John Frampton's Distributed Reduplication provides an excellent representation of the state of our current understanding of reduplication. Although Frampton does not directly develop the threads of discussion, his book demonstrates where a general unstated consensus has been reached on certain issues in reduplication and where pockets of disagreement remain. Just as Anderson (Reference Anderson1985) lays out the main line of questioning in contemporary phonological theory being defined by the tension between rules (i.e. computations) and representations, so our understanding of reduplication follows a similar path. Many aspects of consensus around reduplication are due to the field of phonology arriving at the position that computational derivations are a fundamental aspect of phonology. This review will focus on the computational and derivational aspects of Frampton's proposals, while leaving questions about the representations involved in reduplication aside, because space does not permit their development. Three particular examples from the book will be highlighted to demonstrate important aspects of computation and derivation in reduplication.

As a general introduction to the nature of reduplication, Frampton suggests that there is a connection between the repetition of segments in a reduplicated form and lexical access, in that both are instances of a copying process. Lexical access copies a phonological representation from the mental encyclopedia to phonological working space, and reduplication copies a phonological representation within the phonological working space. According to Frampton, this connection between lexical access and reduplication is the explanation for reduplication being so ubiquitous in human language. Although Frampton does not discuss the connection between reduplication and lexical access beyond this introductory point, it appears that reduplicative copying is different from lexical access in that there are multiple distinct stages of computation in the creation of a reduplicated form. This is a key component of the theory of Distributed Reduplication (DR), where the explanation for different aspects of a reduplicated form are spread across different parts of the grammar.

To begin to understand the distributed nature of explanation in DR, we will consider Frampton's analysis of reduplication in Arrernte (Breen & Pensalfini Reference Breen and Pensalfini1999), with a focus on how the shape of the repeated regions are accounted for by DR. The aspects of this example that we will focus on are how the repetition of segments is delimited and how reduplicative copying occurs. Arrernte has three reduplication patterns, as shown in (1).

1. (1)

   

A neutral description of the three reduplication patterns is as follows. The repeated region in the habitative consists of the last VC(C)VC(C) of a stem, which invariably contains the suffix /e/. Frequentative reduplication consists of a prespecified /ep/ combined with the last VC(C) sequence of the root suffixed to the root. Finally, the attenuative reduplication pattern prefixes a copy of the first VC(C) along with a prespecified /elp/ to the stem. I refer to this as a neutral description, because Breen & Pensalfini (Reference Breen and Pensalfini1999) argue that syllabification in Arrernte is VC(C), so that VC(C) in the descriptions above can be replaced with ‘syllable’. The analysis that Frampton provides demonstrates that we do not have to conclude that Arrernte has VC(C) syllabification to have an elegant description of these reduplication patterns.

What is copied in reduplication in DR is determined by what segments are within transcription-junctures (t-junctures), represented by ‘[ ]’, which are legitimate phonological representations that can be inserted into phonological representations via readjustment rules (in the Distributed Morphology sense: Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993, Harris & Halle Reference Harris and Halle2005). T-junctures can also be memorised as part of underlying representations. Frampton's analysis of Arrernte reduplication patterns uses both inserted and memorised t-junctures. Another type of juncture that Frampton posits is the truncation juncture (‘< >’), which serves the dual purpose of keeping track of reduplicated copying and of marking segments to be skipped in copying. With these two novel types of junctures, Frampton proposes that the phonological content for habitative, frequentative and attenuative morphology is as in (2) (adapted from p. 56).

1. (2)

   

Although the analysis of Arrernte occurs in Chapter 4, ‘Truncated reduplication’, Frampton does not fully discuss juncture-insertion rules until Chapter 5, ‘Sources of variation’. The theory of junction-insertion rules is critical in understanding what shape patterns can be computed in reduplication, so we will discuss them now. (3) presents the template for a juncture-insertion rule.

