2.1. Key proposal

The central claim of the current proposal is that transitions into or out of consonants are represented in the relevant forms, and their correspondence relations are accessible to faithfulness constraints. This claim raises three related questions. First, how are these transitions represented? Second, why should faithfulness constraints evaluate these transitions? Third, and most importantly, how do these faithfulness constraints for transitions solve the syncope problem? This subsection focusses on these questions, while related but less central aspects of the proposal, such as the representational assumptions with regard to features and the grammatical architecture to be adopted, are explored separately in subsequent subsections.

First, regarding how the transitions are represented, it is assumed here, following Flemming (Reference Flemming2002: 23 ff., Reference Flemming2006: 19, Reference Flemming2008b: 16 ff.), that they are specified as subsegments occupying time intervals starting or ending at consonant–vowel boundaries. These subsegments can contain relevant auditory features, such as F2 frequencies. Additionally, it is assumed that subsegments are subordinate to their host phonological segments: no subsegment can stand alone without a host segment.Footnote ¹ The schema in Figure 1 illustrates this with a hypothetical configuration of a medial cluster composed of two stops differing in place of articulation (cf. Flemming Reference Flemming2002: 24). Dotted lines indicate associations between subsegments and their host phonological segments.

Figure 1 The auditory subsegments and features of the sound string [atpu] (NF = noise frequency; NL = noise loudness).

Each segment illustrated in Figure 1 comprises at least one subsegment. For example, the initial vowel [a] consists of two distinct subsegments: V_mid, which represents the vowel’s steady state, and the closure transition, which is the transitional phase from the vowel to [t]. Likewise, the final vowel [u] also has two subsegments: the release transition, the transitional phase from the preceding consonant [p] to the vowel, and V_mid.Footnote ² Crucially, these subsegments, rather than the host segments themselves, serve as the repositories of features, the building blocks of phonetic detail. This implies that changes in feature values are computed in the grammar at the subsegmental level rather than the segmental level.

This assumption that features are computed at the subsegmental level answers the second question: why do faithfulness constraints need to refer to transitions? As illustrated above, most consonant features are represented in the adjacent transitions rather than the phonological consonant itself. Thus, changes in consonant properties due to cluster simplification or other phonological processes are likely to be manifested as changes in feature values within the transitions. For instance, when cluster simplification occurs, a sequence like V₁C₁C₂V₂ is neutralised to V₁C₂V₂, invoking changes in the closure transition carried by V₁ due to the removal of C₁. If the sequence [atpu] in Figure 1 neutralises to [apu], the closure transition into [t] in the source representation differs from its output correspondent aiming at [p], specifically in terms of F2 frequencies, which provide primary place cues to a consonant (Dorman et al., Reference Dorman, Studdert-Kennedy and Raphael1977). Thus, for faithfulness constraints to properly assess the changes in consonantal place of articulation resulting from cluster simplification, they must be capable of referring to the F2 values present in the transitions.Footnote ³

To illustrate how faithfulness constraints that refer to auditory features can effectively regulate cluster simplification, suppose that changes in release and closure transitions are evaluated by constraints of types schematised in (7) and (8). Adopting Correspondence Theory (McCarthy & Prince Reference McCarthy and Prince1995), I use Ident constraints for featural changes while reserving Max and Dep constraints for the deletion and insertion of segmental and subsegmental units.Footnote ⁴

The tableau in (9) illustrates how these Ident [transition] constraints evaluate the relevant changes (the transitions are denoted by superscripts). When a cluster consisting of two consonants, such as a [tp] sequence, is simplified, C₂ deletion would change the transition from [p] to [u] into a transition from [t] to [u], thus violating Ident [release transition: F2]. On the other hand, C₁ deletion would alter the transition from [a] to [t] into a transition from [a] to [p], consequently incurring a violation of Ident [closure transition: F2].

The C₂ dominance effect in cluster simplification can be captured by the hierarchical relationship between the Ident [release transition] and Ident [closure transition] constraints. The P-map idea is crucial here, stating that a constraint evaluating a perceptually more significant change must outrank a constraint evaluating a less significant change. The ranking Ident [release transition] ≫ Ident [closure transition] can be set based on the fact that listeners weigh onset signals more heavily than offset signals in perception (e.g., Fujimura et al. Reference Fujimura, Macchi and Streeter1978; Phillips et al. Reference Phillips, Hall and Boehnke2002; Wright Reference Wright, Hayes, Kirchner and Steriade2004: 43). For instance, release cues are perceptually more important than closure cues in detecting a consonant and its features (Ohala Reference Ohala, Kingston and Beckman1990; Kingston Reference Kingston and Keating1994). These faithfulness constraints formulated in (7) and (8) are thus P-map correspondence constraints, and they are universally ranked based on perceptual differences across distinct contexts:

The tableau in (11) illustrates how this ranking condition captures the C₂ dominance effect:

In the given grammar, candidate (11a) is ruled out by the top-ranked *CC constraint, which prohibits consonant sequences. Then, the consonant to be deleted is determined by the ranking condition in (10). Candidate (11c), which removes the first stop, is preferred over candidate (11b), which deletes the second. The former candidate only violates the lower-ranked Ident [closure transition: F2], while the latter is penalised by the higher-ranked Ident [release transition: F2].

Now that correspondence constraints capable of addressing the C₂ dominance effect have been established, the final question can be tackled: how do these correspondence constraints for transitions solve the syncope problem? To answer this, imagine a hypothetical source sequence [atipu] with a high vowel between two consonants, which ultimately neutralises to [apu]. The tableau in (12) describes how the C₂ dominance effect in this neutralisation can be captured. In this tableau, neither Ident [closure transition] nor Ident [release transition] is violated due to vowel deletion, as both the release transition from the first consonant and the closure transition into the second consonant are removed together with the medial vowel by syncope. In other words, Ident [release transition] is only violated by the changes in the release transition that are present both in the source and the output, as in (12c). Likewise, Ident [closure transition] penalises only the closure transition whose correspondence relationship is maintained, as in (12d). The transitions removed by syncope cannot affect the evaluations since they lack their correspondents on the surface. Consequently, (12d) is selected as the optimal output by the ranking condition Ident [release transition] ≫ Ident [closure transition].

To sum up, the C₂ dominance effect can be captured by the ranking of Ident [release transition] over Ident [closure transition], regardless of whether syncope is involved. In the current proposal, vowel deletion entails the removal of the transitions contained by the deleted vowel; thus, the correspondence in transitions between the source and output representations cannot be maintained. Consequently, Ident [release transition] and Ident [closure transition] are vacuously satisfied with regard to the removed transitions. The remaining transitions are evaluated by the correspondence constraints because they are present in the adjacent vowels even after one member of the cluster is removed. In this sense, syncope provides an environment that is not technically different from the condition where a vowel does not intervene between the two consonants.

