3.1. Introduction

The distinctive features for the consonants listed in 2a,b are given in 3.Footnote ² The seventeen columns have been divided into six traditional manner categories: voiceless stops, voiced stops, affricates, fricatives, nasals, and liquids. The features in 3 are assumed to be privative; hence, a segment bears a feature, in which case it has “√,” or it does not, in which case it lacks that feature altogether. The features in 3 are arranged hierarchically in an approach to feature geometry that I describe in detail in section 3.3 below. All segments in 3 are assumed to bear [consonantal], which I henceforth omit.

1. (3) Distinctive features for IG consonants

   

   

In general, the place features in 3 do not play a role in the present analysis and will therefore be ignored. The one exception is the status of the place features for /r/, which I discuss below. The first six features in 3 require comment. Let us consider first the four manner features [nasal], [lateral], [continuant], and [strident].

All and only nasal consonants bear [nasal]. I adopt the standard view that [lateral] distinguishes /l/ and /r/ (for example, Cohn Reference Cohn1992, Blevins Reference Blevins1994, Kenstowicz Reference Kenstowicz1994, Odden Reference Odden2005, although these phonologists view the feature as binary). My privative treatment of [lateral] is similar to the one proposed by Rice & Avery (Reference Rice, Avery, Paradis and Prunet1991) and other researchers. The feature [continuant] distinguishes fricatives from stops: Rice (Reference Rice1992:88–89) adopts a featural model like the one endorsed here in which [continuant] is privative. A number of phonologists have argued that the same feature can also be used to distinguish /l/ from /r/ (see Mielke Reference Mielke2005 for a recent discussion), but this is not the case in Imst German. The feature [strident] distinguishes affricates from stops (see Jakobson et al. Reference Jakobson, Fant and Halle1952, LaCharité Reference LaCharité1993, Rubach Reference Rubach1994, Clements Reference Clements and Fujimura1999, Kehrein Reference Kehrein2002); that is, affricates bear [strident] and stops do not.

The approach to features adopted in 3 and below differs from treatments that have underspecified matrices with default rules filling in the redundant values, that is, the minus values (for example, Archangelli Reference Archangeli1988). According to the latter theory, these default rules can be freely ordered with other phonological rules. By contrast, Avery & Rice (Reference Avery and Rice1989) and many other researchers see such default rules as belonging to the phonetic component. My treatment is more compatible with the latter approach because there does not seem to be evidence that the negative values of the privative features in 3 play a role in IG phonology.Footnote ³ Evidence that default features do not play a role in the phonology and that they therefore belong to the phonetic component will be discussed below.

The features in 3 are intended to hold for Imst German as described in 1897 by Joseph Schatz (and probably also after that date). I also assume that these same features hold for earlier stages of Imst German as well, in particular the stages of that dialect immediately prior to the changes involving fricatives (see section 6.1), that is, the Bavarian dialect of Middle High German (MHG).

The only significant difference between the consonants in Imst German in 3 and the corresponding consonants in the Bavarian dialect of Middle High German is that surface [r] in the latter stage was coronal (that is, denti-alveolar) and not dorsal (that is, uvular); see Schatz (Reference Schatz1897:94). That IG /r/ is uvular on the surface is captured in the synchronic phonology of Imst German with the default rule in 4.

1. (4) Default Rule

   

The Default Rule adds the feature [dorsal] to the underlying [coronal] /r/ sound in 3, thereby producing a sound that is [dorsal] and [coronal]. That a phonetically dorsal sound can behave phonologically as coronal is not unusual. See, for example, Hall's (Reference Hall, van der Hulst and van de Weijer1995) analysis of Standard German uvular [r], which patterns phonotactically with coronal sounds. It is also instructive to consider Blevins's (Reference Blevins1994) analysis of the velar lateral /l/ because this surface dorsal sound behaves in many languages phonologically as coronal.Footnote ⁴ To capture the phonological patterning of /l/, Blevins (Reference Blevins1994:317) argues that in languages like Yagaria and Kanite it is underlyingly [coronal] and that [dorsal] (and its dependent features) are added by default.

3.2. The Feature [liquid]

The analysis developed in section 5 below depends crucially on the feature [liquid], which is present on both laterals (/l/) and rhotics (/r/). Although [liquid] is not usually assumed in the theoretical literature, it has been defended by Walsh Dickey (Reference Walsh Dickey1997) and is adopted below.

