3.1. The Law of Initials is Unreliable

All five of Pierce's arguments are based on highly dubious premises. Pierce appeals to the Law of Initials to argue that since Gothic does not permit word-initial clusters ending with the palatal glide (**Cj-), one is entitled to conclude that Gothic does not permit such sequences word-internally. However, the Law of Initials is notoriously unreliable when applied to glide-final clusters. For example, the Law of Initials predicts that present-day Standard German will not permit tautosyllabified -V.djV- since no German word can begin with the cluster dj-, as in Diabetes [dia’be:t∊s], Dialekt [dia’l∊kt] (Clark & Thyen 1997:848). Hence the Law of Initials predicts the following syllabifications in Stan-dard German: Idiot [id.jo:t] and Idiom [id.jo:m].

However, such syllabifications run afoul of two very well founded phonological rules: Coda Devoicing and Open-Syllable Vowel Tensing. By dint of the former, all obstruents in codas are devoiced (fortified). This is such a pervasive rule that native German speakers attempting to master other languages have difficulty suppressing it (for example, by pronouncing [g?t] for [g?d] ‘good‘). By dint of Open-Syllable Vowel Tensing, one should expect lax vowels in the initial closed syllables of the two words under discussion. Hence, by dint of these two rules, one should expect counterfactual Idiot **[?t.‘jo:t] and Idiom **[?t.'jo:m] instead of standard [i.'djo:t] and [i.'djo:m] (see Clark & Thyen 1997: 1024).

A similar problem arises in the analysis of Latin sequences exem-plified by the internuclear glide-final cluster in the word sanguis (putatively [sa?gwis]). No Latin word begins or ends with any com-bination of the cluster [?gw]. Thus, the Law of Initials as well as the Law of Finals utterly fails to predict the actual syllabification of such sequences in Latin. Indeed, Pierce himself (p. 289) admits that the Law of Initials fails to predict Pierce's putative heterosyllabification of the labiovelar glide-final sequences (-C.w-) in Gothic. By dint of Pierce's analysis, such sequences were heterosyllabified (-C.w-) in Gothic. Nevertheless, Gothic does indeed permit such word-initial sequences: swistar, twaddje, Þwairhs. Thus, the Law of Initials offers no credible support for Pierce's analysis.Footnote ⁶

3.2. Very Few Glide-Final Clusters Contain No Morpheme Boundaries

Pierce argues that one cannot attribute the regular scribal practice of dividing words immediately before glides to the presence of morpheme boundaries because there are two sources for such internuclear glide-final clusters that contain no morpheme boundaries, but which, in spite of this, are regularly divided before the glides. Accordingly, Pierce argues that the motivation for division must have been phonological, hence a reflection of syllabification.

The first source for such sequences is Holtzmann's Law, by dint of which the long glides, -jj- and -ww- were partially occlusivized in Gothic to -ddj- and -ggw-, respectively. For example, ⁺twajj- > Go. twaddj- and ⁺triww- > Go triggw-. Since, according to Pierce, no morpheme boun-dary appears immediately before the glide in these sequences, the scribal practice of dividing words immediately before the glide is clear evidence of division induced by syllabification, hence -C.j- and -C.w-. The second source for such sequences is found in verbs whose stems end in /ngw/, such as siggwan. Again, the scribes regularly divided such sequences before the glide, which thereby provides more evidence of the syllabification -C.w-.

Pierce's argument here rests on a tacit premise. The occurrence of the forms lacking a morpheme boundary, -CG-, was relatively frequent. Therefore, the scribes were not motivated to follow the pattern of the majority, -C#G-, and did not divide words to produce -C/G- on the pattern of ubiquitous -C/#G-. However, in the EDB, there are only 21 forms fitting Pierce's criterion (see 3, and table 2 below).

1. (3)

   1. a. Forms in -idd/jV- created in accordance with Holtzmann's Law

   2. b. Forms in -ugg/wV- created in accordance with Holtzmann's Law

   3. c. Verb stems in -gg/w- (that is, -ng/w-)

Table 2.

-C/G- forms displaying either Verschärfung or the verbal ending -ggw-.

The total number of forms with word division immediately before glides (-C/G-) is 626 (see Appendix, table II, columns 4 + 6). Hence, Pierce's exceptions to sequences of -C#G- constitute a mere 3.3% (21/626) of the total number of pre-glide (-C/G-) word breaks. It is not likely that this meager number would cause the scribes to deviate from their normal practice of dividing words immediately before the glides. Note also that these 21 tokens actually represent only 5 different morphemes (see 4).

