1. INTRODUCTION

1.1.The replacement of one morphological pattern by another in historical development often follows an indirect route and as such is a provocative problem in the study of language change. A primary example of this is found in the Germanic language family. In the Germanic parent language there was a class of reduplicating verbs, i.e. verbs using REDUPLICATION for indicating the preterite: present = root ∿ preterite = reduplicative element + root. This principle was preserved in Gothic. In North and West Germanic, however, it was replaced by different new formations, especially one applying the well-known ablaut principle. The modalities of the development from reduplication to ablaut have been, since Grimm, a subject of hot debate, with none of the numerous attempts to solve the problem receiving general acceptance.

Although we now concur with Bech (1969) in explaining the development from reduplication to ablaut as a result of morphological REANALYSIS of the reduplicated form, we crucially disagree with him about the form of this reanalysis. Unlike Bech we consider the full implications of the Germanic accent shift for this reanalysis. Furthermore, while he views the different new formations resulting from the reanalysis mainly as successive stages in the development from reduplication to ablaut, we see them as parallel and competing with one another, with the structurally most adequate formation, one based on ablaut, being ultimately the only one to survive. The development of the Germanic reduplicating class offers then a prototypical case of reanalysis and competition in morphological change.

1.2.Germanic had a relatively small class of mainly nonablauting verbs that used the reduplication principle for indicating the preterite, e.g. present-infinitive *Xait- ∿ preterite *Xexait-, *laik- ∿ *lelaik-.¹

1.2.1.This principle was preserved in East Germanic, namely Gothic, e.g. present-infinitive haitan ∿ preterite haihait, aukan ∿ aiauk, fallan ∿ faifall, slepan ∿ saislep, flokan ∿ faiflok; in some cases the verb was also ablauting, letan ∿ lailot.

1.2.2.In the earliest documented stages of the North and West Germanic languages, however, the reduplication principle appears to have deteriorated, and to have been replaced by different mechanisms.

(i)A so-called r preterite formation, a formation with internal r, is documented clearly in Old Norse (ON), Old High German (OHG), and hypothetically in Old English (OE).

(𝛼)In ON it occurred exclusively in verba pura, i.e. in verbs with vowel final roots, e.g. rōa ∿ rera, sā ∿ sera, grōa ∿ grera, snūa ∿ snera. Such verbs were limited in number. In a case like sā ∿ sera, when compared to Gothic saian ∿ saiso, the original reduplication principle is easily recognizable. Considering its productivity - (e) ra was in such forms obviously felt as a suffixal formation, in which the conjugation endings, a, etc., were those of the weak preterite (singular). As a suffixation the formation is somewhat comparable with the dental preterite of the weak conjugation. Yet it was unusual, and as such must have been subject to shift and analogy; cf., e.g. the fact that sā was weak in Old Swedish (Noreen 1904 : 449).

(ß)In other Germanic languages the verba pura were treated differently and were often weak. They could develop or generalize glides y,w between the verbal root and the inflectional endings, and thus acquire a more regular consonant final root type (Van Coetsem 1956:66-9, Rauch 1972). In OHG the r preterite formation is found in one original verbum purum; this, however, developed a glide and became a verb with consonant final root. It is documented only in biruuuis/biruuis ∿ būwan/būan (South- Rhenish Franconian, Otfrid). We should note that the regular preterite form was būta. Otherwise in OHG the verba pura (bluoen gruoen, sāen ) were generally weak (Braune-Mitzka 1959:284).

Besides biruuuis, biruun, and outside the limited group of verba pura, there were some other OHG r preterites, which are documented only in pleruzzun, oapleruzzi ∿ bluozan (normally weak, plōzta, past participle strong, kaplōzan; pi- Alemannic for bl- ), kisorerot (documented only once) ∿ scrōtan; steroz/sterozun ∿ stōzan (cf. also Kögel 1892:500-1, Schatz 1907:152). These forms were Upper German, mainly Alemannic. Since South-Rhenish Franconian (biruuuis, biruun) constitutes a transition area to Alemannic, the OHG r preterites were by and large found in Upper German. The formation is very scantily documented and clearly residual, disappearing in the later development of German. The verbs scrōtan and stōzan regularly had the new ablaut pattern as a preterite formation [see (iii) below], while bluozan was normally weak (Braune-Mitzka 1959:284).

(ii)OE had some forms in which the reduction of the second, nonprominent syllable resulted in the disappearance of that syllable. These forms generally appear to be the outcome of CONTRACTION of the original reduplicative element and the verbal root. They were Anglian [cf., however, heht under (𝛼) below] and of restricted usage (e.g. poetic). Co-occurring with this formation were forms showing a new ablaut formation [see (iii) below], the latter being the regular formation in West Saxon, and also weak forms (cf., e.g. Sievers-Brunner 1951:335-40). The shift to the weak conjugation became more pronounced in the subsequent development of English. By the time of Middle English the contracted forms had disappeared.

(𝛼)Some cases are fairly obvious examples of contraction: heht (occurring also in West Saxon texts, beside hēt ) ∿ hātan, leolc ∿ lācan, beoftun ∿ bēatan and speoft/speaft, speoftun, probably belonging to spātan.

(ß)Similar cases have generally, though not always, been viewed as contracted forms as well: reord/hreordun (the latter as mentioned in Sievers-Brunner 1951:336, beside rēdon and some weak forms) ∿ rēdan, West Saxon rœdan/ ondreord/ondreard (in West Saxon also weak, ondrœdde ) ∿ ondrēdan, West Saxon ondroedan, and leort/forleortun (beside lēt ) ∿ lētan, West Saxon lœtan. The form leort is ordinarily assumed to result from *leolt with dissimilation. If leort is considered an analogical formation after, e.g. reord, though, we could regard these forms as r preterites in Old English (Anglian). Actually, Bech (1969:23) considers such forms r preterites, but he does so in a perspective quite different from ours.

(iii)There is also a formation showing a new ablaut pattern, whose origin remains unexplained. It became the dominant usage throughout the North and West Germanic language area, and is the so-called NORMAL TYPE (Bech 1969: e.g. 15). The original, basic form of the new ablaut pattern has been generally reconstructed as follows (cf., e.g. Hirt 1932:144) :

The following examples from OHG, ON, and OE reflect the ORIGINAL, BASIC ablaut pattern: OHG heizan ∿ hiez, lāzan ∿ liez, (h) loufan ∿ liof,(h) ruofan ∿ riof, fallan ∿ fiel, ON heita ∿ hēt, lāta ∿ lēt, hlaupa ∿ hljōp, falla ∿ fell, OE hātan ∿ hēt, lētan/lœtan ∿ lēt, hlēapan ∿ hlēop, hrōpan ∿ hrēop, feallan ∿ fēoll/feoll.

The above-mentioned pattern is called BASIC because certain phonic variants and forms of a more incidental nature may have been present alongside it from the start. It is also called ORIGINAL because it was subsequently affected by changes, shifts, or analogical extensions (for more details, cf., e.g. Van Coetsem 1956:56-69 and 1963). Since minor aspects of these distinctions are subject to discussion, the reconstruction of the original basic form of the new ablaut formation itself may be to a certain degree debatable. In general a combination of criteria (time, degree of expansion, and nature of the occurrence of the form) allows us to establish what is original and what is not.

(iv)An example of a much more sporadic formation is seen in cases where the forms of the present and the preterite were completely identical (no reduplication, ablaut, or other differentiating factor), e.g. ON heita ∿ heit, sveipa ∿ sveip, OE gangan ∿ gang.

1.3.Thus, the three most important new preterite formations in the North and West Germanic reduplicating class were the v preterites, the OE contracted forms (including possible examples of the r preterite formation), and the new ablaut formation. In areas where they coexisted, they were in competition. Yet the r preterites and the OE contracted forms were moribund phenomena. They appear to be peripheral in the North and West Germanic area,² and to exhibit only residual and restricted usages next to the prevalent new ablaut formation, which was used virtually throughout the whole area. Outside the reduplicated class, the weak conjugation appears to be the fourth competitor. In later development it became more and more influential, and in certain cases it totally supplanted the new ablaut formation.

1.4.In the present study we hope to provide at least an acceptable basis for explaining the development, occurrence, and interrelationship of these new preterite formations, especially the new ablaut formation. For this we will follow the general language evolution step by step, considering the chronological order of the changes involved and their respective impact on the development of the reduplicating class. In conjunction with this we will carefully examine the different formations, paying equal attention to each of them; in this we basically follow an approach advocated by Bech (1969: e.g. 15, 54). These formations constitute precisely the main focus of our study, not the changes, shifts, or analogical extensions which reduplicating verbs have undergone in the subsequent development of the separate Germanic languages.

1.5.We will view and discuss in greater detail in our further treatment the development of the Germanic reduplicating class. The following, however, may serve as a general outline.

As a result of changes, of which the accent shift was the most determinant, the reduplication principle became increasingly weakened and obscured. In Gothic the development was REVERSED, that is, the reduplication principle was preserved or reinstated. In North and West Germanic, speakers may occasionally have resorted to the same procedure, but in general, REANALYSIS WAS APPLIED to the reduplicated form. In this reanalysis the reduplicative element, e.g. *le- in *lelaik-, was reinterpreted as the initial part of the verbal root, while the initial part of the verbal root, as -la- in *lelaik-, was felt not to belong to the root any more, although it did not necessarily or immediately acquire affixai (morphemic) status. To indicate this, we will use the term SPECIFIC NONROOT MATERIAL; it is specific insofar as it is differentiated from other nonroot elements such as affixes.³ Bech (1969: 9) does not make such a distinction and considers the reanalysis in question a direct change from a reduplicated or prefixed form to an infixed form (the latter becoming suffixed in the case of the ON verba pura).⁴

In spite of certain lexical and distributional readjustments that seem to have affected the reanalyzed preterite form over the whole of the North and West Germanic area, this form remained structurally inadequate due to the combination of two mutually supportive factors. These were its generally BISYLLABIC ROOT STRUCTURE and the SPECIFIC NONROOT MATERIAL. They were extremely unusual features within the paradigm of the strong verb; their deviant character is readily apparent when the reanalyzed preterite form is compared with the unmarked present (or nonpast), whose monosyllabic root structure was representative of the general root type of the strong verb.⁵ Yet, through the reanalysis the constant e of the original reduplicated form (i.e. e as in *lelaik- ) was brought within the bounds of the verbal root, yielding a new ablaut pattern, present (a in *laik- ) ∿ preterite (e in *lelaik) .Ablaut was precisely the dominant principle for differentiating principal parts within the paradigm of the strong verb. Through further corrective or “remedial” changes (Malkiel 1969:26), PARALLELING and COMPETING formations developed. In this way the r preterites, the OE contracted forms, and the formation showing the new ablaut pattern arose. While the r preterites analogically generalized one form of the specific nonroot material, namely r, the OE contracted forms, preserving the specific non-root material in its different forms, reduced the original bisyllabic preterite to a monosyllabic structure. On the other hand, the new ablaut formation resulted from a consistent structural re-formation of the reanalyzed preterite form based on the model of the present; the new preterite form evolving this way was monosyllabic and did not contain the specific nonroot material.

1.6.The three formations reflect to different degrees a well-known basic principle of language development aiming at structural generalization and adequacy. This principle promotes generalization, the PRESERVATION of what is structurally relevant or motivated, and the LOSS or REJECTION of what is structurally irrelevant or unmotivated. The operation of this principle is, however, relativized by the indeterminate and gradual nature of such notions as RELEVANCE or MOTIVATEDNESS. It may also be counteracted by a number of things, such as the high frequency of the lexical items or patterns involved. It goes without saying that the degree of, e.g. relevance or motivatedness of a given element may vary from one developmental stage to another.

Preservation and rejection reveal themselves in certain circumstances as “defense mechanisms” of the language. They are directed at maintaining, as much as possible, the existing structural properties of the language, in particular its distributional characteristics and root structure. It is well known that when a language is (suddenly) confronted with deviations from its structural properties, as may be the case with borrowing or reanalysis, it will tend to eliminate or adapt these deviations (the notion of DEVIATION being again a relative and gradual one). Uhlenbeck’s work (1966 [1949]) on the Javanese morpheme is in this respect illustrative.

