2.1. Assuming Grammaticality in 9th-Century German

Generative syntax is a framework that relies heavily on native speaker intuitions, which presents challenges for the study of historical varieties. Crucially, this view of syntax disconnects the token (or instance of production) from the underlying grammar (or competence) that generated it. The fact that a particular structure is attested in itself does not constitute evidence of its grammaticality; tokens are instead established as accurate reflections of underlying competence, that is, as grammatical, only through the intuitions of native speakers. This method cannot be applied to historical varieties. This notion of grammaticality means that historical linguists must always be aware that their data are imperfect reflections of competence. Distorted competence must be considered a very real possibility for 9th-century German data: All of its major texts are either translations of a Latin source or works of poetry. Received wisdom suggests that one should treat the data with care.

To date, I have encountered in the literature two strategies to distinguish the “good” data from the “bad” in 9th-century German syntax. The first dates back to Ruhfus (Reference Ruhfus1897) and applies to works of translation, such as Tatian’s Evangelienharmonie; this strategy only analyzes structures that deviate from the source. Thus, the unacceptably compromised data are those for which one cannot rule out the possibility of syntactic transference. Identifying bad data in a poetic text also involves comparison, in this case performed against 9th-century prose texts rather than a Latin source. For example, based on the observation that asyndetic verb-late clauses occur more frequently in Otfrid than in Tatian or Isidor, Axel (Reference Axel2007:77–78) concludes that the structure is extragrammatical.Footnote ¹¹ Axel’s logic is based on the assumption that the prose of the 9th-century—even if it is a translation—is more grammatical and a better reflection of actual competence than its poetry. So uncontroversial is this conclusion that scholars often do not defend or even explicitly express it, as is the case in Axel’s (Reference Axel2007) study of OHG, which is based on a corpus of prose texts (especially pp. 3, 16–22).Footnote ¹²

However, the question is where this view comes from. I argue that the reasons behind the prevailing assumption that 9th-century poetry is less grammatical than 9th-century prose are faulty and do not constitute actual evidence of the ungrammaticality of poetry. The first is the discomfort scholars feel about the variation in surface order patterns in texts such as the Evangelienbuch. Robinson (Reference Robinson1997:2–3), for example, chooses to examine Isidor over Otfrid, because a more variable verbal syntax makes the latter text difficult to work with. I am less inclined to believe that Robinson does not like a good challenge than I am to conclude that he is equating more regular, easily identifiable syntactic patterns with grammaticality. Unless one wants to maintain that, for example, written standard languages, with their regularized syntax, are inherently more grammatical than nonstandard oral varieties, one must reject this equivalence.

More generally, it seems this assumption stems from modern notions of the relative grammaticality of poetry and prose. Most modern literates have been exposed to verse whose lyrical modes of expression are removed from our daily idiom, while prose seems closer to everyday discourse.Footnote ¹³ However, it is important to remember that early medieval writing occurred in a fundamentally different environment—one in which a new Latinate literacy was developing, while vernacular culture was still overwhelmingly oral (see, for example, Green Reference Green1994:47, 49; Edwards Reference Edwards1994:141)—and its prose and poetic expressions are also different from modern prose and poetry. The product of the oral tradition, Ong (Reference Ong2012:34) reminds us, was crucially poetic, not ornamentally so. Poetic language was mnemonic language: It was rhythmic, it alliterated or had assonances; it was rooted in patterns of speech, more planned and formal than everyday discourse, but thoroughly dependent on it as well. That is, any poetic language that deviated significantly from more spontaneously produced language would tax the memory. Otfrid is not bound by these cognitive constraints, but he is still composing verse in a world in which mnemonic, or poetic, language is the primary vehicle of cultural memory, the public form of the vernacular.Footnote ¹⁴ The poetic scene in 9th-century Weissenburg was vastly different from a modern one, in which poetry occupies a more peripheral cultural and linguistic space.

