2. Old Norse R-Umlaut

Historically, the phonology of North Germanic was characterized by variousumlaut processes (a-umlaut, i-umlaut, etc.) described in standard works(Noreen Reference Noreen1923:§58ff, Gordon Reference Gordon1956:271–273, Heusler Reference Heusler1962:§15ff, Nedoma Reference Nedoma2001:§§7–9.3). The phonetic origin of most of these sound changesisregressive coarticulatory assimilation. One such development, R-umlaut, is the fronting of short and long back vowels before the phonemic runic segment algwiz, <R>, as shown in 1 (see Noreen Reference Noreen1923: 66–67, Gordon Reference Gordon1956:272, Nedoma Reference Nedoma2001:28–29, among others).

1. (1)

   

R-umlaut in Old Norse has been traditionally viewedas a “palatalizing” or “fronting”sound change (for example, Gordon Reference Gordon1956:272, Heusler Reference Heusler1962:21, Steblin-Kamenskij Reference Steblin-Kamenskij1963:366), where the coarticulatory effect of <R> is the tongue's shift forward in the oral cavity that causes fronting of back vowel tokens. Translated into acoustic terms, Denton (Reference Denton2003:19–20)has argued that the effect of the segment <R> was “… the raising of the preceding vowels’ F2 with little effect on F1, which generally correlates with vowel height.”

3. Historical Issues with the<R> of R-Umlaut

Despite the seemingly straightforward phonetic explanation outlined above, R-umlauthas traditionally been a source of scholarly contention because, as with many cases of consonantally conditioned vowel mutations in historical phonology, it is uncertain what phonetic value is to be reconstructed for the phonemic speech sound represented bythe futhark rune algwiz<R>. The minimal facts that define the parameters for reconstructing the rune include the following:

1. (i) PGmc⁺/z/ derives from PIE ⁺/s/ via Verner's Law, and the algwizrune <R> is a reflex of this segment, as in PGmc⁺sū-z NGmc⁺sū-R ON sýr ‘pig’ (see Krahe Reference Krahe1956:§24, Nedoma Reference Nedoma2001:28). Based on reflexes found comparatively, PGmc⁺/z/ may be recon-structed asa voiced dental or alveolar sibilant fricative. The algwizrune of the Elder Futhark thus may represent a conservative articu-lation such as [z] (Antonsen Reference Antonsen1975, Telemann Reference Teleman1980, Lass Reference Lass, Davenport, Hansen and Nielsen1983,Faarlund Reference Faarlund and Woodard2008).

2. (ii) PGmc⁺/z/ undergoesrhotacism (z > r) in North and West Germanic, merging with reflexes of PGmc⁺/r/ (< PIE ⁺/r/), the latter is generally reconstructed as a voiced apical trill, [r] (see Krahe Reference Krahe1956:§24, Denton Reference Denton2003).Footnote ¹ How the merger of ⁺/z/ and ⁺/r/ occurred has always been controversial, but a logical assumption is that (at least some) reflexes of ⁺/z/ must have had an articulation that could develop via a natural phonetic pathway into [r]. Traditional ac-counts claimed abipartite progression of fricative [z] toapproxi-mant [ɹ] via lenition, followed by merger of [ɹ] and [r] (=PIE ⁺/z/ and ⁺/r/, respectively; see Gordon Reference Gordon1956, Steblin-Kamenskij Reference Steblin-Kamenskij1963, among others; see Nielsen Reference Nielsen1998:130ff for a detailed discussion of the merger based ona thorough analysis of the corpus of runic inscriptions).

3. (iii) R-umlaut in Old Norse only occurs before the segment <R> (=PGmc⁺/z/) and not before reflexes of PGmc⁺/r/, represented by the rune raido (see examples in Noreen Reference Noreen1923, Steblin-Kamenskij Reference Steblin-Kamenskij1963). Therefore, the runes algwiz<R>and raido<R> must have represented different speech sounds in orthographic practice, at least initially, prior to any subsequent merger of the phonemes repre-sented by the two runes.

The last point is crucial as it shows that the reflex of PGmc⁺/r/ (< PIE ⁺/r/), commonly reconstructed as an apical trill [r], does not condition R-umlaut, whereas reflexes of PGmc⁺/z/, whatever their precise phonetic value, did. Thus, the sound change of R-Umlaut in Old Norse itself furnishes a possible avenue for insight into the reconstruction of PGmc⁺/z/.