In the reanalyzed preterite form the generally bisyllabic root structure and the specific nonroot material were both structurally deviant and deficient in structural relevance or motivatedness, ablaut being the primarily relevant principle for differentiating, e.g. the preterite from the present within the paradigm of the strong verb. This made the bisyllabic root structure and the specific nonroot material susceptible to loss. It is a significant fact that the above-mentioned basic principle was only partly effected in the r preterites and the OE contracted forms (the bisyllabic root structure and/or. the specific nonroot material being preserved), but was consistently carried through in the new ablaut formation (the bisyllabic root structure and the specific nonroot material being lost). This difference underscores the fact that the r preterites and the OE contracted forms were, at the time of documentation, residual and peripheral, while the new ablaut formation became the dominant usage, being precisely THE MOST COMPATIBLE WITH THE STRUCTURE AND DEVELOPMENT OF THE LANGUAGE.

1.7.The development of the reduplicating class in North and West Germanic thus offers a highly illustrative case of competition in the realm of morphological change, with the structurally most adequate formation being ultimately the only one preserved. In such a perspective one cannot but compare with a TRIAL AND ERROR procedure. The notion of COMPETITION has long been recognized in various forms of language change. With reference to sound change, the term was coined by Wang (1969) together with a theoretical discussion of the notion; illustrations with a discussion of the phenomenon are found earlier, e.g. in Van Wijk (1936:70), Van Coetsem (1956:38).

1.8.The above also shows that we have modified our earlier stand on the topic (a stand that employed the notion of inverted analogy), as first set forth in our study of 1956. We will discuss this further in Appendix I. In that earlier work we reviewed rather extensively previous research. In the present treatment (especially Appendix II) we take particular note of more recent work, primarily Bech’s (1969), which, although unsatisfactory as a whole, may be credited with some useful ideas.

2. THE WEAKENING OF THE REDUPLICATION PRINCIPLE

2.1.The reduplication principle in the Germanic reduplicating class is of Indo-European origin. Like ablaut in other verbs it distinguished verbal morphological categories. Three changes in the development of the Germanic parent language, each acting in a different way and with different impact,weakened the reduplication principle. These were the ā̆-ō̆ merger, the accent shift, and Verner’s Law. The last two changes were closely related, in that Verner’s Law may be considered a segmental reflex of the accent shift.

2.2.The ā̆-ō̆ merger appears to have chronologically preceded the accent shift, or at least the latter’s completion. The following can support this assumption.

2.2.1.First, there may be system-internal evidence. Raising and lowering changes (umlaut- and consonant-conditioned changes) seem to have operated with or directly after the accent shift (Van Coetsem 1956: e.g. 38). The ā̆-ō̆ merger counterfed one of these changes; specifically, the ā̆-ō̆ merger (short vowels) preceded u > o. We have then the following diachronic order:

If the changes had occurred in the reverse order, namely,

the resulting a would show reflexes of Indo-European u, which is not the case. Of course, such argumentation is not completely watertight, because the o’s involved in the two changes may have been of different quality and status.

2.2.2.Second, there is system-external evidence, which, although not compelling, has confirming power. It concerns the areal-linguistic fact that the ā̆-ō̆ merger shows a wider spread within the Indo-European language family (Germanic, Slavic, Baltic, etc.) than the accent shift, the latter being a shift of both the nature and the place of the accent (Germanic and Old Irish). Thus we can assume that the ā̆-ō̆ merger preceded both the accent shift and Verner’s Law.

2.3.We will now discuss the effects of the three changes on the reduplication class.

2.3.1.The ā̆-ō̆ merger introduced a paradigmatic rearrangement of verbs into a so-called a group, based on the vocalism of the present. Aside from the reduplicating class [ *Xait-, *Xlaup-, *fall-; cf. OHG heizan, (h) loufan, fallan) this a group then contained the traditional 6th ablaut series (*far-, *grab-;⁶ cf. OHG faran, graban). Although its constituents are clearly of Indo-European origin, the a group is a Germanic formation. It tended to organize itself along the lines of the older e group, which had e as vocalism of the present. It contained the traditional ablaut series 1 to 5 (*reid-, *leug-, *Xelp-, *stel-, *geb- ; cf. OHG ritan, liogan, helfan, stelan, geban). Each of the two groups had specific structural properties. In addition to the e and a groups, there were a number of ē and ō (originally å) verbs (*slēp-, *lēt-, *Xwōp-/*Xwåp-; cf. OHG slāfan, lāzan, Gothic hwopan), with some of the ē verbs being not only reduplicating, but also ablauting (ē ∿ ō) (for more details, see Van Coetsem 1980). Within this overall, cohesive system of the so-called strong verbs, reduplication would clearly constitute a minority principle and be therefore strongly subject to pressure from ablaut.

2.3.2.The accent shift contributed more directly, and with far more impact than the ā̆-ō̆ merger, to the weakening process of the reduplication principle. It led to the obscuring and even disruption of that principle. The accent shift involved two changes, one of NATURE and one of PLACE. The change of nature concerns a change from a NONDOMINATING to a DOMINATING PROMINENCE (Van Coetsem-Hendricks-McCormick 1981), which is traditionally but improperly called a change from a PITCH to a STRESS ACCENT. The change of place refers to a change from a FREE to a FIXED PROMINENCE, namely an initial or root syllable prominence.

(i)The sound change referred to as Verner’s Law, i.e. the voicing of the Germanic fricatives f, ƥ, X, s to b, d, g, z [Ѣ, đ, ǥ, z], under accent conditioning, presupposes only a change of nature and not a change of place, since Verner’s Law is conditioned by an Indo-European accent placement. For example, the alternation ƥ ∿ d [-đ-] for Indo-European t, under accent conditioning, is shown by Gothic broƥar [-ƥ-] ∿ fadar [-đ-] as compared to Greek ɸρᾱτηρ/ Sanskrit bhrātar- vs. Greek πατήρ, Sanskrit pitár-.

It is worth noting that Verner himself had already come to that conclusion concerning the order of the changes of nature and place. This is clearly apparent in his own words: “Erst nachdem sich das germanische von seinem nächsten verwandten, dem slavo-litauischen geschieden und sein sonderleben angefangen hatte, treffen wir den accent in seinem wesen etwas verändert: er war expiratorisch geworden oder vielleicht, da er wohl noch seinem chromatischen character behielt, chromatisch-expiratorisch. Aber die zweite characteristische eigenschaft, die frei-heit, hatte die urgermanische accentuation in wunderbarer Vollständigkeit behauptet. Der dann folgende übergang zur gebundenen accentuation (wurzelbetonung) ist eine gründlich durchgeführte analogiebildung.” (Rooth 1974:28).

While the present form of a reduplicating verb has one anlaut, the corresponding reduplicated form of the preterite may be analyzed as having two. Here we will follow Bech (1969:4) in calling the anlaut of the reduplicated form the ABSOLUTE ANLAUT and the anlaut of the verbal root, as also found in the present, the THEMATIC ANLAUT. For example, in the preterite *XeXlaup-, the absolute anlaut is X while the thematic anlaut is Xl, also found in the present *Xlaup-.

The reduplication principle derives the absolute anlaut from the thematic anlaut by specifying that: (1) Before vowels the reduplicative element consists of the vowel e (Gothic ai, e.g. *auk- ∿ *eauk-; cf. Gothic aukan ∿ aiauk). (2) Before one or more consonants of the thematic anlaut the reduplicative element consists of one consonant (the first) + e (e.g. *Xait- ∿ *XeXait-, *frais- ∿ * fe frais ; cf. Gothic haitan ∿ haihait, fraisan ∿ faifrais). However, when the thematic anlaut is st or sk (sp not being documented) the reduplicative element consists of this cluster + e (e.g. *staut- ∿ *stestaut-; cf. Gothic stauten ∿ staistaut). In Gothic, hw (˂ƕ˃) is treated as one consonant (monophonematic) (e.g. hwopan ∿ hwaihwop). It is important to note that the vowel e is the ONLY CONSTANT in the reduplicative element.

As we have seen, Verner’s Law emerged with or after the change of nature of the accent. That it also was applied to the thematic anlaut of the reduplicated forms can hardly be doubted and has long been recognized (cf. e.g. Streitberg 1896:328). This implies that the accent was on the root and not on the reduplicative element. (Perhaps we cannot exclude the possibility that the application of Verner’s Law was already in the reduplicated forms of the Germanic parent language somehow restricted or even levelled at a later stage.) Witnesses to the application of Verner’s Law are the various r preterites (e.g. ON sā ∿ sera) and possibly Gothic saizlep of slepan. The form with -zl- does not occur consistently in the Gothic corpus, which has gasaizlep, gasaizlepun, but also saislep, anasaislep, anasaislepun, each of them documented once; Gothic also does not show *saizo, comparable to ON sera, but only saiso, saisost (Streitberg 1965 :lexicon, De Tollenaere-Jones 1976) . This is not so surprising in light of the overall restricted representation of Verner’s Law in Gothic, a fact which is usually ascribed to levelling (Van Coetsem-Hendricks-Siegel 1981:172-3).

Insofar as Verner’s Law was applied, the thematic an-laut became differentiated from the absolute anlaut, e.g. *Xait- ∿ *Xegait- [-ǥ-] (X ∿ X/g for earlier X ∿ X/X), *Xlaup- ∿ * Xeglaup- [-ǥl-] (X ∿ X/g for earlier X ∿ X/Xl), *fall-*feball- [-Ѣ-] (f ∿ f/b for earlier f ∿ f/f) . However, at this point, the voicing alternation was purely phonologically (suprasegmentally) conditioned and, as such, automatic. In other words, the underlying representation for X ∿ g, XI ∿ gl and f ∿ b must originally have been X, XI, and f, respectively (e.g. *Xegait- was thus a phonetic realization of *XeXait-).

Assuming that Verner’s Law was consistently carried through in the reduplicated forms of the Germanic parent language, Bech (1969:6) submits the following (here slightly adapted) diagram (Table 1). It shows the verb types that were subject to the alternations of Verner’s Law (B.I-II) and those that were not (A.I-II). The diagram also illustrates how in the four verb types, the absolute anlaut and the thematic anlaut differed from one another. The rather rare verb type with a vowel initial root (cf. Gothic aikan ∿ aiaik, aukan ∿ aiauk) has not been included in the diagram; if necessary, it could be listed in A.I.

The application of Verner’s Law to the reduplicated forms suggests that their thematic anlaut was not felt as a word anlaut.⁷ This is further supported by the fact that in the verbal pattern ‘reduplicative element + verbal root’ the link between the two constituents was definitely stronger than in the general pattern ‘prefix + verbal root’. For this we can refer to, e.g. Gothic afaiaik ∿ afaikan, gastaistald ∿ gastaldan, in which af- and ga- did not separate the reduplicative element from the verbal root.

(ii)The change in accent placement, particularly the concentration of the accent on the reduplicative element, is clearly evidenced by such cases as OE heht of hātan, leolc of lācan, and ON sera of sā. It has two further consequences:

(α)The alternations produced by Verner’s Law were no longer phonologically (suprasegmentally) conditioned. This caused a change in underlying representation.

(β)In general, the verbal compound ‘prefix + root’ had the prominence on the root (cf. Gothic fra’letan), while the corresponding compound had the prominence on the prefix (cf. Gothic ’fralets) . The best interpretation of this is that in verbs the prefix was still unbound, while in nominalization it was really compounded. We can then consistently speak of an initial accentuation, with the implication, however, that in verbs the initial accentuation was also a root accentuation. In other words, the verb was always root-accented. A deviation from this principle arose as a result of the concentration of the prominence on the reduplicative element. This accentuation was related to the closeness in composition of the pattern ‘reduplicative element + verbal root’, as stated above. In this blending process the status of thematic anlaut became obscured. Such a situation was subject to regularization, and was in fact regularized through reanalysis, as we will discuss in Section 3. Regarding this we agree fully with the strong emphasis that Meid (1971: 97) places on the determining nature of the accent shift, namely a shift “vom beweglichen zum festen, an die Anfangssilbe des Wortes gebundenen Akzent.”⁸

(iii)In conclusion, the change in accent placement was the primary determinant in the blending process of the reduplicative element and the verbal root. After the change in accent placement, the results of the earlier operation of Verner’s Law became contributing factors to this process.

