2. An Ingenerate View of i-Umlaut across Germanic

Many of the best scholars take the philological facts at face value as evidence for the original patterns of Norse umlaut, as shown in 1.

1. (1)

However, this contradicts the usual picture of how sound change unfolds. In the many various life cycle of language change models (Ohala 1993, Kiparsky 1995, Iverson & Salmons 2009, and many others), regular sound change grows from phonetic seeds typically nurtured by coarticu-latory and/or perceptual biases. Phonetic effects are then reinterpreted by later generations as phonological generalizations, and these are inte-grated into the broader grammatical fabric of the language and over time often lead to new morphological generalizations, such as association with particular inflectional categories. Later changes eventually obscure earlier generalizations but may leave clear traces in exceptional or mar-ginal patterns. Umlaut, as a quintessential assimilatory process, can be expected to follow such a path, as argued for West Germanic umlaut by work in the tradition of the “Wisconsin School” over the last 20 years, such as Iverson, Davis, & Salmons 1994, Iverson & Salmons 1996, Howell & Salmons 1997, and more recent work, all inspired by Buccini 1992 (see also Buccini 1995).

On a whole set of points, that work makes an account of umlaut sensu lato as restricted to long stems improbable. For instance, phone-tically, assimilation in long stems is obviously interrupted both by more intervening material and more distance between the triggering i and the target stem vowel. If distance assimilation is sensitive to intervening material, we expect it to operate over less and not over more intervening material. Blocking in Old High German umlaut (like many other assimilatory patterns in the languages of the world) behaves in ways thoroughly consistent with this: aCi forms umlaut in OHG regularly, while umlaut-blocking strings are typically of the shape aCCi. Phono-logically, short vowels appear more susceptible to umlaut, as in primary umlaut, while umlautless residues involve either complex intervening structures, as just described, or long vowels, for instance in the lack of secondary umlaut of /a:/ in Dutch, as in kaas ‘cheese’ (German Käse). Finally, the broad tendency across Germanic is for reduction and loss to proceed earlier and faster in long stem forms than in short stem forms, so that umlaut triggers remain in the grammar later in the latter than in the former paradigms. Given the chronological interaction between trigger loss and umlaut, this generalization provides more opportunity for um-laut to have operated in short stem forms than in long stem forms.

Also, directly within the historical and modern Nordic languages, we find weight-related patterns consistent with the basic principle that short stem forms are more susceptible to V-to-V assimilations. Nordic dialects, especially varieties of Swedish, exhibit phenomena known as vowel balance and vowel harmony (Riad 1998, 2005). In these, short stems show more general assimilations than long stems, as illustrated here with data from Älvdalen in western Sweden for the bidirectional process known as vowel harmony ii (from Riad 2005:1110):

1. (2)

Suffix vowels in the light forms assimilate to the quality of root vowels, while suffix vowels in heavy forms do not, as indicated by the bolded vowels.

As we have argued elsewhere (Iverson & Salmons 2009), the story of Old Norse umlaut is one of a very late life cycle change, which suggests that we should see strong morphological and analogical effects in its distribution. More generally, anchoring the emergence of Norse umlaut in a life cycle view pushes us to make explicit not only the phonological interpretation but also the earlier phonetic and later morphological context into which the phonology fits.

We hasten to add that the philological record differs between West Germanic, especially Old High German, and North Germanic: While we have some data from the key time period in Old High German, early i-umlaut in Norse falls between earliest Runic and the textual attestation of Norse. We have only sparse evidence from the key period, as surveyed by Riad (1992). In terms of ultimate outcome, then, Norse stands squarely between its West Germanic cousins English and German. In the former, i-umlaut is overwhelmingly ousted from the grammar; in the latter, it has seeped into every morphological nook and cranny. Norse has kept it, but in a contained way, deeply embedded in the inflectional morphology but not spread across it as far as in German. This, too, is consistent with a late stage life cycle.

