Gothic <r> and Old High German <r>: Implications from phonological patterning

This article reconstructs the archaic Germanic rhotic by examining a natural class pattern common to Gothic and Old High German (OHG). Specifically, I argue that the sounds represented by the graphemes <r> and <h> patterned as [high] segments. Due to the Obligatory Contour Principle, those [high] consonants triggered dissimilatory lowering of high vowels in Gothic. On account of the No-Crossing Constraint, the same (i.e. etymologically related) consonants blocked the OHG process known as Primary Umlaut. That is, <r> and <h> inhibited the height features of [i] and [j] from spreading regressively onto a preceding low vowel. These novel analyses not only offer insight into patterns which have been poorly understood for the better part of two centuries, but also add clarity to our understanding of the phonological and phonetic properties of Early Germanic rhotics.


Introduction
Owing to their significant phonetic diversity, rhotic sounds are difficult to characterize phonologically (see inter alios Wiese 1996;Walsh-Dickey 1997;Chabot 2019). This article aims to navigate that multiplicity in the absence of a phonetic record through an examination of two independent developments involving an interaction between rhotics and vowels. The first is Gothic Lowering, the second is Old High German Primary Umlaut. I claim that the interactions observed in these sound changes elucidate a natural class pattern, namely, that the rhotic sounds of these early Germanic languages pattern with segments that are marked with the feature [high]. 1 Due to Gothic Lowering, all instances of Gothic [e] and [o] were regular reflexes of Proto-Germanic (PGmc) *[i] and * [u], respectively, before the graphemes <r> (a rhotic < PGmc *r) and <h> (a dorsal fricative < PGmc *x 2 ), e.g. Gothic w[e]r 'man ', h[o]rn 'horn', sl[e]ht 'smooth', [o]hs 'ox' < PGmc *w[i]r-, *h[u]rn-, *sl[i]xt-, *[u]xs-. It is my contention that PGmc *r and *x were specified with the vocalic feature [high] and, thus, PGmc forms like *wir-, *hurn-, *slixt-, *uxs-contained sequences of adjacent [high] segments. Those sequences resulted in an OCP violation (Leben 1973), which Gothic Lowering repaired by shifting the high vowels to corresponding mid vowels.
sented, consistent with traditional assumptions, in which Gothic Lowering is assimilatory. Section 2.3 discusses how to evaluate the two analyses when no Gothic data are available to make such an evaluation possible.

Data
Gothic Lowering refers to the shift of the East Germanic (EGmc) short high vowels *[i] and * [u] to the corresponding Gothic mid vowels, [e] and [o], 4 before the sounds represented by the graphemes <r>, <h>, and its labialized counterpart <ƕ> (pronounced 'hwair'). Some examples are presented in (1). EGmc reconstructions are presented in the first column; attested Gothic forms are presented in the second one. 5 A comparison of the EGmc forms with the Gothic reconstructions in (1a) indicates that EGmc *[i] lowers to Gothic [e] before the sound represented by the grapheme <r>, which is shown with bold-facing for emphasis. The data in (1b) exemplify lowering of the high back vowel in a pre-rhotic context. The examples of Gothic Lowering in (2) show the graphemes <h> and <ƕ> as additional conditioners of the change. The data in (2a) illustrate the shift from EGmc *[i] to Gothic [e] in this context; the examples in (2b) demonstrate the lowering of EGmc * [u] to Gothic [o]. 4 It is well-understood that the Gothic mid vowels [e] and [o] were represented with the digraphs <ai> and <au>, respectively (see Collitz 1918, fn. 2). Grimm (1822) originally called the process Brechung 'Breaking' (i.e. 'diphthongization') due to a misinterpretation of those digraphs. Whether the Gothic mid vowels were tense or lax is left open here and does not bear on the analysis. 5 The EGmc forms are based on PGmc reconstructions from Orel (2003). They simply alter the PGmc stems to reflect the context-free raising of PGmc *[e] to EGmc *[i] (see discussion of this development in Heidermanns & Braune 2004: 32ff). In this way, the EGmc forms are intended to be informal reconstructions that aid the reader to account for that early raising process. While Gothic Lowering is an excellent example of a neogrammarian regular sound change, its interpretation, despite that regularity, is not at all clear. In the section below, two contradicting analyses are presented, followed by a discussion on how to disambiguate those analytical treatments.