There is widespread agreement that (sonorant) laterals like /l/ and (sonorant) rhotics like /r/ form a natural class. The cross-linguistic evidence for the natural class of liquids consists of at least three pieces of evidence (from Walsh Dickey Reference Walsh Dickey1997:142ff.): (a) phonotactic restrictions on laterals and rhotics as a class, (b) the patterning together of laterals and rhotics in phonological processes, and (c) OCP restrictions on laterals and rhotics.Footnote ⁵

Evidence for (a) derives from the avoidance of laterals and rhotics together in a particular context. For example, a number of languages are attested in which liquids are banned from occurring in word-initial position, for example, Diyari, Djabugay, Kuman, and Mongolian. An example illustrating (b) is the deletion of the retroflex lateral before any liquid consonant in Djaru. OCP restrictions on [liquid] (see (c) above) involve languages that avoid adjacent liquids. There are a number of such languages. For example, in Yidiny (an Australian language with the three phonemic liquids /l r ɹ/) there are various consonant clusters, including those with liquids. However, there are no consonant clusters consisting of two liquids in sequence. Significantly, Walsh Dickey (Reference Walsh Dickey1997) demonstrates that this gap cannot be reduced to restrictions on place of articulation (that is, the avoidance of coronals or the avoidance of apicals), nor can it be reanalyzed as a sonority-based restriction. Similar examples involving the avoidance of adjacent liquids can be found in Gumbaingar and Javanese.

The feature [liquid] (Dixon Reference Dixon1972, Walsh Dickey Reference Walsh Dickey1997) has the advantage of capturing the natural class of /l/ and /r/ directly. Given this feature, the phonotactic restrictions described above require a statement to the effect that [liquid] is disallowed word-initially and the context in the rule in Djaru refers consequently to [liquid] segments. OCP restrictions can also be captured directly with a feature-specific constraint, for example, OCP-[liquid]: *[liquid] [liquid]. I show below that this constraint is active in Imst German and that it triggers the change in 1b.

Walsh Dickey (Reference Walsh Dickey1997) argues that [liquid] is on par with other major class features like [consonantal] in the sense that it is a part of the Root node. As I demonstrate below, I depart from this treatment because I analyze [liquid] as a manner feature that displays autosegmental behavior (that is, delinking). Ultimately, I show that [liquid] is a daughter of SV.

Consider now two obvious alternatives to the feature [liquid]. In traditional phonological theory the natural class of (sonorant) laterals and (sonorant) rhotics is captured with a set of three features, namely [+consonantal, +sonorant, –nasal]. However, as pointed out by Walsh Dickey (Reference Walsh Dickey1997:151), there are two problems with this analysis. First, it is not clear how this proposal can capture the contrast in certain languages (for example, Igbo) between plain and nasalized rhotics. If liquids are inherently non-nasal, then the existence of nasalized liquids is problematic. Second, there is excellent cross-linguistic evidence that nasality is a privative property, meaning that there is no evidence for the lack of nasality (that is, [−nasal]). The privative nature of [nasal] has been defended by Avery & Rice (Reference Avery and Rice1989) and Piggott (Reference Piggott1992), among others. Assuming that the negative value of [nasal] does not exist, the traditional way of capturing the natural class of liquids is simply not an option.

According to another approach (for example, Clements Reference Clements, Kingston and Beckman1990:292ff.), liquids are characterized as [−vocoid, +approximant, +sonorant]. The feature [vocoid] is the converse of [consonantal], so [−vocoid] is equivalent to [+consonantal]. In Clements's featural system [+approximant] sounds subsume all liquids (for example, /l r/), all vowels and all glides (for example, /j w/). The reason [liquid] is not the same as [approximant] is that the latter feature includes vowels and the former one does not.Footnote ⁶ While this approach does not suffer the drawbacks described above for the traditional analysis of liquids, it is unclear how the OCP-constraint and liquid dissimilation described below in section 5 would translate into a system in which liquids are captured with three features as opposed to one.Footnote ⁷

3.3. The SV Node

A model of segment structure relying crucially on the node SV (“Sonorant Voice”) has been developed by a number of researchers, in particular Rice & Avery (Reference Rice, Avery, Paradis and Prunet1991), Rice (Reference Rice1992, Reference Rice1993), Brown (Reference Brown, Hulst and Weijer1992), Piggott (Reference Piggott1992), and Frigeni (Reference Frigeni2005). According to these authors, SV replaces the traditional feature [sonorant] and is underlyingly present in all sonorants as well as obstruents that function phonologically as sonorants, that is, what Rice calls “sonorant obstruents.” In 3, we can see that not only the nasals and liquids (that is, the traditional sonorant consonants) bear SV, but also the three voiced stops /b d g/. By contrast, voiceless stops, (voiceless) affricates, and (voiceless) fricatives all lack SV.

The features in 3 are arranged hierarchically in the feature-geometric representation in 5. The Root node dominates Place, SV, and [continuant], while SV dominates [nasal], [liquid], and [lateral]. I tentatively assume that [strident] is dominated directly by the Root.