1. (4)

   1. a. -ïddj- (the suppletive preterite of gangan)

   2. b. -waddj-

   3. c. -bliggw- ∼ -blaiggw-

   4. d. triggw-

   5. e. -siggw- ∼ -suggw-

The fact that none of these forms contains any readily discernable internal morpheme boundaries in a synchronic grammar makes it more plausible that scribes would simply follow their general habit of dividing words before the glides. Nevertheless, all such sequences were probably tautosyllabified in accordance with the Contact Law: -C.C/G-.Footnote ⁷

3.3. Very Few -/PL- Sequences Contain Morpheme Boundaries

Pierce maintains that those who posit a morpheme boundary as the basis for the scribal practice of dividing before glides, -C/#G-, are inconsistent because they do not explain why the scribes did not follow the same procedure in dividing stop (or Þ) + liquid sequences separated by a mor-pheme boundary, -P/#L-. Instead, they produced historical -/P(#)L-.

Again, Pierce's argument rests on a tacit premise, but—in this case—on a premise that is even more dubious than the one at the basis of the previous argument. Pierce assumes that the occurrence of such sequences containing a pre-liquid morpheme boundary was frequent enough that the scribes would have remained mindful of them and not have been motivated by habit to follow the general pattern presented by the majority of forms containing the sequence -PL- (-or ÞL-) where no morpheme boundary separates the stop (or Þ) from the liquid.

In the entire EDB, there is only one unquestionable example of a word divided before the stop (or Þ) + liquid sequence containing an intervening morpheme boundary: Arg EDB 340 framal/d[#]rozei, Lk 1:18f, the comparative nom.sg.fem. form of framaldrs ‘more advanced in age‘. This form is clearly related to ald ‘time, age‘ in the synchronic grammar of Gothic, and, therefore, exhibits a morpheme boundary im-mediately before the liquid /r/.

There are only two other candidates for such word division, and both are questionable. First, Feist (1939:577) relates the adjective (Amb A EDB 241) wul/Þ[#?]rais ‘of worth, consequence, gen.sg.fem.‘ to the noun wulÞus ‘glory‘. Second, Feist (1939:77) relates the noun (Amb B EDB 274) bai/t[#?]rei ‘bitterness‘ to the verb beitan ‘bite‘. These two forms, wul/Þrais and bai/trei, ostensibly contain morpheme boundaries from a diachronic point of view. Whether they do synchronically is questionable given the semantic divergence from their putative etymolo-gical roots. In any case, there is only one form, framal/d[#]rozei, which can be considered clearly probative of Pierce's argument. However, it is highly unlikely that a single form—or even three such forms—would motivate scribes to deviate from their standard practice of dividing obstruent + liquid clusters before the obstruent, in conformance with the Syllable Contact Law.

Let us consider these one to possibly three tokens of -/P(#)L- displaying word breaks regardless of morpheme boundaries with the number of tokens of the same structure where there is no intervening morpheme boundary. According to the EDB (table I, column 2 in Appen-dix), this total is 41. It is not unreasonable to assume that the scribes would be more likely to follow the general pattern reflected in the 41 cases than to deviate from that pattern for the one, or possibly three, exceptions.

3.4. The Older Germanic Dialects and Heterosyllabification of -CG-

Pierce argues that heterosyllabification of -CG- clusters is corroborated by the other older Germanic dialects and cites Frey (1989:288) without providing any data to support this conclusion (and neither does Frey, at least for the -CG- clusters). Nevertheless, it is true that the meter of Old Icelandic Stabreim is based on the heterosyllabification of -C.jV- (Murray & Vennemann 1983). However, this heterosyllabification in Old Icelandic Stabreim may be an artifact of the slow, deliberative method of recitation of that genre (Barrack 1998:67–80). In any case, the evidence from the other Germanic dialects lends support to Sievers‘ position, namely, that such sequences were tautosyllabified in Proto-Germanic (see Barrack 1998, chapters 3, 4). Space does not permit a thorough dis-cussion of the issue here.

Nevertheless, one of Pierce's (2006:282) assertions on this subject requires a response:

Given the context of this argument, the implication here is that word breaks in Old English also support the heterosyllabification of glide-final clusters, -C.j- and -C.w-. However, this is patently false: word breaks in the Old English manuscripts do not support such a premise at all. In her tally of Wetzel's (1981) compilation of these breaks on microfiche, Lutz (1985:234) examines 30,780 word breaks of the type -VC/CV- of which only 38 comprise -C/w-, viz., -l/w- (n = 7) and -r/w- (n = 31). Note that neither of these poses a serious violation of the Contact Law since both members of the contacts are of near equal strength.

Let me now turn to scribal breaks of the type -C/j-, the most germane to the present discussion as they make up 24.5% (509/2074) of the breaks in internuclear clusters in the EDB.Footnote ⁸ Lutz finds none to list—and for good reason. By dint of West Germanic Gemination, all such contacts were eliminated by the doubling of the first member of the contact and, in prehistoric Old English, the loss of the palatal glide: PG ⁺-VC_ajV- > WGmc. ⁺-VC_aC_ajV- > OE -VC_aC_aV-. For example, PG ⁺sitjan- > WGmc. ⁺sittjan- > OE sittan. The only exception to this historical development is presented by the weak contact -rj- in which /r/ failed to geminate, and there was no loss of the palatal glide, as in nerian (Brunner 1965:144f, 186ff.).