Analysis of Gothic Lowering
Because Gothic Lowering involves a shift of high vowels to mid vowels, an analysis of the process depends crucially on representations of vocalic height. High Vowels like [i] and [u] can be represented, more or less uncontroversially, as in (4). (4) The representation in (4) shows that high vowels are marked with an autosegmental feature [high]. For the purpose of this article, it is not crucial whether that autosegment is thought of as a binary feature '[+high]', a unary feature '[high]', or as an abbreviation for elements like '|i| and |u|'. What is important -and unanimous in feature theories -is that all high vowels are marked with an active feature.
It is unclear, however, if mid vowels are also marked with active features, due to an ageold debate. On the one hand, Chomsky & Halle (1968) argue that mid vowels are marked by inactive features. In privative approaches, e.g. Lahiri & Reetz (2010), that has been interpreted such that mid vowels are unmarked for height features. That is, mid vowels are nonhigh and non-low, as in (5a). On the other hand, Element Theory (Backley 2011) and the related frameworks of Dependency Phonology (Anderson & Ewen 1987) and Government Phonology (Kaye et al. 1985;1990;Harris 1994) have long argued that mid vowels are combinations of elements. That is, they are marked with two active features, as in (5b).
These two representations for mid vowels mean that there are two interpretations for the lowering of a high vowel to a mid vowel. These are stated in (6).
b. Additive Lowering: high vowels become 'ET-like' mid vowels (as in (5b)) by adding the feature [low] to the representation in (4).
As I elaborate in section 2.2.1 and 2.2.2 below, both interpretations of Gothic Lowering -absent further data -are possible and depend crucially on the features ascribed to <r> and <h>. For this reason, sound changes that involve <r> and <h> as a natural class are significant because they offer a way to disambiguate the interpretations in (6) Because the sounds represented by <r> and <h> condition Gothic Lowering, these segments must be marked with a feature that can interact with high vowels and trigger the lowering of those high vowels to corresponding mid vowels. In this section, I propose that the sounds represented by <r> and <h> are marked with the feature [high], as in (7): As a result of the feature marking in (7), forms like EGmc *wir 'man' have two adjacent segments that are specified with the feature [high]. As made explicit in (8), those contiguous [high] segments are marked on account of the Obligatory Contour Principle (OCP), in the sense of Leben (1973).
To resolve the OCP violation, Gothic deleted the feature [high] from the EGmc high vowel (as in (9a)). Due to the deleted feature, a vowel obtained that was unmarked for height features, as shown in (9b). That new vowel was an 'SPE-like' mid vowel, consistent with the representation in (5a).
If Gothic <r> and <h> represent sounds marked with the feature [high], an apparent shortcoming of the analysis in (9) is that Gothic has many forms containing the sequences <ri>, <ru>, <hi>, and <hu>, e.g. Gothic riqiz 'darkness', runnuþ 'you all ran', hinder 'behind', and huzd 'treasure'. Those sequences are also characterized by adjacent [high] segments and could therefore be expected to trigger Gothic Lowering as well. For this reason, I argue that dissimilation, like assimilation can be progressive or regressive and that Gothic Lowering is a kind of regressive dissimilation. As such, [high] vowels dissimilate from following [high] consonants. Because the process is not progressive, however, [high] vowels do not dissimilate from preceding [high] consonants. For space, I do not discuss formal approaches that capture regressive dissimilation. However, there are analyses that can capture dissimilatory directionality. For example, Kostakis (2015: 72ff) argues for an OCP constraint that is contextualized to vowel-consonant sequences to account for regressive denasalization in French language history. This is precisely the kind of contextualization that should be extended here to capture the regressive nature of Gothic Lowering. Note also the specific nature of the OCP constraints proposed by Coetzee and Pater (2006) as well as Hall (2008); such OCP constraints can also capture directionality.