1. (5) Feature geometric representation

   

The structure in 5 is illustrated for the six coronal ([cor]) consonants in 6. These sounds represent six of the seven of the manner categories in 3. I omit the affricate [ts] (representative of the seventh manner category in 3) because it is not relevant for the ensuing analysis. All Root nodes in 6 and below are assumed to consist of the feature [consonantal] (McCarthy Reference McCarthy1988).

1. (6) Representations for six coronal consonants

   

As obstruents, /t/ and /s/ have no SV node; the latter bears [continuant] ([cont]) and the former lacks this feature altogether. Sonorant obstruents like /d/ have an SV node, but in contrast to true sonorant consonants, the SV node for segments like /d/ has no dependent features. True sonorant consonants like /n r l/ have by definition not only SV, but also at least one dependent feature. For nasals like /n/ this dependent feature is [nasal], for /r/ it is [liquid], and for /l/ it is [liquid] and [lateral] ([lat]).Footnote ⁸

The contrast between sounds like /p t k/ and /b d g/ is often assumed to involve the feature [voice] (for example, Wiese Reference Wiese1996 for Standard German), but [voice] is not necessary given my use of the feature SV. Note too that SV in 3 is not exactly the same as [+voice] because the latter feature is usually assumed to be present underlyingly for sounds like /b d g/ and redundant for sonorants (see Wiese Reference Wiese1996). Alternatively, several linguists have argued that German /p t k/ vs. /b d g/ contrasts are captured with the (privative) feature [spread glottis], for example, Iverson & Salmons (Reference Iverson and Salmons1995) and Jessen & Ringen (Reference Jessen and Ringen2002). On that view, /p t k/ are underlyingly [spread glottis], while /b d g/ are unmarked for that feature. Assuming that there is evidence from Imst German that /p t k/ behave phonologically as [spread glottis] sounds, one could conceivably modify the features in 3 by including the Laryngeal node and [spread glottis] for /p t k/. Thus, I see my treatment as being compatible with the [spread glottis] analysis. The nature of the laryngeal phonology of Imst German is a question I do not investigate here and hence I leave this topic open for further research.Footnote ⁹

4.1. Properties of Dissimilations

Two properties that have been argued to hold for dissimilations are: (a) dissimilations are sporadic and not regular, and (b) dissimilations are structure preserving. I show below that (b) but not (a) is a consistent property that characterizes dissimilations.

A common statement involving dissimilations in textbooks is that this type of process is irregular, or sporadic (for example, Hock & Joseph Reference Hock and Joseph1996:125–126, 141), although it needs to be said that this statement is made in reference to dissimilatory sound changes, as opposed to synchronic processes of dissimilation. While many dissimilatory sound changes are indeed sporadic, it is important not to lose sight of the fact that there are also many well-attested examples of regular dissimilations (see Campbell Reference Campbell2004:30–32 for discussion). Examples of regular dissimilations mentioned by Campbell include Grassmann's Law in Greek and Sanskrit, Dahl's Law in Bantu, and the change of /k/ to [h] before /t d/ in Finnish. (For an additional example of a process that was obviously dissimilatory and regular, see section 6.1 below.) Because regular diachronic dissimilations are attested, I hold that property (a) above is a tendency at best.

A second property that has been argued to characterize dissimilations is Structure Preservation (see, for example, Ohala Reference Ohala and Jones1993:255–256 and Kiparsky Reference Kiparsky and Goldsmith1995:658). Structure-preserving changes are changes in which the output is a phoneme of the language, whereas non-structure preserving processes are ones that introduce a non-phoneme (that is, an allophone). Allophonic processes are by definition non-structure-preserving. By contrast, the studies referred to above have demonstrated that all attested dissimilatory changes are structure preserving—a finding that is confirmed for processes argued to be dissimilatory in sections 5.2 and 6.1. Since the structure-preserving property is a consistent cross-linguistic property, I adopt it in my treatment below.

4.2. Dissimilation and the OCP

Following Yip (Reference Yip1988) and Cohn (Reference Cohn1992), I analyze dissimilations as a repair mechanism applying as a response to a feature-specific OCP constraint (see also the discussion in Alderete & Frisch Reference Alderete, Frisch and Lacy2007). The general approach adopted here is outlined in 7. The feature [F] in 7a and 7b is assumed to be privative and not binary.

1. (7) Approach to dissimilations adopted below

   

Dissimilation itself is schematized in 7b as the deletion of a feature [F] when adjacent to the same feature. Languages specify in 7b (i) whether the first or second [F] deletes, and (ii) whether the root nodes A and B are adjacent. In the case of root nodes A and B which are non-adjacent, the delinking in 7b involves a long-distance dissimilation, for example, the dissimilation of /l…l/ to [l…r] in Latin; see [nav-alis] ‘naval’ vs. [sol-aris] ‘solar’ (Steriade Reference Steriade1987:351).