1. (3)

   

(3) provides the parameter space of juncture-insertion rules based on well-established differences among the categories of plain ×-slots (X), consonants (C) and two types of vowel calculations. The ‘v’ option indicates a calculation that respects geminate inalterability, meaning that a geminate long vowel cannot be split when the juncture is inserted, while the ‘V’ option allows for geminate long vowels to be split by a juncture. (3) is very much in the spirit of proposals on pivot points in Yu (Reference Yu2007) and anchor points in Raimy (Reference Raimy2000a, Reference Raimy, Raimy and Cairns2009). The specific juncture-insertion rules in (2a, b) are ‘insert [ before V, rightmost in the word’, when converted to the options in the template in (3). The juncture-insertion rule in (2c) invokes a domain parameter, ρ, set to ‘initial foot’, which has the effect that a juncture can be inserted with reference to the ‘second’ vowel from the left (i.e. the rightmost vowel in the leftmost foot; Frampton uses /_V_second as a provisional specification for (2c)). There is no discussion or argumentation as to why there is a prosodic domain parameter, as opposed to simply positing a parameter of 1 or 2 (as in Nevins Reference Nevins2010) for right/leftmost. This is a potentially interesting question about the role which prosody plays in defining reduplicative templates. The fact that Frampton only invokes the prosodic domain of ‘foot’ throughout the book suggests that phonology being able to count to two may be the fundamental process which produces the prosodic category of foot, as opposed to Frampton's implicit suggestion that the prosodic category of foot allows phonology to count to two.

With the morphology of reduplication in Arrernte settled in (2), we can now turn to how reduplicative copying occurs in DR. Applying the morphology from (2) to forms in (1) will create representations that contain t-junctures and truncation junctures, as in the examples in (4). Note that there is a question as to the ordering of affixation and juncture insertion. I present an opinion in (4) below that juncture insertion occurs before affixation, which allows the elimination of reference to ‘root’ or ‘stem’ as part of the juncture-insertion rule. Frampton appears to favour the opposite position, where operations can be set to only consider ‘root’ or ‘stem’ and thus ignore affixes. There may be ways to tease apart these positions, but space prevents this being done in this review. See Embick (Reference Embick2010) for discussion of issues of locality in morphology.

1. (4)

   

These three examples from Arrernte provide critical cases that demonstrate Frampton's proposals on how reduplicative copying occurs. Two crucial observations that the reader should make at this point is that (4a) does not contain any truncation junctures, while both (4b) and (4c) do, aligning with either the right t-juncture, (4b), or the left t-juncture, (4c). These representational differences create different derivations for reduplicative copying.

Reduplicative copying occurs due to transcription (discussed in full detail in Chapter 4), which has three subprocesses: Initialisation, CopyShift and Cleanup. Initialisation is the first stage of transcription and inserts a matched pair of truncation junctures, < >, aligned with either the left or right t-juncture. This left/right decision is a parameter which must be set and will also determine the direction of copying. For (4a) the initialisation parameter will be set to ‘right’, which will insert truncation junctures inside the right t-juncture as in (5). Note that the derivations require the explicit representation of ×-slots, so the shorthand representations in (1)–(3) must be expanded to include them.

1. (5)

   

The necessity of explicitly representing the ×-slots is apparent when we see how the CopyShift operation works. CopyShift moves a truncation juncture across an ×-slot, copies the ×-slot and places it adjacent to and outside of the t-juncture at the edge selected by the direction parameter. For the example in (5), this creates (6a).

1. (6)

   

As (6a) shows, only the ×-slot is copied and shifted to a place within the representation, but the melody associated with the original ×-slot is also associated with the new ×-slot. CopyShift is iterative, and (6b) shows that it will move the truncation juncture across the next ×-slot, copy that ×-slot and infix the new ×-slot between the t-juncture and the previously CopyShifted ×-slot. (6c) shows the results of iterating CopyShift until the left truncation juncture, <, is adjacent to the left t-juncture, [. Once t-junctures and truncation junctures are aligned at each end, a process of Cleanup occurs which deletes all juncture types, producing the final representation in (6d). The representation in (6d) is the crucial novel representation in DR that provides insights into the interaction between reduplication and phonological rules. Frampton names these types of representations crossed structures: we will discuss the role they play in a moment, after completing our discussion of the relevance of the way that transcription operates.

(4b) and (4c) are representations that contain truncation junctures prior to the start of transcription. If truncation junctures are present in a representation at the start of transcription and align with a t-juncture that is the same as the direction parameter, then the Initialisation process is skipped. CopyShift will then apply in a regular fashion, moving the ‘outside’ truncation juncture towards the other t-juncture. Consequently, pre-initialisation truncation junctures produce the surface effect of infixing prespecified material between reduplicated segments. More concretely, for (4c), [<elp>it]ir-em, the direction of transcription is set to ‘left’, so if we shorthand the melody/×-slot representation, iteration of CopyShift will produce the representation it[<elpit>]irem, with Cleanup (and other processes) resulting in the correct surface form itelpitirem, (1c). (4b) works too, by simply setting the direction to ‘right’.