Just as distinct from its modern counterpart is 9th-century prose. In the OHG corpus, there are only a few short examples of vernacular prose writing, such as the Wessobrunner Gebet. Practically all prose works are translations of Latin texts, in which the possibility of syntactic distortion is obvious. This is particularly true for Tatian, a mostly line-by-line translation of Latin. However, the question is whether or not the same applies to Isidor, which scholars see as an excellent translation, or “our best early source of Old High German prose” (Robinson Reference Robinson1997:2). Is this text the best snapshot of a 9th-century speaker’s competence? The answer must be no: Prose writing is the default for the modern literate, but only after years of schooling in the standard language. In contrast, prose vernacular writing in 9th-century German was a completely new phenomenon, while poetic vernacular writing—also new and surely challenging in that the poet must find ways to visually represent what had existed previously only as sound—at least had an autochthonous oral tradition to draw on. Producing a vernacular prose translation required consciously developing a written language appropriate for the task. Matzel (Reference Matzel1970:517, 519; see also Green Reference Green1994:45 for a summary) describes the Isidor translation in this way, noting that Latin provided a model for how a written language ought to look with a regularized grammar and orthography. In sum, its syntactic system is a constructed one, informed by the vernacular but also necessarily deviating from it. It is a fallacy to assume Isidor is a more accurate reflection of a natural native speaker competence than the Evangelienbuch.

To be clear, I do not argue here that 9th-century poetry is the better source of grammatical data. I only point out that discounting its structures on the grounds that they are not attested in 9th-century prose is based on dubious assumptions. The extant prose texts are not an appropriate stand-in for native speaker judgments, and the potential for the distortion of competence is present in either genre. I submit that the only real strategy historical linguists have is to try to identify and explain the patterns attested in the data, which must include analysis of the potential influence of pragmatic factors. Given the small size of the corpus, it is more productive simply to accept all data as grammatical, unless a compelling empirically grounded, text-specific case can be made for their pure artificiality. Whether there is compelling evidence of the asyndetic verb-late clause’s artificiality in Otfrid is the topic of the next section.

2.2. Accepting the Grammaticality of the Asyndetic Verb-Late Clause

In this section, I argue against the characterization of asyndetic verb-late clauses as poetic phenomena, which can be found even in basic descriptions of OHG syntax, such as Schrodt’s (Reference Schrodt2004:204) Althochdeutsche Grammatik II: “Die Späterstellungen bei Otfrid sind reimbedingt und damit keine aussagekräftigen Belege” [Cases of verb-late are conditioned by rhyme and, thus, are not meaningful pieces of evidence]. This conclusion is often accompanied by an example or two, like the one in 9.

1. (9)

   

According to Schrodt (Reference Schrodt2004), the verb-late placement of ríat simply mirrors the subordinate clause structure in the b-verse. Note that data such as these do not actually preclude the structure’s grammaticality: The verb-late placement could be grammatical and satisfy the end-rhyme scheme simultaneously. A more accurate phrasing would be that, unless one can exclude the possibility that end-rhyme influenced syntax, one cannot conclusively state that the verb-late syntax of the clause in the a-verse is the reflection of a speaker’s underlying competence.

Axel’s (Reference Axel2007:70) statement is stronger: Because instances of verb-late syntax generally satisfy poetic exigency, none should be accepted as grammatical. However, if verb-late syntax is only there for poetic purposes, most, if not all, asyndetic verb-late clauses should have rhyming finite verbs. The data clearly show that verb-late verbs are not confined to rhyming position: Of the dataset’s 73 asyndetic verb-late clauses, 20 clauses, which is over a quarter of all tokens, do not feature finite verbs that rhyme:

1. (10)

   

Among these 20 clauses there are 11 asyndetic verb-late tokens that function as main clauses, as in 10a, and 9 that function as dependent, as in 10b.Footnote ¹⁵

As discussed in section 1.1, it is the verb-late main clause that is most relevant to the debate between Lenerz (Reference Lenerz1985) and Axel (Reference Axel2007). That is, the verbal syntax of tokens such as 10a should be attributable to extragrammatical pressures. In this case, however, it is clear that end-rhyme is not to blame, as the finite verb is not in verse-final position. Strengthening the argument that the verb’s placement was not a poetic choice is the fact that a verb-second configuration would have yielded the exact same metrical cadence, as shown in 11.