The authors cited above see dissimilations as in 7b, although both Yip (Reference Yip1988) and Cohn (Reference Cohn1992) assume that the feature that dissimilates is binary, in which case the delinking in 7b is followed by a default rule adding the negative value, that is, [−F]. As explained above, the approach I adopt eschews default rules that fill in negative values.

The basic analysis of dissimilations as described below is also the one often adopted in OT analyses of dissimilations (see Alber 2000 and Hall & Scott Reference Hall and Scott2007 for examples from German). Some work in the OT framework has identified various problems with the analysis of dissimilation as a response to an OCP violation; for example Alderete Reference Alderete and Kusumoto1997 and Suzuki Reference Suzuki1998. The former author argues that dissimilations are triggered by a local conjunction and the latter that the trigger is due to a “generalized OCP” constraint violation (Suzuki Reference Suzuki1998:27). Suzuki (Reference Suzuki1998:16–26) lists five problems (from the cross-linguistic perspective) with the traditional OCP analysis I adopt above, but I do not see these problems as being relevant to the IG examples I discuss below.Footnote ¹⁰ A question I leave open for further study is whether or not the approach taken by Alderete (Reference Alderete and Kusumoto1997) or Suzuki (Reference Suzuki1998) can offer a deeper explanation for the phonology of Imst German than the one I propose.

5.1. Adjacent Non-Identical Liquids

Examples illustrating the historical change from MHG /r/ to IG /d/ before a tautomorphemic /l/ are presented in 8 below (Schatz Reference Schatz1897:94). The Standard German orthographic forms (which can be equated with MHG) are presented in the first column and the corresponding IG forms in the second column.Footnote ¹¹

1. (8) (8) MHG /r/ > IG /d/ / ___ /l/ within a morpheme

   

The sound change in 8 did not simply involve a restructuring of the underlying representations because it caused many morphemes to arise with [d] ∼ [r] alternations. This point can be illustrated with the additional examples in 9. In 9a the same historical change in 8 took place across a morpheme boundary and in 9b it occurred across a word boundary. In both data sets we see a word ending in [r] in isolation which is realized in Imst German as [d] if the next sound is [l]. The page numbers in the final column in 9 refer to Schatz Reference Schatz1897. The final example in 9a was obtained from Hathaway Reference Hathaway1979:31.

1. (9) G [d] ∼ [r] alternations

   1. a.

   2. b.

The alternations in 9 require /r/ to be analyzed as the underlying sound with /r/ → [d] __ /l/ as a synchronic rule of Imst German. For example, the first two words in 9a are /ʃwɑːr/ and /ʃwɑːr-lig/ and /r/ becomes [d] in the second example. The reason the alternations in 9 cannot be analyzed with an underlying /d/ and a process converting this sound into [r] word finally is that Imst German has many words with a word-final surface [d], for example, Rad [roːd] ‘wheel’ (p. 141), which would incorrectly be converted into [r].

Schatz's (Reference Schatz1897) description of the development in 8 and 9 leaves little doubt that this was a regular, neogrammarian sound change (as opposed to a sporadic one): “Today every r is pronounced as d before l within words and between sentences.”Footnote ¹² It is also worth emphasizing that the change from /r/ to [d] occurs only before /l/ and nowhere else. Examples in which [r] surfaces before other sounds abound in Imst German; for example, gewahr werden [gwoːrnə] ‘become aware of something’ with [rn] and Dirne [diərnə] ‘prostitute’ (pp. 25, 64); warmer [wørmər] ‘warmer’ (p. 25) and Wurm [wurm] ‘worm’ with [rm].Footnote ¹³

It is perhaps worth noting that the [dl] cluster created by the development in 8 and 9 was not a new cluster, since it was inherited in other words, for example, the surface [d] in Stadel [ʃtoːdl] ‘repository’ (p. 89) and Mädchen [mɑːdlə] ‘girl (dim.)’ (p. 89) goes back to [d] in Middle High German. Thus, the development in 8 and 9 did not result in clusters that were novel to Imst German.

Schatz (Reference Schatz1897:94) observes that a development parallel to the one in 8 and 9 involved the shift from MHG /r/ to IG [d] after /l/, as in 10. In these examples, the Standard German orthographic form is presented in the first column, the MHG form in the second column, and the IG equivalent in the third column. Note that the two liquids in the original sequence /lr/ in Middle High German were tautomorphemic.

1. (10) /r/ > /d/ / /l/ ___ in Imst German

   

Schatz does not provide examples of the change in 10 across morpheme or word boundaries parallel to the ones in 9. Perhaps one reason for this omission is that there do not appear to be r-initial suffixes in Imst German that would have an alternant beginning with [d] after a stem ending in [l]. This being said, examples in connected speech in which one word ends in /l/ and the following word begins with /r/ should not be difficult to come by. Although Schatz does not give examples of such across-word alternations, it is noteworthy that there do not appear to be any examples in his 1897 grammar (nor in Hathaway Reference Hathaway1979) of words (or sentences) containing surface [lr] clusters. Assuming that this gap is systematic, I adopt the view that the change in 10 is a synchronic process of Imst German.