Truncation junctures and transcription interact in a different way if pre-existing truncation junctures are not at the edge where Initialisation would insert them. Frampton uses this situation to provide an analysis of the pattern found in Ndebele (pp. 57–60), where the root is reduplicated and suffixed after a suffix, lim-a ‘cultivate’→lim-a-lim. The key here is that the truncation junctures introduced as part of the readjustment are at the right edge, while initialisation occurs at the left edge. Thus, the morphology creates a representation of [lim<a>], where the /a/ suffix is enclosed in truncation junctures. Initialisation occurs at the left edge, producing [<>lim<a>]. CopyShift will eventually produce a representation lim[<lim><a>]. At this point there are two sets of truncation junctures, and Frampton posits an additional rule within transcription, CopyDelete, to deal with this situation. CopyDelete takes the contents of the pre-existing set of truncation junctures, copies it to outside the t-junctures and then deletes the original content. Applying CopyDelete to lim[<lim><a>] produces lima[<lim><>], with Cleanup producing limalim, the correct surface form. CopyDelete can produce morphological metathesis through the general representation [<a><b>]. Since there are two truncation juncture regions, CopyDelete will apply to one of them, moving the contents to the other side of the other segment. Cleanup rules will then remove the junctures, leaving ba.

We will now turn to how DR captures aspects of the interaction between reduplication and phonological rules. Frampton discusses Korean Consecutive Reduplication (pp. 28–31; data from Chung Reference Chung1999), which provides an excellent example of the complexities involved in phonology–reduplication interaction. He adopts a now standard computational stance on the phonology–reduplication interaction (see Raimy Reference Raimy, van Oostendorp, Ewen, Hume and Rice2011 for discussion), where phonological rules can show under- or overapplication effects if they apply before reduplicative copying occurs, normal application after reduplicative copying occurs and backcopying effects (McCarthy & Prince Reference McCarthy, Prince, Beckman, Dickey and Urbanczyk1995) if the phonology occurs ‘during’ reduplicative copying.

1. (7)

   

(7a) shows that this pattern of reduplication in Korean follows an AABB pattern, where both syllables are repeated separately. (7b) shows that some rules of Korean show normal application effects. Frampton can easily account for normal application by having a rule apply after NoCrossingConstraint (NCC) repair has occurred. NCC repair (p. 16) is a process that eliminates crossed structures through the fissioning of melodies that are associated to multiple ×-slots through transcription. Given this architecture, we can understand the presence of the tensed velar stop, [k'], in (7b) as the result of Tensification (Sohn Reference Sohn1999: 173) applying after NCC repair, as in the derivation in (8).

1. (8)

   

The key aspect of the derivation in (8) is that the Tensification rule which causes a lax consonant to become tense if it follows a plain p, t or k must apply to the representation in (8c) for the correct surface form, (8d), to be produced. It cannot apply to the pretranscribed representation in (8a), because the environment triggering the rule has not yet been created. This would produce an underapplication effect. It cannot apply to the crossed structure in (8b), because although the triggering environment is present (i.e. the first /k/ is linked to an ×-slot that follows an ×-slot associated with a plain consonant) it would produce the form /sik'akk'ak/, which is an overapplication effect. Technically, this hypothetical overapplication effect is an example of backcopying, because the environment to trigger the application of the rule is created by the reduplication structure itself.

(7c), /hilak/→/hihinaŋnak/, is a crucially important form, because it demonstrates the existence of backcopying overapplication of ‘n/l shifting’ (Sohn Reference Sohn1999: 169), which causes an /l/ in word-initial position or following a consonant other than /l/ to become /n/. In the root in (7c), the /l/ only follows a vowel, so this form itself does not provide the environment to trigger l→n shift. Reduplication creates new environments through the repetition of phonological material, so it can be the cause of the application of l→n shift that we see in (7c). Consider the derivation in (9).

1. (9)

   

(9b) is the crucial step in the derivation, where we have the crossed structure creating the environment where /l/ follows /k/. This is sufficient to cause the l→n shift rule to apply, changing the single /l/ melody attached to two distinct ×-slots to /n/. The ×-slot under 1 in (9b) is associated to the /l/ melody, and follows an ×-slot associated to an /i/ melody; this environment does not trigger n/l shifting. The ×-slot under 2, on the other hand, is also associated with the /l/ melody, and crucially follows an ×-slot associated to /k/, satisfying the l→n shift rule's environment. Although the /l/ melody is associated with two distinct phonological environments, the fact that one satisfies the structural description of l→n shift is sufficient to cause the /l/ melody to change to /n/, as in (9c). This is a case of backcopying, because the environment that triggers this rule is created by reduplication itself. (9d) shows the application of NCC Repair to split the multiply associated melodies into distinct ×-slots, and (9e) shows the application of nasalisation (Sohn Reference Sohn1999: 172–173), which causes any consonant other than /l/ to become nasalised if it precedes a nasal sound. Nasalisation occurs after NCC Repair, so it shows normal application, where only the first /k/ changes to /ŋ/.