2.2. Representation of the features

In (7) and (8), I formalised the Ident constraints for subsegmental features without detailing their nature. Since the crucial aspect of the proposal is the presence of consonantal features in transitions, the proposal is compatible with various feature specifications, provided that transitions are represented in relevant forms and referenced by faithfulness constraints. For instance, transitions may be represented with conventional SPE-style binary features like [±labial] or [±nasal] instead of the auditory features in Figure 1. Using these conventional features raises no critical issue in addressing the syncope problem.

Consider in detail how such binary features can be employed with the current proposal. For this purpose, it is assumed that release and closure phases are inserted during consonant adjacency, inheriting relevant binary features from it. Specifically, in some existing subsegmental theories, the proposal could be implemented with the assumption that featural inheritance is controlled by surface correspondence relationships, as in Agreement by Correspondence (ABC) theories (e.g., Walker Reference Walker2000; Rose & Walker Reference Rose and Walker2004). The following schema illustrates a potential subsegmental representation within a framework drawing on ABC+Q-theory, which combines ABC with Q-theory (Shih & Inkelas Reference Shih and Inkelas2019).Footnote ⁵ Consonantal subsegments and transitions are coindexed, allowing features to be inherited from a consonant by its adjacent subsegmental transitions. This inheritance occurs when two conditions are met: first, a constraint enforcing surface agreement between these subsegments is ranked high; and second, a change in transition is preferred over a change in the consonantal subsegment itself.Footnote ⁶

With these binary features and surface correspondence relationships, the analytic consequences are the same as in (12), as illustrated in the tableau in (14):

Alternatively, auditory dimensions like F1 or F2 can be represented using simple discrete values (Flemming Reference Flemming2002). For instance, the F2 feature in a transition adjacent to [t] can be specified with a higher value (e.g., 4), while a transition adjacent to [p] can be assigned a lower value (e.g., 1), as in (15). These values reflect the F2 loci of the relevant consonants. This strategy remains compatible with the current proposal, provided that changes in these discrete values are penalised by the relevant P-map constraints.

Transition features may not have specific values but rather exist as unary phonetic cues. For example, coronal transitions can be manifested by [high F2 frequency], while labial transitions are represented by [low F2 frequency], without the need for specific feature values. Within the Bidirectional Phonetics and Phonology (BiPhon) framework (Boersma Reference Boersma2007, Reference Boersma, Benz and Mattausch2011, Reference Boersma, Noske and Botma2012; Boersma & Hamann Reference Boersma and Hamann2008, Reference Boersma, Hamann, Boersma and Hamann2009; Boersma & Leussen Reference Boersma and Leussen2017), cue constraints that refer to such transitional cues can theoretically capture the C₂ dominance effect in the syncope context. This is illustrated in the tableau in (16), where [^t] represents [high F2 frequency], and [^p] denotes [low F2 frequency].

In (16), the BiPhon model is implemented within parallel OT. For the underlying form |atipu|, potential pairs of a surface form (in slashes) and phonetic form (in square brackets) are given as candidates. The faithfulness constraints presented here are of two types: those evaluating the mapping from an underlying representation to a surface form (e.g., Max-C) and cue constraints regulating the mapping from a surface form to a phonetic form (e.g., */pV/[_^tV] and */Vt/[V^p_]). Similarly, markedness constraints operate on different representational levels: some evaluate surface forms (e.g., Syncope), while others evaluate phonetic forms (e.g., *[V^t_^pV]). With these constraints, the grammar correctly selects candidate (16d) as optimal. Details are as follows. For |atipu|, its surface representation should be /atpu/ due to the high-ranked Syncope constraint and Max-C. This surface form is then phonetically realised as [a^p_^pu] under the pressure of the constraint that demands the same frequency feature on both sides of the acoustic silence (i.e., *[V^t_^pV]). The alternative phonetic form [a^t_^tu] in (16c) is ruled out due to the P-map ranking between the cue constraints, in which */pV/[_^tV], which protects the release transition, outranks */Vt/[V^p_], which preserves the closure transition. Once again, the crucial role is played by the distinction between the release transition and the closure transition, as determined by the constraint ranking reflecting the P-map hierarchy.

Although there are various potential ways to specify transitional features, including those previously mentioned, I assume that considerable phonetic detail is represented in relevant forms for the complete formulation of the P-map constraints. Given that subsegmental transitions are assumed to represent real acoustic events at consonant–vowel boundaries rather than being abstract units, it is natural to suppose that features encoded by such subsegmental units reflect detailed acoustic events. Therefore, I assume that transitions encode auditory features like F1 or F2, rather than abstract features.

In addition, it is assumed that these auditory features are context-dependent and continuous. This is based on the phonetic observation that there are no clear invariant acoustic cues for some features, a phenomenon known as the lack of invariance in the speech signal (see Delattre et al. Reference Delattre, Liberman and Cooper1955). For example, a consonant’s place of articulation can be detected by F2 transition properties, but there are no absolute values serving as invariant landmarks for each place feature, since relevant F2 frequencies vary as a function of the vocalic context. In fact, not only the F2 transition but also many other acoustic cues are context-dependent, while invariant acoustic cues are rare (Reetz & Jongman Reference Reetz and Jongman2020: 227; cf. Blumstein & Stevens Reference Blumstein and Stevens1980). Note that this assumption has additional advantages: grammatically controlled language-specific phonetic events, such as coarticulation, can be consistently analysed within the same framework as their phonological counterparts, such as place assimilation or cluster simplification (e.g., Flemming Reference Flemming2001, Reference Flemming, Embarki and Dodane2011). While this article focusses on phonological processes with contrastive outcomes, the boundary between such processes and their corresponding phonetic events is often unclear. Given this, I maintain the position that phonological processes, such as cluster simplification, can be derived through the computation of continuous auditory features, leaving open the possibility of a unified model of phonetics and phonology.Footnote ⁷ However, as mentioned earlier, the assumptions about the nature of these features are not central to the proposal. Readers should have no problem understanding the analyses from §3 onward, even if they assume one of the alternatives mentioned above instead of the specific suggestion that the features are continuous and context-dependent.

On the assumption that features are considerably detailed, I propose that the correspondence constraints for the transitions, previously introduced in (7) and (8), can be more precisely defined as follows:

These constraints evaluate the perceptual differences in the release and closure transitions between representations i and j. Assuming that the transitions contain only features representing the first three formant frequencies (F1, F2 and F3) at the onset or offset of the trajectory, these specifications can be quantified as in (19).Footnote ⁸ The d-value denotes the perceptual distance between two transitions.