In sum, <r> and <h> are marked with the feature [high]. Thus, Gothic Lowering emerges as a dissimilatory process of subtractive lowering: in order to resolve the OCP violation that emerges from contiguous [high] segments, the feature [high] is deleted, and a mid vowel is produced by virtue of that deletion. The implication of this analysis is that Gothic mid vowels are SPE-like, as in (5a) and that <r> and <h> were [high] sounds.
2.2.2 Analysis 2: Gothic Lowering is additive lowering (to be rejected) It is possible that the graphemes represented by <r> and <h> were not marked with the feature [high], but rather the feature [low], as in (10) If the sounds represented by the graphemes <r> and <h> are marked with the feature [low], Gothic Lowering falls out from a process of feature spreading. Accordingly, it is an assimilatory process of additive lowering. For this to be the case, not only do <r> and <h> need to be marked with the feature [low], but Gothic mid vowels must also be 'ET-like'.

Evaluating the best analysis
As the analyses in sections 2.2.1 and 2.2.2 suggest, the rhotic and dorsal fricative sounds of Gothic can be taken as [high] (as in (7)) or [low] (as in (10)). What is needed, then, is a way to disambiguate these specifications. Because there appear to be no other Gothic processes which point to one analysis or the other, the only additional information that can be used comes from related languages. For this reason, I turn to OHG.
There are three main reasons why OHG is a good language to consider for gaining insight into Gothic. Firstly, OHG is genetically related to Gothic. As the (simplified) family tree in (13a) makes clear, both languages descend from PGmc. Secondly, the sounds represented by <r> and <h> are inherited from the PGmc rhotic and dorsal fricative, respectively, as indicated in (13b). 6 In (13c), the development of the PGmc verbs *faranan 'to travel' and *slaxanan 'to strike' is presented as a concrete example of a common, inherited rhotic and dorsal fricative sound.
The third and most important reason why OHG can shed light on Gothic phonology is that, like Gothic, <r> and <h> pattern together as a natural class (elaborated in the next section). When these facts are taken together, there is a fairly strong syllogism that emerges: if the rhotic and dorsal fricative of Gothic pattern as a natural class and are inherited from PGmc and if the rhotic and dorsal fricative of OHG pattern as a natural class and are inherited from PGmc, then it is highly probable that Gothic <r> and <h> are phonologically the same as OHG <r> and <h>.
While it is clear that the comparison of these sounds is less convincing as more and more time elapses -and to be sure, Gothic and OHG have a great deal of temporal and geographic separation -the natural class patterning is very difficult to explain unless these later stages perpetuate some phonological attribute of the earliest common stage.

Old High German Primary Umlaut
In this section, I argue that <r> and <h> in OHG pattern as a natural class of [high] segments because they inhibit the process known as Primary Umlaut. Primary Umlaut is discussed and analyzed in section 3.1. Section 3.2 follows with a discussion and analysis of the blocking effect that <r> and <h> exact on Primary Umlaut. Because OHG <r> and <h> pattern as [high] sounds, the Gothic analysis in section 2.2.1, as dissimilatory lowering, is most consistent with the emerging picture of early Germanic phonology.