DR's crossed structures are a solution to the parallelism paradox that reduplication presents. McCarthy & Prince (Reference McCarthy, Prince, Beckman, Dickey and Urbanczyk1995) suggested that backcopying effects (and other aspects of reduplication) presented evidence that phonology should be done in a computationally flat manner, where there were no intermediate forms or serial derivation. This drove the majority of OT practitioners to pursue parallel correspondence theory for roughly the following 15 years. Within the last few years, the field has rejected parallel correspondence theory in favour of more serially based models. Harmonic Serialism (McCarthy Reference McCarthy2010) and Stratal OT (Kiparsky Reference Kiparsky, Uyechi and Wee2010, Reference Kiparsky, Cairns and Raimy2011) are contemporary approaches that have derivations within OT architecture. Reduplication was shown to not need parallel derivation (Raimy Reference Raimy2000a, Reference Raimyb), but the price to be paid was an expansion of representational power. DR pays the representational price through the persistence of a crossed structure for a period of time in the derivation. The key aspect of the representational expansion necessary to account for reduplication is the notion that a phonological segment is in multiple phonological environments at a single point in the derivation. This is what crossed structures do, and why they account for backcopying phenomena.

The rejection of parallel correspondence has caused some researchers to question whether reduplication provides evidence for any modification to phonology beyond a copying process. Inkelas & Zoll (Reference Inkelas and Zoll2005) and Kiparsky (Reference Kiparsky, Uyechi and Wee2010) specifically argue that serial derivations can account for all reduplication data except for backcopying phenomena, and that backcopying phenomena do not exist. Consequently, both sources argue that a model of reduplication that can account for backcopying effects is too powerful and unconstrained. Presumably, this conceptual argument would be extended to DR. The existence of the Korean data already discussed (and other known examples of backcopying effects; see Raimy Reference Raimy, van Oostendorp, Ewen, Hume and Rice2011) presents an empirical argument that trumps conceptual arguments about the ‘power’ of an analysis.

The final example from the book that I will discuss here demonstrates the empirical argument that there are backcopying effects. Frampton discusses imbrication in Ndebele, based on data from Hyman et al. (Reference Hyman, Inkelas, Sibanda, Hanson and Inkelas2009). Hyman et al. describe imbrication as a ‘a special fusion process triggered by the perfective suffix -ile’ (2009: 291). The surface effects of this can be seen when we compare (10a) with (10b).

1. (10)

   

(10a) shows the suffixation of -ile to form a perfective. (10b) shows the effects of imbrication, where the -ile suffix fuses with the root, causing the /a/ to change to /e/ and the /il/ part of the suffix not to appear on the surface. Forms that undergo imbrication can also be reduplicated, yielding very interesting interactions.

1. (11)

   

(11) shows that there are three attested types of imbrication reduplication interaction: overapplication, normal application and underapplication. The overapplication and normal application patterns are readily accounted for in any model of reduplication which uses serial derivation. Overapplication is produced by having imbrication apply before transcription occurs. The opacity of overapplication is produced through reduplication, creating new phonological environments that do not satisfy the structural description of imbrication. The first /e/ in the reduplicated form in (11a) appears to be unconditioned, because it is not followed by the remnants of the -ile suffix. This is just a simple case of opacity. Normal application in (11b) is produced by having imbrication apply after NCC Repair has occurred. NCC Repair causes the reduplicated ×-slots to have independent melodies associated with them. This allows a rule to alter melodies in a reduplicated form, separately producing a non-opaque interaction between imbrication and reduplication. Frampton's identification of (11c) as an underapplication effect is very important, because it is an example of backcopying (possibly better called ‘backblocking’) underapplication, where the reduplicated structure itself blocks the application of the rule in question. (12) presents the representations that imbrication applies to in order to produce the patterns in (11).

