In Locality in vowel harmony, Andrew Nevins presents a Minimalist account of vowel harmony. Readers need not be familiar with either vowel harmony or the Minimalist Program (Chomsky Reference Chomsky1995) in order to understand the claims of this book, making the work accessible to a wide range of linguists. Nevins's book makes three major claims. The first claim is that vowel harmony is the result of a search.and.copy procedure; a ‘needy’ vowel (i.e. a vowel that has no featural specification) receives its features from searching for the nearest vowel that can supply a feature value, and copies that feature value. This creates harmony – vowels in a given domain sharing feature values. The second claim is that the search procedure must follow a notion of relativized locality. The ‘needy’ vowel must copy a feature value from the closest possible source, but the definition of ‘closest’ is relative. A vowel feature can therefore be copied from a non-adjacent vowel. This creates long-distance harmony effects, such as transparent vowels. The third major claim of the book is that the search-and-copy procedure is constrained by principles and parameters. All vowel harmony patterns can be explained by a set of principles that parallel the minimalist Agree function. The proposed principles constrain the direction of the search, the copying of harmonic feature values, and locality requirements (e.g. how far a search may proceed until it fails).
In Chapter 1, ‘What is vowel harmony, how does it vary, and why study it?’, Nevins uses quirky metaphors to explain the concepts of harmony and relativized locality. Relativized locality is the idea that what is closest depends on how distance is measured and on the method of transportation. In Nevins's subway metaphor, the closest subway station in terms of distance is not necessarily the fastest stop to get to. The closest stop ‘as the crow flies’ may be on another line, or, if you happen to be on an express train, not a regular stop. The same principle applies in harmony. Looking for the closest stop on the red line likens to searching for the closest high vowel. The harmony recipient searches for the closest possible donor for its feature value, but this may not necessarily be an adjacent vowel. The idea, then, is that vowel harmony follows strict principles of locality, but apparent violations of locality result from the fact that the notion of locality is relative to the search parameters.
The premise of the search-and-copy procedure, described in Chapter 2, ‘The search principle’, is that vowels can be either donors or recipients of feature values. Harmony occurs when a recipient vowel gets its feature value(s) by searching for the closest suitable donor and then copying the feature values of that donor. This view of harmony is recipient-centric in that only recipients may seek out donors; donors may not seek out recipients. The search-and-copy procedure assumes a feature-filling approach to harmony, and is thus highly similar to Mailhot & Reiss's (Reference Mailhot and Reiss2007) proposal. It is unclear whether Nevins claims that all harmony processes must be feature-filling, and whether search-and-copy procedures are compatible with feature-changing harmony processes.
Nevins makes three arguments for a recipient-centric approach to harmony. The first argument comes from the typology of exceptions to vowel harmony. Vowels may exceptionally fail to undergo harmony, but there are no vowels that exceptionally fail to be a harmony donor. The second argument is based on the fact that there are cases of harmony in which the copied feature value is decided in terms of distance to the harmony donor. This suggests that there are languages in which the vowel harmony property is defined in terms of the recipient, making it difficult to capture harmony solely in terms of the donor. The third argument for a recipient-centric approach to harmony is that when harmony applies to more than one feature (e.g. backness and rounding), the donor may copy one feature from one vowel and another feature from a different vowel. A donor-centric approach to harmony would require that a single vowel spread both harmonic features, leading to an unwanted result.
Nevins poses convincing arguments that a recipient-centric approach to harmony is possible, but it is not clear that recipient-centric harmony is necessary in all cases, and that a donor-centric approach would not be possible. First, the argument that vowels never exceptionally fail to be a harmony donor appears to be correct, but there are vowels that exceptionally are exceptional donors to harmony (Finley Reference Finley2010), which would require a donor-centric approach to harmony. In addition, the argument that different segments can spread multiple features to a single recipient does not rule out a donor-centric approach. One possibility is that each harmonic feature is represented separately, and each feature spreads via a separate rule. Further, a donor-centric approach to harmony provides a straightforward account of harmony systems that are best characterized in terms of the donor (Walker Reference Walker2005). Nevins states three restrictions on the donor in Chapter 3, ‘Contrastiveness, markedness, and feature-based locality’. First, if a potential donor is unspecified for the given harmonic feature, it cannot be a donor and will be skipped. Second if a potential donor fails to bear a contrastive value of a harmonic feature, that vowel may be skipped. Third, some vowel harmony languages require the donor to be marked for the harmonic feature. If a potential donor bears the unmarked feature value, that vowel could be skipped. These three restrictions are used to account for a variety of restrictions on harmony donors, including transparent vowels. Nevins argues that there is not one single account of transparent vowels. Rather, transparent vowels emerge from different combinations of the above restrictions on donors and recipients.
