of suffixes. Alternating suffixes are needy, lacking the harmonising feature(s), and must seek a value for [±back] and/or [±round]. When such a suffix is attached, it initiates a leftward search for a licit value-source. The search is commanded by a specific set of instructions encoded in the lexical representation of the suffix, stipulating the direction of the search and the feature value(s) to be copied. The search terminates when the closest segment that bears the relevant feature value is reached. Turkish consonants are normally transparent to backness and roundness harmony, because they are generally not contrastive for [±back] or [±round]. However, the feature [±back] is contrastive for three pairs of Turkish consonants (i.e. /k/ vs. /kʲ/, /g/ vs. /gʲ/, /l/ vs. /lʲ/) and, predictably, these consonants participate in backness harmony. They supply a needed [±back] value to a following vowel, while being ignored in any search for [±round] (pp. 53–57). The analysis of Turkish also illustrates how the Search-and-Copy model deals with ‘exceptions’ to harmony; some morphemes do not undergo harmony. Since a search must be initiated by a specific set of lexical instructions, an ‘exceptional’ morpheme is one that contains no such instructions (p. 32). Significantly, ‘exceptional’ morphemes may still participate in harmony as value-sources. Nevins contends that the fact that morphemes can be ‘exceptional’ as non-undergoers and not as non-triggers ‘is an automatic consequence of the asymmetrically dependent recipient-centric nature of vowel harmony’ (p. 32). The analysis of Turkish also provides interesting evidence from a reduplication game that non-initial root vowels may be subject to vowel harmony. Chapter 2 also includes cases studies of harmony patterns in Woleaian, Kalenjin and Barra Gaelic, strengthening the claim that the search must always terminate with the closest donor. Woleaian bidirectional harmony is an interesting illustration of the explanatory power of the model. A needy suffix seeks a [±low] value from the left and from the right, but, since two sources are equally close, copying is possible only when they have the same specification for [±low].
Chapter 3 deals with one of the most difficult problems in the analysis of vowel harmony, the problem of segment transparency. In the simplest case, segments are invisible to harmony, because they lack the relevant feature. The more difficult cases are segments that bear the relevant feature but are nevertheless transparent. Nevins attributes these to one of two properties, the contrastive or marked status of the harmony feature. A search for contrastive [±F] ignores non-contrastive instances of this feature in the search path, while a search for the marked value for a feature ignores the unmarked value. In Finnish vowel inventory, the front vowels /i e/ do not contrast with a [+back] counterpart. Therefore, a leftward search for a contrastive [±back] value skips these vowels. Since contrast can be positionally restricted (Beckman Reference Beckman1997, Reference Beckman1998), it is not surprising to find cases where harmony ignores segments in a position where they do not contrast. The Turkic language Karaim provides an excellent example of this type of transparency. A leftward search for contrastive [±back] ignores non-initial vowels, because contrast for backness is restricted to initial position. Interestingly, ‘all consonants (except glides) participate in harmony, because they are contrastive for [±back]’ (p. 84). The role of markedness in accounting for segment transparency is illustrated by a harmony pattern in the Tungusic language, Sibe. Nevins considers the feature [±low] to be contrastive in the vowel system of this language. However, the [−low] value is invisible to a process that copies [+low] from a preceding vowel and produces an alternation between velar and uvular consonants; the latter are specified as [+low]. For Nevins, suffixes that display the alternation initiate a leftward search for a [+low] value, identified as the marked value. Relativised locality then allows the search to skip any non-low vowel, because [−low] is irrelevant to the search. The choice of [+low] as the marked value in Sibe is considered to be language-specific (p. 90). If a search for a contrastive or marked feature value fails to find a source, the derivation does not crash. The insertion of a language-specific default value comes to the rescue. For example, [+back] is default in Uyghur, but [−back] is in Finnish. The parametric choices that account for segment transparency produce some interesting microvariation patterns, illustrated by dialect differences in Kirghiz, Yoruba and Finnish. Similar choices are also supposed to account for variation in obstruent transparency in nasal harmony systems (pp. 105–109). For example, non-consonantal segments in Warao must initiate a leftward search for any instance of [±nasal], regardless of the contrastive or marked status of this feature. Since [+consonantal] segments are either [+nasal] or [−nasal], any consonant may provide a value for nasality. Obstruents are therefore participants in Warao nasal harmony (Piggott Reference Piggott1992). In contrast, Nevins postulates that the leftward search for [±nasal] in Barasano is restricted to contrastive values only. Only sonorants contrast for nasality in this language. Therefore, obstruents are ignored by the search; they are transparent.
Arguments in Chapter 3 that sensitivity to contrastive specification cannot be expressed by underspecification are quite convincing. Unlike Steriade (Reference Steriade1987), Dresher (Reference Dresher2009) and others, Nevins assumes that segments are fully specified for both contrastive and non-contrastive values, since non-contrastive values that are ignored by harmony in the computation of locality may be visible to other processes in the language. Finnish assibilation (p. 80) is considered to be evidence that the vowel /i/ must be specified for [−back], although this specification is not taken into consideration in the computation of locality in backness harmony.