This models the perceptual distances of Bark-scaled formant frequencies (Traunmüller Reference Traunmüller1990) between i and j. To illustrate, consider English preconsonantal t/d deletion, such as /bæd bɔɪ/ → [bæ^bb^bɔɪ] (Dilley & Pitt Reference Dilley and Pitt2007). In English, the onset frequencies of the first three formants in Barks for [bɔ] are 3.80 (F1), 9.26 (F2) and 14.17 (F3), while the values for [dɔ] are 3.53 (F1), 11.66 (F2) and 14.63 (F3) (Kewley-Port Reference Kewley-Port1982). Using these measurements and the metric in (19), the d-value between [bɔ] and [dɔ] at the onset of [ɔ] can be calculated as 1.42, assuming $a = b = c = 0.\bar {3}$ (summing to 1.00). Thus, the change from [bɔ] to [dɔ] violates Ident [release transition] > 0.5, which penalises onset changes with d-values exceeding 0.5, but not Ident [release transition] > 1.5, as the d-value is less than 1.5. Similarly, the d-value for the perceptual difference between [æb] and [æd] can be obtained as 0.94, given the formant values in Barks: 4.32 (F1), 11.21 (F2) and 13.51 (F3) for [æb] and 3.73 (F1), 12.25 (F2) and 14.62 (F3) for [æd]. In principle, the English cluster simplification case can be analysed based on these d-values, as in the tableau in (20).

In (20), the faithfulness constraints are ranked hierarchically according to the P-map principle. In other words, this stratification means that more perceptually distinct changes incur more severe faithfulness violations. In addition to this basic principle that larger changes in the transition correspond to a larger d-value, considering that onset changes are perceptually weighted more than offset changes, the P-map principle implies the following hierarchical relationship among the Ident [transition] > $d_{ij}$ constraints:

The condition in (21) states that changes in the release transition are perceptually more significant than those in the closure transition for the same d-value, allowing the possibility that a minor change in the release transition can be overridden by a more significant change in the closure transition.

This formalisation directly implements the P-map by being sensitive to actual acoustic-perceptual differences. The core proposal of the P-map is that it guides speakers to identify the smallest perceptual change from the input, minimising overall deformation (Steriade Reference Steriade, Hanson and Inkelas2009). This idea can, in theory, be grammatically encoded using correspondence constraints in various formalisations. However, the extent to which these different forms of faithfulness constraints reflect the P-map can vary. For instance, a P-map ranking of contextual faithfulness constraints such as Ident (place) /__V ≫ Ident (place) /__C suggests that the perceptual difference for the same place contrast is invariably greater in the prevocalic context than in the preconsonantal context. However, in fact, small differences in CV transitions can be overshadowed by significant differences in VC transitions. As Steriade (Reference Steriade, Hume and Johnson2001) points out, the t/ʈ apical contrast relies primarily on formant differences in the closure transition, particularly in F3, while the release transition’s contribution is negligible because the tongue tip slides towards the alveolar position during the retroflex closure. Consequently, in a Swedish word like /væɖslig/ ‘worldly’, realised as [væɖʂɭig] (Eliasson Reference Eliasson and Andersen1986: 280; Hamann Reference Hamann2003), apical assimilation shows a progressive direction, indicating a preference for a minimal change in the release transition over significant changes in the closure transition.Footnote ⁹ In contrast, the Ident [transition] constraints, which reflect the exact acoustic difference between forms, more effectively explain such progressive apical assimilation. One hypothesis consistent with (21) is that changes in the closure transition take precedence over changes in the release transition when the d-value associated with the closure transition is significantly larger. For example, while Ident [closure transition] > 0.5 should be outranked by Ident [release transition] > 0.5 according to the condition in (21), Ident [closure transition] > 2.5 does not necessarily need to be outranked by Ident [release transition] > 0.5. A ranking in which Ident [closure transition] > 2.5 dominates Ident [release transition] > 0.5 could lead to progressive assimilation. Therefore, this formulation, which directly refers to the d-value, more realistically reflects the P-map.

However, despite these considerations, it is true that (21) is too intricate for analytical convenience, requiring exact d-values for each change and the threshold where strict onset dominance can be subverted. I henceforth assume that the perceptual effects of potential onset changes from cluster simplification generally exceed those of offset changes, as shown in (22).Footnote ¹⁰

This allows further simplifications of the relevant constraints, abstracting away from detailed measurements. Consider again the English cluster simplification case analysed in (20). Given the condition in (22), the tableau in (20) can be reduced to the simpler grammar in (23):

This is similar to (11), except that more features, that is, F1 and F3, are considered here according to the metric in (19). By assuming the ranking condition in (22), it is possible to reduce the effort involved in measuring d-values and thresholds while maintaining formal rigour.

Considering the above, for the remainder of this article, I will adopt two simplifications for analytical convenience. First, given that the relevant onset changes are, in general, perceptually more significant than the offset changes in the cluster simplification contexts, the d-values are not fully specified. Second, all relevant feature specifications are calculated by a metric, so features are not manifested on constraints unless necessary. These strategies allow capturing C₂ dominance with a simple ranking of Ident [release transition] over Ident [closure transition], abstracting away from detailed measurements.

2.3. Framework: the Realised Input model

I adopt the Realised Input model of phonological grammar (Flemming Reference Flemming2008b) as a framework for the current proposal. This decision is not based on the superiority of this model over alternatives in treating the syncope problem with subsegmental features. Rather, it is due to the fact that the Realised Input model is specifically designed to allow P-map correspondence constraints to reference phonetic details, thereby connecting the P-map principle with the roles of phonetic details in phonology. Since the current proposal argues for the role of subsegmental details in conjunction with the P-map principle, it would be a good choice to rely on a framework designed to address faithfulness to phonetic details, without further assumptions. Additionally, adopting the Realised Input model provides an ancillary benefit in the context of cluster simplification, because cross-linguistic patterns of cluster reduction processes, including cluster simplification, are determined by phonetic generalisations beyond simple phonological considerations: cluster reduction is not reported in languages where C₁ has a phonetically audible release (e.g., Jun Reference Jun2002; Flemming Reference Flemming2008b).Footnote ¹¹ If cluster reduction is at least partially determined by such phonetic factors, the grammar needs to refer to those facts. Thus, introducing subsegments alone may be insufficient for modelling cluster reduction, as it only provides a solution to the syncope problem, which is just one of several issues related to cluster reduction processes. Although this article focusses on solving the syncope problem, it is ideal if the solution to the syncope problem is carried out within a framework that is claimed to provide a solution to other issues concerning cluster reduction processes.Footnote ¹²

Figure 2 illustrates the basic architecture of a simplified version of the Realised Input model. For the current purposes, the model’s details are simplified by assuming there are only two grammatical components: Phonetic Realisation and Phonotactics. For the full version of the model and a detailed discussion, see Flemming (Reference Flemming2008b) or Kim (Reference Kim2024c: ch. 2).