5.2. The Analysis

Given the representations for consonants in 6, I analyze the change from /rl/ to [dl] (as well as the mirror image development) as a dissimilation of the feature [liquid], which itself applies as a response to the feature-specific OCP constraint in 11. The prose statement of the change in 12a is intended to express both changes depicted in 12b. The OCP constraint in 11 needs to specify that the two segments are Root-node adjacent because there are many examples of IG words containing non-adjacent liquids in which no change occurs, for example, traurig [traurig] ‘sad’ (p. 93).

1. (11) OCP-[liquid] (Root nodes are adjacent)

   

1. (12) Liquid Dissimilation

   1. a. Delete [liquid] when adjacent to [liquid]

   2. b.

Liquid Dissimilation in 12 simply states that [liquid] should be deleted, but it does not specify whether or not the feature deletes from /l/ or /r/. I show below that my analysis offers a principled reason why [liquid] can only be eliminated from the latter segment.

Consider now the derivations from /rl/ to [dl] in 13a and from /lr/ to [ld] in 13b. In these derivations I consider all and only the node SV and its dependents.

1. (13) [rl] → [dl] and the mirror-image

   1. a.

   2. b.

In 13a we can observe that the OCP constraint in 11 bans the sequence of two liquids to the left of the arrow and that Liquid Dissimilation applies as a response to this OCP violation (recall 7). The mirror image development is depicted in 13b. In both 13a,b the deletion of [liquid] from the /r/ results in a segment with the SV node but without any dependent manner features. An examination of 6 reveals that this segment can only be interpreted as the (homorganic) stop /d/ because /d/ (as opposed to /t/) is analyzed as a sonorant obstruent with an SV node without any dependents.Footnote ¹⁴

Liquid Dissimilation simply states that the feature [liquid] should be eliminated, but it does not say whether or not the feature deletes from /l/ or /r/. In fact, this need not be specified in the rule because the correct result is obtained given the structure preserving property of dissimilations. Given Structure Preservation, the [liquid] on /r/ can delete because the output is a sound with SV without any manner features, which can only be the stop /d/. Were the feature [liquid] to delete from /l/, then the result would be a segment with SV that bears only one manner feature, namely [lateral], but there is no such segment in Imst German.

In contexts other than before or after /l/, IG /r/ surfaces as a uvular (that is, [dorsal]) sound (recall 4). I follow Schatz (Reference Schatz1897:93), who holds that the historical change in 8 and 9 took place at a point in time before /r/ shifted to a uvular sound.

In section 3.1, it was stated that default rules like the one in 4 are assumed to apply in the phonetic component and that they therefore do not interact with phonological processes. My treatment of IG /r/ supports this approach. /r/ is underlyingly [coronal] and must remain so throughout the postlexical phonology because Liquid Dissimilation is postlexical. The domain of Liquid Dissimilation must be the postlexical component because the process applies across words; recall 9b. In the phonetic component, all /r/'s not specified for place features receive [dorsal] via 4.

Although it is often claimed that only lexical processes can be structure preserving (for example, Kiparsky Reference Kiparsky1985, and much subsequent work in Lexical Phonology), my analysis crucially depends on analyzing Liquid Dissimilation as structure preserving. I follow Rice (Reference Rice, Inkelas and Zec1990) and Hyman (Reference Hyman, Hargus and Kaisse1993), who show on the basis of data from non-Germanic languages that non-structure preserving processes may persist into the postlexical phonology. My analysis therefore suggests that it is not the component (that is, lexical vs. postlexical) that tells us which rules are structure preserving, but instead it is the rule type: dissimilations are always structure preserving even if they are postlexical.

5.3. Adjacent Identical Liquids

Since the analysis of Liquid Dissimilation in section 5.2 depends crucially on the constraint OCP-[liquid], the implication is that sequences of adjacent identical liquids should be avoided in Imst German as well. In fact, there is excellent evidence that this is the case.

Consider first tautomorphemic geminate liquids, that is surface [ll] and [rr], which I pointed out in section 2 are allophones of /l/ and /r/ respectively. At face value, [ll] and [rr] appear to be problematic if they consist of two separate identical segments, each bearing the feature [liquid]. Particularly [rr] might seem to be troublesome because it is not pronounced [dr], for example, Karren [kxorrə] ‘cart’ (Schatz Reference Schatz1897:94). In fact, the behavior of [ll] and [rr] is not surprising if such tautomorphemic geminates consist of a single Root node whose length component is expressed with skeletal slots or moras (see Hayes Reference Hayes1986, and much subsequent work by other authors). The nonlinear representations just described are presented in 14a, where “X” can be interpreted either as a mora or skeletal slot. This representation contrasts with the one in 14b, which is usually assumed to hold for heteromorphemic geminates.