The focus of research on the reconstruction of <R>and the phonetics of PGmc⁺/z/ has disappointingly focused on fact (ii) at the expense of (i) and (iii). The debate on the rhotacism and merger of ⁺/z/ with⁺/r/ re-volves around the observation that the corpus of runic inscriptions in Scandinavia show that the rune algwiz<R> (=PGmc⁺/z/) was kept distinct from the rune raido<R> (=PGmc⁺/r/) until about 900 C.E. when rune-carvers began to use the two runes interchangeably (see Nielsen Reference Nielsen1998:130). This later confusion led researchers to believe that reflexes of the phoneme /z/ developedarticulatorilyinto a rhotic segment, so-called jüngeres-r ‘younger r’, which allowed it to become confused for and interchangeable with reflexes of the otherrhotic segment in the picture, ⁺/r/,so-calledälteres-r‘older r’ (see, for example, Noreen Reference Noreen1923).

Germanic linguists have taken the noncontrasting usage of the two runic symbols as evidence of rhotacism in action: In practice, R is understood as ashorthand for z > r, an orthographic cover term for anypossiblearticulation on the cline from /z/ to /r/. For instance, Noreen (Reference Noreen1923:33–34)viewsthe phonetic value of the phoneme as “…wo[h]l einfrikativesdorsalesr”, presumably an articulation such as an uvular fricative, IPA [ʁ], or a uvular (fricative) trill, IPA [ʀ], whereas Gordon (Reference Gordon1956:268), among others, speculated that <R> was an “r-colored [z]” or a palatalized approximant [ɹʲ],either of whichwould be distinct from the other liquid phoneme in Old Norse, the apical trill [r]. Gordon (Reference Gordon1956:268) proposes the developmental sequence in 2a, with possible IPA values given in 2b.

1. (2)

   

Gordon's palatalized stage [ɹʲ] is ad hoc: <R> is understood to be palatalized at that stageprecisely to explain R-umlaut, that is, after rhotacism but before the youngerr merges with the older r.

This account can be argued against on phonetic grounds as follows. The lenition of an alveolar fricative [z] to alveolar approximant [ɹ] involves the tongue tip gradually lowering and retracting to produce an approximant; just such an articulatory progression is documented for rhotacism phenomena in Spanish and Catalan (see Solé Reference Solé, Dressler, Luschützky, Pfeiffer and Rennison1992) and elsewhere crosslinguistically due to well-understood articulatory and aerodynamic reasons (see Ohala & Solé Reference Ohala and Solé.2008). However, Gordon's progression requires the tongue tip to raise again in palatalization, the opposite coarticulatory shift required for lenition. The proposed se-quence, and numerous accounts like it (see Heusler Reference Heusler1962:21, Steblin-Kamenskij Reference Steblin-Kamenskij1963:366, among others),are basedon the traditional assumption that PGmc⁺/z/ developed in the following order: (1) Rhotacism of PGmc⁺/z/ > [ɹ] in early North Germanic, (2) R-umlautconditioned by [ɹ] or something like it in early North Germanic, (3) Merger of [ɹ] reflexes with /r/ (<PGmc⁺/r/) in later North Germanic.

There is a competing tradition onrhotacism of ⁺/z/ > /r/ in North Germanic that supports an alternative view that R-umlaut precedes rhotacism, that is, the progression occurred (1) R-umlaut, (2) Merger (= rhotacism). Proponents of this sequence argue thatthere is no clear evi-dence for any intermediate stage of development between PGmc⁺/z/ and its merger with ⁺/r/ in North Germanic; an increasing number of linguists studying the historical phonology of Germanicunderstandthe outcome of rhotacism/lenition of PGmc⁺/z/ (> [ɹ])as merger of /z/ with /r/ (see Antonsen Reference Antonsen1975, Teleman Reference Teleman1980, Lass Reference Lass, Davenport, Hansen and Nielsen1983, Faarlund Reference Faarlund and Woodard2008, among others). Stemming from PIE ⁺/s/ via Verner's Law, PGmc⁺/z/ is often taken to have a dental or alveolar realization (for ⁺/s/ in PIE, see Szemerényi Reference Szemerényi1996:51–53; for PGmc⁺/z/, see Faarlund Reference Faarlund and Woodard2008:220), though some scholars posit a postalveolar realization, such as [ʒ], as this place of articulation may have facilitated merger with reflexes of PGmc⁺/r/ on acoustic and auditory grounds (see Teleman Reference Teleman1980, Lass Reference Lass, Davenport, Hansen and Nielsen1983, Painter Reference Painter2011).