Figure 2 Simplified version of the Realised Input model of phonology.

In this model, raw inputs with abstract sound strings are first sent to the Phonetic Realisation component, which assigns phonetic substance to a raw input and returns a realised input. This component is called Phonetic Realisation in that it determines language-specific phonetic patterns, resulting in phonetically enriched representations. The Phonotactics component receives the realised input and returns the surface output. This component essentially determines a language’s surface phonological patterns through the interplay between markedness constraints that encode phonotactic requirements and faithfulness constraints that refer to phonetic details. This component is referred to as Phonotactics to differentiate it from Phonetic Realisation, as it regulates surface phonotactic patterns. Although these two components share constraints and their rankings, they differ in that Phonetic Realisation activates only a subset of those constraints, while Phonotactics activates all constraints. Following Kim (Reference Kim2024c), it is assumed here that Phonetic Realisation activates, in particular, two types of constraints: phonetic articulatory constraints that determine gestural coordination or segmental duration, and cue realisation constraints that demand the realisation of target phonetic cues.

To illustrate how this architecture works and how phonology can refer to language-specific phonetic details, consider a hypothetical example where a coronal stop is deleted before another stop (as in /atpa/ → [a^pp^pa]). As already mentioned, empirical evidence suggests that cluster reduction processes, including simplification, largely depend on the presence or absence of an audible release of C₁. For a language to allow cluster reduction, it should not have a release of C₁, possibly due to the fact that such a release strongly signals the presence of C₁ (e.g., Jun Reference Jun2002; Flemming Reference Flemming2008b). The tableau in (24) illustrates how such phonetic details are provided to /atpa/ by the Phonetic Realisation component:

When the two consonants are not overlapped, subsegmental transitions could be realised by connecting them to the adjacent consonants even without a host vowel (see fn. 1). However, in the given grammar, such an open transition is not allowed because the Overlap constraint, which requires adjacent consonants in the input to be articulatorily overlapped, outranks the cue realisation constraint for transitions, ruling out candidate (24a) (cf. Gafos Reference Gafos2002). Although the cue realisation constraint requires the realisation of release and closure transitions with target feature values, such as F2 target values, for each consonant in the input, the enforced gestural overlap between two consonants prevents /t/ from realising its release transition and /p/ from realising its closure transition.Footnote ¹³ Consequently, the grammar selects (24b) as optimal, where the consonants are articulatorily overlapped without the realisation of the transitions between them.

Then, constraints in Phonotactics, including the P-map correspondence constraints and *CC, can access the relevant phonetic detail: whether C₁ is released or not. This component is where our main interest lies: all processes that have contrastive and phonological consequences, such as neutralisation, deletion, or insertion, take place as a result of interactions between the relevant markedness constraints and the faithfulness constraints in this component. The tableau in (25) shows how cluster simplification can occur in Phonotactics when C₁ is not released. Suppose that Ident [release transition] is universally undominated, in that there is no convincing evidence demonstrating that cluster reduction processes effectively remove buccal C₂ or alter its major place of articulation.Footnote ¹⁴ This implies that other constraints, including *CC, are generally dominated by Ident [release transition], or at best, tied with it as in (25). Suppose that *CC is sensitive only to clusters without C₁ release, reflecting the observation that cluster reduction processes do not affect clusters where C₁ is released. Given these assumptions and the fact that Overlap outranks Realise [transition], when *CC triggers cluster reduction by dominating Ident [closure transition], the outcome must be (25d), which removes C₁.

In contrast, the Phonotactics component does not yield cluster simplification when the given realised input form has an open transition between the two consonants. Given that a realised input form with a released C₁ suggests that the Realise [transition] constraint outranks the Overlap constraint in Phonetic Realisation (cf. (24), where Overlap outranks the cue realisation constraints), Realise [transition] is also ranked over Overlap in Phonotactics. Under this ranking condition, although the relative ranking of other constraints, including *CC, remains unchanged, the result is now different: cluster simplification is blocked by Realise [transition]. Since Realise [transition] requires the realisation of the release and closure transitions of the consonants in the input forms, the grammar opts to maintain the open transition of the realised input form, ruling out candidates that do not meet this requirement.

Although a serial implementation of the grammatical architecture has been shown thus far for illustrative purposes, it is, in principle, possible to perform a parallel implementation without separating the two components of Phonetic Realisation and Phonotactics, as these two components share the same constraint ranking (see also Kim Reference Kim2024a,Reference Kimb,Reference Kimc). The tableau in (27) shows an example of such a parallel implementation.

Constraints in the tableau are categorised into two groups: phonetic realisation constraints (marked with a superscript pr) and general phonological markedness and correspondence constraints. Phonetic realisation constraints, such as Overlap and Realise [transition], were the only activated constraints in the Phonetic Realisation component in the serial implementation (see (24)). It is worth noting that the correspondence constraints here evaluate the relationship between the designated candidate (here (27a), marked with D) and other candidates rather than directly referring to the input (cf. McCarthy Reference McCarthy1999; Jun Reference Jun2002). The designated candidate is chosen only by phonetic realisation constraints from candidates that have not undergone contrastive changes like neutralisation, deletion or insertion (Flemming Reference Flemming2008b; Kim Reference Kim2024a,Reference Kimb,Reference Kimc). In this tableau, (27a) and (27b) are such candidates. Between these two, which differ only in the phonetic release of C₁, (27a) is selected as the designated candidate because it satisfies the higher-ranked Overlap constraint among the phonetic realisation constraints. The optimal candidate is then determined by evaluating all constraint violations, with (27d) preserving the release transition specified in (27a).

Within this Realised Input model, whether serial or parallel, the introduced constraint set generates a stable prediction about potential typological patterns involving syncope. When the top-ranked Syncope constraint enforces interconsonantal vowel deletion, only three patterns can emerge from the given constraints and their P-map ranking: [a^tt^t◦^pp^pa] (syncope with released C₁), [a^tt˺p^pa] (syncope with unreleased C₁), and [a^pp^pa] (syncope with C₁ deletion). The grammar does not predict the potential output with C₂ deletion, *[a^tt^ta], due to the ranking Ident [release transition] ≫ Ident [closure transition]. For instance, when Ident [release transition] is ranked highest with Syncope, and Ident [closure transition] is dominated by the constraints in the top and second tiers, it is possible to posit the factorial typology in (28) based on the relative ranking of the remaining constraints (Max-C is assumed to be ranked lowest, ensuring cluster simplification as a response to the *CC constraint).Footnote ¹⁵ The model is implemented in a parallel way, as in (27).