3. COUNTERDEVELOPMENTS TO THE WEAKENING OF THE REDUPLICATION PRINCIPLE, ESPECIALLY THE REANALYSIS OF THE REDUPLICATED FORM WITH THE EMERGENCE OF A NEW ABLAUT PATTERN

3.1.After the ā̆-ō̆ merger, the accent shift and the alternations produced by Verner’s Law were in their cumulative effect particularly efficient in weakening, obscuring, and even disrupting the reduplication principle. There appear to have been two counterdevelopments to this weakening process, developments which were complementary and which operated in opposite directions. They were ultimately realized in two different areas of the Germanic language community. The two developments have been recognized by Bech (1969:9 ff.).

3.2.In Gothic, the alternations produced by Verner’s Law were systematically levelled, but the reduplication principle was preserved or reinstated. In other words, by ruling out factors that contributed to the weakening of the reduplication principle, a REVERSAL occurred in Gothic. For example, for *Xegait- and *feball- we find accordingly in Gothic haihait and faifall.⁹

3.3.On the other hand, in North and West Germanic the alternations produced by Verner’s Law were generally maintained, but the reduplication principle was given up. It is, however, quite possible that a reversal procedure similar to that in Gothic had occasionally been utilized. Yet, REANALYSIS seems to have been the more general procedure. We realize that the different possibilities regarding Verner’s Law may have resulted in co-occurring forms like *XeXait-/*Xegait-, where the first might be considered not as good a candidate for reanalysis as the second. It is crucial to keep in mind, however, that the concentration of the accent on the reduplicative element was the overriding factor in reanalysis, in which case both forms were good candidates. In view of this, rather than referring always to, e.g. * XeXait-/*Xegait-, the form reflecting Verner’s Law, e.g. * Xegait-, will be used to refer generally to both possibilities.

3.3.1.While the Germanic verb was root-accented, the reduplicative element in North and West Germanic was accentually prominent. The pattern ‘reduplicative element + verbal root’, then, could be reinterpreted or reanalyzed in such a way that the reduplicative element became the initial part of the verbal root; the unusual nature of reduplication in the Germanic verb was undoubtedly a contributing factor. It is in any case not uncommon for a reduplicative element to be reinterpreted as a root (cf. German beben, zittern and see also Henzen 1957:258-60 [173]). This reanalysis brought the original reduplicated form in line with the general accentuation pattern of the verb. However, with the reanalysis, the preterite form generally acquired a BISYLLABIC ROOT STRUCTURE, e.g. *Xegait- (∿ present *Xait-), and along with this, the so-called SPECIFIC NONROOT MATERIAL, which continued the initial part of the verbal root of the earlier reduplicated form, e.g. -ga-. In addition, in a number of cases a lexical anlaut discrepancy between the present and the preterite form resulted, e.g. *Xl ∿ *X in *Xlaup- ∿ *Xeglaup-.

The yardstick for establishing what belonged to the root and what was the specific nonroot material, was the root of the present. Not only was the present the unmarked tense (the nonpast), but its monosyllabic root structure was compatible with the general root type of the strong verb in Germanic. Therefore, there can be no doubt about the deviant character of the generally bisyllabic root structure of the reanalyzed preterite form, and about the nonroot status of the specific nonroot material. Also, the form of the present must have served as the standard in the case of lexical anlaut discrepancy.

The prominent part of the reanalyzed preterite form, e.g. *Xe- in *Xegait- was, then, quite normally identified or equated with the initial part of the root; *Xe- of the reanalyzed preterite form corresponded to *Xa- of the present *Xait-, yielding a new ablaut pattern, present a ∿ preterite e, and thus bringing the form in line with the dominant principle for differentiating principal parts within the paradigm of the strong verb.

The second, nonprominent part of the reanalyzed preterite form comprised the specific nonroot material and the final part of the root. The distinction between the root and the specific nonroot material was determined by the root structure of the present, so that, e.g. -gait- in *Xegait- must ideally have been reanalyzed as -ga-, specific nonroot material, and -it-, final part of the root; -it- of the reanalyzed form corresponded then to -it- of the present form * Xait-.¹⁰ However, the vocalism of the second, nonprominent part of the root must have been directly subject to reduction, as we will discuss in Section 4.4.

The difference in morphological analysis between the reduplication and the subsequent reanalysis can be schematically illustrated by the following examples (Table 2), which also show the different aspects of the emerging new ablaut pattern.

3.3.2.In the reanalysis verba pura came to be analyzed as suffixed. In, e.g. *sezō-, *se- was then the initial part of the verbal root with e as an ablaut alternant, z a suffix, while ō became identified with a conjugation ending, e.g. ON sā ∿ sera [cf. 1.2.2 (i) (α)].

3.3.3.Bech (1969:9 ff.) identifies the pattern ‘reduplicative element + verbal root’, e.g. *Xe + gait-, with the pattern ‘prefix + verbal root’. He does not see the reanalysis as *Xe + ga + it-, like we do, but as *X + eg + ait- (his notation being *h + eg + ait-), i.e. *X-ait-, verbal root, and -eg-, infix. In his view the change is basically one of prefixation to infixation (or suffixation).

Bech’s conception of the reanalysis ignores the full implications of the accent shift, and is therefore insufficiently founded and unsatisfactory. Indeed, in his view, the prominence continues to be on the affix and not on the verbal root, and his analysis does not take into consideration the discrepancy or differentiation within the verbal accentuation pattern. Our conception of the reanalysis, however, does take into consideration the full implications of the accent shift. At the same time it accounts for the introduction of the constant e of the original reduplicative element into the bounds of the verbal root. The divergent conceptions of the reanalysis are the source for crucial differences between Bech’s view and ours.

(1) Reduplication

(2) Reanalysis

TABLE 2.

4. STRUCTURAL AND DEVELOPMENTAL IMPLICATIONS OF THE REANALYSIS

4.1.As stated, the general bisyllabic root structure and the specific nonroot material as combined characteristics of the reanalyzed preterite form were structurally deviant; in the presence of ablaut they could not be the primarily relevant markers of the preterite. The lexical anlaut discrepancy between the present and the preterite that resulted from the reanalysis also yielded an unstable situation in a number of cases. Following the general principle of structural generalization and adequacy (Section 1.6), these deviations brought on readjustments or changes, which we will now discuss further.

In coping with these deviations, the reanalyzed preterite form must have undergone certain lexical and distributional readjustments over the whole of the North and West Germanic area. These adjustments, which for the sake of convenience we will call PRIMARY READJUSTMENTS, are found in or presupposed by all the formations described in Section 1. Further readjustments or changes from which these formations emerged seem to have had a more specific character.

In our treatment of these deviations, that is of the structural and developmental implications of the reanalysis, we will consider three points which bear on both the primary readjustments and the further readjustments or changes of the reanalyzed preterite form. Together with these general considerations we will discuss in this section the primary readjustments, while the further readjustments or changes will be studied in Section 5. The three points in question are: (1) the principles available for distinguishing between the present and the preterite, (2) the relation between the absolute and thematic anlauts, and (3) the bisyllabic root structure of the reanalyzed preterite form.

4.2.Our first point concerns the principles available for distinguishing between the present and the preterite.

4.2.1.Prior to the reanalysis there was in general one principle available for distinguishing the preterite from the present, namely reduplication (e.g. preterite *Xegait- ∿ present *Xait-).¹¹ This was true of the a verbs which constituted the greater part of the reduplicating class. Only a few ē verbs of that class exhibited both reduplication and ablaut (e.g. present *lēt- ∿ preterite *lelōt-, Gothic letan ∿ lailot). It is important to note that the reduplicative element contained with its constant e a potential ablaut alternant, i.e., the potential ablaut alternant e and the reduplication constituted an amalgam.

4.2.2.On the other hand, the reanalysis of, e.g. *Xe + gait- to *Xe + ga + it- produced a preterite *Xe-it- vs. a present *Xait-, i.e. an ablaut alternation present a ∿ preterite e (which was only one aspect of the new ablaut pattern), with the specific nonroot material ga. In other words, the reanalyzed preterite form contained both the ablaut alternant e and the specific nonroot material ga (the latter in combination with a bisyllabic root structure to be discussed in Section 4.4), which differentiated it from the present; the ablaut alternant e and the specific nonroot material ga were, however, separate entities. How then did ablaut relate to the specific nonroot material in the reanalyzed preterite form, and what were their respective statuses?

Ablaut was, within the paradigm of the strong verb, the formal principle for the differentiation of principal parts. In other words, given that paradigm it was the principle necessary for the maintenance of the differentiation between, e.g. the present and the preterite, and as such it was structurally primarily relevant and fully motivated. On the other hand, the specific nonroot material was within the same paradigm associated with the preterite only in a minority group of verbs. It was quite susceptible to loss, being at best a concomitant signal of the preterite; at the same time it was in the majority of cases directly instrumental in preserving another deviant characteristic of the reanalyzed preterite form, namely the bisyllabic root structure. Considering this, the potential for loss is understandable. Furthermore, the specific nonroot material exhibited a structurally unmotivated variety of lexical forms (cf. ga/la/sta, etc., in Table 2), which made it lack structural relevance or mo-tivatedness not only in its occurrence but also in its form; this is precisely the original impetus for the emergence of the v preterite formation. Through its productivity the latter formation also illustrates the potential affix status of the specific nonroot material.

4.2.3.We can schematize the above differences between the present and the preterite in the reduplication and in the reanalysis as follows:

1. Reduplication

*Xe -: potential ablaut alternant e and reduplicative element constituted an amalgam

2. Reanalysis

*Xe + ga-: ablaut alternant e (I) primarily relevant and fully motivated, and specific nonroot material ga (II) structurally deviant and deficient in structural relevance or motivatedness, were separate entities

4.3.Our second point examines the relation between the absolute and thematic anlauts in the reduplicated form, and between the absolute anlaut and the specific nonroot material in the reanalyzed preterite form.

4.3.1.In the reduplicated form, e.g. (*XeXait->) *Xegait- and (*XeXlaup->) *Xeglaup-, there was a lexical (phonological) relationship of dependence of the absolute anlaut on the thematic anlaut. In other words, a form such as * feXait- (∿ *Xait-) would have violated the reduplication principle.

4.3.2.With the reanalysis, however, the thematic anlaut of the reduplicated form lost its status as anlaut and became the specific nonroot material. Concomitantly, the absolute anlaut and the specific nonroot material were no longer in a lexical (phonological) relationship to one another. As a result, they were free to change and develop independent of one another. Yet, the absolute anlaut remained lexically (phonologically) dependent on the thematic anlaut as found in the present, and this triggered lexical readjustment of the anlaut of the preterite to that of the present whenever necessary. Compare OHG present bluozan ∿ preterite pleruzzun, that is *blōt- ∿ *blerot-, the latter with analogical spread of the r infix (Section 5.2) and lexical readjustment of the anlaut from *beblōt- to *bleblōt-.

4.3.3.The described anlaut relations, in particular the lexical anlaut readjustment, can be schematically illustrated as follows:

1. Reduplication

2. Reanalysis

4.4.Our third point concerns the fact that the reanalysis produced a generally bisyllabic root structure (cf. OHG bluozan, stdzan ∿ pleruzzun, steroz). Such a bisyllabicity was extremely unusual within the paradigm of the strong verb, and lacked structural relevance or motivatedness in the differentiation between the present and the preterite; it was therefore not only developmentally but also structurally subject to reduction.

4.4.1.Given the prevailing accent type (dominating prominence) and the structural (distributional) requirements of the language, reduction of the second, nonprominent syllable of the reanalyzed preterite form, that is reduction of the vocalism involved, was a quite normal reaction.