Primary Umlaut: The basic pattern
OHG Primary Umlaut is a sound change that refers to the raising and fronting of short [a] to a short, tense [e] when an [i] or [j] (i.e. an umlaut trigger) followed in an adjacent syllable. Some examples are presented in (14). The first column pre-6 Unlike Gothic, where virtually all instances of <r> and <h> are old and descend from PGmc *r and *x, OHG has both old and new instance of <r> and <h>. The new cases of <r> emerge from the rhoticism of PGmc *z (compare Gothic miz 'me' with OHG mir). The new instances of <h> come from PGmc *k (compare Gothic ik 'I' with OHG ih). With respect to Primary Umlaut, discussed in section 3, the old and new instances of OHG <r> and <h> pattern consistently. Their phonological representation is therefore taken to be consistent. sents PGmc reconstructions from Orel (2003), using reconstructed inflections from Ringe (2006). Each of these forms has a short [a] followed by an umlaut trigger. The OHG forms in the second column are the reflexes of the PGmc forms. In each example, OHG [e] in the initial syllable corresponds to PGmc [a]. For space, the additional sound changes that separate the OHG forms from PGmc are not discussed.  (14) are diachronic in nature, Primary Umlaut also produced synchronic alternations in the language. Some examples are given in (15). OHG forms with non-umlauted stems are shown in the first column. Umlauted forms are presented in the second column. The latter forms are umlauted because their inflectional endings begin with an umlaut trigger, namely, [i]. Stem allomorphy among compositionally related verb forms is exemplified in (15a). The data in (15b)  Umlaut processes (in Germanic languages and elsewhere) are unequivocal examples of distance assimilation. All versions of Autosegmental Phonology model such processes as the result of feature spreading. Adopting privative features, there is only one possible height-based analysis: 7 the umlaut trigger, which is specified with the feature [high], spreads that feature onto the target stem vowel, as in (16). [high] In (16a), the active height feature spreads regressively from the umlaut trigger ([i]) onto the target ([a]). In consequence, a new mid vowel forms, as shown in (16b), which has two active height features. In other words, the resulting mid vowel is an 'ET-like' mid vowel, which is simultaneously [high] and [low]. 8 Given that the feature [high] is the only active feature which can trigger the process of Primary Umlaut (see also fn. 12), we stand to learn about the phonology of the OHG rhotic (<r>) and dorsal fricative (<h>), since these segments can inhibit Primary Umlaut. Examples are presented and analyzed in section 3.2.

Primary Umlaut blocking
OHG Primary Umlaut was blocked in certain dialects. This point is demonstrated in (17). The first column presents PGmc reconstructions that are expected to serve as inputs to Primary Umlaut in OHG. The second and third columns are representative of the two kinds of OHG dialects. 'Blocking' dialects are exemplified by the forms in the second column. Here we see that PGmc *[a] unexpectedly corresponds to OHG [a], despite the presence of an umlaut trigger in the following syllable. 9 In contrast to the 'Blocking' dialects, 'Non-Blocking' dialects are characterized by the expected pattern of umlaut, as in the third column. Refer to Reiffenstein & Braune (2004: 29-31) for more detailed information about the dialect regions where blocking does and does not occur. There are three segments which block the application of Primary Umlaut: coda rhotics, as in (17a), dorsal fricatives (represented by <h>), as in (17b), and coda laterals, as in (17c) For ease of comparison to Gothic <r> and <h>, the analysis below will focus only on the blocking effect imposed by OHG <r> and <h>. The fact that <l> in OHG is 8 The difference between the mid vowel representation for OHG and the one for Gothic is discussed in section 4.3.2. 9 Or former presence of an umlaut trigger, as in OHG warmen < warm[j]an. 10 The blocked forms in (17) (Goldsmith 1976), as in (18). [high] The diagram in (18a)  Because <r> and <h> pattern as a natural class in Gothic and in OHG, the best analysis of both languages is the one that can treat the natural-class pattern in a consistent way. In OHG, <r> and <h> were [high]. If the etymological connection is to be maintained, and if these sounds are assumed to pattern together for the same reasons, it follows that the best analysis of Gothic Lowering is the one that sees that change as a process of dissimilatory lowering, as in section 2.2.1.

Discussion
In this section, I discuss some selected points pertaining to the analyses presented above. Section 4.1 considers possible phonetic implications of the feature [high] for the sounds represented by <r> and <h>. Section 4.2 finds further support for the "rhotics-arehigh" analysis by looking to instances where rhotics trigger postalveolarization. In section 4.3, I discuss some of the phonological implications that the analysis has for segments like velar stops, laterals, and mid vowels. Previous accounts of Gothic Lowering and OHG Primary Umlaut are addressed in section 4.4.