1. (14) Representations of [ll] and [rr]

   1. a. Tautomorphemic geminates

      

   2. b. Heteromorphemic geminates

      

Given the representation of ‘true’ geminates in 14a, Liquid Dissimilation cannot affect the “first” /r/ in [rr], because the representation just described has only a single instantiation of [liquid].

Consider now adjacent heteromorphemic sequences of identical liquids. As noted earlier, Imst German apparently has no r-initial suffixes; hence the question is how the language deals with suffixes beginning with /l/ which potentially attach to stems ending in /l/. The data in 9a above illustrate that Imst German possesses two /l/-initial suffixes (Schatz Reference Schatz1897:70–71), namely the adjectival suffix [lig] and the diminutive suffix [lə]. Schatz does not discuss what happens when a stem ending in /l/ occurs before the adjectival suffix [lig], although it appears doubtful that such words exist.Footnote ¹⁵

There is an extended discussion in Schatz (Reference Schatz1897:70–71) concerning the distribution of the diminutive suffix [lə] and its allomorph [ələ]. In 15a we can see that the former allomorph occurs after stems ending in any consonant other than /l/, while the latter allomorph surfaces after stems ending in singleton [l] or geminate [ll] (in 15b). Note that in Standard German the diminutive suffix -lein is avoided and that -chen is used in its place after stems ending in /l/ (see Plank Reference Plank1981:156).

1. (15) Diminutive suffix allomorphy in Imst German

   1. a.

   2. b.

What the data in 15 suggest is that the constraint OCP-[liquid] in 11 also prevents the concatenation of /…l-l…/. The implication is that such /l-l/ sequences cannot have the representation in 14a as “true” geminates; otherwise they would not violate OCP-[liquid]. Instead, at the level of grammar where word formation takes place (that is, in the lexical component) heteromorphemic (that is, “fake”) geminates must be represented as a sequence of identical segments (with one instantiation of [liquid] each), as in 14b.

The behavior of /l-l/ and /r-r/ in connected speech (that is, at the postlexical level) is also compatible with the present analysis. After words are concatenated into sentences, fake geminates with two identical root nodes, as in 14b, arise. Schatz (Reference Schatz1897:24) observes that /l-l/, /r-r/ as well as the corresponding nasals, undergo degemination after long vowels, for example, Qual leiden /kwoːl # lɑidə/ ‘suffer agony’ surfaces as [kwoːlɑidə]. I sketch briefly what an analysis of such examples might look like: The change from /l-l/ to [l] suggest that the fake geminates in 14b are transformed first into the real geminates in 14a and that the root node in this structure bears a mora associated with the preceding syllable. Assuming that Imst German bans trimoraic syllables, then a surface form like *[kwoːllɑidə] cannot be correct because the first syllable would contain three moras. The true geminate in examples like this one are subsequently repaired by reducing the geminate to a singleton (which is no longer moraic).

5.4. Alternative Analyses

An obvious alternative to Liquid Dissimilation in section 5.2 is to reanalyze this process as an assimilation, although it is not clear how this should be done. Virtually all approaches to phonological features view /l/ and /r/ as [+sonorant] and /d/ as [−sonorant]. Given that the output [d] is a stop (that is, [−continuant]), it might be reasonable to analyze /l/ as [−continuant] (see Mielke Reference Mielke2005) in order to analyze the change from /r/ to [d] before /l/ as an assimilation. Given the features just described, the reanalysis of Liquid Dissimilation as an assimilation might be expressed linearly as in 16. That the output is [d] and not [t] could be captured by stipulating that the /r/ in the input has already been assigned [+voice] by default. Alternatively, if [spread glottis] is assumed to distinguish /t/ from /d/, then /r/ (which is unmarked for laryngeal features) would correctly shift to [d] because this sound is unmarked for laryngeal features.

1. (16) Hypothetical assimilation

   

Note that the change from /r/ to [d] entails not only a change in continuancy, but also in sonorancy. However, while the rule in 16 assimilates the former feature, the latter feature simply changes its plus into a minus in a non-assimilatory way. The obvious problem with 16 is that it requires a process to assimilate one feature ([continuant]) and change the value in another ([sonorant]).Footnote ¹⁶ This formal problem does not arise in the analysis defended above in section 5.2 since the deletion of [liquid] in /r/ automatically entails that the output is [d].

Note too that the present analysis—with its reliance on OCP-[liquid]—sees a direct connection between the instability of /r-l/ and the avoidance of identical liquids described in the preceding section. It is not clear how the alternative in 16 can establish a link between the change expressed in that process and the examples involving the diminutive suffix allomorphy in 15.