From the typology presented above, it is evident that cluster simplification is possible only when both Overlap and *CC dominate Realise [transition]. The ranking of Overlap over Realise [transition] means that a consonant cluster could be realised without C₁ release if there were no cluster simplification, aligning with the observation that cluster reduction processes only affect clusters without C₁ release. In other words, the typology predicts that cluster simplification, even in the syncope context, can only take place in a grammar where it could result in a cluster without C₁ release if there were no cluster simplification (i.e., only in the grammar where Overlap ≫Realise [transition]). In contrast, a grammar that would result in a cluster with C₁ release never undergoes cluster simplification, even with a higher ranked *CC, as in (28e). When Realise [transition] outranks Overlap, the expected outcome is a cluster with C₁ release, even after the application of syncope.

Given that our primary focus is on the interaction between cluster simplification and syncope, subsequent sections will operate under the assumption that in the full grammar *CC is undominated and Overlap outranks Realise [transition], invariably resulting in cluster simplification, as in (28f). Under this assumption, I will not present constraints that are not central to the discussion, such as Overlap and Realise [transition], and will instead focus on the P-map ranking of the Ident [transition] constraints and their interaction with phonotactic markedness constraints.

In summary, the Realised Input model provides a formal apparatus for P-map constraints that require reference to substantial phonetic details. Regardless of whether the model is implemented serially or in parallel, it allows phonological patterns to be determined based on phonetic details such as the realisation of C₁ release or formant transitions. In this respect, the Realised Input model is beneficial in explaining the syncope problem within a broader context involving phonetic details such as C₁ release. It can be used not only to solve the syncope problem but also to provide a grammatical account of the phonetic generalisation underlying cluster reduction processes.

3.1. Medial cluster simplification

As a motivation for cluster simplification, I reformulate the *CC constraint to penalise only sequences of supralaryngeal heterorganic consonants, excluding a cluster with a glottal consonant.

Languages may remove one consonant in the target cluster to satisfy this constraint unless a potential violation of Max-C is fatal. However, this markedness constraint can also be satisfied by a geminate or a homorganic cluster, which preserves the total duration of the consonants without a violation of Max-C. Therefore, it is necessary to consider factors that favour deletion over forming a geminate or homorganic cluster. The constraint in (30) penalises lengthened gestures formed by geminates and homorganic clusters:

Ranking this constraint over Max-C enables cluster simplification instead of place assimilation. For example, consider the Basque cluster simplification in (2a), repeated in (31):

In this language, neither heterorganic clusters nor geminates are preferred over consonant deletion, suggesting that both *CC and *LongC outrank Max-C.Footnote ¹⁶ The tableau in (32) illustrates this:

As presented in the tableau, consonant deletion results in changes in the relevant transitions.Footnote ¹⁷ Deleting [t] in the realised input can change the closure transition, as in candidate (32e), while removing [k] can alter the following release transition as in (32d). The perceptual significance of these changes determines which consonant is targeted for deletion. The ranking of Ident [release transition] over Ident [closure transition] reflects the perceptual difference between release and closure transition, favouring (32e) over (32d).

Note that potential geminate formation can also result in changes in transitions. This suggests that the directional asymmetry in place assimilation can also be accounted for by the ranking of the same Ident constraints. This possibility will be discussed in detail in §4. In the present case, the C₂ dominance effect is accompanied by cluster simplification. Candidate (32c), which preserves consonant duration, is ruled out by *LongC, although both (32c) and (32e) satisfy Ident [release transition].

An example of cluster simplification fed by syncope can be found in Tunica. In this language, word-final high vowels are syncopated when they are followed by another word with initial-syllable stress, creating a potential VʰkCV cluster, which is illicit in the language. Thus, to avoid this illegal cluster, a further application of the simplification process is enforced, removing the velar stop. If the second consonant is another stop, only [k] is deleted, leaving its pre-aspiration as in (33a). If the second consonant is a sonorant, then both [k] and the pre-aspiration are removed as in (33b).Footnote ¹⁸

The key to capturing the C₂ dominance effect in the syncope context is identifying the relevant correspondence relationships. Since syncope removes the interconsonantal high vowel, its associated transitions are also absent from the surface. The Ident [transition] constraints only evaluate changes in transitions with correspondences between the relevant representations, meaning that the removed transitions do not affect the evaluations given by these correspondence constraints. In other words, the Ident [transition] constraints do not penalise the loss of the transitions in the syncopated vowels, although they may be penalised by other constraints such as Max [transition] or Realise [transition].

In (34), the vowel deletion process is controlled by the top-ranked Syncope constraint, which eliminates candidate (34a), and *CC disallows an intervocalic heterorganic cluster as in (34b). The remaining grammar determines which consonants are affected and how the cluster is repaired. The ranking of Ident [release transition] over Ident [closure transition] identifies C₁ as the target based on changes in transitions present in both the realised input and the provided surface candidates. The ranking of *LongC over Max-C favours cluster simplification over place assimilation as a repair strategy.

3.2. Simplification of triconsonantal clusters

The P-map principle predicts that, in VC₁C₂C₃V clusters, the second consonant is the target for deletion because it is less distinct from $\varnothing $ than the first and the third consonants are. Experimental literature supports this prediction (Haggard Reference Haggard1973; Klatt Reference Klatt1973), as discussed by Steriade (Reference Steriade, Hanson and Inkelas2009: 169). When VC₁C₂C₃V clusters are shortened, they are confused with VC₁ːC₃V and, in turn, with further simplified VC₁C₃V. This prediction is derivable from the current implementation of the P-map theory. To illustrate this more clearly, consider optional interconsonantal stop deletion in Hungarian, in which a stop surrounded by two non-approximant consonants may undergo optional deletion, reducing triconsonantal clusters to biconsonantal clusters (Côté Reference Côté2000; Siptár & Törkenczy Reference Siptár and Törkenczy2000).

To analyse this triconsonantal cluster simplification, following Steriade (Reference Steriade, Hanson and Inkelas2009: 170), I assume that the process is motivated by the size-of-cluster constraint C//V, defined in (36):

With this markedness constraint, the default ranking of Ident [transition] constraints predicts C₂ to be the target consonant, since its deletion does not yield a change in transitional properties. In other words, C₂ deletion in the triconsonantal cluster does not violate any Ident [transition] constraints because this consonant lacks transitions into or out of it.Footnote ¹⁹

Here, the Ident [transition] constraints are specified for F2 and nasality, meaning that the perceptual difference between transitions is evaluated mainly based on factors such as nasality and second formant frequencies. The emphasis on nasality is due to the homorganic nature of the first two consonants in the cluster, which have similar formant values, with the main difference being their potential to nasalise adjacent transitions. As can be seen in the tableau, the optimal output is (37d), because deleting C₂ does not result in any changes to the F2 or nasality of the transitions.