The focus of Chapter 4 is on parasitic harmony, cases in which a segment x may copy a value [±F] from a segment y only if x and y bear the same value for another feature [±G]. Feature copying in this type of harmony always fails if a searcher and a potential donor are separated by a segment bearing the relevant harmonic value [±F] but lacking the orthogonal requirement [±G]. Nevins describes this as defective intervention, and derives the failure from the computation of locality as regulated by the minimal distance principle. Roundness harmony in Yawelmani is presented as an example of how defective intervention works. Vowels in this language display a contrast for [±high], and a suffix vowel harmonises with the roundness of a preceding vowel, provided that donor and recipient have the same specification for [±high]. When a [±round] vowel precedes a needy suffix vowel but differs from it in height, the harmony requirement is not satisfied. Furthermore, the search for another value-source cannot proceed further. A defective intervener always blocks harmony. Height harmony in some Bantu languages and roundness harmony in Khalkha Mongolian are presented as other examples of defective intervention at work. In the better-known cases of parasitic harmony, the orthogonal requirement is phonological. However, Nevins argues that the Australian language Jingulu provides evidence that the additional requirement may be morphological. Defective intervention, like the relativisation of locality, has a syntactic instantiation, providing further support for the hypothesis that all long-distance dependency relations have a fundamental unity.
Chapter 5 is the last substantive chapter. Its focus is on how the domain of harmony is delimited. In the clearest cases, the domain of harmony is the prosodic word. However, it can be smaller. Nevins identifies two extrinsic factors that may determine the boundedness of a harmony domain: (i) distance parameters, and (ii) sonority hurdles. Distance parameters may limit the number of segments or syllables that may intervene between a searcher and a value-source to one or none, or a search may unbounded. For example, long-distance nasalisation in Kikongo and Lamba is regulated by different distance requirements. It is unbounded in Kikongo; any number of syllables may intervene between the suffix that needs [+nasal] and the donor element. In contrast, the needy suffix in Lamba must seek a donor in the immediately preceding syllable; failure to find a donor in such a location terminates the search. Backness harmony in certain Hungarian dialects provide evidence that maximally one syllable may intervene between recipients and donors. Boundedness by segmental restrictions is illustrated by Yucatec Maya and Assamese. Different settings of the distance parameters provide insights into microvariation in the realisation of Dahl's Law in Gikuyu. With regard to sonority, Nevins identifies a number of cases where the barriers are correlated with the relative sonority of a particular segment or class of segments. It is well known that the low back vowel /a/, among the highest on the sonority scale, is a barrier in many [±ATR] harmony systems. ‘The patterning of [+low] /aː/ as a nontransparent sonority hurdle is widespread throughout the [±ATR] vowel harmony systems of West Africa’ (p. 180). Wolof is one of these languages. Sonority barriers to harmony are erected in a range of non-African languages, including Classical Manchu, Finnish and Menominee. Nevins recognises an important difference between distance and sonority bounding. A segment that constitutes a sonority hurdle blocks a search but its value for a harmonic feature is still copied. The demarcation of bounding domains in harmony systems is equated with the restrictions that apply to Wh-movement and the concepts of islands in syntax.
The final chapter reviews some of the implications of the Search-and-Copy theory regulated by relativised locality, inventory-constrained feature search, defective intervention and bounding parameters. It identifies the set of possible harmony languages (p. 206), and, more appropriately, a set of impossible harmony languages (p. 204). Nevins speculates that an artificial grammar learning paradigm might be employed to test the predictions that some systems are possible and learnable, while others are impossible and unlearnable.
This monograph is truly original. With respect to the analysis of vowel harmony, the target-centric approach is not championed by any others except for Mailhot & Reiss (Reference Mailhot and Reiss2007). It shares with Government Phonology (Kaye et al. Reference Kaye, Lowenstamm and Vergnaud1985, Charette Reference Charette1991, Harris Reference Harris1994) the thesis that some relations in syntax and phonology are governed by the same or similar principles. The arguments that harmony and Agree are fundamentally the same type of dependency relations are quite persuasive. If Agree is a post-syntactic phenomenon, as argued by some (e.g. Bobaljik Reference Bobaljik, Harbour, Adger and Béjar2008), the idea that harmony and Agree are both optimal solutions to interface conditions is easier to envisage. Certain aspects of the theory advocated in this monograph have no counterparts in the dominant OT framework. For example, inventory-determined feature values cannot be mirrored in OT. To the extent that these are crucial to the determination of what is a possible harmony system, the Search-and-Copy model achieves a level of explanatory adequacy that is not replicated by its competitors. The monograph also contains some compelling arguments and evidence against a phonetically grounded view of harmony (pp. 8–10, 30–32).