1. (11)

   

Though the source of Otfrid’s metrical cadences is unclear, it is generally accepted that he aimed for an undulating lift-dip beat structure, with the verse’s dips held to one syllable (see Bostock Reference Bostock1976:210 for more details). In this instance, however, the two variants are identical. The only other possible motivation for the verb-late placement—outside of it being a grammatical option, that is—is a phonotactic one: Perhaps Otfrid wanted to avoid having two rhotics in a row.Footnote ¹⁶ There is evidence against this argument as well. Namely, it is not difficult to find er+r-sequences elsewhere in the Evangelienbuch: Er ríchisot githíuto (I 5 29); Er ríat thaz man biwúrbi (III 26 27a); so er rúarta imo thaz óra (V 17 23b). It is unlikely that Otfrid would produce an ungrammatical structure simply to avoid a phonotactic sequence attested elsewhere.

As was the case in 10a, there is no evidence that the metrical cadence influenced verbal syntax in 8 of the remaining 10 verb-late main clauses. These clauses show one of the following: They have metrically viable verb-second alternatives, but Otfrid opted for the verb-late placement instead (L 89; I 1 3; I 1 8); they do not conform to their metrical cadence but also have no metrically viable verb-second alternatives (L 19; L75; S 5; III 2 29-30); they do not conform as attested, but could have, had the poet opted for verb-second syntax and cliticization, a common process attested throughout the Evangelienbuch (Somers Wicka Reference Somers Wicka2009:87–99).Footnote ¹⁷

1. (12)

   

Example 13a is a made-up sequence with verb-second (and pronominal encliticization), whereas 13b,c are similar verb-second examples.

1. (13)

   

Note how the imagined verb-second version of 12 in 13a eliminates the extraneous unstressed beat. Only two clauses (L 91 and I 1 13) show a conformance to their cadence that would have been undone by the verb in second position. These two clauses might matter more if there were evidence that Otfrid was loath to violate a strict lift-dip rhythm, but this was clearly not the case: Of the 516 lines and 1,032 verses in my dataset, by my count only about 57% of verses conform to their metrical cadence; 47% do not. Thus, metrical conformity in Otfrid is better described as a tendency, rather than an imperative. Though the grammaticality of asyndetic verb-late dependent clauses has not been called into question in the literature, it is worth noting that the dropping of thaz ‘that’ also does not seem to be conditioned by the meter. Below is an example of how the line could have incorporated a clause-initial complementizer—as the oft attested clitic group theiz ‘that it’—into the exact same cadence.Footnote ¹⁸

1. (14)

   

Once again, metrically viable alternatives are available to Otfrid that would have allowed for a clause that appears more canonical.

With no access to native speakers, there is no way to conclusively demonstrate that asyndetic verb-late clauses were grammatical—as defined in generative syntax—in Otfrid’s variety of South Rhenish Franconian. What can be said is that there is no evidence supporting Schrodt (Reference Schrodt2004) and Axel’s (Reference Axel2007) assertions that the structure is created entirely by rhyme and meter. I have also argued in this section against anachronistically assuming that Otfrid’s is a poetic syntax in the modern understanding of the phrase, namely, that it exhibits an idiom-stretching lyricism that routinely creates clauses so different from everyday discourse that one must understand them as adhering to a different set of grammatical rules. This conception of 9th-century German vernacular poetry ignores the early medieval cultural context in which it was produced. It also ignores Otfrid’s own words found in the preface (Ad Liutbertum), in which he discusses his decision to opt for Frankish idiom even when it conflicted with prescribed Latin norms: “ut morum se locutio praebuit, dictare curavi” [I have taken pains to write as customary speech has showed itself (to be)] (translation from Magoun Reference Magoun1943:885). Thus, he produced clauses that contained structures such as double negation, “ob usum tamen cotidianum” [on account of everyday usage] (Magoun Reference Magoun1943:885). The general didactic purpose of the poem (section IV) is to make the Latin Gospels more accessible to monolingual Frankish speakers so that one could “in his own language become familiar with the most holy words …, understanding in his own language the Law of God” (Magoun Reference Magoun1943:876). Though Otfrid discusses here Latin influence rather than poetic influences, the point is still relevant: The poet aimed to produce idiomatic language that less educated Franks could understand and thus consciously chose structures from everyday discourse.

I move forward, then, with the understanding that asyndetic verb-late clauses are grammatical, even idiomatic, that they are produced by the same grammar that yields other more canonical-seeming clauses, not some poetic grammar. In the next section, I discuss how this clause type can be functional within a clausal system, that is, used as main or dependent, even though its verb is always final and complementizer absent.