Possible phonetic implications
In phonological theory of the 1970s, 80s, and 90s, features were thought to be universal. The two most important reasons for assuming universality were that (1) features could be defined with respect to articulation (and to a lesser extent also acoustics), and (2) groups of sounds with a common feature patterned in natural classes, which turned up again and again in the synchronic and diachronic processes of diverse languages and language families.
More recent work on features and feature theories, most explicitly Mielke (2008), argues that features are not truly universal (in the sense of Universal Grammar), but rather emergent. An emergent feature system is newly constructed by every first language learner based on everything from non-language-specific categorizing of sounds to social factors. An important piece of supporting evidence for emergent theories is that unnatural classes are quite common. Since Emergent Feature Theory maintains that a learner can ascribe any feature onto any sound -so long as there is evidence for that feature assignment somewhere in the phonology, or even the linguistic community -any group of sounds can potentially enter into a natural class relationship. 13 Nevertheless, the general facts about producing (and perceiving) a sound like [i] or [u] are similar from language to language. Thus, Emergent Feature Theory predicts that acquirers of a language will have a fairly high probability of encoding such sounds with consistent phonological features (e.g. [high]). In this way, 'universal features' and 'emergent features' can and should make the same predictions most of the time. Indeed, Mielke (2008: 118) finds that the SPE feature system is able to capture phonologically active natural classes in 70.97% of the 6,077 cases investigated. Put differently, features correlate to 12 It might be possible to imagine an approach to Primary Umlaut blocking that sees <r> and <h> as [+low] segments. Such an approach might argue that Primary Umlaut is actually the spreading of [-low] and that the [a] and <r> in a word like arbi share a linked [+low] autosegment. In this approach, the feature [-low] would be blocked due to the so-called linking constraint (Hayes 1986 (Blaho 2008, and sources therein) represents an alternative approach to account for phonological features that do not accord with their phonetic properties, thereby licensing unnatural classes.
articulation with some degree of success. Even if articulation does not cause feature structure, the correlation between articulation and feature structure suggests that analyses like the one in section 2 and section 3 likely reveal characteristics of pronunciation. The Chomsky & Halle (1968: 304) definition of [high] is presented in (21). 14 (21) "High sounds are produced by raising the body of the tongue above the level that it occupies in the neutral position; 144 non-high sounds are produced without such a raising of the tongue body." Applying the findings of this article to the definition in (21), we can make stronger hypotheses about the phonetic nature of <r> and <h>, as summarized in (22). . Alveolar and uvular consonants, by contrast, do not involve tongue raising above the neutral position. Thus, they are non-high sounds.
From an articulatory perspective, the IPA symbol [ɹ] refers to a collection of different rhotic sounds that are produced along an exceptionally broad range of places. For this reason, it is set apart from the other rhotic symbols in (22), which are more precise articulatorily. Quoting phrases from Ladefoged & Maddieson (1996: 234), [ɹ] can be "alveolar or postalveolar." It can be "more or less retroflex." If bunched, [ɹ] will have constrictions "at the center of the palate" (and in the lower pharynx 16 ). Zhou et al. (2008: 4466) further clarify that "the locations of constrictions, and the degrees and lengths of constriction significantly differ, especially along the palate." Since the center of the palate is the dividing point for palatal and velar sounds, it follows that bunched [ɹ] could be palatal or velar. In sum, [ɹ] is a rhotic symbol for numerous rhotic articulations that overlap entirely with the places of articulation corresponding to the feature [ There are a number of independent reasons to reconstruct PGmc *[ɹ]. Firstly, [ɹ] is found throughout the Germanic languages into the present day. In American English and Faroese, [ɹ] occurs in all environments. Árnason (2011: 115) describes the Faroese rhotic as "typically (post-) alveolar or retroflex, and most of the time…more like an approximant than a trill." Additionally, [ɹ] is frequent in Dutch (e.g. Goeman & Van de Velde 2001;Sebregts 2014). It is found in Frisian (cf. the transcription in Walker 1990: 27). It is also a feature of German (which has declined significantly in recent decades). Consider, however, the map of German rhotic pronunciations in (23) from Göschel (1971). (23) It can be seen from the map that, in the mid 1930s, the German of far eastern dialects employed a retroflex rhotic or alveolar approximant [ɽ]/ [ɹ]. 18 This is observed also by Bellmann (1961: 21-23 Low and Central German dialects. 19 In sum, the collection of sounds captured by the symbol [ɹ] may best represent a '[high] rhotic'. Since such [ɹ]-sounds are found in North Germanic (e.g. Faroese) and West Germanic (e.g. English, Frisian, Dutch, and German), the analyses from section 2 and section 3 suggest that the [high] rhotics of Gothic and OHG were similar to the [ɹ]-sounds of North and West Germanic languages. One area for future research might consider whether the diverse articulations characteristic of [ɹ] can shed light on the rhotic diversity of present-day Germanic languages.