3.5. Putative “Errors” Support Tautosyllabification of -CG-

Finally, Pierce (2006:288) asks:

Indeed, there is ample evidence of such “errors.” In fact, most of the “errors” in the four manuscripts in the EDB appear to be precisely of the type Pierce claims should occur if -Cj- were indeed tautosyllabified. Moreover, it is ironic that Pierce cites the form twaddje in his attempt to demonstrate that the scribes divided words before glides even when the morpheme boundary was elsewhere: “Consider forms affected by Holtzmann's Law, such as twaddje ‘of the two‘ (where the word break/syllable boundary occurs before [sic] the j […]” (p. 288). However, Pierce's description of the word break in this word is false because twaddje occurs only once in the EDB and presents the only example of Holtzmann's Law that exhibits a scribal “error:” it is divided before the long stop, twa/ddje Skeireins EDB 45 (Bennett 1960:62; III.d.3).

One must ask: could this “error” be probative of other factors? Consider once again the other divided sequences displaying Holtzmann's Law (see table 2 above):Footnote ⁹

1. (5)

None of these forms provide any obvious reason to posit an internal morpheme boundary in a synchronic grammar of Gothic. Hence, it is not surprising that in dividing these forms at the end of lines the scribes fell back on their habit of dividing immediately before the glide where a morpheme boundary occurs in virtually every other such sequence in Gothic, except perhaps for ïz/w- discussed in notes 3 and 8 above.

If we now turn to twa/ddje, we find a completely different situation. There is indeed a morpheme boundary before the long stop as revealed in the other occurrences of the morpheme twa- ∼ twai- ∼ two- (see table 3; compare Braune/Heidermanns 2004:127.)

Table 3.

The declension of the cardinal twai ‘two‘.

Moreover, there is at least one other occurrence of the allomorph twa-, which, of course, bolsters the positing of a morpheme boundary after twa-, as in twalif ‘twelve‘. Hence, the word break twa/ddje appears to serve as good prima facie evidence of morphological influence on scribal practice. Where no clear morpheme boundary is discernible in the forms resulting from Holtzmann's Law, the scribes fell back on the only well-established pattern: breaking the word before the glide.

Of course, the influence of sonority via the Contact Law may also be a contributing factor in the division of twa/ddje. Moreover, there are other “errors” that provide evidence of the tautosyllabification of glide-final clusters based mainly on such influence. “Errors” of this type—if indeed there are such—would also evince premature word breaks, that is, breaks made earlier than expected in the glide-final sequence; in other words, breaks reflecting tautosyllabified -.CGV-.

Indeed, if we consider only premature breaks, we find that over 73.7% (14/19) take place in internuclear glide-final clusters (see forms in bold face in table 4). In other words, the proportion of premature breaks

Table 4.

Premature word breaks in the EDB.

in glide-final clusters is significantly higher than the number of pre-mature breaks in nonglide-final clusters: a chi-square test conducted on the two-by-two contingency table of the data (table 5) reveals that it is highly unlikely that this difference arose by chance (p = 0.00015).Footnote ¹¹ This corroborates the hypothesis that in these instances the scribe was uncon-sciously motivated by considerations of consonantal strength, yet had no established scribal tradition to appeal to and was unsure of precisely where to make the break earlier in the sequence.

Table 5.

Early vs. regular breaks in glide vs. nonglide-final clusters (percentages are column percents).

Three of the remaining five cases of “errors” are also interesting from the point of view of consonantal strength: Whereas Amb B EDB 304 Þau/rftais and Skeireins EDB 133 Þwai/rheins appear to be errors, pure and simple, the other three “errors” also could very well reflect the influence of consonantal strength since all three are divided before one of the strongest fricatives, /s/, and a following weaker consonant: Arg EDB 337 hun/slastadis, Skeireins EDB 61 gareh/snais and EDB 137 ain/shun.

The fundamental weakness of Pierce's analysis stems from his fail-ure to view the scribal practices in their totality—a totality which reveals itself only if one studies the so-called errors. The scribes divided words across consonant clusters in accordance with consonantal strength except where ubiquitous consonant-initial suffixes terminated words. Thus, the scribes followed the Contact Law faithfully, dividing words between consonants except where the strength differential was relatively greater, as in the case of stops (or Þ) followed by liquids (and ignoring the one to possibly three cases where a morpheme boundary was present). In such cases, they usually heeded consonantal strength and divided words before the obstruent. The same tendencies are revealed in 17 of the 19 so-called errors representing premature word breaks. The only significant deviation from the sonority scale is the regular word division before ubi-quitous glide-final clusters, where a morpheme boundary precedes the glide in at least 88.4% of the breaks.Footnote ¹²

In conclusion, it appears that Sievers‘ (1892:22f.) original assertion that internuclear clusters ending in the palatal glide were tautosyllabified in Germanic has withstood the test of time.