Taking the examples *Xegait- (*Xe + ga + it -) and *Xeglaup- (*Xe + gla + up-), the prominence being in each case on the first syllable, the sequences ai and au of the second, nonprominent part of the reanalyzed preterite form were directly subject to reduction on purely distributional grounds (cf. OHG stōzan ∿ steroz). It is, however, not certain in what way the reduction could have been implemented in the case of a diphthong. It may have been achieved by assimilative contraction of the two components (ai > e, au > o: *Xeget-, *Xleglop-, the latter with lexical anlaut readjustment as described in Section 4.3). It may also have been realized by total reduction of one of the components, presumably the first (ai > i, au > u: *Xegit-, *Xleglup-). The resulting reduced vocalism i/e, u/o in the reanalyzed preterite form was then short and nonlow, and at the time regularly subject to umlaut and/or consonant conditioning. It was also comparable to the second component of a diphthong, which in Germanic could behave like a vowel under umlaut and/or consonant conditioning, and is then to be characterized as “vocalic” but “nonsyllabic” (Van Coetsem 1979): *Xait-/*Xaet- ∿ *Xegit-/*Xeget-, * staut-/*staot- ∿ *stestut-/*stestot-. While umlaut and/or consonant conditioning of the second component of a diphthong is clearly evidenced by eu/eo/iu (cf. OHG beotan ∿ biutu/biutis), it is less well or only indirectly attested in the case of ai/ae and au/ao (Van Coetsem 1968, especially 521 ff., and compare Van Coetsem 1975a).

It does not appear valid to view such reduced vocalism, i/e, u/o, as possible reflexes of an originally Indo-European zero grade of the preterite plural, as has been done in earlier research (e.g. Boer 1924:110). Such an idea depends on cases like ON sveip ∿ svipom, which are, however, clearly analogical extensions from the e verbs (cf. Noreen 1904:444 and see Van Coetsem 1956:56, 60-1 with further references). We should also keep in mind that the absence of vocalic alternation opposing singular and plural in the root of the preterite was an important structural property of the a verbs, as opposed to the e verbs, where such a vocalic alternation was the rule (Van Coetsem 1980:290-1, 327).

Not only in a diphthong (e.g. *Xait-), but also in the case of a long vocalism (e.g. *slēp-), the reduction seems to have been of one mora, that is, of one minimal unit of length (cf. OHG bluozan ∿ pleruzzun); in the case of a long vocalism (X̅) as well as in the case of a diphthong (X̆X̆) the reduction would thus again have produced a short and nonlow vocalism (X̆), comparable to what may have happened in a diphthong.

In the case of a short vocalism (e.g. *fall-), the reduction of one mora may have been carried through, producing zero vocalism (*fall- ∿ *fe + b + II-). However, on general distributional grounds the short vocalism could have been maintained as well (*fall- ∿ *fe + b + all-).¹²

4.4.2.There was an obvious difference between the OHG r preterites and the OE contracted forms insofar as the former show partial reduction and the latter total reduction. The partial reduction in OHG would reflect the primary readjustment described in Section 4.4.1.

(i)The OHG forms maintained their bisyllabic root structure: bluozan ∿ pleruzzun, capleruzzi, i.e. uo ∿ u (the latter possibly reflecting umlaut of o before u, i of the following syllable), stozan ∿ steroz, i.e. ō ∿ o, būwan ∿ biruuuis, i.e. ū ∿ u. Because of the similarity, it would seem that the reduced vocalism of the second, nonprominent syllable of such forms was felt as (part of the) root vocalism and not as the specific nonroot material. (This would imply that the reduced vocalism could be readapted depending on change in the root vocalism of the present.) We should also note that the total vocalism of the preterite in each of the examples under consideration was: e-u in pleruzzun, e-o in steroz and i-u in biruuuis; these remind us of the familiar realizations of the Germanic eu diphthong in OHG, namely eu/eo/iu.

(ii)The OE contracted forms displayed total reduction of the vocalism of the second, nonprominent syllable of the original bisyllabic form. The resulting form was monosyllabic (although still containing the specific non-root material): hātan ∿ heht, lācan ∿ leolc, bēatan ∿ beoftun. This is very possibly a secondary development as compared to the partial reduction in OHG discussed in (i) (further on this and related questions in Section 5.3).

(iii)The documented OHG r preterites show back vocalism, Germanic au, ō, ū (*staut-, *blōt-, *bū (w) -; OHG ō (before dental), uo, ū, stozan, bluozan, būwan, būan). There are no examples with Germanic ai, ē vocalism (OHG ei/∿ē, ā). This is probably coincidental. If it is not, it could suggest that the lack of a sharp distinction between the e vocalism in the initial part of the root of the preterite and the reduced vocalism i/e (from ai, ē) in its final part did not favor the preservation of r.¹³ Furthermore, the OE contracted form beoftun ∿ bēatan reflects a case of a Germanic au vocalism (*baut-). It does not, therefore, appear valid to assume with Bech (1969: 32) an earlier differentiating development between front and back vocalism, the front vocalism being completely reduced, the back vocalism, however, only partially.

(iv)In the case of a short vocalism, e.g. *fall- (Section 4.4.1), if there was reduction at all, it was total by the very nature of things, *fe + b + ll-. A distinction between partial and total reduction is then not applicable. In such cases unusual (unacceptable) or unstable distributions or clusters might also have developed, which were subject to change and/or structural readjustment.

4.4.3.As we have observed (Section 3.3.2), verba pura came to be reanalyzed as suffixed, e.g. ON sā ∿ sera. In this case the verbal root was again monosyllabic.

4.5.In conclusion, given the general structural and developmental implications of the reanalysis, there were, we may assume, two primary readjustments of the reanalyzed preterite form, namely the lexical anlaut readjustment applied whenever necessary, and the partial reduction of the second, nonprominent syllable. In spite of these readjustments the reanalyzed preterite form remained deviant insofar as it had a bisyllabic root structure and/or the specific nonroot material. Three different strategies were applied to the (readjusted) reanalyzed preterite form. This produced the three different preterite formations of North and West Germanic.

5. THE SPECIFIC STRATEGIES OF NORTH AND WEST GERMANIC, IN PARTICULAR THE EMERGENCE OF A NEW ABLAUT FORMATION

5.1.Let us now return to the formations themselves. As stated (Section 3.3), in North and West Germanic a reversal procedure similar to that in Gothic may have occasionally been utilized; it might even be considered the basis for the OE contracted forms. A preserved or restored reduplication could also account for some sporadic forms which appeared without reduplicative element, e.g. ON heita heit, sveipa ∿ sveip, OE gangan ∿ gang.¹⁴ However, at least some of these preterites can also be considered analogical forms modelled after the e verbs, cf. ON bida beid (preterite singular) bidom (preterite plural), sveipa ∿ sveip ∿ svipom, in which case the plural svipom is the best evidence for analogy (Van Coetsem 1956:56-6, 65). ON sveip ∿ svipom are analyzed by Bech (1969:46) as suppletive forms of a Germanic verb continued in Middle High German swifan. Old Frisian hēta ∿ hēt [discussed in Appendix II. 4.2 (ii) (β)] is clearly the result of change, and analogy and has nothing to do with a loss of the reduplicative element.

In general, the reanalysis serves as a starting point for a description of the North and West Germanic development of the reduplicating class. The (readjusted) reanalyzed preterite form gave rise to three corrective strategies from which PARALLELING and COMPETING preterite formations originated. On the basis of the principle of generalization and adequacy (Section 1.6), we have found the generally bisyllabic root structure along with the specific nonroot material to be structurally deviant and deficient in structural relevance or motivatedness; this made the reanalyzed preterite form susceptible to further readjustments or changes.

Taking into consideration the development of suffixation with the verba pura, and the analysis of the OE contracted forms as possibly resulting from preserved or restored reduplication, we can schematically distinguish and characterize the three strategies as follows:

1. Analogical generalization of one form of the specific nonroot material (infix or suffix), as found in the r preterite formation.
2. Preservation of the specific nonroot material (or of the thematic anlaut in case of a reduplicated form) with reduction to a monosyllabic structure, as found in the OE contracted forms.
3. Consistent structural re-formation of the reanalyzed preterite form on the model of the present, resulting in a preterite form which was monosyllabic and did not contain the specific nonroot material. This preserved the new ablaut pattern which had emerged with the reanalysis.

Keeping in mind that the weak conjugation was also a possible recourse for the reduplicating verbs, we will now examine points 1-3 in greater detail.

5.2.The first strategy discussed is the analogical generalization of one form of the specific nonroot material (infix or suffix), as found in the r preterites. As pointed out (Section 4.2.2), the specific nonroot material appeared in a structurally unmotivated variety of lexical forms. A reaction to this was the spread of one of these forms by analogical extension outside of its original lexical domain; with this, other forms of the specific non-root material were levelled. This produced the so-called r preterite formation.

5.2.1.The analogical spread is nicely illustrated by the ON verba pura. Besides rōa ∿ rera and sā ∿ sera we find, e.g. grōa ∿ grera, snūa ∿ snera, in which the r was a suffix. With the suffixation and the reinterpretation of the endings, a monosyllabic root structure evolved. The analogical spread is also exemplified by the OHG r preterites, e.g. būwan ∿ biruuuis, stōzan ∿ steroz, in which the r was an infix;¹⁵ in this case the bisyllabic root structure was preserved. Thus, this formation contained forms in which the r infix or suffix occurred analogically, and in the broader sense also forms from which the analogical extension started.

5.2.2.Another question here concerns the origin of the r in the r preterites.

(i)For the ON verba pura some have assumed (e.g. Boer 1920:190) that the analogical spread originated in a verb with r anlaut. Here the thematic anlaut of the original reduplicated form was r, rōa ∿ rera (*rerō-). A similar origin, that is an analogical extension from a verb with r anlaut (verbum purum or not), may be posited for the OHG r preterites.

(ii)It has been more generally assumed that the r developed from a verb with s anlaut (verbum purum or not), in which the thematic anlaut of the original reduplicated form was voiced to z (Verner’s Law) and subsequently rhotacized to r. We can again refer to an ON verbum purum, namely sā (Gothic saian) ∿ sera (*sezō-, Gothic saiso).¹⁶ Actually, this second assumption has a better foundation; an analogical spread was better motivated in a case where the anlaut and the specific nonroot material were different.

Based on the requirements of his theory, Bech (1969) assumes the analogical spread to have started from a verb with s anlaut. He considers an ez infix in accordance with his conception of the reanalysis (cf. our Section 3.3.2). He posits an analogical extension of ez to all the verb types of the reduplicated class, and views the new ablaut pattern as a development from this ez infix as well. There are several serious shortcomings in his theory, which we will discuss in greater detail in Appendix II. 3.

5.2.3.A question related to the previous discussion of the origin of r in the r preterites concerns the choice of z (> r) as a model for analogical extension, rather than another consonantism, such as l (cf. Gothic lailaik), m, or even a consonant cluster. In trying to answer this question we can only speculate. Verb types with single initial consonants were probably more plausible candidates for the analogical spread in question than those with initial consonant clusters. Following Bech (1969: 20-1), we may perhaps assume that of the verb types with single initial consonants, those occurring in both the verba pura and impura were the most probable candidates, and these were then precisely the verb type with s anlaut (*sē-, e.g. ON sā, *salt-, e.g. OHG salzan) and the one with r anlaut (*rō-, e.g. OE rowan, *red-, e.g. ON rāda). As we have noted, of these two verb types the former would have offered a better motivation for analogical extension of its specific nonroot material.

5.3.Preservation of the specific nonroot material (or of the thematic anlaut in case of a reduplicated form) in its original different forms is found in OE (Anglian) heht of hātan, leolo of lācan, etc.; these were evidently the result of contraction.