The current proposal also captures the interaction between syncope and simplification of triconsonantal clusters. Consider cluster simplification in Tangale, a West Chadic language spoken in Nigeria. In Tangale, a stem-final vowel is removed when it is followed by an inflectional suffix or an NP complement. As illustrated in (38), when this syncope process results in a triconsonantal cluster whose first two members are homorganic, cluster simplification occurs as a repair strategy.

The analysis of this process is illustrated in the tableau in (39).

In the analysis, the top-ranked Syncope constraint removes the stem-final interconsonantal vowel, eliminating candidate (39a) and the C//V constraint disallows the resulting triconsonantal cluster in (39b). The consonant to be deleted is determined by the hierarchy of the Ident [transition] constraints. C₃ cannot be removed, as in (39c), as it causes a change in F2 values, violating Ident [release transition: F2 & nasal]. Similarly, removing C₁ in (39d) yields a change in the nasality of the closure transition, violating Ident [closure transition: F2 & nasal]. Since removing the second consonant does not violate any Ident [transition] constraints, candidate (39e) is selected as the optimal output.

Note that the contextual faithfulness analysis fails to predict the Tangale pattern. Although a ranking of the relevant P-map constraints for segmental contexts – such as Max-C/V__V ≫ Max-C/C__V ≫ Max-C/V__C ≫ Max-C/C__C – can generally capture the simplification process targeting triconsonantal clusters, these constraints lose their analytical power when the target cluster is fed by syncope, as in Tangale. The tableaux in (40) show that both input- and output-oriented contextual faithfulness constraints fail to predict the correct optimal output.

3.3. Apocope and final cluster simplification

A final cluster can also be targeted by simplification processes, as seen in Lardil. In this language, a non-apical second consonant of a final cluster cannot surface due to restrictions on complex final clusters. This final cluster simplification can be fed by apocope, which targets nominals with more than two morae. The examples in (41) demonstrate that final vowels and C₂s in clusters, which are present in future and non-future forms, are absent in uninflected forms. In these uninflected forms, the vowels and C₂s are not followed by a suffix and are therefore subject to apocope and final cluster simplification.

Because final cluster simplification in Lardil involves apocope, conventional contextual faithfulness constraints face again the same problem as in syncope contexts. If the relevant faithfulness constraints are input-oriented, the grammar predicts that C₁ will be the target consonant rather than C₂, since C₂ can be protected by higher-ranked Max-C/__V _input as the final vowel is present in the input. This is shown in (42):

The output-oriented alternative also cannot rescue the contextual faithfulness constraints. The tableau in (43) shows that output-oriented faithfulness constraints cannot identify the target consonant, since whichever of C₁ and C₂ is retained in the surface form will be postvocalic and word-final, as in candidates (43c) and (43d).

In contrast to the contextual faithfulness account, the interaction with apocope is not problematic in the current approach, as illustrated in (44):

As shown in (44), since the release transition hosted by the final vowel in the realised input form is removed by apocope, there is no surface correspondent of the release transition out of the derived final consonant. Therefore, the removal of the final consonant in candidate (44c) does not critically violate any Ident [transition] constraints. In contrast, a potential deletion of the first consonant of the cluster, as in candidate (44d), is correctly ruled out because it fatally violates Ident [closure transition].

It is worth noting that what happens in Lardil is, in fact, more complicated than what has been analysed above. In Lardil, labial and velar consonants are deleted in final position even when they are not part of final clusters (e.g., /wuŋkunuŋ/ → [wuŋkunu] ‘queenfish’; Staroverov Reference Staroverov, Assmann, Bank, Georgi, Klein, Weisser and Zimmermann2014: 431).Footnote ²⁰ This invokes a complex opacity problem called ‘fed counterfeeding’ by Kavitskaya & Staroverov (Reference Kavitskaya and Staroverov2010), in which apocope and final deletion have both feeding and counterfeeding relationships. As a result of this opaque interaction, apocope underapplies when it and final consonant deletion have already applied, as in (45).

The mapping from /dibiɾdibi/ to [dibiɾdi] is opaque because both the input and output forms end in a vowel, demonstrating the underapplication of the apocope process. In other words, although the final vowel in [dibiɾdi] meets the environmental condition for apocope, it remains unaffected, resulting in opacity. This is because apocope has already occurred before this final vowel emerges in the derivation. While such an opaque interaction is challenging to analyse within classic OT, the current approach, in which correspondence constraints can reference phonetically detailed realised inputs, handles it effectively. In recent studies (Kim Reference Kim2024a,Reference Kimb,Reference Kimc), I have demonstrated that faithfulness to phonetically enriched representations provides a comprehensive solution to opacity. For instance, in the Lardil case, it can be assumed that phonetically longer vowels are more strongly protected by the relevant faithfulness conditions than phonetically shortened vowels, which may instead undergo complete deletion. Indeed, cross-linguistically, vowels in final positions frequently have shortened durations due to factors such as devoicing (e.g., Myers & Hansen Reference Myers and Hansen2007). If the Phonetic Realisation component derives this vowel shortening, the resulting shortened vowels can be clearly distinguished from regular vowels that fully maintain their duration in realised input forms. The tableau in (46) illustrates how these shortened vowels are derived within the Phonetic Realisation component:

Suppose that the shortened vowel has a duration between zero and that of the full vowel [i] (e.g., [i] = 80ms and [i̯] = 40ms). This phonetically shortened final vowel is subsequently eliminated on the surface by the *V̯ constraint, which penalises overly shortened vowels. Ident [V: duration] > 30ms, which penalises changes in length of the degree of the difference between [i̯] and $\emptyset $ or [i̯] and [i] (40ms), is ranked lower than Ident [V: duration] > 60ms, which penalises changes of a greater degree, such as a change from [i] to $\emptyset $ (80ms), according to the P-map hierarchy. Consequently, in Phonotactics, short vowels (i.e., underlyingly word-final vowels) are deleted, whereas vowels of normal length (including vowels whose word-final position is derived) are preserved. The tableau in (47) illustrates how this surface opacity is derived. Note that a transparent output like candidate (47e) could be optimal if Final V Shortening were ranked above all Ident [V: duration] constraints.

4.1. Place assimilation

Cross-linguistically, place assimilation in medial clusters is typically regressive, targeting the first member of the cluster rather than the second (Webb Reference Webb1982; Mohanan Reference Mohanan and Goldsmith1993; Jun Reference Jun1995, Reference Jun, Hayes, Kirchner and Steriade2004). This directional asymmetry in place assimilation is illustrated in (48) with Korean examples, where non-velar nasals and stops assimilate in place to the following stops.