A second alternative treatment might attempt to analyze Liquid Dissimilation as a change motivated not by the violation of OCP-[liquid], but instead by the avoidance of bad syllable contacts. Assuming that /l/ and /r/ are equally sonorous in Imst German, the change from /ʃwɑːr-lig/ to [ʃwɑː.dlig] (see the second example in 13a) might make sense because the sonority plateau in the incorrect surface form *[ʃwɑːr.lig] would violate the Syllable Contact Law (Hooper Reference Hooper1976, Murray & Vennemann Reference Murray and Vennemann1983, Green Reference Green1997, Ham Reference Ham1998, Harris Reference Harris1994, Davis & Shin Reference Davis and Shin1999, and Rose Reference Rose2000). Note that this alternative requires [dl] in [ʃwɑː.dlig] to be tautosyllabified because the heterosyllabic parse would surface with a bad syllable contact.

While it does appear that Liquid Dissimilation has the function of eliminating bad syllable contacts, this analysis cannot replace OCP-[liquid] because Imst German allows for many “bad” syllable contacts which are not repaired, for example, Bäumlein [pɑ̃ːm.lə] ‘tree (dim.)’ (Schatz Reference Schatz1897:25) with a heterosyllabic nasal-liquid. Equally inexplicable is the fact that the /r-l/ contact being repaired is better in terms of its sonority profile than ones like /m-l/ which are not repaired.

A third alternative to the present proposal is discussed briefly by Schatz (Reference Schatz1897:94), who suggests that the process I have analyzed as Liquid Dissimilation really involves the epenthesis of [d] between /r/ and /l/ (and between /l/ and /r/). He holds that this epenthesis analysis in sequences like /rl/ and /lr/ derives support because the epenthetic [d] is observable in other contexts, for example, between [n] and [l] (Männlein [mɑndlə] ‘man (dim.)’ (from /mɑn-lə/), p. 45).

I see three arguments against an epenthesis analysis. First, it is not clear what would motivate the epenthesis of /d/ between /r/ and /l/, /l/ and /r/, and /n/ and /l/, but not between /r/ and /n/ (recall the [gwoːrnə] example mentioned above). Second, it is not clear that the epenthesis Schatz describes was a regular development in Imst German; for example, Hathaway (Reference Hathaway1979:19) has the original /nl/ sequence in the word wahrscheinlich ‘probably’ with [nl] and not [ndl].

But the most serious drawback with an epenthesis analysis is that it is not clear how speakers of Imst German since 1897 have been able to deduce that this is the correct process given the data in 9. If the [r] ∼ [d] alternations in Imst German were to be captured with a rule of d-epenthesis, then the treatment would also require a rule of r-deletion; for example, an input like /ʃwɑːr-lig/ would undergo d-epenthesis to /ʃwɑːrdlig/, followed by r-deletion to [ʃwɑːdlig]. However, there is no evidence in Imst German for a synchronic process of r-deletion. It is certainly conceivable that the [r] ∼ [d] alternations in 9 had their origins in the epenthesis of [d], but this historical analysis does not necessarily imply that this is the correct synchronic treatment. In fact, since the examples in 9 involve /r/ becoming [d] and not /d/ being inserted between /r/ and /l/, I would argue that the burden of proof lies on the shoulders of linguists who want to uphold an epenthesis treatment in the synchronic phonology of Imst German.

6.1. OCP-[continuant] and Dorsal Fricative Dissimilation

In 17a we can observe that the MHG dorsal fricative [x] shifted to IG [k] in the context before [s]. This was a very regular development in most regional varieties of High German (Schatz Reference Schatz1897:107). The additional examples in 17b show that MHG [x] remained before other sounds (that is, [t]). (For a more detailed analysis of examples like the ones in 17 than the one offered below, see Hall Reference Hall2008.)

1. (17) Development of MHG [xs] and [xt] into Imst German

   1. a.

   2. b.

An examination of the examples in 17a reveals that [x] and [s] can be tautosyllabic (for example, in vu[ks]) or heterosyllabic (for example, wa[k.s]en). [s] was the only fricative that could follow [x] in Middle High German and [x] was the only fricative in Middle High German that surfaced before [s]. The change in 17a only took place if the adjacent fricatives belonged to the same morpheme because the modern inflectional suffixes beginning with [s] (for example, the second person singular ending in verbs) was schwa-initial in Middle High German, that is, -est [əst]. According to Moser (Reference Moser1951:281), the change in 17a was attested in late Middle High German (for example, since the end of the 13th century in Bavarian), although some evidence suggests that it might have even begun in Old High German (OHG).