For instance, Painter (Reference Painter2011) arguesthat certain realizations of voiced coronal fricatives [z, ʑ, ʐ] such as those found in Polish could closely resemble realizations of rhotic approximant [ɹ] and trill [r] in theiracous-tic and auditory properties. This suggests that realizations of PGmc⁺/z/ may have been reanalyzed as tokens of PGmc⁺/r/ directly via hyper-correction (Ohala Reference Ohala, Masek, Hendrick and Miller1981, Reference Ohala and Jones1993) without anyintervening stage of develop-ment, such as an observable period where /ɹ/ (<PGmc⁺/z/) and /r/ (<PGmc⁺/r/) contrasted in the language. Under this analysis, rhotacismof ⁺/z/ is the equivalent of merger with ⁺/r/ (for the theoretical discussion of this view, see Solé Reference Solé, Dressler, Luschützky, Pfeiffer and Rennison1992, Painter Reference Painter2011). Importantly, from this perspective, the NGmc segment <R>which conditioned the R-umlautcould not have been a rhotic articulation but was more likely an unchanged reflex of PGmc⁺/z/, that is, an alveolar or postalveolar voiced fricative.

Recent phonetically-informed research onhistorical phonology of PGmc⁺/z/ and ⁺/r/ seems to delineate plausible developments from im-plausible ones (for example, Denton Reference Denton2003), but even newer accounts remain speculative and open to debate since there is no empirical data to support descriptive phonetic claims. With few further phonetic parameters to be adduced from comparative reconstruction of possible value of <R>, the state of knowledge is a surfeit of theories and no hard data. Consider that under various accounts, the<R> segment, whichconditions R-umlaut in North Germanic, may be any of the articulations in 3.

1. (3)

   1. a. voiced dental or alveolar sibilant fricative [z] corresponding to an unchanged pronunciation from the archaic PGmc⁺/z/ (see Antonsen Reference Antonsen1975 among others)

   2. b. voiced postalveolar or retroflex fricative, [ʒ] or [ʐ] (Teleman Reference Teleman1980, Lass Reference Lass, Davenport, Hansen and Nielsen1983)

   3. c. voice dapicalrhotic with central approximant [ɹ] realizations (Denton Reference Denton2003)

   4. d. voiced “lisped” apical or uvular fricative, [r] or [ʁ] (Noreen Reference Noreen1923:33)

The acoustic experiment reported below explores the issue of phon-etic reconstruction of R-umlaut in Old Norse within the framework of experimental historical phonology (Ohala Reference Ohala, Anderson and Jones1974, Reference Ohala, Solé, Beddor and Ohala2007). We attempt to test the degree of positive movement of F2 (which correlates broadly with vowel fronting) of the segments [z], [ʒ], and [ɹ] on the immediately preceding vowel, and examine the movement of F1 (which inversely correlates with vowel height). This laboratory investigation seeks to address empirically the question of what phonetic value might best be reconstructed for NGmc<R> through studying the acoustic effects of segments often cited as candidates for such a reconstruction. The results may thus partially clarify a long-standing problem in the historical phonology of Old Norse.

The remainder of this paper is structured as follows. The next section, section 4, outlines some phonetic assumptions and hypotheses that informed our experimental design. In sections 5 and 6, we present the methodology and results of our acoustic experiment on segments that condition vowel fronting. Section7discusses the implications of the experimental findings, and section 8 provides a model for the sound change, and presents our conclusions on the phonetic triggers of this sound change within the historical phonology of North Germanic.

4. Phonetic Hypotheses

In a recent discussion on what kind of rhotic in Germanic could condition R-umlaut, Denton proposes an alveolar approximant /ɹ/: “The ‘bunched’ or ‘light’ articulation of American English /ɹ/ has just the sort of tongue dorsum raising toward the palate and concomitant low front tongue position which can centralize vowels” (Denton Reference Denton2003:20; see Lass Reference Lass, Davenport, Hansen and Nielsen1983:79–82). The view that a species of rhotic approximant might con-dition R-umlaut has a long pedigree (going back at least as far as Noreen Reference Noreen1923). Laboratory investigation on the effect of [ɹ] tokens on back vowels shows that their primary effect on the vowel quality is central-ization, not fronting. Clark & Hillenbrand Reference Clark and Hillenbrand2007, an acoustic and perceptual study on the effects of /ɹ/ on back vowels in the American English dialect of the Inland North, show that, in the production of the sequence /aɹ/ as in the word star, there is significant fronting of the vowel nucleus, that is, the rhotic indeed causes F2 to risemarkedly; however, the rhotic also lowers F1, making the offset of the vowel nucleus more raised, and thus the vowel is centralized (Clark & Hillenbrand Reference Clark and Hillenbrand2007:283).