3. Old Norse i-Umlaut

The key point of this overall analysis is that the analogical changes ensuing from the morphologization of phonetically expiring umlaut over the course of the 8th and 9th centuries caused umlauted vowels to associate morphoprosodically with long stem words like gestr < ⁺gest+ir < ⁺gast+ir ‘guest’ rather than with short stems like staÐr < ⁺steÐ+ir < ⁺staÐ+ir ‘place’. The full attested paradigms as given by Noreen (1970: 266ff.) and recapitulated by Iverson & Salmons (2004:85) are presented in 3.

1. (3) Attested Old Norse masculine i-stem nouns

   

Thus, the familiar distinction here is between short and long stems: The short stem nouns, whose stem rhyme consists of short vowel plus single consonant, have the unumlauted vowel throughout (/a/), whereas the long stem nouns, whose stem ends in short vowel plus two consonants or long vowel plus single consonant, have an umlauted vowel throughout (/a/ > /e/). As both of these noun classes would have had a thematic -i- in each of their case forms in Proto-Norse, for example, ⁺staÐ-i+r ‘place-nom.sg’, ⁺staÐ-i ‘place-acc.sg’, ⁺gast-i+r ‘guest-nom.sg’, ⁺gast-i ‘guest-acc.sg’, etc. (following Iverson & Salmons 2004:85), it would be expected that umlauted stem vowels should appear throughout the short stems as well as the long stems. However, this is famously not the case. The solution we laid out in 2004 was that the various losses of suffix vocalism, or syncope, took place after umlaut had exercised its effects to begin with only after long stems, as in the traditional view. These effects can be charted as in 4 (Iverson & Salmons 2004:93).

1. (4) Umlaut and syncope in early Proto-Norse nouns

   

At some point, syncope generalized so as to occur after short stems, too, of course. However, at the restricted early stage summarized in 4, when umlaut was still a purely phonetic process, long stem syncope introduced a measure of opacity inasmuch as, superficially, an umlauted vowel shows up in some words (namely, [gestr]) where the trigger for umlaut (an [i] in the next syllable) no longer appears. We reasoned that the transparent status of umlaut at this early stage induced follow-on generations of speakers to rethink the way a word like [gestr] could acquire an umlauted stem vowel even in the absence of a phonetic trigger for umlaut. With umlaut still phonetically very much alive, and with umlauted stem vowels occurring throughout the paradigm (trigger or no trigger), young learners could produce outputs identical to those of older speakers, with a regular occurrence of umlauted vowels throughout the long stem paradigms, by restructuring, or relexifying, the base form of a stem like /gast-/ to /gest-/. Maintaining an allomorphy-free paradigm by sticking with the stem representation /gast-/ would not do, since umlaut still changes /gast-/ into /gest-/ in the nominative plural and other words in the paradigm which still have an [i] in the suffix, and to preserve transparency of umlaut as a purely phonetic process meant that the umlauted vowel in the nominative singular had to have come about by some other means. That means was the restructuring of lexical /gast-/ to /gest-/, resulting in outputs identical to those of the older generation, just by different means. This development is laid out in 5 (Iverson & Salmons 2004:94).

1. (5) Relexification of long stems in later Proto-Norse

   

However, restructuring, we supposed, did not take place in the similarly uniformly umlauted paradigms of the short stems because here umlaut remained transparent: Stem vowels are umlauted where and only where an [i] vowel follows inasmuch as syncope had not yet generalized to take place after short stems as well as long. Thus, the presence versus absence of umlaut in short stems at this point was still fully predictable, as in nominative singular ⁺steÐir versus genitive singular ⁺staÐar, moti-vating retention of the unumlauted vowel in the underlying form. This contrasts with the situation in the long stem paradigms, where syncope has disturbed umlaut's phonetic determination, prompting its lexical-ization. Yet at some later point in this still early period of pre-Old Norse, syncope did generalize so as to apply irrespective of stem weight, re-sulting in the loss of suffix vocalism in the short stems as well.