The avoidance of adjacent fricatives is due to what I see as the effects of the feature-specific OCP constraint in 18—a constraint which was active in the phonology of Middle High German, if not before. The Root nodes are specified as adjacent because there was no avoidance of fricatives separated by intervening sounds. According to the matrices in 3, adjacent [continuant] segments include only sequences of adjacent fricatives because no other sound is marked for that feature; hence, OCP−[continuant] only affected adjacent fricatives and nothing else.

1. (18) OCP-[continuant] (Root nodes are adjacent)

   

I see the development in 17a as a dissimilation of the feature [continuant]—a sound change I refer to below as Dorsal Fricative Dissimilation (see 19). Dorsal Fricative Dissimilation was a neutralization because the contrast between [x] and [k] was suspended to [k] before a fricative.

1. (19) Dorsal Fricative Dissimilation (Root nodes adjacent)

   

According to 19 the first of two adjacent [continuant] specifications deleted. I see this deletion as a repair mechanism (that is, 7b) when OCP-[continuant] (that is, 7a) became activated. Since [xs] was the only sequence of adjacent fricatives at this stage in the language, there is no need to specify a place of articulation for either fricative in 19.Footnote ¹⁷

Dorsal Fricative Dissimilation simply restructured underlying representations and therefore did not have the status of a synchronic rule that was passed on to future generations. For example, when MHG [vuxs] was pronounced as [vuks], the underlying representation of this morpheme changed from /vuxs/ to /vuks/ as well. Put differently, the change in 20 did not cause alternations to arise between [x] and [k] that might motivate a synchronic rule of the form /x/ → [k] / __ /s/. In this respect, Dorsal Fricative Dissimilation was very different from Liquid Dissimilation in 12, which survives to the present day in Imst German as a synchronic rule.

Although there may be no direct evidence for the claim that Dorsal Fricative Dissimation is active as a synchronic process of Imst German, there is good reason to believe that OCP-[continuant] is retained in Imst German as a morpheme structure constraint. As in other varieties of German, Imst German allows for consonant clusters consisting of two or more obstruents. Within morphemes, these two-obstruent clusters can consist of (a) two stops, for example, Magd [mokt] ‘maid’ (p. 20); (b) a fricative and a stop, for example, stampfen [ʃtomfə] ‘stamp’ (p. 20); or (c) a stop and a fricative, for example, Achsel [okslə] (p. 18). By contrast, there are no tautomorphemic fricative-fricative sequences.Footnote ¹⁸

6.2 On the Status of Other Manner Features

There appears to be evidence that the other manner features of Imst German play a role in OCP related OCP constraints as well. Consider the evidence for OCP-[nasal] (that is, *[nasal] [nasal]). Like OCP-[continuant], OCP-[nasal] derives support as a morpheme structure constraint. Although adjacent tauto- and heteromorphemic sonorants can occur in Imst German (for example, Bäumlein [pɑ̃ːmlə] ‘tree (dim.)’, warmer [wørmər] ‘warmer’, and Wurm [wurm] ‘worm’, as mentioned above), there are no tautomorphemic sequences of adjacent nasals (for example, */mn nm nŋ ŋn/). Schatz (Reference Schatz1897) presents occasional examples of IG words with adjacent nasals—for example, lange mir [loŋŋmər] ‘long to me’ (p. 18)—but they all appear to involve examples of two adjacent words in a syntactic phrase. However, within morphemes, Imst German does not tolerate adjacent nasals.Footnote ¹⁹

Evidence can also be adduced for OCP-[strident]. Since the only [strident] sounds in 3 are affricates, a constraint *[strident] [strident] would derive support if adjacent affricates are avoided. This appears to be the case. As noted in section 6.1, sequences of adjacent obstruents occur with a high degree of frequency in Imst German, but these sequences involve either (a) two stops, (b) a fricative and a stop, or (c) a stop and a fricative. By contrast, sequences of adjacent affricates (for example, /pfts tspf kxts tstʃ…/) do not seem to occur.Footnote ²⁰

In 20, I have listed the four OCP constraints for IG manner features in the first column. In the second and third columns we can observe their respective domain and the effects. OCP-[liquid] is the only one of the constraints in 20 that still maintains an active function in the sense that it triggers processes (Liquid Dissimilation and allomorphy selection in 15). It is also the only constraint with a lexical (that is, within words) and a postlexical (that is, across words) domain. The final three constraints in 20 are necessary as morpheme structure constraints (MSC) to capture phonotactic regularities.

1. (20) OCP manner constraints in Imst German

   

The one manner feature in 3 not present in 20 as an OCP constraint is [lateral]. It was demonstrated in 15b above that Imst German systematically avoids adjacent laterals, but these examples were captured with OCP-[liquid]. Since OCP-[liquid] is independently necessary, it is impossible to tell whether or not Imst German also has OCP-[lateral].