Chapter 4, ‘Defective intervention: When search comes back empty-handed’, discusses cases in which harmony is blocked and the search fails. This includes a discussion of parasitic harmony, in which the donor searches for a vowel that shares an orthogonal feature value (e.g. back harmony occurs only if the donor and recipient agree in height features). Interestingly, the search is bounded such that it cannot ‘skip’ vowels until it finds a match. Once the search encounters a vowel marked for the orthogonal feature value, the search terminates regardless of whether there is a match. The restriction is required because there are no cases of parasitic harmony that skip vowels that differ in the orthogonal feature value. While Nevins claims that the search parameter avoids these unattested cases of non-local parasitic harmony, the search parameters must also account for cases in which the search skips vowels that are marked for a given feature. How the parameters are able to achieve and constrain both types of searches is unclear. Further, it is a mystery why this difference should exist.
Nevins presents two additional parameters on the harmony search in Chapter 5, ‘Domain limitations on search’: sonority and distance. These parameters allow the search to terminate if the search runs into a vowel of sufficient sonority (typically a low vowel), or has reached a particular distance. The sonority restriction appears to be a novel claim, but it is unclear whether these restrictions are caused by the sonority of the vowel, or because low vowels appear to have a special status as harmony blockers. Low vowels have unique articulatory and acoustic properties created by the vocal tract that may contribute to why low vowels often fail to participate in harmony. This is an interesting question for future research, and an opportunity to test the role of phonetics in phonology. While Nevins appears to be on the side of a substance-free approach to phonology (Hale & Reiss Reference Hale and 2000) (using analyses that completely abstract away from phonetic grounding of phonological processes), understanding how phonetic properties of individual vowels lead to differences in behavior is important for understanding where particular restrictions come from, and whether and how they should be instantiated in the grammar.
The locality parameter introduced in Chapter 5 constrains the number of elements that the search may encounter past the recipient (even if a suitable donor appears one or two syllables beyond the designated ‘stopping point’). Locality constraints follow a strict typology. The search may be unbounded, restricted to an adjacent element (syllable or segment), or may skip a single element. There are no languages that restrict the search to skip a specific value greater than two elements (e.g. up to four syllables). While the locality parameter is a variable that could be set to any number, Nevins restricts the parameter to the quantification variables none, only one and any amount. This raises the question of why and how these quantifications become connected to vowel harmony, and to what extent locality principles are bound by these quantifications.
Chapter 6, ‘Minimalist computation of vowel harmony: Implications’, ties together Nevins's parallels between syntactic phenomena and vowel harmony through an argument for a Minimalist approach to phonological representations. By incorporating ideas from minimalism into phonology, Nevins argues, the problem for the language learner becomes that much easier. However, it is not clear how parameter setting for syntactic agreement should make parameter setting for phonological agreement easier. It is also unclear whether the parallels Nevins offers between syntactic phenomena and vowel harmony are on target. While it may help readers with a syntactic background understand the importance of the problems encountered in vowel harmony, it is hard, as a phonologist, to understand the importance of these relationships.
The parameters for the search procedure are written as morpheme-specific constraints. For each morpheme that undergoes harmony, there is a specific rule stating the parameters for the search. While formalizing the harmony search for each morpheme makes it possible to account for morphemes that fail to undergo harmony, it leaves open the question of why languages do not have more variation between morphemes. For example, Nevins's system makes it possible for each morpheme to have a slightly different vowel harmony rule: one morpheme with parasitic harmony, another morpheme with transparent vowels, and another searching in the opposite direction. This opens the question of how much variation is tolerated within a given language across morphemes, and how many different types of harmony processes a language can tolerate.
Nevins's simple analyses of harmony provide a tool for understanding the classic issues in vowel harmony, as well as raise issues that have yet to be addressed. For example, despite the fact that previous analyses of consonant harmony (Hansson Reference Hansson2001, Rose & Walker Reference Rose and Walker2004) argue for a difference in formalism between the two types of harmony systems, Nevins analyzes vowel harmony and consonant harmony (as well as consonant–vowel interactions) with a single search-and-copy procedure. This leads to the question of why and how consonant and vowel harmony patterns are similar and different, and whether these similarities and differences should be captured in the grammatical formalism. While there is a tendency to prefer a unified account of all types of harmony and assimilation phenomena, such a unified account may not accurately represent human knowledge. Further typological research, along with psycholinguistic experiments, may help uncover answers to these questions.
Nevins's book provides an important look at a range of vowel harmony processes, proposing interesting and innovative solutions to classic problems, and raising novel issues. While Nevins does not give a large background describing previous models of vowel harmony (nor providing comparisons to similar proposals), this book is an important addition to the corpus of work on vowel harmony.