The other key event, which we would date to about this same time, was the demise of umlaut as a purely phonetic process, initiated in the familiar way by the loss of trigger vowels due to syncope and other opacity-inducing adjustments. Associating the inevitable demise of phonetic umlaut temporally with the emergent generalization of syncope to all stem types results in a natural reversion of the (unattested) superficially umlauted vowels in short stems to unumlauted vowels: As the underlying form of the short stems had not changed, the removal of suffix vowels more or less simultaneously with the expiration of phone-tically triggered umlaut automatically surfaces the base form of the stems, as charted in 6 (Iverson & Salmons 2004:95).Footnote ² (As described below, this also creates other relevant morphological alternations, such as competing suffixes that would or would not have triggered umlaut, as in —isk/-sk.)

1. (6) Umlaut loss and generalization of syncope to short stems

   

Thus, this maintains the chronology in which umlaut occurs prior to syncope, which itself—following Riad's (1992:113–114, elsewhere) well-known, detailed survey of the Runic evidence—began around 625 in long stem words, then extended to medial environments in short stems ca. 675 and to final syllables in short stems ca. 830. The chronology of Norse syncope in 7 is adapted from Riad 1992, with his orthography.

1. (7)

To recapitulate, since umlaut remained predictable in short stem words until the later losses of trigger vowels, short stems retained, and thus surfaced, their underlying unumlauted vowels upon the death of umlaut as a phonetic process. In contrast to the reversion in short stems, underlying vowels in long stems restructured to umlauted variants with the earlier disappearance of umlaut-inducing vocalism because in these words syncope took place when umlaut was still phonetically alive, forcing generalization to the umlauted forms of paradigms. The general result in the literary language, then, is that long stems have umlauted vowels, while short stems do not.

4. Paradigm Resolution via Rule Loss: Restructuring versus Reversion

Thus, the course of i-umlaut in Old Norse took two paths, both emanating from loss of the rule as other changes, namely, those of syncope, made its effects unrecoverable. On the one hand, in long stems umlaut generalized throughout paradigms because that was the only way to achieve parity in umlauted vowels between a generation of speakers who could derive them phonetically (prior to syncope) and another who could not (after syncope). On the other hand, in short stems umlaut re-mained phonetically determined even as its effects were lexicalized elsewhere. However, with umlaut no longer derivationally active in long stems, its utility to the grammar reduced to a point where its dynamic function could be dispensed with altogether, resulting in its removal from the phonology per se.

The loss of phonological i-umlaut was primed by the opacity of the process that arose in long stems, but its consequent coincidental dismissal from the grammar affected short stems (where it had still been operating transparently prior to the generalization of syncope) in quite the opposite way, namely, with reversion to the unumlauted status quo ante. The loss of phonetically determined umlaut in short stems was thus precipitated by its syncope-induced invalidation in long stems—a passenger, as it were, that sank on the same ship.

It has often been observed that languages do not lose transparent rules, only opaque ones (King 1973, among others), and the phono-logical demise of umlaut due to its opacity in long stems had the automatic consequence of stopping its operation in short stems as well. We refer to loss of a rule under this condition as parasitic, presumably the only circumstance under which transparent loss may occur, that is, as part of a rule which is otherwise opaque. Our basic understanding of the effects of rule-related sound change on paradigm resolution is this: In paradigms where a phonological innovation has been rendered opaque by the operation of other sound changes, restructuring of the base form incorporates rather than continues to derive the results of the innovation as it dies out; but if the innovation remains transparent in other para-digms, its expiration enables reversion to the antecedent phonological form.