Another important direction for future research will be in rhotic allophony. The rhotic sound of many Germanic languages stands in complementary distribution. In Walker's (1990: 27)  There are two additional pieces of evidence supporting that <h> was [x] and not some other fricative. The first is that conservative Bavarian and Alemannic dialects preserve the velar articulation. For example, Russ (2002: 77) notes that [x] is pronounced in the Alemannic dialect of Bosco Gurin. Berroth (2001) transcribes Middle Swabian (Alemannic) words with a velar fricative. Bachmann (2000) offers a particularly detailed phonetic description of the Bavarian dialect spoken in Eslarn. According to Bachmann, an advanced velar fricative [x̟ ] is even found after back vowels.
The second piece of evidence that <h> was [x] is that, before [s], the sound represented by <h> dissimilated its continuancy feature and shifted back to [k]. For example, OHG ahsa 'axis', flahs 'flax', fuhs 'fox', lahs 'salmon, lox', sehs 'six' > Modern German Achse , we would expect this pre-sibilant shift to produce a non-velar sound. 22 Accordingly, the analysis presented above, in addition to evidence from conservative dialects and the development of the <h>-sound before [s] makes any reconstructed sound other than [x] untenable. This finding is significant because it is frequently assumed that PGmc *[x] retracted to [χ] in early Germanic languages (see discussion in Barrack 1987 andHall 2009). Some researchers even argue for early debuccalization, whereby <h> represented [h] (Howell 1991;Iverson et al. 1994;Howell & Somers Wicka 2007). Counter to these claims, the evidence above indicates that <h> in Gothic and OHG represented [x]. 20 Hall (1993) finds that the uvular [ʀ] of modern Standard German patterns as a coronal sound in the synchronic phonotactics of the language. One interpretation of Hall's finding is that the coronal feature was supported by an earlier pronunciation. When that pronunciation shifted to [ʀ], the coronal specification continued, even though it did not accord with its phonetic implementation. Future research will need to investigate how long it is possible for a feature specification to be retained when that specification is no longer supported by a sound's phonetic characteristics. 21 WGmc *[k] is rendered in OHG orthography as <k> and <c>, where it is retained, but as <h>, where it has shifted to a fricative. If <hs> sequences entered into OHG from an earlier *

Further evidence from postalveolarization
The present section examines instances of rhotic-conditioned postalveolarization as further support for the "rhotics-are-high" analysis. Section 4.2.1 discusses postalveolarization in Germanic language history, and section 4.2.2 considers how [high] rhotics may also advance our understanding of so-called 'ruki-rules' outside of Germanic languages.

Early New High German Postalveolarization
Early New High German (ENHG) Postalveolarization is exemplified by the forms in (24) (24) can be analyzed as an example of progressive height assimilation. This change is presented in (25).