Unlike cluster simplification, the emergence of place assimilation as a repair strategy for *CC can be attributed to the ranking of Max-C over *LongC. To satisfy the higher-ranked Max-C constraint, this language prefers forming a geminate consonant or homorganic sequence over deleting a consonant, as shown in (49):

In (49), place assimilation is triggered by the high-ranked *CC constraint, which rules out the faithful candidate (49a), similarly to typical cluster simplification cases. The C₂ dominance effect is also captured by the same Ident constraints, and their ranking, that were implemented for the analyses of cluster simplification (cf. Flemming Reference Flemming2008a, who employs similar correspondence constraints to explain the manner asymmetry in place assimilation). C₁ is the target of place assimilation, since changing its place incurs only a non-fatal violation of Ident [closure transition], as in the winning candidate (49c), while changing the place of C₂, as in (49b), is penalised by the higher-ranked Ident [release transition] constraint.

Place assimilation fed by syncope can also be captured by the same analysis. Keley-i (Austronesian, spoken in the Philippines) presents such a case. In this language, when an infix with a nasal attaches to referent-focus past roots with CeCeC or CeCaC forms, the first root vowel is removed, and the infix-final nasal assimilates in place to the following consonant.

The same set of Ident constraints accounts for the C₂ dominance effect in both cluster simplification and place assimilation, so the regressive direction of place assimilation conditioned by syncope should also be explainable by these constraints. This is depicted in the tableau in (51). The ranking of the Ident [transition] constraints determines the direction of the process, and Max-C outranking *LongC prevents deletion from being a potential repair for heterorganic clusters.

4.2. Debuccalisation

In some languages, cluster reduction is achieved by debuccalising the first member of the cluster, showing the C₂ dominance effect (McCarthy Reference McCarthy2008). For instance, when a medial cluster undergoes debuccalisation, it always results in the form [VʔCV], not [VCʔV], as observed in languages such as Arbore (Hayward Reference Hayward1984), Cariban languages (Gildea Reference Gildea1995), Malay, Thai (Lodge Reference Lodge1992) and Toba Batak (Hayes Reference Hayes1986). This subsection demonstrates that debuccalisation affects adjacent transitions and can be regulated by the Ident [transition] constraints, resulting in the C₂ dominance effect. This effect is consistent regardless of the presence of syncope.

Before performing a specific analysis, let us consider the acoustic characteristics that make glottal consonants distinct from buccal ones. As is well-known, glottal stops are usually not realised with complete closure in most languages (Ladefoged & Maddieson Reference Ladefoged and Maddieson1996: 74–76). These articulatory properties of glottal stops also influence their acoustic realisation. Although they inherently impose characteristics such as creaky voice or stiff phonation, in intervocalic positions they are likely to display formant structures mediating between surrounding vowels. Similarly, in consonant clusters like [VʔCV], glottal stops exhibit evident formant transitions reflecting the characteristics of the adjacent supralaryngeal consonant (cf. Foulkes & Docherty Reference Foulkes, Docherty, Ohala, Hasegawa, Ohala, Granville and Baily1999). Glottal fricatives also show similar transitional characteristics. The vocal tract shape of the glottal fricative is usually similar to its neighbouring vowels, and its transitional characteristics reflect its surrounding environment (Keating Reference Keating1988). Although [h] is voiceless, major formant landmarks are apparent during its production, suggesting that glottal fricatives in clusters can also mediate transitions between vowels and adjacent buccal consonants. These acoustic facts demonstrate that debuccalising a consonant can affect the properties of the adjacent transition by projecting the adjacent consonant’s characteristics to a vowel. For instance, when [a^ttp^pa] undergoes C₁ debuccalisation, the properties of C₂ [p] can be projected across the resulting glottal consonant, as in [a^pʔp^pa] or [a^php^pa].

These acoustic properties of glottal consonants suggest that debuccalisation can also be regulated by Ident [transition] constraints. To illustrate, consider the debuccalisation process in Toba Batak, exemplified in (52). In this case, a word-final stop undergoes debuccalisation when it is followed by another consonant.

To analyse this pattern, I introduce a constraint that penalises voiceless transitions. This constraint reflects the need to ensure that the transitions are as amplified as possible, keeping the higher perceptual resolution. To put it more simply, this constraint prohibits consonantal properties from being projected onto transitions in a weakened manner as they pass through the glottal consonant.

With this constraint, debuccalisation in Toba Batak can be analysed as in (54). In this tableau, the ranking of *LongC over *VlsTrans determines that reduction is achieved through debuccalisation rather than place assimilation. Then, the ranking of Ident [release transition] over Ident [closure transition] captures the C₂ dominance effect.

Syncope does not pose a problem for the analysis of debuccalisation. Consider the Panare examples in (55), where syncope feeds debuccalisation. In this language, when the verb -utu- ‘give’ is followed by a CV-initial suffix, the stem’s final high vowel is deleted, feeding debuccalisation of the preceding stop.

The tableau in (56) analyses this pattern. First, candidate (56a), which preserves the realised input structure, violates the top-ranked Syncope. Since this syncope removes all correspondence relationships the deleted vowel could host, further deletion of a consonant does not affect the evaluation provided by Ident [release transition] or Ident [closure transition]. Then, C₁ is determined as the target by the violations related to the remaining surface vowels. C₁ undergoes debuccalisation, not place assimilation, since *LongC outranks *VlsTrans.Footnote ²¹

6.1. Harmonic Serialism

McCarthy (Reference McCarthy2008, Reference McCarthy2009, Reference McCarthy2011) argues that C₂ dominance effects in cluster reduction processes and the associated syncope problem can be accounted for within the framework of HS. As outlined in McCarthy (Reference McCarthy2011), P-map correspondence constraints in the form of contextual faithfulness constraints can be implemented in HS to derive the expected results correctly. The tableaux in (61) illustrate how McCarthy’s suggestion addresses the syncope problem.Footnote ²⁵

In the HS grammar illustrated in (61), Syncope must outrank *CC to allow syncope to create a cluster targeted by *CC. At Step 1, the grammar selects [atpa] as the output, which serves as the input for Step 2. The optimal output at Step 2 is a candidate with C₁ deletion, that is, [apa], by the ranking condition Max-C/_V ≫ Max-C/_C. Crucially, syncope applies at Step 1, and because the Syncope constraint is ranked highest, it can condition the environment for further cluster reduction processes.