Both patterns might be understood as instances of the traditional notion of analogy, as they result in allomorphically uniform paradigms, but the former generalizes a sound change to contexts in which it never occurred naturally, whereas the latter actually undoes, or reverses, a sound change. In the case of gestr < ⁺gest+ir < ⁺gast+ir, opacity arose via the early effect of syncope in long syllable contexts. That is, the chronologically prior fronting of /a/ to /e/ due to umlaut was caused by a vowel that no longer exists, namely, syncopated /i/, rendering umlaut qua rule opaque with respect to gestr. At this stage, learners seeking to match the output of speakers who continued to derive gestr from /gast+ir/ through sequenced interaction of umlaut and syncope, chose instead to arrive at gestr directly from /gest+r/, there being no longer reason for positing underlying /i/ in the increasingly invariable, syncopated form of the nominative singular suffix, /r/. At the same time, umlaut is functioning transparently in short stems, before syncope had extended to all syllable types; hence, /staÐ+ir/ > ⁺steÐ+ir.

However, this is the point at which umlaut became opaque in other classes, namely, long stems like gestr, causing learners to give up on it derivationally in these cases and to restructure the base form. As umlaut loses its derivational vitality, the further loss of its conditioning factor through the generalization of syncope to short stems, where umlaut had been operating transparently, further removes reasons for positing its phonological existence and leads to the rule's demise generally (residually restricted now to stem-internal /i/, as in /katil+r/ > ketill) and to automatic reversion to the base form: /staÐ+ir/ (> ⁺steÐ+ir > ⁺steÐ+r) > staÐr. Moreover, the syncope pattern of vowels in short stems paral-leled as well as followed that in long stems, that is, the two types were not distinguished, in the manner of Hesselmann 1945, as to whether unaccented vowels reduced (short stems) or not (long stems); rather, syncope proceeded in a uniform, chronological way.

5. Umlaut in Weak Verb Paradigms

The situation in verb paradigms is basically the same: Long stem class I weak verbs like dœma ‘to deem’ < ⁺dom+jan (compare dom ‘judgment’, also ‘doom’) have umlauted vowels throughout their paradigms, whereas short stems, like telja ‘to tell’ < ⁺tal+jan (compare tal ‘tale’), do not show umlaut in preterit indicatives (talÐa ‘told.1.sg’). However, short stem verbs do have umlauted vowels elsewhere (subjunctives and present indicatives). Paradigms exemplifying this difference are listed in 8, adapted from Gordon (1957:303).

1. (8)

Kiparsky (2006) points out that these and many other verbs in the weak class are denominal, so that a derivational association would still exist between the noun forms without umlaut and the verb forms with. Yet with umlaut no longer phonetically active, but rather morphologized so as to apply only in specific word categories, the question naturally arises as to how the difference in extent of umlaut between long and short stem weak verbs might have come about. The paradigms in 8 reveal that the synchronic morphophonological umlaut rule required for Old Norse weak verbs is restricted because stem vowels are umlauted everywhere except in short stem preterit indicative forms. The syn-chronic morphologized remnant of once phonetically productive umlaut must then be something like 9, a “crazy rule” (Bach & Harms 1972, Blevins 2004, 2006) rising out of the vicissitudes of fading historical changes.

1. (9) Weak verbs of the first conjugation have umlauted stem vowels in the present tense and in subjunctives, and, for long stems, also in the past tense.

A rule like 9, peculiar as it may be, is necessary if denominal verbs like telja and dœma are still synchronically derived from their source nouns tal and dom. Of course, derivational morphology is often easily separated from source words, so it could be that the psychological association between Old Norse tal and tel- or dom and dœm- was syn-chronically no closer than that between modern English tale and tell or doom and deem. Consider, for another parallel, the once phonetically identical German words Weg ‘path, way’ with [e:] and weg ‘away’ with [ε]—the former has analogized to the long vowel of the plural, while the latter has retained its historical short vowel (Fourakis & Iverson 1984, Page 2007). Here, even very close semantics and zero derivation did not prevent one member of the pair from changing while the other remained the same. At any rate, even assuming that telja still derived from tal and dœma from dom, a morphologically restricted umlaut rule for Old Norse to the effect in 9 would be necessary in order to account for the syn-chronic relationship between noun and verb paradigms.