3.1. Distinguishing Asyndetic Verb-Final Main and Dependent Clauses

Holler (Reference Holler2008:190–91) demonstrates the difficulties in distinguishing main and dependent clauses even in Modern German: Many clauses are easy to categorize, but noncanonical clauses, such as Gärtner’s (Reference Gärtner2001) V2 relative clauses (see note 4), undermine the idea of a neat binary distinction. Unfortunately, many of the pragmatic and grammatical criteria that could identify clause type in modern languages will not be effective in analyzing the asyndetic verb-late clauses of OHG. The formal cues—verbal syntax and presence of complementizer—obviously cannot elucidate their status. Neither does information structure, if one accepts Schlachter’s (Reference Schlachter2012:144) conclusion that verb-late clauses, be they asyndetic or not, fulfill the same discourse function of providing background information. Features like intonation and prosody are unrecoverable for a historical variety. Instead, I rely mostly on “interpretational aspects” Holler (Reference Holler2008:191) and look to the context of the asyndetic verb-late clause to assess whether or not there is a logical connection between it and surrounding clauses, one that could be made explicit through the addition of a complementizer.

Looking to context allows for the categorization of asyndetic verb-late clauses into two groups: those that function as arguments in a preceding clause, as in 15, and those that do not, as in 16.

1. (15)

   

1. (16)

   

The clause in 15a is an argument of the preceding predicate weiz, while the asyndetic clause in 15b with the predicate bigínnen connects to correlative thaz and explains what the Franks have been forgoing. The second asyndetic clause, in turn, is an argument in its preceding clause. In contrast, 16a is the work’s opening clause and so can have no connection with its preceding clause. The clause in 16b connects to preceding discourse, but it is impossible to imagine a dropped comple-mentizer thaz, especially with an initial sar signaling the new and emphatic information that follows. All in all, there are 45 asyndetic verb-late clauses that can be interpreted as main, 28 as dependent.

The two categories exhibit surface similarities but are different with respect to grammatical mood. Beginning with the surface similarities, a verb-third analysis of this same dataset (see Somers Reference Somers2018:90) identifies the absolute position of finite verbs in asyndetic main and dependent clauses. Those data are represented in table 2.

Table 2. Verb position in asyndetic verb-late clauses.

These distributions are not significant (p=0.0915).Footnote ¹⁹ Somers (Reference Somers2018:92) also identifies the types of sentential constituents that occur before the finite verb in main and dependent asyndetic clauses: Both show a wide array of prefinite constituents, including subject and object pronouns, adverbs, noun phrases, and prepositional phrases. The only identifiable difference is that clauses in the former category tend to be in the indicative mood, the latter in the subjunctive (table 3).

Table 3. Mood differences.

These distributions are significant (p<.00001). This strong correlation points to two conclusions. First, it suggests that Otfrid is sensitive to the distinction between main and dependent clauses, and second, that the subjunctive mood was one way to signal this distinction. The fact that the subjunctive mood serves as the only characteristic distinguishing the asyndetic dependent clauses from the main implies that Otfrid saw dependency as a condition that he could mark through various means, not just through verb-late syntax and overt complementizers.

3.2. The Subjunctive Mood as a Marker of Dependency

Comparing asyndetic dependent clauses to clauses marked as dependent through a verb-late syntax combined with thaz strengthens the argument that Otfrid used the subjunctive mood as another dependency marker. That is, asyndetic dependent clauses show a higher rate of subjunctive verbs than do dependent clauses with complementizer thaz.

Table 4. Thaz versus asyndetic dependent clauses, part I.

These data indicate that Otfrid was more likely to mark a dependent clause with the subjunctive when the dependency marker thaz was absent. However, the number of tokens is too small to establish this preference as significant (p=0.0919). Expanding the dataset to include all relevant tokens from two new chapters—Ad Hartmuat et Werinbertum and Book I, chapter 5—adds 19 tokens: 16 thaz-clauses and three asyndetic dependent clauses. The expanded dataset suggests that a similar preference may hold throughout the work.Footnote ²⁰

Table 5. Thaz versus asyndetic dependent clauses, part II.