In an earlier experiment, Clark & Hillenbrand (Reference Clark and Hillenbrand2003) had determined that front vowels such as /i/ and /e/ are also lowered and slightly retracted due to the presence of a following /ɹ/. Thus, the global effect of various forms of approximant /ɹ/ is centralization of vowels, that is, they draw vowels inward from the periphery of the vowel space (Clark & Hillenbrand Reference Clark and Hillenbrand2003; Reference Clark and Hillenbrand2007). This centralizing effect of approximant /ɹ/ furnishes a possible explanation for the lowering of high vowels before reflexes of <R> in North Germanic and Old English: OE meord ‘reward’ (Go. mizdo); ON ero (PGmc⁺izunÞ), OE pronouns we, ge, he, me, Þe and ON pronouns véR, séR, éR, méR, ÞéRin comparison with Gothic weis, jus, is, mis, Þusand OHG wir, ir, mir, dir(examples from Noreen Reference Noreen1923, Campbell Reference Campbell2003; compare Denton Reference Denton2003:19). These examples make a suggestive case that <R> was a form of rhotic approximant that could trigger bothvowel-fronting and vowel-lowering coarticulatoryassimi-lations. However, it is clear from the examples in 1 that the R-umlaut vowel mutation involves more than modest fronting, such as /u/ > /y/ and /o/ > /ø/, which suggests that vowel fronting and vowel lowering/centralization are perhapsdistinct changes in the history of Old Norse.

We hypothesize that vowel fronting in R-umlaut was conditioned by an alveolar fricative /z/ (a conservative reflex of PGmc⁺/z/),and that this was an independent development from these later vowel lowering changestriggered by a rhotic /ɹ/ after rhotacism in North Germanic. It is worth considering experimentally whether other nonrhotic articulations exhibit positive F2 movement, particularlythe voiced dental sibilant fricative [z] or a voiced postalveolar sibilant fricative [ʒ], the two articu-lations reconstructed for PGmc⁺/z/ beforerhotacism discussed above (see also Teleman Reference Teleman1980). The main research question considered in the experiment reported below is simply this: Do reflexes of a voiced fricative /z/—phonetically alveolar [z] or postalveolar[ʒ]—condition the fronting of back vowels significantly better than tokens of the AE rhotic approximant /ɹ/?

Based on the previous phonetic research, it is extremely likely that a segment articulated with the tongue tip raised at the dental or alveolar place of articulation would induce the fronting of back vowels, since such a change is fairly well attested, for instance, the fronting of /u/ to /y/ before [t] (several earlier studies are cited in Ohala Reference Ohala, Masek, Hendrick and Miller1981, and more recent empirical evidence is found in Harrington et al. Reference Harrington, Kleber and Reubold2008). In the following acoustic experiment, we investigate, first, what phonetic segments in our test set show the greatest raising of F2 of the preceding vowel, that is, which trigger the most significantvowel fronting. Second, we look at which segments condition movement of F1, that is, which articulations affect vowel height significantly. Lastly, the experiment investigated which phonetic segments create the most distinctive coarti-culatory shift in the value of F2 over the duration of the vowel, since it has been demonstrated that rapid F2 movement can be a useful percep-tual cue for the identification of vowel quality (Ohala Reference Ohala, Masek, Hendrick and Miller1981).

It is predicted that the alveolar fricative /z/ and postalveolar /ʒ/ will induce higher mean values of F2 in an immediately preceding back vowel than the approximant /ɹ/. In view of previous work on vocalic fronting induced by alveolar obstruent segments (Ohala Reference Ohala, Masek, Hendrick and Miller1981, Harrington et al. Reference Harrington, Kleber and Reubold2008), we predict that alveolar [z] will show the greatest degree of fronting of an exemplar back vowel /u/, and that this effect will be greater in faster speech than in citation speech due to an increase in coarticulation. This would suggest that the voiced fricative /z/ is more likely to cause fronted variants of the high back vowels and support the argument that <R> involved in umlautwas phonetically realized as alveolar fricative [z] or something phonetically similar and had not yet undergone rhotacism in Old Norse. In other words, reflexes of PGmc⁺/z/ and ⁺/r/ were relatively conservative in the early stages of North Ger-manic as an alveolar fricative /z/ and an apical trill /r/, respectively.

5. Experimental Design