Although implementing P-map constraints as contextual faithfulness constraints in HS seems to work for the core patterns, there are situations where HS appears to be incompatible with the P-map principle. One such case arises when dealing with instances of irreducible parallelism, a phenomenon known to be challenging for HS (Flemming Reference Flemming2013; Adler & Zymet Reference Adler and Zymet2021). An illustrative example of this problem is root fusion in Sino-Japanese vocabulary (Itô Reference Itô1986; Itô & Mester Reference Itô, Mester, Otake and Cutler1996; Kurisu Reference Kurisu2000; Adler & Zymet Reference Adler and Zymet2021, among others). In Japanese, a sequence of one- or two-syllable roots fuses, deleting the first root’s final vowel and resulting in a voiceless geminate with place assimilation. However, syncope is blocked if it conditions place assimilation that yields a voiced geminate. For example, /betu-kaku/ ‘different style’ undergoes both syncope and place assimilation, resulting in [bekkaku], while /betu-bin/ ‘separate carrier’ does not undergo syncope, as it would create an illegal voiced geminate. Thus, it is realised as [betsubin] instead of *[bebbin] (examples from Itô & Mester Reference Itô, Mester, Otake and Cutler1996: 15). This type of interaction is recognised as being problematic for HS because of the restriction that only a single operation can be applied at each derivational step.Footnote ²⁶

The issue can be illustrated as follows. As demonstrated in the tableaux in (62), Syncope must be ranked high enough to create the environment for place assimilation between voiceless obstruents.

However, the same grammar fails to capture the case where syncope is blocked to avoid creating a voiced geminate. Since Syncope is ranked high and Gen in HS allows only one derivational change per step, the grammar returns a candidate with vowel deletion at Step 1, failing to block syncope. The constraint that penalises voiced geminates, *VdGem, cannot play any role in the evaluation.

In principle, this issue could be circumvented by allowing /u/ before a voiceless obstruent and that before a voiced obstruent to interact differently with syncope through constraint rankings rather than derivational steps. For instance, an undominated contextual faithfulness constraint like Max-V/__C _[+voi] could block the vowel deletion process before a voiced obstruent, resulting in (63a) rather than (63b). However, while such an analysis is technically possible, it remains unclear how such a constraint and its ranking would be motivated under the P-map principle: why, for instance, is Max-V/__C_[−voi] outranked by Max-V/__C _[+voi], rather than vice versa? As long as the P-map principle is assumed to be the key element for analysing the syncope problem, introducing such a constraint becomes problematic. Given that this constraint refers to perceptually arbitrary contexts, its justification under the P-map principle is difficult.

Contrary to HS, such a situation is not a problem for the current model, where syncope and place assimilation can be applied simultaneously, as shown in (64):

A similar situation where the explanatory powers of HS and the P-map principle are difficult to reconcile can be found in opaque contexts. Recall from §3.3 that in Lardil, /dibiɾdibi/ is realised as [dibiɾdi], with underapplication of apocope. In HS, the apocope constraint must be top-ranked to induce final vowel deletion in the underlying form at the first step, but this hierarchy forces the deletion of [i] in [dibiɾdi] as well. Although introducing a hierarchy of contextual faithfulness constraints (e.g., Max-V/_C ≫ Apocope ≫ Max-V) may avoid this issue in HS (cf. Hauser & Hughto Reference Hauser and Hughto2020), such an approach encounters a similar problem to the one identified for the irreducible parallelism case: it is unclear why this preconsonantal-specific faithfulness constraint should dominate other constraints. Justifying this ranking proves difficult. In particular, granting legitimate status to P-map faithfulness constraints as an HS solution to the syncope problem, as illustrated in (61) above, directly conflicts with attempts to incorporate contextual faithfulness constraints into HS to resolve the opacity problem. This conflict arises because it is difficult to establish perceptual justification for assigning greater protection to underlyingly preconsonantal vowels than to final vowels.

In summary, the aforementioned issues show that, in certain contexts, the compatibility between the HS model and the P-map principle is questionable. It is worth noting, however, that this discussion does not preclude the possibility of reconciling P-map faithfulness constraints with HS in contexts of irreducible parallelism or opacity through alternative means. The focus of the argument is not to demonstrate an inherent structural flaw in HS, but rather to highlight that the current approach, centred on a set of proposed P-map constraints, offers greater flexibility in analysing cluster reduction processes across diverse environments, without introducing additional assumptions or mechanisms to deal with irreducible parallelism or opacity.

6.2. Contiguity

As discussed throughout this article, subsegments are crucial in analysing the C₂ dominance effect in the syncope context. However, one may ask whether the proposal can be more formally implemented without referring directly to transitions, using a P-map version of Contiguity constraints. This is a valid question, and at least for cluster simplification, it is possible to account for the C₂ dominance effect by ranking Contiguity (CV), which specifically preserves a CV sequence, above Contiguity (VC), which protects a VC sequence, as shown in (65) (cf. Kenstowicz Reference Kenstowicz1994; McCarthy & Prince Reference McCarthy and Prince1995: 123; Lamontagne Reference Lamontagne1997). Suppose that these constraints assign a violation mark specifically when an input consonant originally adjacent to a vowel surfaces adjacent to a different consonant or vowel, thus losing its original adjacency relationship.

However, this analysis has a critical problem. With this type of Contiguity constraints, cluster simplification is enforced in the syncope context regardless of the relative ranking of *CC. This predicts unconditional cluster simplification whenever syncope occurs because, for an input form like /atipu/, when /i/ is deleted, a surface output like [atpu] will inevitably violate both Contiguity (CV) and Contiguity (VC). First, when the CV sequence /ti/ loses its vowel, [t] becomes adjacent to [p], violating Contiguity (CV). Similarly, the loss of the vowel in the VC sequence /ip/ leads to a violation of Contiguity (VC), as [p] becomes adjacent to [t]. The tableau in (66) illustrates this:

This result is undesirable because not all languages with a syncopated interconsonantal vowel undergo cluster simplification. Using Contiguity constraints sacrifices explanatory power for cases of syncope without simplification, in return for resolving the syncope problem.

To address this issue while retaining the Contiguity constraints, one might consider a scenario where Max-C is also ranked high, thereby excluding candidates with cluster simplification such as (66c) or (66d). However, this approach proves ineffective, as the top-ranked Max-C constraint actually results in place assimilation rather than preserving a cluster without reduction, as demonstrated in (67).

Again, in response to this, one could propose introducing a new faithfulness constraint, such as Ident [place], to make (67b) optimal. However, the introduction of Ident [place] engenders additional complexities. For instance, the Ident [place] constraint fails to capture the C₂ dominance effect in place assimilation: it cannot differentiate between candidates (67e) and (67f). Attempts to resolve this issue by refining Ident [place] through stratified contextual faithfulness constraints (e.g., Ident [place]/_V ≫ Ident [place]/_C) would reintroduce the original problem, in which these contextual faithfulness constraints function inadequately in syncope contexts.