These results are significant (p=0.0266) and indicate that the subjunctive mood alone could mark a clause as dependent. Otfrid is sensitive to a main-dependent clause distinction and provides for its grammatical marking in all clause types.

That Otfrid’s use of a complementizer to mark dependency is not invariable points to the conclusion that he had various means by which to signal it and sometimes used one, sometimes more than one to do so. It is still unclear why Otfrid opted for the asyndetic variant when he did. I am skeptical that the meter created the phenomenon (section 2.2), though it could be that the poet opted for the perfectly grammatical asyndetic dependent clause when he thought it achieved a more desirable rhythm. There is also no evidence that Otfrid used asyndetic dependent clauses in instances of indirect speech, as Axel-Tober (Reference Axel-Tober2012:189–191) claims: This dataset exhibits more variation in the types of predicates that take asyndetic dependent complements:

1. (17)

   

1. (18)

   

Some of these predicates can introduce complements containing indirect speech, for example, quedan and sagên, but many others do not, such as wesan adeilo and biwankôn. Wunder (Reference Wunder1965:194–197) also places verbs such as gibietan, fergôn, and bitten in a different category: Though they are verba declarandi, whose dependent clauses express a pronounce-ment, they generally do not introduce indirect (or direct) speech in Otfrid, for example: Emmizen nu ubar ál ih druhtin férgon scal, mit lón er iu iz firgélte joh sínes selbes wórte ‘I shall always ask the Lord by all means (that) he may bestow unto you in accordance with his promise’ (S 17–18). Wunder identifies additional predicate types, including verbs of believing or thinking (gilouben, wizzan, and drahtôn), verbs of preparing, doing, or the absence of the same (flîzan, biwankôn, and wesan adeilo), which are not associated with indirect speech.

An easier question to answer is why Otfrid would use the subjunctive to indicate dependency; this association makes particular sense in light of similar patterns in Ecclesiastical Latin (see Collins Reference Collins1985:224), the one (written) language in which Otfrid was educated.

1. (19)

   

In Ecclesiastical Latin, the subjunctive could still convey that the proposition of a clause was hypothetical, but also functioned as the “mood of subordination” (Collins Reference Collins1985:224) or a grammatical marker of subordination. The indirect and direct questions in the examples have the same indicative semantics, and the subjunctive mood in 19b instead signals the “syntagmatic dependence,” to use Schrodt’s (Reference Schrodt2004:187) term, of the dependent clause.

German still shows traces of a correlation between the subjunctive and dependency, but it is unclear whether or not this early association reflects a native speaker’s intuition or if it was borrowed into early written German from a Latin model. It is easy to imagine both being true: There could have been a preexisting association in the vernacular, but any systematicity in the recognition and grammatical signaling of dependency markers strongly implies a Latin source. The syntax of today’s spontaneous spoken language exhibits less grammatical subordi-nation, less embedding, and more fragmentation than its highly developed written standards. Speakers link clauses together, though less tightly and not exclusively through formal grammatical markers (Miller & Weinert Reference Miller and Weinert1998:20–24). Given the fact that 9th-century Frankish was an overwhelmingly spoken phenomenon with hardly any written tradition and no standard form, it is highly improbable that the Evangelienbuch is simply capturing a wholly autochthonous, systematic, formal gram-matical marking of dependency that was not influenced by the one model of a written language known to the poet, in which this systematicity was long established. This section’s evidence of subjunctive as one of several dependency markers demonstrates that Otfrid had a sensitivity to the main-dependent clause distinction but not a fully regularized way of signaling it.

4.1. The Information Structural Account of Asyndetic Verb-Late Clauses

Though Lötscher (Reference Lötscher2009) and Schlachter (Reference Schlachter2012) both argue that information structure influences OHG clause structure, they each identify different patterns in their data and, thus, analyze verb-late main clauses differently.Footnote ²³ I begin with Lötscher’s (Reference Lötscher2009:313–316) description of these clauses in Otfrid. He asserts that an overwhelming number of them begin with a heavy, stressed constituent, which should be a “contrastive or new topic,” which is followed by at least one weak, anaphoric constituent that functions as a “continued topic,” or “background.” Weak constituents attach enclitically to the preceding heavy element. Lötscher’s example 38c, reproduced in 20, illustrates this configuration.