5.3.1.We could hardly talk of a “corrective strategy” here, if it were not that the second, nonprominent syllable of the reanalyzed preterite form had in the OE contracted forms CONSISTENTLY disappeared [for this reduction cf. Sections 4.4.2 (ii) and (iii)]. In OE bisyllabic roots with a short vocalism in the first, prominent syllable (that is, in trisyllabic words when inflectional endings are involved), the second, nonprominent syllable was either not so consistently dropped or it remained altogether (Sievers-Brunner 1951:141 [159,c]). The OE contracted forms seem, therefore, to reflect primarily a readjustment mechanism, eliminating the structurally deviant and unmotivated bisyllabicity of the root, i.e. producing a monosyllabic root structure in accordance with the root structure of the present and the general root type of the strong verb. Yet, with the preservation of the specific nonroot material a structurally deviant and unmotivated factor remained.

In comparison with the OHG r preterites, which show partial reduction of the vocalism of the second, nonpromi-nent syllable, the OE readjustment mechanism resulted in total reduction of that vocalism. It seems reasonable to assume that this total reduction (as opposed to partial reduction) represents a second step diachronically, and is accordingly not a primary readjustment. Perhaps there is, then, a relationship of developmental complementarity between OE on the one hand and OHG on the other; in its contracted monosyllabic forms OE preserved the specific nonroot material in its different forms, while the OHG r preterites maintained the bisyllabic root structure with analogical extension of the r infix.¹⁷

5.3.2.Some further remarks are necessary.

(i)In assuming a virtually generalized spread of the ez infix after reanalysis, Bech (1969:23) can explain such cases as heht only as the direct continuation of a reduplicated form. This may be possible. The question is, however, whether reduplication is compatible with the occurrence of the strong reduction in the OE contracted forms. We should also note that in a case like heht, the second h may have developed from g as well as from x.

(ii)Through regular development and/or structural readjustment in the case of unusual (unacceptable) or unstable distributions or clusters, the contraction process may have occasionally produced forms with modified specific nonroot material or forms showing the new ablaut pattern in a way similar to those discussed in Section 5.4.

(α)If some verbs with s anlaut (s cluster) met the conditions, an occasional form with ez + voiced consonant could have resulted in the pattern ‘ē + voiced consonant’ [cf. Appendix II. 3.3 (iv) (ε)].

Similarly, in a case like OE hwōpan [assuming that the alternation produced by Verner’s Law had been erased; cf., however, OE wōpian (weak) and wēpan (red)], a preterite like *hwe (h) wp- could have developed (proposed by Sacks 1977:17). Whether this actually happened cannot be stated with any certainty. The assumption of such a development is also plausible for, e.g. OHG wiof ∿ wuofan, if Pre-OHG knew the formation with total reduction of the second, nonprominent syllable.

(β)Structural readjustment in the case of an unusual (unacceptable) or unstable distribution or cluster may be evidenced by OE speoft/seoftun, with ft from pt adapted from spt in * spespt-, probably belonging to spātan.

(γ)We have noted [Section 4.4.2 (iv)] that in the case of a short vocalism total reduction of the vocalism of the second, nonprominent syllable, e.g. *fe + b + ll- ∿ *fall-, could have occurred over the whole of the North and West Germanic area, with, perhaps in certain cases, change and/or structural readjustment of unusual (unacceptable) or unstable distributions or clusters.

(iii)The form hēt of hātan, which will be discussed further (Section 5.4), cannot have developed from heht, as Prokosch (1939:176) quite correctly points out: “If heht was the regular phonetic development of *hehāt = Go. haihait, it seems improbable that at a comparatively late period it should have gone through the altogether irregular development to hēt ; if the form had been preserved at all, it would have remained heht in Anglian, as in reht, cneht, or become *heoht in West Saxon, as in reoht, cneoht.”

(iv)Aside from lācan, the roots of the OE contracted forms ended in t or d. Considering the scarcity of the examples, this is probably coincidental; it could also reflect some form of phonetic preference.

(v)Finally, if leort of lētan actually had an analogical r after the model of, e.g. rēdan (verb with r anlaut), the latter represents the prototype in which the analogical extension must have originated. In virtue of our earlier discussion (Section 5.2.1), we can then consider all of them r preterites. Bech (1969:23) does this as well, but within the bounds of his own analysis. Since, however, all of the involved OE (Anglian) forms, e.g. heht, leolc, reord, can uniformly and plausibly be accounted for by simple contraction (with leort being considered a dissimilation from *leolt), there are no unambiguous examples of r preterites in OE. There is then little evidence for assuming that OE ever had an r preterite formation.

5.4.In spite of the applied readjustments or changes, the r preterites and the OE contracted forms continued to show structural inadequacies, i.e., characteristics which were structurally deviant and deficient in structural relevance or motivatedness, namely the bisyllabic root structure and/or the specific nonroot material (affix). These formations were unfavored ones compared to the new ablaut pattern (and the weak conjugation). The formations showing this new ablaut pattern resulted from a third corrective strategy applied to the reanalyzed preterite form. We will examine this strategy now.

5.4.1.In order to properly approach the problem, we must first consider how FORMAL DIFFERENCES comparable to that between the present and the reanalyzed preterite form generally manifest themselves, and how they develop further. We thus focus on the occurrence and development of such differences between, e.g. the present and the preterite (and the past participle) within the paradigm of the strong verb.

(i)What types of differences (which do not include phonologically conditioned variations) can we isolate?

(α)Our primary concern being the reanalyzed preterite form, we have seen until now one type. Here one or more segments (S) alternates with zero (ø), i.e. ø ∿ S, e.g. ø ∿ ga: Present *Xa + ø + it- ∿ Preterite *Xe + ga + it-. Such an alternation or difference between the present and the preterite is found in e.g. ø ∿ n: OHG fāhan ∿ fieng and in reverse order S ∿ ø, or in e.g. n ∿ ø: Gothic standan ∿ stop, English stand ∿ stood.

(β)In another type none of the alternants is ø, i.e., S ∿ S’, e.g. z ∿ r: Dutch verliezen ∿ verloor. More recent developmental stages of ablaut show the type S ∿ S’ (∿ S’) [e.g. i ∿ a (∿ u), German trinken ∿ trank ∿ getrunken, English drink ∿ drank ∿ drunk ]. Ablaut was also an S ∿ ø type, namely in Proto-Germanic (Van Coetsem 1980: e.g. 290-1).

(γ)The distinction between (α) and (β) presupposes PARTIAL difference (and partial identity) between the present and the preterite forms. This distinction is neutralized in cases of TOTAL difference, that is in cases of suppletion (e.g. Gothic gaggan ∿ iddja, English go ∿ went). In relation to our problem, the reanalyzed preterite form, partial difference is the pertinent one.

(ii)How do such differences relate to the notion of DEVIATION (as described in Section 1.6)?

Such differences correspond to degrees of deviation from the structural properties of the language. When the difference falls completely within the bounds of structural properties of the language, its structural deviation is zero. As S ∿ ø or S ∿ S’, ablaut represents a zero deviation. In the Germanic languages in general the type S ∿ ø (cf. Gothic standan ∿ stoƥ) appears structurally more deviant than the type S ∿ S¹ (e.g. alternations of Verner’s Law).

(iii)How do such differences relate to the question of RELEVANCE or MOTIVATEDNESS?

(α)In principle, the types, ø ∿ S (S ∿ ø) and S ∿ S’, may be primarily relevant or fully motivated, secondarily relevant or less motivated, and even irrelevant or unmotivated. Both types may occur together within the same lexical item (e.g. English stand ∿ stood, i.e. a ∿ oo, S ∿ S’, n ∿ ø, S ∿ ø) or the same type may occur more than once in the same lexical item (e.g. Dutch vevliezen ∿ vevloov, i.e. ie ∿ oo and z ∿ v, S ∿ S’).

Of these differences we can now clearly single out one, namely ablaut, which, as pointed out in Section 4.2.2, was the necessary and dominant marker of the differentiation between, e.g. the present and the preterite within the paradigm of the strong verb. As such, the difference (alternation) brought about by ablaut was primarily relevant or fully motivated and structurally nondeviant; it was thus quite distinct from the formal difference between the present and the reanalyzed preterite form.

(β)From the above examples we can see that a difference between the present and preterite forms may emerge in a variety of ways. For example, in Gothic standan ∿ stop it resulted from the occurrence of a presential n infix. In OHG fahan ∿ fieng there was a phonological change in the present form. Such differences are, in the development of a language, likely to change in relation to their degree of relevance or motivatedness and deviation.

(iv)What can happen to such differences?

(α)Independent of the fact that the two types, ø ∿ S (S ∿ ø) and S ∿ S’, may replace one another in language development, such differences are often ruled out by remodelling the preterite form on the basis of the present. For example, OHG stantan ∿ stuot (cf. Gothic standan ∿ stoƥ) developed to stantan ∿ stuont (Braune-Mitzka 1959: 280). That in such a case n was integrated into the preterite form and became part of the root indicates that the original presential n infix had lost its status and had become irrelevant to the differentiation between the present and the preterite. The other example, OHG fahan ∿ fieng, developed more occasionally to fahan∿ fieg [Braune-Mitzka 1959:283 and cf. (β) below]. The fact that in such a case n was dropped from the preterite form indicates again that it was irrelevant to the differentiation between the present and the preterite; actually it probably never had affix status.

There are more such examples. E.g., while OHG had lāzan ∿ liez, Middle High German developed a contracted form lān (root lā -; cf. present lān, lāst/laest, lāt/laet, lān, lāt, lānt) ; the preterite lie was formed on this model. Similar readjustments of the preterite form to the present in Middle High German were gān ∿ gie, vān ∿ vie, hān ∿ hie (Weinhold-Ehrismann-Moser 1965:118); such cases apparently did not involve any form of affixation. Middle High German offers a further example of this in the preterite forms kom/kōmen (Bavarian, since 11 C.), kam/ kāmen (e.g. Alemannic), and Modern German kam/kamen, which had kw anlaut in OHG quam/quāmen (cf. also Gothic qiman/ qam/qemum/qumans, OHG queman, etc.), but acquired k anlaut as a readjustment to the present komen in Middle High German (Braune-Mitzke 1959:277, Weinhold-Ehrismann-Moser 1965:94 and cf. Modern English come ∿ came, but also komen ∿ kwam in Modern Dutch). This can also be considered a repetition of the primary lexical anlaut readjustment discussed in Section 4.3. In previous work we have emphasized as well the dominant role of the present in similar and related cases (Van Coetsem 1980: e.g. 325, 329-30).

The above are examples of the type difference ø ∿ S (S ∿ ø). These are nearest to the difference between the present and the reanalyzed preterite form. Levelling of differences of the type S ∿ S’ is also well documented. The levelling of the alternations of Verner’s Law in the Gothic strong verb (mainly on the basis of the alternant of the present) is a famous case in point (Van Coetsem-Hendricks-Siegel 1981:172-3). Another well-known example is the preterite form brang- ∿ brung-/brong- (for braht-/ bracht-, cf. German bringen ∿ brachte, gebracht, English bring ∿ brought/brought, Dutch brengen ∿ bracht/gebracht), formed after the present bring-/breng-, and following the appropriate ablaut pattern. It is a repeatedly introduced and suppressed innovation found in old and new Germanic languages or dialects (for further discussion see Van Coetsem 1975b: 280-1).

One might like to speak of INSERTION in the case of OHG stuont, and of DELETION in the case of OHG fieg. A-side from the fact that this represents an unnecessary restriction to the type ø ∿ S (S ∿ ø), the forms stuont and fieg did not result either from a direct insertion process, stuot > stuont or a direct deletion process, fieng > fieg. The insertion or deletion of n in these examples was not determined by the forms stuot and fieng themselves, but resulted from a readjustment of the preterite to the present. If, nonetheless, we speak of insertion or deletion in such cases, we do so on the basis of a comparison of the competing preterite forms, namely stuot vs. stuont and fieng vs. fieg.

(β)Less usual but certainly not uncommon is the levelling or readjustment of the present form on the basis of the preterite (and past participle). For example, it is generally assumed that the present-infinitive fangen in German (cf. OHG fāhan) and vangen in Dutch (cf. Middle Dutch vaen) owe’their forms with ng to the preterite and past participle, fingen, gefangen in German and vingen, gevangen in Dutch. Also, there are a number of well-known cases with alternations where the alternant of the preterite (plural) and the past participle has been introduced into the present. In addition to the example of ng in German fangen, Dutch vangen, compare, e.g. r in German verlieren, verlor, verloren with z ∿ r in Dutch verliezen ∿ verloor, verloren, the latter representing an earlier situation.