1. (12)

   

Frampton's insight with respect to the underapplication in (11c) is that if imbrication is a rule that is constrained by geminate inalterability then a crossed structure will not satisfy the structural description of the rule. The variable behaviour of imbrication in Ndebele is then attributed not to variable application of the rule but instead to variable ordering of the rule in the derivation. ‘Early’ application of imbrication before transcription, (12a), produces overapplication, and ‘late’ application of imbrication after NCC repair, (12b), produces normal application. It is when imbrication is ordered to apply to the crossed structure that exists between early and late application that the underapplication effect is produced. The crossed structure in (12c) is directly created through reduplication, and produces the situation where imbrication is blocked because it is subject to geminate inalterability. This backblocking underapplication vindicates DR and other models of reduplication which allow for segments to occur in multiple environments.

Overall, Frampton does an excellent job of laying out and investigating all of the different aspects of grammar that should play a role in understanding reduplication. The name ‘Distributed Reduplication’ could not be better as a guiding principle. The three examples discussed in this review demonstrate how different computations are required for different aspects of reduplication. The shapes found in reduplication are computed in a way that connects with work by Yu (Reference Yu2007) on infixation and Nevins (Reference Nevins2010) on how vowel harmony operates. The common theme amongst these lines of research is that prosodic effects are produced through computation, using linear searches that are sensitive to prosodic landmarks. Prosodic output targets no longer appear to be a necessary component in understanding reduplication or infixation. The other two examples discussed here demonstrate that a serial model of phonology is not only sufficient to account for reduplication phonology interaction but also necessary. This result can be understood as the now standard position, since the principal contemporary theories of reduplication, as represented by Raimy (2000, Reference Raimy, Raimy and Cairns2009), Inkelas & Zoll (Reference Inkelas and Zoll2005), Kiparsky (Reference Kiparsky, Uyechi and Wee2010) and McCarthy et al. (Reference McCarthy, Kimper and Mullin2010), all assume at least some serial processing in reduplication. Since it has been established that there is serial derivation, the question now turns to how much. Frampton's position (along with Raimy's) appears to be a centrist position on this point, with Inkelas & Zoll and Kiparsky being at the sparse derivation end (i.e. parallelism within strata) and McCarthy et al. with Harmonic Serialism in OT at the radically derivational end. There is obviously much work to be done both in reduplication and phonology in general to understand the role of derivation in grammar.

In many cases the long-term value of a work is determined by the responses it generates. There are some presentational and stylistic flaws in Frampton's work that may prevent its impact from being as great as it could be. One flaw is that almost all of the data that Frampton analyses is presented in an extremely minimalist way. This limits the ability of the reader to evaluate and appreciate Frampton's contribution. To alleviate this problem, I suggest that the reader consult the primary (or cited) literature while considering Frampton's analysis. The second flaw is closely related, and it is that there is no explanation or framing provided for the examples that Frampton picks to analyse. This prevents the reader from appreciating what the larger point or issue is that is being addressed with respect to reduplication. This too can be alleviated with familiarity with the larger literature on reduplication. The three examples that I have chosen to discuss here are prime examples of this type of flaw. Frampton does not explain the importance of accounting for reduplication in Arrernte without invoking VC syllabification, the importance of the Korean data as being a better-documented example of backcopying than the contested Malay data (see Kiparsky Reference Kiparsky, Uyechi and Wee2010 for discussion) or the previously unrecognised backblocking pattern found in Ndebele. Within the realm of research on reduplication, these three results are extremely important, but one would not gather this from reading the book. These flaws are purely presentational though, and if one is familiar with the literature on reduplication then it is clear what Frampton's contribution is. If one is unfamiliar with the literature on reduplication or hoping to become familiar with it through Frampton's monograph then there will be much work to be done.

To conclude, Frampton's Distributed Reduplication is an excellent snapshot of contemporary thought on reduplication in language. There is general agreement that serial derivation plays a central role in how reduplication and other parts of the grammar interact. There is an emerging consensus that the prosodic effects seen in the shapes of repeated regions in reduplication are produced through non-prosodic linear means. More careful work needs to be done in this line of research, because the linear calculations do reference prosodic landmarks, so the most likely best understanding of prosodic shapes in reduplication will result from a synthesis of the linear and prosodic points of view. Finally, Frampton's work clearly indicates that there is no consensus on representational issues in reduplication. Many will reject his proposals simply because he makes such bold and novel claims about representations in phonology. This would be very unfortunate, because the interaction between computations and representations is extremely complex and elegant. OT has recently adopted derivational computation as a requirement, and it is now time to return to the question of what representations are necessary to understand reduplication and phonology in general. Frampton does an excellent job of pushing this question forward.