1. (20)

   

Lötscher’s assessment of druhtin Kríst as heavy seems to be based on the constituent’s status as a full determiner phrase/noun phrase (DP/NP); he does not explain how it functions as a contrastive or new topic within discourse.Footnote ²⁴ These main clauses are no different from other main clauses with the finite verb in clause-initial position, which Lötscher argues is the basic position of the finite verb in the main clause (p. 312). A constituent may but need not be fronted. If it is preposed for grammatical reasons—as opposed to information structurally motivated ones—fronting is restricted to one constituent. However, more than one constituent may be fronted for pragmatic, or “information structurally motivated,” reasons. Verb-final main clauses fall into the latter category: The late position of the verb is due to this process of “expanded fronting,” through which a heavy element is preposed and the light element (or elements) is placed in second position, in accordance with Wackernagel’s Law. This process yields trivially verb-late clauses.

Yet, the particular arrangement of heavy and light constituents that suggests expanded fronting in Lötscher’s analysis is attested less frequently than other configurations in my dataset: Among my 45 verb-late main clauses only 17 (less than 38%) are consistent with the expanded fronting analysis, while 28 (62%) are not.Footnote ²⁵ Of the 28 clauses that are inconsistent, 23 clauses (51%) have initial light elements, as in 21a—the pattern that Lötscher characterizes as exceptional, and 4 clauses have more than one heavy constituent preceding the finite verb, as in 21b.Footnote ²⁶ Finally, one clause has an initial discontinuous constituent, as in 21c.Footnote ²⁷

1. (21)

   

Note that the light pronominal iz in 21a could have been placed after the heavy constituent iuer húgu with no disruption to the end rhyme, which undermines the claim that poetic considerations drive such orderings. Lötscher provides no evidence, nor discusses the claim, that verb-late main clauses with initial light constituents are particularly affected by meter and rhyme. It would be impossible to treat óstarrichi in 21b as a light, enclitic, anaphoric constituent, particularly as this is the work’s opening sentence. The pronoun ih’s placement into a linear/surface verb-second position creates a discontinuous constituent in 21c; it is difficult to see how the discourse value of constituents can be crucial in driving verbal syntax, when the heavy new topic is divided in half.

Lötscher (Reference Lötscher2009:316, note 40) suggests that data such as 21a, which has an initial light element, could still fit into the expanded fronting hypothesis. Consider the example in 22.

1. (22)

   

Here the argument becomes circular. The author states that the pragmatic function of sie in 22 is unclear and that “we could assume … that the pronoun has a special topic function within the context,” which means the clause would be consistent with the expanded fronting hypothesis. Lötscher presents no evidence that sie is a heavy, new, or contrastive topic. In fact, such a reading seems impossible: In the chapter’s first 14 lines Otfrid notes that God the Father and Son have always existed and then observes that they both created everything together; indeed sáman, which itself occurs preverbally—though not initially—is the most likely candidate for new topic emphasis. Beyond the assertion that there is a predominant heavy-light constituent pattern in verb-late main clauses, there is no reason to assume that sie is a heavy topic.

I can only speculate on why Lötscher’s (Reference Lötscher2009) pattern for asyndetic verb-late clauses is not widely attested in my dataset. Lötscher’s analysis presents hardly any numbers: The author never says how many clauses he looked at for his study, only that examples were drawn mostly from Book II (p. 319). The absence of indications that the analysis is based on a defined dataset undermines any statements made regarding frequency. One would need to know, for example, how many verb-late main clauses there are in Book II and what proportion of those clauses exhibits the heavy-light pattern before accepting conclusions on the frequency of that pattern. Such presentations of data are absent in the work, which implies that its conclusions are not empirically grounded and instead based on a general impression of the data. In contrast, the current study draws on a defined dataset (see section 1.2), consisting of clauses taken from multiple samples of continuous text. Thus, its conclusions are more reliable.