(γ)In some instances the difference remains. For example, n ∿ ø in Gothic standan ∿ stop is still present in Modern English stand ∿ stood. In such cases several preserving factors may be at work, such as the high text frequency of the lexical items involved. Occasionally a difference along these lines is not only maintained, but may even become relatively productive. In Middle Dutch staen developed sting (stong) for regular stont on the a-nalogy of gaen ∿ ging (gong), vaen ∿ ving (vong) (Schönfeld-Van Loey 1964:180).

OHG fāhan ∿ fieng is interesting in that, next to the dominant ablaut alternation ā ∿ ie, the differences ø ∿ n (ø ∿ S) and h ∿ g (S ∿ S’) occurred between the present and the preterite. In the development of fāhan ∿ fieng to fāhan ∿ fieg, the difference ø ∿ n was ruled out while h ∿ g remained intact. This illustrates possible distinctions in the degree of relevance or motivatedness and the degree of deviation. Differences of frequency should not be discounted either.

(δ)We can schematize the interrelationships of such differences in the following way:

However, preserving factors may be at work:

5.4.2.We now return to the (readjusted) reanalyzed preterite form. Referring to Section 4.4.1 (i), where the treatment of the second component of a diphthong is discussed, we can posit, e.g. present * Xait-/*Xaet- ∿ preterite *Xegit-/*Xeget- (for earlier *Xegait-) .

(i)The formation showing the new ablaut pattern resulted from a consistent structural re-formation of the reanalyzed preterite form on the model of the present. As this strategy was applied, a form without the specific nonroot material and with a monosyllabic root structure evolved. Because the formation was based on ablaut, it was in accordance with the fundamental principle governing the distinction of principal parts within the strong verbs. It was thus completely compatible with the structure and development of the language. As such, it is not surprising that it became the dominant usage and supplanted the two other competing formations.

In comparison with the examples discussed in Section 5.4.1, the need for structural re-formation was all the more urgent in the reanalyzed preterite form, since the latter involved characteristics that without any doubt were structurally deviant and deficient in structural relevance or motivatedness. Furthermore, that the reformation occurred on the basis of the present and not on that of the preterite is all the more understandable with the reanalyzed preterite form, since the preterite was in this case different from both the present and the past participle, unlike the examples discussed in Section 5.4.1 (iv) (β).

The structural re-formation of the reanalyzed preterite form can be illustrated as follows:

[Compare here the realizations e-u/e-o/i-u in the OHG -r- preterites pleruzzun, steroz, bivuuuis, Section 4.4.2 (i).]

[In the latter example there may also have been a development -bll- to -ll-, Sections 4.4.2 (iv) and 5.3.2 (ii).]

The new ablaut formation thus clearly contained the constant e of the original reduplicative element. Although he does not account for the development itself, Hirt (1932:145), for one, has recognized the presence of that vowel in ē²/eo/e of the new ablaut pattern. He states: “Mir scheint es vor allem auffallend, dass in dem Perfektum, mag es einer Reihe angehören, welcher es will, immer ein e-Vokal erscheint, as. hēt, *hleop, feng, und das legt doch die Vermutung nahe, dass es sich um reduplizierende Formen handelt.”

We can conclude, then, that the new ablaut formation derived from the maintenance of a primarily relevant difference (ablaut). At the same time, less motivated, deviant differences were not preserved here. On the other hand, the r preterites and the OE contracted forms both maintained less motivated, deviant differences; this is comparable to cases like OHG fāhan ∿ fieng vs. fāhan ∿ fieg. With this analysis it is not surprising that the formation which was most compatible structurally with the language, that is, the new ablaut formation, was exactly the formation to become the dominant usage.

(ii)We will now consider some further questions that relate to the new ablaut formation.

(α)In comparison with the examples examined in Section 5.4.1, one specific aspect of the reanalyzed preterite form has not been discussed. This is the fact that the specific nonroot material occurred in intervocalic position, i.e. in so-called “interlude” or “ambisyllabic” distribution. This is a highly uncommon position for non-root elements in Germanic. As such, it made the specific nonroot material structurally all the more deviant and susceptible to loss. Yet this does not make it ESSENTIALLY different from such examples as OHG fāhan ∿ fieng (> fahan ∿ fieg). As a matter of fact, the reanalysis caused the unusual distribution, which is unambiguously reflected in the OHG r preterite formation. The fact that in this formation the specific nonroot material developed to an r infix shows how much the intervocalic position may be compared, in the question at hand, to other distributions in the word. Moreover, as we will discuss under (β), the reformation could also have applied to a monosyllabic structure in which the specific nonroot material did not occur in intervocalic position.

(β)The re-formation proposed in Section 5.4.2 started from a bisyllabic root structure that had partial reduction of the second, nonprominent syllable, as reflected in the OHG r preterites. This is, in our view, the most probable analysis. However, the re-formation could also have applied to a bisyllabic structure before any reduction of the second, nonprominent syllable took place. It must also have applied in specific cases to a structure exhibiting total reduction of the second, nonprominent syllable [Section 4.4.2 (iv)]. If a more generalized total reduction of the second, nonprominent syllable of the reanalyzed preterite form came about early [although this was probably a second step in the reduction process; see Section 5.3.1, but cf. also Sacks (1977) discussed in Appendix II.4] the re-formation could even have applied to contracted forms such as Old English heht [cf. 5.3.2 (ii)]. The re-formation would in such a case have depended even more on the form of the present. Cf. *Xait-/ *Xaet- ∿ *XeXt- with i/e ∿ X > i/e ∿ i/e, yielding *Xait-/ *Xaet-> Xeit-/*Xeet-; preterites as slē²p- (OHG slief) would then be best explained as analogical.

In any case we should be well aware of the fact that a formation does not have to derive from one single form. As Hirt (1932:156) has stated in a general perspective: “Es ist ein Kōhlerglaube, dass eine scheinbar einheitliche Bildung immer auf eine einzige Grundform zurūckgehen muss.” And as far as our specific problem is concerned, we should also heed Fourquet’s words (1962:67): “On a plus de chances d’ētre dans le vrai, si l’on admet que la systématisation westique et nordique de la Vile classe a pour base une multiplicité de formations.”

We should also remember that once the ablaut formation was established, it must have been productive and expansive. The forms of the other formations were then gradually supplanted. There is no need to assume that the same re-formation was utilized again and again. In fact, forms like OE heht and OHG steroz apparently went out of style, rather than being eliminated thrpugh a repeated re-formation process.

(γ)Umlaut and consonantal conditionings operated with or after the accent shift, that is, at the time of the evolution of the above forms. The role of umlaut (Section 4.4.1) is not in all respects clear, but consonantal conditioning before checked nasal is unambiguously attested, e.g. *feng-> *fing- (subtype of *fell-), although not in a consistent way. In West Germanic the form with e is found most commonly, e.g. Old Saxon feng. This has been called a “restoration of e-vocalism” (Kerns 1937:14) following the model of *fell-. Actually, this is only one example of a clearly marked and more general preference for the “mid” vocalism ē²/eo/e (as compared to i/iu/i) in the preterite part of the new ablaut formation; in other words, the prevailing distinctions or alternations between such “mid” and “high” vocalisms in the present are not found in the preterite. Likewise, the jū/ jō split in ON, based on consonantal conditioning (in general jū before labials and velars, e.g. krjūpa, and jō before dentals, e.g. bjōda) is not upheld in the preterite, and jō (for earlier eo) is generalized (e.g. hljōp). For further discussion and more examples see Van Coetsem (1956:65-6).¹⁸

(δ)The original, basic form of the new ablaut formation was subsequently affected and to a certain extent obscured by changes, shifts, or analogical extensions (Van Coetsem 1956:56-69 and 1963).

For example, ē ² seems to have spread from the *Xait- type to the *fall- type in OHG fiel. In the development of Dutch we see a strong expansion of ē² (or its developments) even outside of the reduplicating class (Van Coetsem 1951 and 1980:336-7). Perhaps we should not rule out the possibility that a form like OHG fiel goes back to a variant of the reanalyzed preterite form in which the vocalism of the second, nonprominent syllable had not been reduced to zero (Section 4.4.1).

The ē² of ON blēt of blōta is apparently a good example of analogical extension as well. The verb blōta, the only one of this type in the reduplicating class of ON (the verba pura have to be considered separately), has adopted the ē² of the preterite of the more frequent verb types with ai, ON heita ∿ hēt, leika ∿ lek, and ē¹, ON blāsa ~ blēs, grata ∿ grēt (cf., e.g. Karstien 1921:114); the regularly expected vocalism eo is found in OE blēot.

A third example is OE fallan ∿ fēoll/feoll. It may be that the vocalism eo (old eo) was analogically extended to the *fall- type, paralleling the assumed analogical extension of ē² in OHG. Yet eo may be a direct development of e as well; cf. ON, Old Saxon fell, and see Van Coetsem (1956:65) with further references. An example of analogical extension in Old Frisian is discussed in Appendix II. 4.2 (ii) (β).

(ε)The verba pura could develop or generalize glides (y, w) between the verbal root and the inflectional endings, and thus acquire a regular consonant final root type. In such a case, aside from a strong tendency to switch to the weak conjugation, they could follow the general development of the reduplicating verbs. A form like OE sēow can then be directly derived from *sezow- (< *sezō-); cf. Old Saxon sāian (with y glide) but preterite obar-seu (otherwise regularly weak), Middle Dutch saien ∿ zieu (otherwise regularly weak), Modern Dutch zaaien ∿ zaaide (weak).

(Ç)The ē² in ON OE lēt, OHG liez, etc., can have developed from *lelet- < *lelēt- (cf. Gothic saislep ∿ slepan) occurring beside *lelot- ˂ lelōt- (cf. Gothic lailot ∿ letan); it may also be analogical [see (β) above].

(𝜂) The verb with vowel initial root, a rare type (cf. Gothic aikan ∿ aiaik, aukan ∿ aiauk), must have favored the emergence of the new ablaut formation (*e + aik-> *e + ek-, *e + auk- > *e + ok-).

5.5.While the formal prehistory of the three formations may be in our perspective basically clarified, their respective areal-linguistic expansion (with the possible social or stylistic connotations) previous to their attestation remains more hypothetical. As we have seen (Section 1), at the time of their documentation they exhibited not only areal-linguistic but also stylistic differences. We cannot, however, simply project that picture into the past. Were the r preterites and the new ablaut formation each originally confined to a specific area? Or did they areally overlap, mutually and with the formation from which the OE contracted forms derived; were they then used with differences in social or stylistic connotation? Although the r preterites and the OE contracted forms may originally have known a larger lexical and areal expansion than the materials reveal, the only thing we are really certain of is that at the time of documentation they were fighting a losing battle against the new ablaut formation.

I. OUR OWN RESEARCH AND THE MIRROR IMAGE THEORY

I.1.First we will discuss the mirror image view as presented in our earlier work (e.g. 1956, 1963). We assumed there that the new ablaut pattern in the reduplicated class had originated from a so-called INVERTED ANALOGY between the present and the preterite of the e and a groups of the strong verbs:

that is,

We interpret inverted analogy here strictly as a mechanism that effected or created the mirror image alternation in question, not as a possible lexical expansion of an already existing mirror image alternation (productivity).

I.1.1.The above analogy was based on the fact that reduplication in the a verbs (*Xait-, etc.) was a minority principle within the paradigm of the strong verbs. As such, it was subject to replacement by the prevailing ablaut principle. The assumption of such an inverted analogy seemed warranted at the time, as clear examples of REGULAR analogy between the e and a groups had already been found (Van Coetsem:1956, e.g. 61). The question of the validity of inverted analogy, however, even in the presence of regular analogy, was not addressed. In order to distinguish this view from other theories on the origin of the new ablaut formation (contraction, Indo-European origin, etc.), this explanation has come to be known as the MIRROR IMAGE THEORY, e ∿ a/a ∿ e, especially by Dutch linguists.