Our approach, parallel to traditional descriptions, is that umlaut applied before syncope historically, first after long stems like /do:m-/, later after short stems like /tal-/. In view of the -j- (or -i-) affix originally present in all forms of the weak class I conjugation (compare paradigms of Gothic nasjan ‘to save’, sokjan ‘to seek’, Heusler 1964:101–102), there would have been umlauted vowels throughout both stem classes at the earliest stage when umlaut was still phonetic, as sketched in 10.

1. (10) Umlaut and syncope in early Proto-Norse class I weak verbs

   

Except for the preterit indicatives in short stems, umlaut pervades the paradigms, even in forms that no longer have an umlaut conditioning suffix, such as tel < ⁺tal+ja ‘tell-1.sg.pres.ind’ and telr < ⁺tal+jir ‘tell-2.sg.pres.ind’. Of course, these are accounted for if umlaut preceded the syncope of finals but would also fall into place via the analogizing effects of rule 9 even if that sequence were reversed. However, the question remains as to why particularly the preterit indicatives, and only in short stems, should have emerged without umlaut in the literary lan-guage.

On the traditional view that umlaut took place before syncope, the special status of the preterit indicative short stems becomes clear once it is recognized that regular syncope in long stems—which, as we and most others claim, occurred earlier than in short stems—resulted in restructuring of the suffix /-iÐa/ to just /-Ða/. Thus, the syncope of medial /i/ in long stems like /dom+iÐa/ happened after otherwise phonetically transparent umlaut to result in dœmda. This, in turn, led speakers to adduce a morphologically triggered version of umlaut operating in the long stems because, though still generally based on the occurrence of a following /i/ or /j/, umlaut in the long stem paradigms could no longer be phonetically predicted everywhere, just as was the case in the develop-ment of long stem nouns charted in 4 and 5.

Ultimately, the regular absence of syncopated vowels had the effect of synchronically separating some of the suffixes in this conjugation according to stem weight, so that short stems had -Ø, -r, -r in the first, second, and third persons of the singular present indicative, respectively, versus -i, -ir, -ir in the long stems, as shown in 8. We doubt that these variants continued to be determined by synchronic syncope during the literary period; rather, we believe that they came to be associated with existing forms directly, as allomorphs conditioned by stem weight in this class. Similarly, at the stage exemplified in 10, the -iÐa of short stem preterits corresponded superficially to -Ða (or -da) in long stems, which, we expect, led to synchronic separation of these affixes according to stem weight, too, even at the point where syncope still had not extended its effects to short stems. This is exemplified in 11.

1. (11) Umlaut in Proto-Norse class I weak verbs after effects of syncope in long stems

   

With the generalization of syncope to all stem types, umlaut would have become opaque in the short stem preterits, too, as shown in 12.

1. (12) Umlaut in Proto-Norse class I weak verbs with syncope in all stem types

   

We suppose that the allomorphy with respect to historical /-iÐa/ (sometimes -iÐa, sometimes -Ða) combined with the emergent extension of syncope in short stem medials to restructure the suffix to simply /-Ða/ everywhere. This was a relatively early development, too, taking place while umlaut was still subject to phonetic triggering (though also to morphological determination in long stems). However, with no basis any longer for positing /i/ in the preterit suffix, which, at this juncture, was always surface [-Ða] ([-da]), learners came to interpret the suffix consis-tently as underlyingly /-Ða/, too. The result of this restructuring was that preterit indicatives had no /i/ vowel to continue to induce umlaut, as represented in 13 (Noreen 1970:§159).

1. (13) Umlaut in later Proto-Norse with /-iÐa/ restructured to /-Ða/ in all stems

   

At this point, in long stems umlaut was already being implemented by a morphological generalization imposing a front vowel throughout the paradigms; indeed, this was the only transparent way, following syncope of the umlaut trigger, to generate vowel fronting in the syncopated forms. However, with syncope extending to short stems and the consequent restructuring of /-iÐa/ to /-Ða/—umlaut still being a phonetically active process, too—the motivation for retaining umlaut in ⁺telÐa, now from /tal+Ða/, simply disappeared. As a result, /tal+Ða/ came to surface as tal-Ða, much in the same manner as /staÐ+r/ came to surface as staÐr despite earlier pronunciation as ⁺steÐr.