In (25a), the feature [high] spreads progressively from the rhotic sound onto the following alveolar sibilant. As a result of that spreading, [s] (which is alveolar and therefore non-[high]) shifts to [ʃ] (which is postalveolar and [high]), as illustrated in (25b). 23 The characterization of [ʃ] as a [high] sound is not controversial. Beyond the fact that a 'high ' [ʃ] is consistent with the definition presented in (21), there is overwhelming crosslinguistic evidence to support the claim. For example, there are many instances of postalveolarization, which are conditioned by high vowels (see Kochetov 2011 for a discussion of such cases, which comprise a subset of coronal palatalization processes). These instances make clear that [ʃ] and high vowels frequently pattern together as a natural class.
In contrast to high vowels, Kochetov (2011) finds no instances of postalveolarization, which are triggered only by low vowels. Accordingly, it is unclear how a 'rhotics-arelow' analysis of these data could be reconciled with broader cross-linguistic facts about postalveolarization.
In sum, the 'rhotics-are-high' analysis captures high vowels, postalveolar consonants, and the rhotic of early Germanic languages as a natural class. That natural class pattern, unlike any other proposals that I am aware of, points to the feature [high] as a common phonological structure involved in Gothic Lowering, Primary Umlaut Blocking, and ENHG Postalveolarization. As I discuss in the next section, it is precisely this group of sounds that are also involved in so-called 'ruki-rules'.

An application beyond Germanic
An example of a ruki-rule is observed in the Avestan reflexes of Proto-Indo-European (PIE) *[s]. In Avestan, PIE *[s] is reflected as a post-alveolar [ʃ] following the segments [r, u, k, i]. Some examples of the change from Hall (1997: 211) and Martínez & de Vaan (2014) are presented in (26) The analytical sketch in (27), which is not intended to be a detailed account, may be helpful to better understand data that have not been easy to capture in phonological terms. Blevins (2017: 49) points out that ruki-rules have not been satisfactorily accounted for with reference to any traditional approach to features. Blevins sees the lack of a featural explanation as a failure of formal phonological approaches to these data. While there are undeniably 'crazy rules' that speak to Blevins' point, the sketch here suggests that rukirules might not be among them. The important point to be drawn from the discussion above is that the mid vowels of Gothic and OHG never have a common origin. As such, it is possible, as the analysis indeed suggests, that the two languages developed new mid vowels that were typologically distinct. Gothic mid vowels were SPE-like mid vowels. They were non-[high] and non- [low]. OHG mid vowels, by contrast, were ET-like mid vowels, which is to say, they were simultaneously [high] and [low]. Because the OHG mid front vowel is sometimes inherited from PGmc, an implication of the analysis may be that PGmc, like OHG, had ET-like mid vowels. I leave this open to further research. 28
Concerning Primary Umlaut, Braune (1877: 552) first noted that <r> and <h> (and <l>) blocked the sound change from occurring when they were situated between the trigger and target of Primary Umlaut. In Braune's opinion, <r> and <h> (in words like OHG arbi 'inheritance' and mahti 'powers') were 'dark' sounds, meaning that they were closer in articulation to back vowels. As such, they counteracted the umlauting effect of [i]. Beyond that initial speculation, later work did not supply much insight into the inhibitory nature of <r> and <h>. For instance, Voyles (1992: 212)  'conspiracy' to refer to a linguistic commonality, which turns up in temporally and geographically disparate contexts. 31 Vennemann (1972) argues that Gothic Lowering is a kind of 'relative' assimilation. From this perspective, <r> and <h> have a relative feature {low} 32 since, relative to the high vowels [i] and [u], the tongue posturing of <r> and <h> is claimed to be lower. In a similar manner, the mid vowels [e] and [o], relative to [i] and [u], are also marked with the relative feature {low}. In this way, Vennemann (1972) can formalize Gothic Lowering as a kind of relative assimilation: Gothic high vowels shift to {low} mid vowels owing to the influence of the {low} consonants <r> and <h>. 33 Relative features are clearly language specific. For example, in OHG, Vennemann argues that all dental consonants are {low} with respect to the high back vowel [u]. If Gothic were similar to OHG, dental consonants in Gothic in words like brust 'breast' and gulþ 'gold' should be {low} and, as such, trigger Gothic Lowering. In the end, it is unclear how the relative feature responsible for Gothic Lowering might account for the blocking of Primary Umlaut in OHG.