While Nevins makes a case for his target-centric, Search-and-Copy model, the approach engenders some unintended and, perhaps, undesirable consequences. Consider first the characterisation of a needy morpheme as one that lacks a particular feature value and initiates a search for it. However, the initiation of a search is also an idiosyncratic property of a morpheme. If a search is unsuccessful, a default value is assigned. For example, Turkish has needy and harmonising suffixes (e.g. /ler/~/lar/ (plural)) contrasting with non-needy and non-harmonising suffixes (e.g. /gen/ (nominaliser)). However, the model seems to allow for a third possibility, a needy and non-harmonising suffix (e.g. /gVn/). The latter would be needy because it lacks a harmonic feature, but would not undergo harmony because there are no instructions to do so. It would receive a default feature value. Assuming that a default feature is inserted as a last resort, this hypothetical needy and non-harmonising suffix in a Turkish-type language should be invisible to harmony. Transparency would then be limited to cases in which a donor and recipient are separated solely by a needy but non-harmonising morpheme. This hypothetical Turkish-type language is unattested. Consequently, the Search-and-Copy seems to be vulnerable to the claim that it overgenerates. This problem is not directly addressed in the book, but it might be solved if there was a principled way to ensure that a needy, non-harmonising suffix received its default value before a search is triggered by a following needy and harmonising suffix.
The overgeneration problem has an additional source, not inherent to the Search-and-Copy model, but linked to the fact that certain analyses proposed by Nevins depend crucially on language-specific feature specification. For example, Nevins assumes that Nawuri labial consonants [p b m f] are specified as [+round], thereby accounting for the fact that they are defective interveners in a search for roundness in a following vowel or glide. Since defective interveners are possible donors, we therefore expect to find cases in which labial consonants are donors in long-distance roundness harmony. There is no Turkish-type system in which labial consonants are value-sources for [±round] harmony. The unintended consequences of the Nawuri analysis can be eliminated by abandoning the assumption that labial consonants are ever [+round]. The rounding effects associated with these segments in a number of languages can be and have been derived from their specification for a labial feature.
The treatment of the harmonising behaviour of an epenthetic vowel is also not fully accounted for. It is not clear how the instructions about Barra Gaelic epenthetic vowels (p. 58) are integrated into the grammar. Does an epenthetic vowel carry a feature indicating that it is epenthetic? If ‘the epenthesis process furnishes only a timing slot with the feature [+vocalic]’ (p. 58), what is responsible for triggering a search? The apparent problem with the treatment of epenthetic vowels contrasts strikingly with the routine treatment of such vowels from a trigger-centric perspective. Triggers assign values to recipient segments, regardless of the morphological or phonological source of the latter.
A search for an inventory-dependent feature value, crucial to the computation of relativised minimal distance, is also a potential weakness of the Nevins theory. How does a search ‘know’ when it encounters a contrastive value? Since this knowledge is inventory-dependent, it cannot be locally determined. It appears that the theory has to appeal to global knowledge. As Embick (Reference Embick2010) reminds us, theories with global power are invariably too powerful. Notice that an underspecification theory such as the one advocated by Dresher (Reference Dresher2009) would not have this problem, because only contrastive values are visible to harmony processes. Local determinism can be achieved in the Search-and-Copy model by marking contrastive values in some way, thereby incorporating some elements of the Dresher theory into the Search-and-Copy model.
Nevins considers his Search-and-Copy model to be adequate for all harmony systems, but it would have to incorporate some modifications to accommodate nasal harmony patterns as described by Piggott (Reference Piggott1992, Reference Piggott2003) and Walker (Reference Walker1998). The proposal in §3.10 (pp. 105–109) that nasal harmony patterns reflect the choice between a search for all [±nasal] values and a search for only contrastive [±nasal] is inadequate. While a leftward search for contrastive [±nasal] in Barasano would account for the transparency of obstruents, it does not account fully for the nasalisation of sonorants. Among sonorants, only vowels can be analysed as contrastive for nasality in Barasano (Piggott Reference Piggott2003). A leftward search for [±nasal] would generate nasal sonorants to the right of a value-source but not to the left. Nevertheless, sonorant consonants immediately to the left of a value source are always nasalised. The Barasano-type might be accommodated by recognising a parameter allowing a search to be initiated by either a segment or a syllable. Syllables in Barasano would therefore be either nasal or oral, as defined by Piggott (Reference Piggott2003).
The few weaknesses identified in a monograph that proposes a truly radical change in our thinking about vowel harmony and phonology in general should not detract from its contribution to the field. This engaging book is highly recommended as supplementary reading in an intermediate level phonology course. It is also recommended to syntacticians, because it would positively influence their view about the relevance of phonology to the generative enterprise. Phonology cannot be marginalised, if the principles and parameters that regulate all long-distance dependencies are fundamental to the overall organisation of grammar.