I.1.2.In the MIRROR IMAGE ALTERNATION, e ∿ a/a ∿ e, that we posited for the new ablaut formation, there were two problems, namely a synchronic problem concerning functionality and a diachronic problem dealing with the origin of the alternation.

(i)As we have just pointed out, we did not address the latter question, i.e. the one concerning the validity of the mirror image alternation as a form of language change; we did not have empirical evidence for substantiating it. Consequently, the explanation of inverted analogy was merely a hypothesis which tried to account for the subsequent evolution in a coherent and simple way. This latter fact implies that the positing of e in the preterite of the reduplicating class can be motivated by reconstruction and must therefore be distinguished from the problem of origin of that [See the assumption of e in ē²/eo/e by Hirt (Section 5.4.2 (i) above.]

(ii)As to the synchronic problem of functionality, we made the following remark: “dieser Ablaut wurde zur Unterscheidung der Tempora angewandt, ohne dass dabei e und a charakteristisch zu sein brauchen für das Präsens bzw. das Präteritum; nicht der Vokalismus an sich, sondern der betreffende Unterschied und Ablaut war funktionell. Das Nebeneinander von z.b. ie/oo in ndl. ik schiet ‘ich schiesse’ und ik schoot ‘ich schoss’ (Präteritum) und ik loop ‘ich laufe’ (Präsens) und ik liep ‘ich lief’ (Präteritum) dürfte eine beweisende Parallelle dafür bieten.” With the Dutch example ie ∿ oo/oo ∿ ie we merely meant to illustrate the functional validity of a mirror image alternation, and not its origin, as Bech (1969:52) erroneously thinks. The fact that ie ∿ oo/oo ∿ ie belongs historically to the ablaut alternation e ∿ a/a ∿ e cannot then be an objection.

I.2.Let us now turn to the question of the mirror image alternation with the knowledge and the perspective that more recent research has provided us.

I.2.1.In the SYNCHRONIC perspective, we find that mirror image alternations are not at all uncommon (Chomsky-Halle 1968:355-6), and that mirror image properties have been attributed to natural languages (Langacker 1969:575-98, 844-62 and see also Chomsky 1973:233).

Chomsky-Halle examine mirror image alternations and mention that as early as 1912, Meinhof coined the term POLARITY RULES for that kind of alternation. Chomsky-Halle discuss these as a general phenomenon which can be formulated in a simple rule: [𝛼 feature x] → [-𝛼 feature x], being itself an abbreviation of two rules that apply disjunctively, namely [- feature x] → [+ feature x] and the reverse [+ feature x] → [-feature x]. This formulation shows the greater generality allowed by a mirror image alternation. Chomsky-Halle mention an example of mirror image alternation (low ∿ nonlow) from Biblical Hebrew, which exhibits certain features similar to our Germanic example:

There has been a discussion about the nature of polarity rules and whether they are exclusively “morpho-lexical” or also phonological (Stephen Anderson, Wayles Browne, James McCawley, etc.; see Zonneveld 1976 also for a summary of the discussion and further references), but it is not necessary to go into that here.

I.2.2.As to the DIACHRONIC aspect, i.e. the origin of mirror image in language, there is no doubt that on the phonic level (i.e. the one of “mere otherness”) inversion directly producing a mirror image of an earlier situation is not at all uncommon. There is, for example, a well-known type of inversion represented in, e.g. g ∿ z > z ∿ g, in French magazin > popular French mazaguin (Grammont 1939: 348-52), but this is not a form of analogy.¹⁹

There are more complex forms of inversion based on analogical hypercorrection (with possible spelling influence) and involving stylistic (social) connotation or function (cf. Anttila 1972:90-1). Both phonological and lexical factors are involved. An example of this is the syncope of intervocalic d and its hypercorrect restoration in the history of Dutch (Zonneveld 1978).

However, when grammatical categories like present and preterite are involved, an inversion or inverted analogy would imply inversion of function, e.g. present e ∿ preterite a/present a ∿ preterite e. For the origin of such a mirror image alternation unambiguous proof in the way of clear examples of direct inversion has not been found, and therefore the notion of inverted analogy has not only been doubted, but also flatly rejected. Bech (1969:52) for one insists that analogy follows a direct proportion (ax:ay = bx:by), never an inverted one (ax:ay = by:bx). Yet proponents of the generative approach, although generally opposed to the notion of analogy, tend to regard an inversion that produces a mirror image alternation as possible. Chomsky-Halle (1968:356), in their discussion of mirror image alternations and polarity rules, state that “polarity rules may arise in a language in a great many ways IN ADDITION TO BEING ADDED DIRECTLY TO THE GRAMMAR” (our emphasis). The latter points to the fact that the authors believe in the possibility of a direct inversion. Earlier Kiparsky (1965:2-38) noted: “It is reasonable to suppose that changing fixed polarity values to variably specified polarity values in a rule should make it more general. If that initial assumption is right, then in view of the idea of analogy as a simplificatory process inverted analogy is just what we would expect to find.”

Unfortunately it remains a fact that, as yet, we do not have a single, unambiguous example of such an inverted analogy.²⁰ Thus, the validity of the notion itself may, indeed, be rightfully questioned. This seems to be reason enough not to turn to it for an explanation of the origin of the new ablaut formation. This is especially the case since there may be a better evidenced basis for accounting for it, as we hope to have demonstrated in previous sections of this study.

I.3.In conclusion, while leaving the question of inverted analogy open, we can maintain here the concept of mirror image, as it characterizes an original property of the new ablaut formation. Yet we should adapt it to our current view. The new ablaut formation appears to have originated from the inclusion of the constant e of the original reduplicative element within the bounds of the verbal root. Keeping in mind that the original reduplicative element contained the potential ablaut alternant (Section 4.2), it is very important to note that BY SIMPLE APPLICATION OF THE REDUPLICATION PRINCIPLE THE MIRROR IMAGE ALTERNATION WAS POTENTIALLY PRESENT: e (*greip-) ∿ a (*graip-) /a (*Xait-) ∿ e (*XeXait-). In the reanalysis the latter e became integrated into the verbal root, and with the resulting new ablaut pattern, the mirror image alternation was actualized. However, it soon became obscured in several parts (resulting from, e.g. the change ei > ī in *greip-). As a phenomenon that allowed for greater generality, it may have helped the productivity and expansion of the new ablaut formation.

II. OTHER RESEARCH

II.1.The development of the Germanic reduplicating class, especially the origin of the new ablaut formation in North and West Germanic, is one of the subjects in Germanic linguistics which have been endlessly researched, discussed, and debated. As a consequence, virtually anything said here about the topic has already been stated in some perspective or other. However, previous research has not produced a generally accepted type of explanation of the total development. Such an explanation is precisely what we have been focusing on in the present study.

II.2.In our earlier work (1956:47-52; cf. also 1970: 89-91) we reviewed the different theories that have been proposed, and thus we can restrict ourselves here to a summary of that review.

II.2.1.The first general type of explanation is the contraction theory, which goes back to Grimm and on which subsequently several variants have been submitted. According to this theory the new ablaut formation evolved through contraction of the reduplicative element with the verbal root. Such developments as dissimilation and deletion of the initial consonantism of that root have been assumed. Another suggested origin for the contraction has been the vowel initial verbal roots (*auk- *e + auk-) .In the way it has been generally proposed, the contraction theory is ill-founded and unconvincing.

II.2.2.While the contraction theory brings the North and West Germanic formation together with the Gothic reduplicated forms, a second theory, which goes back to Brugmann (Wood), explains the new ablaut formation as originating from different Indo-European ablaut alternants. The North and West Germanic new preterite vocalism is then derived from Indo-European ē + y/w/l/r/m/n; there is, however, no evidence whatsoever for such an assumption. Moreover, it seems highly improbable that we would have a Pre-Germanic ablaut here, since there is no reflex of it in the derivational morphology (Van Coetsem 1956:50 with further references).

II.2.3.Another theory, proposed by Karstien (1921), has been generally abandoned (Van Coetsem 1956:50). However, we should note that it assumed the loss or dropping of the reduplicative element, an idea which cannot be completely left out of consideration (cf. our Section 5.1). Finally, our own earlier theory of inverted analogy has been discussed in Appendix I. In the following review of the recent literature on the subject we will focus only on those views (Bech 1969, Sacks 1977) that are germane to our analysis.²¹

II.3.On several occasions we have referred to Bech (1969), either in agreement or disagreement. Since a number of his ideas have been adopted here, we will now systematically examine Bech’s argument, comparing it with the explanation offered in our analysis.

II.3.1.A first point concerns the general approach to the problem. It is to Bech’s credit that he has recognized the importance of considering the different formations for explaining the development of the Germanic reduplicating class, and the emergence of the new ablaut formation, i.e. of the so-called NORMAL TYPE. As previous research, including our own, paid little or no attention to anything which was not that normal type, it missed precisely the basis for an overall explanation: “Man hat es von jeher in viel zu hohem Grade versäumt, die gesamte Prät.-Bildung der nwg. (North and West Germanic) 7. starken Konjugation als eine Einheit und die vorhandenen Varianten derselben als verschiedene je nach der Zeit und Ort variierende Produkte einer und derselben Entwicklungskette anzuschauen und hat es schon deswegen zu keiner ūberzeugende Lōsung bringen können” (Bech 1969:54). However, Bech’s view of the relation between the normal type and the other formations is basically different from ours.

II.3.2.Bech’s explanation provides a second positive point which we have adopted. This is the assumption of a systematic levelling in Gothic of the alternations produced by Verner’s Law with the preservation of reinstatement of the reduplication principle and, conversely, the preservation of such alternations with the abandoning of the reduplication principle in North and West Germanic (cf. Sections 3.1 -3.3 above).

II.3.3.A third constructive point in Bech’s treatment, which has become part of our own as well, is his assumption of a reanalysis of the.original reduplicated form in the North and West Germanic area. This reanalysis produced a nonroot element within the bounds of the verbal root called SPECIFIC NONROOT MATERIAL by us and INFIX by Bech; in reviewing his theory we use his term infix. The reanalysis constitutes the point at which crucial differences between his view and ours arise.

(i)The basic differences between Bech's view and ours can be traced back to our respective conception of the reanalysis.

(α)We noted (Section 3.3.3) that Bech s conception (namely *X + eg + ait- with *X-ait- being the verbal root and -eg- the infix), ignores the full implications of the accent shift, and is therefore insufficiently motivated and unsatisfactory. In his view the constant e of the reduplicative element became in the reanalysis part of the infix, not of the verbal root as in our explanation. Since his view, like ours, implies that this e became integrated into the verbal root, he must account for this in the subsequent development of the reanalyzed form. How does Bech see this subsequent development?

He takes into consideration the productivity of the r preterites, but is limited by his own theory in his explanation of their origin [cf. our Section 5.2.3 above and (iii) below]. He assumes r to have originated from z, that is, from verbs with s anlaut (*sē-, *salt-) and suggests that the various forms of the infix were replaced by es in all the verb types of the North and West Germanic reduplicating class (*X + eg + ait- > *X + ez + ait-, *X + egl + aup- > *Xl + ez + aup-, the latter with anlaut readjustment). The emergence of e in ez as a root vocalism must then be seen in relation to the so-called syncope [in our terminology (total) reduction] of the vocalism of the second, nonprominent syllable of the reanalyzed form; in Bech’s opinion this reduction was a total one in the case of ai, a, ē (ǣ), but a partial one, namely to w, in the case of a back vocalism au, ō, ū (cf. *Xezait- > *Xezt-, *Xlezaup- > *Xlezwp-). Subsequently ez developed to ē through so-called Ersatzdehnung ‘compensatory lengthening’ (*Xezt- > *Xēt-, *Xlezwp- > *Xlewp-). Thus, for Bech the basic development is a succession of four steps, namely (1) reanalysis (with anlaut readjustment whenever necessary), (2) generalization of ez, (3) syllable reduction, and (4) compensatory lengthening. This can be illustrated by the following examples:

(β)In our view, however, the constant e was, with the reanalysis, directly integrated into the verbal root. This allows for a simpler solution than Bech’s, as we have seen.