A formal treatment of OHG Primary Umlaut is proposed by Iverson et al. (1994). In their approach, <r> and <h> are vowel-like. That vocalic quality is responsible for blocking Primary Umlaut (for space, the specific formal reasons are not discussed here in detail). Evidence for the vocalic character of <r> is rhotic vocalization in modern dialects. The authors follow Howell (1991) in that they believe <h> represents [h]. Since [h] can be understood as a voiceless vowel, <r> and <h> can pattern together as a natural class.
As discussed above, however, there is no evidence that <h> ever debuccalized in words like OHG mahti 'powers', which retain a dorsal fricative into the present day. Additionally, the pronunciation of <h> as [k] before [s] strongly indicates that <h> had a velar articulation, not a glottal one.
The additional problem with the account by Iverson et al. (1994) is that there is no attempt to explain why <r> and <h> in Gothic also pattern together as a natural class. Vocalized forms of <r> and <h> are clearly not conditioners of Gothic Lowering (especially since onset and coda consonants serve as triggers). Thus, their account of OHG Primary Umlaut leaves us to conclude that this historical conspiracy is coincidental; two separate accounts would be needed to analyze the OHG facts and Gothic facts. If each historical change were to require a unique account, there is little hope of aggregating insight into broader characteristics of early Germanic phonology.

Conclusion
One of the difficulties in individually analyzing archaic phenomena like Gothic Lowering and OHG Primary Umlaut Blocking is that the assortment of possible theoretical representations (e.g. the representation of rhotics, dorsal fricatives, and mid vowels) corresponds to an array of plausible analytical treatments of the phenomena in question. 34 This article aims to evaluate the different analytical treatments of one language by looking to a sister 31 An uncontroversial example of a linguistic commonality would be a palatalization rule like [k] > [tʃ]/_[i], which as Blevins (2004: 138) points out, is characteristic of Indo-Iranian, Bantu, Chinese, Cowlitz Salish, and Mam language history. 32 I use curly brackets to distinguish the relative feature {low} from the absolute feature [low]. 33 It is unclear if Vennemann's notion of feature relativity holds up to closer scrutiny. In its original conception, Vennemann's (1972) idea doubles the phonological system into "true" phonology and "relative" phonology without careful discussion of how the two systems might or might not interact. I am aware of no other study arguing for relative features. 34 This is equally true for phonetic research. Because rhotics, for example, can interact with vowels in an assortment of ways, it is essential to examine rhotic effects that are embedded within a larger natural-class pattern. Future research that aims to retrofit the phonetic properties of sounds in modern languages to archaic developments (as in Denton's 2001Denton's , 2003 work) needs to better account for natural class patterns.
language. It finds that the phonology of one sister language can elucidate the phonology of another when both languages exhibit a common natural class pattern. I have argued that the graphemes <r> and <h> in Gothic and OHG have in common that they are [high] sounds. With that representation, both the analysis of Gothic lowering and the analysis of OHG Primary Umlaut Blocking fall out in a straightforward way. In Gothic, the height feature of <r> and <h> triggered dissimilatory lowering of preceding high vowels due to an OCP violation. In OHG, that height feature blocked the application of OHG Primary Umlaut due to the No-Crossing Constraint. Further evidence for a [high] rhotic in Germanic language history was observed in the rhotic-conditioned postalveolarization of MHG [s].
Because the rhotic and dorsal fricative sounds of Gothic and OHG are reflexes of PGmc *r and *x, their patterning as [high] segments in Gothic and OHG implies that they also patterned as [high] sounds in PGmc. In this way, the approach here aggregates insights from Gothic and OHG into a broader characteristic of early Germanic phonology.
Insofar as feature representations generally correlate to phonetic implementations, the collection of articulations subsumed under the IPA symbol [ɹ] represents the most prototypical kind of [high] rhotic. That is the case because they can be produced anywhere from the postalveolar ridge to the soft palate -precisely the range of places that, according to Chomsky & Halle (1968)