The consequence of these developments is that short stem class I weak verbs do not show umlaut in preterit indicative forms, but all other forms of verbs in this class do. This is the “crazy,” morphologized generalization in 9 that speakers in the literary period appear to have arrived at and to have followed, and the steps outlined here chart the path that led their linguistic predecessors to morphologize umlaut this way in the face of its impending phonetic demise. Unlike short stem nouns, then, which generalized to a uniform, unumlauted allomorph throughout the paradigm once the umlaut triggers disappeared, short stem weak verbs emerged with the absence versus presence of i-umlaut serving to reinforce a major morphological distinction between preterits and other forms of the verb. The end state of affairs is illustrated in 14, where umlaut is fully morphologized in the broader manner of 9 and no longer functions phonetically, the effects of suffix syncope having been lexi-calized.

1. (14) Morphologized umlaut per 9 in the literary period

   

The reason we find umlaut throughout the class I weak verbs, except in the preterit indicatives of short stems, is that these -jan verbs etymologically had umlaut everywhere by virtue of their suffix forms. Subsequent syncope in the long stems made otherwise phonetically transparent umlaut opaque, giving rise to a morphological generalization distinguishing short stems, where umlaut was still transparent, from long stems, where umlaut now was imposed on the stem class irrespective of phonetic context. As syncope generalized to short stems, however, the transparency of umlaut there came under challenge as well but was maintained (to begin with, at least) by analogical reversion to the basic vowel wherever the trigger for umlaut was no longer in evidence—most notably, in the preterit indicatives. These formed a sufficiently identifi-able subclass to retain their basic rather than derived character as umlaut itself was progressively losing its phonetic motivation, and resulted in the morphologized remnant of umlaut described in 9 exempting just preterit indicatives in short stems from vowel fronting. The pathway to this state of affairs is thus well laid out as the preferred (if not inexorable) choice, and it makes coherent the traditional claim that phonetically conditioned umlaut is of greater vintage than the various forms of syncope.

6. Umlaut in Denominal Adjectives with -isk/-sk

The basic phenomenon described above is part of a broad pattern. Consider now the asymmetry of adjectives derived with the suffix /-isk/, exemplified by the (umlauted) long stem versus (unumlauted) short stem pair, bernskr < ⁺barn+isk+r ‘childish-masc.nom.sg’ versus danskr < dan+isk+r ‘Danish’ (see Kiparsky 2006). Parallel to the above analysis of -iÐa in the verbal alternations, we infer that the historical suffix -isk was reanalyzed to -sk (its now invariant form), and that this restructuring was driven, to begin with, by the early occurrence of syncope in the long stems.Footnote ³ That is, as elsewhere, umlaut took place first, followed by long stem syncope: barn+isk+r > bern+isk+r > bernskr; dan+isk+r > deniskr. With the trigger vowel now gone in long stems, the suffix separ-ated into two weight-conditioned variants: /-isk/ after short stems and /—sk/ after long stems. Though otherwise still phonetically conditioned, umlaut came to be morphologized (parallel to the restrictions that arose in the verbal system, see 9) so as to apply to syncopated denominal long stem adjectives ending in -sk. Of course, the i remaining in short stems continued to trigger umlaut phonetically rather than according to morphological category, but ultimately the suffix generalized to /-sk/ everywhere, thus removing the phonetic basis for umlaut in short stems.

As the morphologization of umlaut did not extend to the other class in these paradigms, the short stems, but rather remained associated with long stems generally (as in the nouns and verbs), the result of the restruc-turing of /-isk/ to /-sk/ was automatic reversion of deniskr > danskr, just as teliÐa > talÐa. The sequenced developments are summarized in 15. The suffix -(i)sk divides into two allomorphs according to stem weight, and umlaut partially morphologizes (“Umlaut denominal long stem adjectives ending in -sk”), Stage II. After generalizing to short stems, the effects of syncope are lexicalized, causing the historical -isk suffix to restructure to /-sk/ throughout, precipitating reversion of umlaut in short stems, Stage III.