(ii)We have stated (Section II.3.1) that Bech’s view of the relation between the normal type and the other formations is basically different from ours.

(α)While Bech sees the formation from which the r preterites evolved (ez) as a stage leading to the new ablaut, his view of the relationship between the ez formation, the r preterites, and the normal type is for the most part not one of parallel formations. In our view, on the other hand, the formations are in more of a relationship of simultaneity, with one of them, the new ablaut, becoming the dominant usages.

(iii)It is now apparent why Bech, in his explanation of the origin of the r preterites and the normal type, must assume an analogical spread starting from ez (verb with s anlaut) and not from er (verb with r anlaut). A form like *Xerait- would have become *Xert- (following Bech’s view of the reduction). No compensatory lengthening to ē could then take place. Cf. Bech 28: “Dass in den ahd. (OHG), an. (ON) und angl. präteritalen r-Formen kein germ. r, sondern germ, z steckt, indem in diesen Prāterita ein verallgemeinertes -ez- vorliegt, und kein germ. -er-, haben wir... einfach behauptet. Die entscheidende Begründung dieser Behciuptung wird erst mit dem ē-Prät. erbracht, indem der Übergang von ez > ē es uns erlaubt, dieses ē-haltige Prät. als Ergebnis einer weiteren Entwicklung synkopierter ez-Formen und somit das gesamte Material des nwg. Prät. der 7. starken Konjugation als eine historische Einheit zu betrachten, was eine Annahme von Formen mit innerem germ. -er- nicht ermöglichen würde. Denn aus germ, -er- vor Kons, entsteht jedenfalls kein nwg. ē, sondern das vorkonsonantische r bleibt ja Überall als r erhalten, z.b. germ. *bergan > got. bairgan, an bjarga, ae. beorgan, as. ahd. bergan.” In order to strengthen his case, Bech continues: “Wenn man z.B. angl. reordon direkt aus germ. *rerddun durch Verlust des ō entstehen lässt, wie man ja dies im allgemeinen tut, so ist ipso facto jede Möglichkeit vernichtet, diese Form den entsprechenden ē-Formen der ūbrigen nwg. Dialekte, an. rēpu, as. rēdun usw., etymologisch gleichzusetzen.”

However, as we have shown, Bech’s is not the only way to account for such forms as ON rēpu within the total development of the reduplicating class in North and West Germanic. Moreover, Bech’s explanation, using ez as the source of the new ablaut formation in North and West Germanic, contains a number of serious weaknesses, flaws, and unnecessary complications.

(iv)Let us consider in some detail some of these shortcomings.

(α)First, Bech’s assumption of an original distinction in the degree of reduction of the second, nonpromi-nent syllable of the reanalyzed form (total or partial), depending on the nature of the involved vocalism, is very questionable. As we have pointed out [Section 4.4.2 (iii)], the OHG r preterites contained only a back vocalism (original au, ō, ū). In view of the strongly limited number of such r preterites, this may well be coincidental; it may also reflect a specific OHG situation. Furthermore, considering OE beoftun (∿ beatan/*baut∿),which was nicely in line with OE heht, leolo, etc., but counters Bech’s assumption, it seems more plausible to view the distinction in question as an areal-linguistic one;.it is then even conceivable that there was some form of complementarity between the development of the OHG r preterites and the OE contracted forms (Section 5.3.1).

Compounding the problem is another of Bech’s hypotheses, namely that the partial reduction of the back vocalism au, ō, ū resulted in the so-called “consonant” w; to make the view more acceptable he refers to cases like Gothic skadwjan, balwjan, triggws. This explanation is necessary to save the theory, since only before a consonant could ez become ē, and the only “consonant” that could result from a reduction of au, ō, ū was the glide w. If the partial reduction had produced a vowel, u or ō, the preceding z of ez would have been rhotacized to r, as the OHG r preterites pleruzzun, steroz, etc., prove. Still, the little evidence we have with the OHG r preterites favors the view of a partial reduction yielding not w, but a vowel, which was apparently felt as (part of the) root vocalism [Section 4.4.2 (i) and also Bech 1969:23]. While Bech concludes that ez must have been followed by w, we may as well dismiss the whole assumption of ezw as a pure construct to serve the theory.

(β)Another flaw in Bech’s argumentation concerns the development of ez to ē before t in *lezēt- [of *lēt- (*lǣt-)] > *lezt- > *lēt-. By all known standards zt would have to develop to st, yielding *lest- instead of *lēt-. Yet Bech (1969:27-8) maintains that this ez became ē before t. To account for this he makes again an ad hoc claim, namely that z of ez be regarded not as a “normal” s, but as a sound of the r type (“ein Laut vom r-Typus”), namely R. But then, why did this R not develop to r, as is normally the case? Bech assumes this for OE (Anglian) leort, but persists in believing that such an eR before t changed to ē in the other Old Germanic languages. Clearly, the assumption of a development *lezt- to *lēt- is again only necessary to maintain the theory.

(γ)Unnecessary complication is evident when Bech (1969:33), consistent with his theory, considers the analogical extension of ez to have applied also to the verb type with a vowel initial root, e.g. *eauk- of auk- developing to *ezauk-> *ezwk-> *ēwk-, and *eaik- of *aik- developing to *ezaik-> *ezk-> *ēk-.

(δ)Similarly, Bech (1969:34 ff.) is consistent with his general explanation when he compares the ON and OE verba pura. Using ON sā ∿ sera, OE sāwan ∿ sēow, he posits a development *sezōwV- > *sezwV-, in which w occurred before a vowel V. He assumes that in *sezwV-, w was dropped in ON (before ez could develop to ē), yielding *sezV-, sera (he compares *sungwun > ON sungu, OE sungon, OHG sungun), while ez developed to ē before w in OE, yielding *sēwV-, sēow.

The question whether *sezwV- can be validly compared with *sungwun aside, there is the argument that the traditional derivation of ON sera directly from *sezō- is far simpler and linguistically more realistic than Bech’s theoretical constructions. Furthermore, because of the integrated character of his theory, Bech would also have to account for the i in Old Saxon sāian (cf. preterite oƀar-seu) which he does not and cannot do; cf. Middle Dutch saeien (preterite zieu), Modern Dutch zaaien (weak). For similar examples with the y glide vs. w, see Van Coetsem (1956:68), Rauch (1972), and especially the Middle Dutch verba pura, Van Loey (1955:82-5). Such glides in hiatus position either have developed from the preceding vocalism or have been generalized; areal-linguistic differences must be taken into consideration.

(ε)Finally, Bech (1969:26) lists a set of equations, which would seem to support his theory and which we reproduce here:

If one looks at this set more closely, the evidence for Bech’s view is no more than an “optical illusion.” The fact that Anglian meord and West Saxon mēd came from *mezd- (Gothic mizdo) does by no means prove that Anglian reord, ondreord, leort and West Saxon rēd, ondrēd, lēt were derived from *rezd-, *lezt-. If Anglian reord came from *rezd-, we would also expect a *Xezt- to produce an Anglian *heort (as a preterite of hātan), but instead we find the regularly contracted form heht. In other words, it is very suspicious that ez would have been analogically extended only to verbs with r (or l anlaut). This precisely justifies the traditional explanation of the preterites reord and leort (*leolt) as contracted forms paralleling heht. If, however, we prefer to account for leovt as an analogical form after the model of a form like reord, we can also speak here of r preterites, but in quite a different way from Bech’s. A form such as West Saxon vēd could be derived from *vezd-, if we have proof that there was such a form. And explaining lēt in West Saxon, etc., as the outcome of *lezt-, is, as we have seen (β), merely a device to save the theory. In conclusion, here again we can better dismiss Bech’s ez hypothesis.

(ζ)As one can see, in spite of its valid foundations Bech’s argument is an accumulation of assumptions. The shortcomings of his treatment ultimately result from his unsatisfactory conception of the reanalysis, which, in turn, is a consequence of his failure to recognize the full implications of the accent shift.

II.4.As we do, Sacks (1977) follows Bech in taking into consideration the various forms and not only the normal type. He suggests that “the confused situation of the Gmc. seventh class seems best explained by a solution ultimately based upon the vocalism of the reduplicating class followed by analogical — not phonological — developments” (1977:250).

II.4.1.His starting point is contraction as represented in the OE contracted forms, e.g. present *rēd- ∿ preterite *rerd- (which he apparently considers the inherited preterite plural replacing *rerōd- of the singular). A number of analogical developments would then have followed. For example, after the model of the present *rēd- ∿ preterite *rerd-, we would have (following his notations): *hait ∿ *heht-, *blōt- ∿ *blelt- [ *wōp- ∿ *weup-; OE had hwōp-, wōp-, wēp-, cf. Section 5.3.2 (ii) (α) above], *hlaup-* ∿ *hlelp (*baut- ∿ *beft-) . Similarly, *hald- would have developed a preterite *hehld-, the latter becoming *held∿. The pattern *hald- ∿ *held- would have generated *rēd-*rē²d- (∿ *rerd-), *hait- ∿ *hē²t- (∿ *heht-), *blot- ∿ *blē²t- (∿ *blelt-), *hlaup- ∿ *hlē²p- (∿ *hlelp-) .

II.4.2.There are some difficulties with Sacks’ view.

(i)In cases like *rēd- ∿ *rē²d- (∿ *rerd-), we would have a kind of analogical lengthening or a sort of "compensatory lengthening," although Sacks admits that "the idea of compensatory lengthening in an analogical process is theoretically a difficult one" (1977:246). Instead of a lengthening in the preterite we would expect the model of *hald-∿ *held- to generate, e.g. *hlaup- ∿ *hleup- (instead of *hlē²p-; Sacks assumes that *hlaup- ∿ *hleup- was made after the model of *wop- ∿ *weup-).

(ii)Another problem with the treatment of Sacks is a lack of distinction in the time perspective.

(α)In view of the OHG r preterites and the areal-linguistic restriction of such forms as OE heht at the time of documentation, one may wonder whether it is dia-chronically justified to use forms with (generalized) total reduction of the second, nonprominent syllable as a general starting point for explaining the new ablaut formation [cf. Section 5.4.2 (ii) (β) above]. It seems hazardous to make the emergence of the new ablaut formation exclusively dependent on either one of the other formations, and that is what both Bech and Sacks do. While the former accounts for the normal type by following the line of the r preterites, the latter starts from the type represented in the OE contracted forms.

(β)While Sacks does not clearly distinguish between what is original and what is secondary in the development of the new ablaut formation, he does not view the analogical processes within the system in which they are supposed to have taken place. For example, when positing a preterite *Xlē²p- for the *Xlaup- type, he must have been primarily considering Old Frisian hlāpa ∿ hlēp (hlīp) [for ON blēt of blōta cf. Section 5.4.2 (ii) (δ) above]. However, such a hlēp form cannot have belonged to the original ablaut formation. The Old Frisian generalization of ē in both the *Xait- and *Xlaup- types is the result of both a specific merger and analogical extension; it will have occurred only after ai and au had merged, each regularly becoming ā and ē (the latter with i umlaut). In the concrete case of hlāpa (*Xlaup-) the occurrence of the preterite hlēp can be readily explained as an analogical extension, for example, from the past participle hlēpen (˂ -in, earlier -en) . For more details see, e.g. Steiler (1928:63-4) and for a discussion of other forms, Van Coetsem (1956:59-60) with further references.

II.4.3.On the other hand, Sacks’ view and ours are conceived on comparable principles in spite of differences in factual interpretation. First, like us, Sacks regards the constant e of the original reduplicative element as the basis for the new ablaut formation, which is an idea already implied in earlier work, as we have seen [Section 5.4.2 (i)]. Second, in our two approaches the further justification of the new ablaut formation is founded on the principle of structural re-formation of the preterite. In Sacks’ view this re-formation is analogically carried out on the model of patterns developed from contraction; in our view it is implemented on the model of the root structure as found primarily in the present, with preservation of the ablaut alternant developed in the reanalyzed preterite form.*