1. (15) Effects of umlaut and syncope in denominal adjectives with -(i)sk

   

Presumably, umlaut continued to function elsewhere as a phonetically conditioned phenomenon for a period; but with the inherited, oddly restricted variant in 15 still at play in the language—a relict of once transparent phonetic processes associated with stem weight but now identified also with morpholexical class—umlaut remained under morphological determination in long stems like bernskr < barn+sk+r, parallel to dœmda < dom+Ða.

This view finds ready support in Noreen (1970:138–139), who writes: “Wo innerhalb eines paradigmas synkopierte und unsynkopierte formen … mit einander wechselten, ist oft ausgleichung—gewöhnlich zu gunsten der synkopierten formen.” (Where, within a paradigm, synco-pated and unsyncopated forms alternated, there is often leveling—usually in favor of syncopated forms.) He exemplifies this point with “danskr statt ^*deneskr nach pl. dansker dänisch.” In other words, Noreen indicates that syncopated forms tended to generalize, reducing the num-ber of surface forms of the suffix as -isk and increasing the numbers showing -sk. Moreover, as a process in derivational rather than inflectional morphology, resolution of the alternation is promoted by increasing distance from the base, sometimes augmented by later developments serving to reduce unwieldy consonant clusters created by the syncope, as in bernskr > berskr (Noreen 1970:214).

7. Umlaut in Nouns of the ketill/katlar Type

Consider now umlaut in a set of nouns containing the derivational suffix -ill (also treated in Kiparsky 2006) that show syncope and no umlaut in some cells of the paradigm (katlar ‘kettles’) but the reverse pattern in others (ketill ‘kettle.nom.sg’). Even the handbooks show considerable complexity, such as Noreen (1970:252, emphasis added):

Even Noreen, in fact, significantly understates how common umlaut is throughout whole paradigms in nouns of this type. The online Old Norse dictionary ONP shows numerous examples of just this type where umlaut is found across the board, like these from the first few letters of the alphabet.Footnote ⁴ In 16, Old Norse nouns derived in -ill show umlaut across the full paradigm.

1. (16)

Indeed, these examples show umlaut, variably or categorically, either in the whole paradigm or in all singulars (that is, including the dative singular). In 17, we illustrate the latter with the entries from Cleasby-Vigfusson 1957, de Vries 1977, and Fritzner 1954. Nonumlauting plurals derived from umlauting singulars are compared to the forms attested in standard reference works.

1. (17) ketill ‘kettle’. A dative singular form ketli is reported in Fritzner, plus occasional late dialectal forms like kætslar (Noreen 1970:252). (Variable)

   fetill ‘strap’. A dative singular form fetli, plural fetlar. Cleasby-Vigfusson does give a problematic dative singular fatla. (Unexpected)

   depill ‘spot’. Cleasby-Vigfusson shows dative singular depli; Fritzner has dative plural deplinum. (Unexpected)

   snepill ‘snip, flap’ eyrasnepill. Fritzner gives a plural form eyrnasneplar. (Unexpected)

   hefill ‘clew-line, bunt-line (nautical)’. Fritzner shows umlauted plurals like heflunum, and Cleasby-Vigfusson as well. (Unexpected)

   lykill ‘key’. Fritzner gives a string of citations with umlauted plurals, for example, hér eru lyklar. Cleasby-Vigfusson gives dative singular with umlaut, while plural without. (Variable)

   tygill ‘string’. Fritzner gives tugla as plural; Cleasby implies dative singular tygli. (Leveling in singular)

   trygill ‘small bowl’. There are no comments or relevant examples in Cleasby-Vigfusson, Fritzner, or de Vries. (Inconclusive)

We have seen above that earlier syncope in some paradigms interacted with still-live umlaut to create these patterns. Ketill-type nouns could have fallen into the same pattern once it was well established.

Crucially, other factors contributed significantly to variability in this class: Old Norse had not only the nominalizing (sometimes diminu-tivizing) suffix -il but also variants in -al, -ul, and -el, which naturally did not trigger i-umlaut (though the middle variant did trigger u-umlaut). Note, for instance, that Noreen consistently gives the citation forms of these nouns with -ell. Even without prosodic reduction of unstressed vowels (which are largely maintained to the present in Faroese and Ice-landic), confusion arose with respect to their distribution in particular words, driving speakers to make generalizations on the basis of the individual lexeme—as probably happened with Open Syllable Lengthening in English (see Iverson & Salmons 2004). A search of nouns in ONP ending in the relevant forms delivers items ending in the string -Vll as follows:

1. (18)

Since not all nouns ending in this string do necessarily contain this suffix, this count is crude and imprecise, but it suggests that fewer than half of -Vll forms probably contained an i-umlaut trigger, namely 42.8% or 556 of 1,299.

This already provides an answer to why a residue of additional words might show no umlaut anywhere in the paradigm: They had or were reinterpreted as having one of the other suffixal variants, -al, -ul, -el. Even in Old Norse we find evidence for suffix variation: drasill is given alongside the form drosull (de Vries and Cleasby-Vigfusson) and the same holds with vaÐill/vöÐull. For the last, Noreen gives veÐill in parentheses. That word, then, shows three different stem vowels: Histori-cal /a/, i-umlauted e, and u-umlauted ö.

For a word like stuÐill, contemporary languages provide the evi-dence: Modern Icelandic and Faroese both reflect the historical -ul form, see Icelandic stuÐull ‘prop’, etc. It is, of course, possible that the couple of remaining words analogized idiosyncratically to derivationally related words, such as skutill to skutr ‘shot’ or svaÐill to svaÐ ‘slippery place’, though we hesitate to make anything more of this than its sporadic character suggests. (See Schulte 1998 on possible analogical models for restructuring of individual lexical items.)

In short, closer scrutiny of attested data reveals a picture quite consistent with our account. If our scenario is right, umlaut should have left traces; ketill, according to Noreen the one word that holds strictly to this pattern, attests some variation in Old Norse and later. Ketill is the last lexical remnant of phonological umlaut, parallel to Modern Icelandic u-umlaut. More broadly, learners and speakers would not have heard consistent umlauted vowels across this class of words.

8. Evidence from Exceptional Forms

If phonetic umlaut once existed across forms with the relevant triggering segments, we should expect to find scattered evidence of its phonolo-gization in some short stem forms. In this section, we present such evidence and argue that it does indeed reflect retention of earlier umlaut and is unlikely to represent a later innovation. We have shown above that variability in the presence of umlaut is often rooted in morphological variability, as in the suffixal variation just described. Lahiri (2000:120) and others note that Old Gutnish shows forms like steþr, related to the iconic nonumlauting short stem lexeme, and scattered place names re-flect an umlauted form as well, like städe (see Andersson 2002:299). Lahiri is right in her view that such forms could be plausibly seen as innovations.

Wadstein 1892, a work apparently since forgotten in the literature, reviews a considerable body of short stem words in Old Norse containing i-umlaut. These are often forms in which analogy could or would not have happened, such as noninflecting forms and Latin loanwords, as in 19.

1. (19)

This securely indicates the phonetic and then phonological process of i-umlaut, as there is no plausible explanation for how a regular sound change could appear widely in some categories, like long stems, and some phonetically and phonologically short stem forms, but not have once existed more broadly, like steþr.

Yet more relevant here is umlaut in mass nouns of the short stem masculine i class, as in 20 (Wadstein 1892:414).

1. (20)

The importance of these cases is that they confirm the earlier presence of umlaut in short stem masculine i-stem nouns, but precisely where morphological processes would not have removed such forms, since, as mass nouns, the examples in 20 had no plurals. It is unclear to us how analogy might have motivated umlaut here.