Now I turn to Schlachter’s (Reference Schlachter2012) information structural analysis of asyndetic verb-late main clauses, which diverges from Lötscher Reference Lötscher2009 in how it distinguishes verb-third from verb-final clauses. That is, Schlachter argues that these configurations are distinct with respect to discourse structure. The study also draws its data from a different text, the OHG Isidor. A quick note on the study’s data presentation: Schlachter maintains that translations of the Biblical passages differ stylistically from the rest of the translation, referred to as the “treatise.” Thus, she keeps data from each separate, a distinction I maintain in the discussion below. Schlachter also includes in the study’s dataset St. Matthew’s Gospel from the Monsee-Vienna Fragments, which she argues were translated by someone else (pp. 20–25). In what follows, I demonstrate that, as was true for Lötscher Reference Lötscher2009, the patterns she asserts for Isidor are not useful for understanding those of Otfrid.

Schlachter (Reference Schlachter2012:136–138, 156) identifies two syntactic configu-rations that she argues connect to distinct discourse structures. The first of these is verb-third (XP-XP-Verb-XP), which comprises “mostly copula constructions with a preverbal adverb” (p. 156). These clauses generally begin with a thematic constituent—one that connects to the preceding discourse “in the broadest sense” (p. 144)—and a focused constituent follows. I reproduce Schlachter’s (Reference Schlachter2012:137) example 143:

1. (23)

   

Schlachter claims this pattern for verb-third clauses in the treatise translation, though she never states how many of its 11 unambiguously verb-third clauses are copula constructions with initial thematic constitu-ents and focused adverbs. Verb-third is poorly attested among the nontreatise clauses in Schlachter’s dataset with only four tokens, though these few tokens mostly exhibit the opposite discourse structure with the focused constituent in clause-initial position and tend not to be copula constructions with preverbal adverbs (p. 138).

There is additional confusion surrounding the information structure that actually characterizes verb-third clauses in Schlachter’s analysis. Namely, the configuration in the various summaries following its initial presentation in section 4.3.1 becomes primarily associated with a focus-initial discourse sequence. Consider the final summary and discussion (p. 209), in which the author asserts that verb-third is “typically” characterized by focus-initial clauses. This conclusion is surprising, as the author never states how many Foc-Top-Verb configurations are attested among the 11 unambiguously verb-third treatise clauses, noting only that such constructions occur more frequently in the Biblical passages and St. Matthew’s Gospel. Given that the Biblical passages’ sole verb-third token is focus-initial, as are two of three verb-third clauses in St. Matthew’s Gospel (p. 138), one can deduce that two, perhaps three, of the eleven treatise clauses are also focus-initial; the rest presumably begin with a thematic constituent, which is followed by the focused constituent.Footnote ²⁸ Given the absence of numbers, however, one can only surmise how many unambiguously verb-third clauses are associated with either discourse pattern. There is no clear empirical demonstration of what verb-third clauses are.

I reach the same conclusion for Schlachter’s (Reference Schlachter2012:129, 144, 153–156, 209) second distinct information structure, which connects to unambiguously verb-final (XP-XP-XP-Verb), or “V-end” clauses. Unlike unambiguously verb-third clauses, which have preverbal focused constituents that sometimes occur in initial position, verb-final clauses never have initial focused constituents. Instead, they begin with a thematic constituent, which can be nominal or pronominal. They contain only familiar information and continue the narration/argument (or provide parenthetical metacommentary), rather than conveying new information. However, Schlachter (Reference Schlachter2012) is not explicit about how many of its 16 unambiguously verb-final clauses (14 treatise clauses, 2 non-treatise) actually fit this description. For example, there is no empirical demonstration that initial constituents tend to be thematic, despite the reference on page 140 to a set of 13 tokens that begin with a “d-pronoun.” Based on the examples provided (pp. 140–141), clause-initial d-pronouns include thematic time adverbials, such as dhar after ‘thereafter’, but also prepositional phrases, such as umbi dhen samun ‘about the/this seed’ and full DP/NPs, such as dhazs himilsca folc ‘the heavenly people’. As is evident in the Otfrid data below, one cannot assume such constituents are discourse-given, as Schlachter implies. Also problematic is that one of the few examples provided is taken from the study’s set of ambiguous clauses (example 153a, p. 140). Because the author does not explicitly identify the 13 tokens as unambiguously verb-final, it is unclear whether they all belong to this category, or whether this number includes some ambiguous clauses that were later analyzed as verb-final.

The question of whether or not Schlachter (Reference Schlachter2012) maintains analytical separation between the unambiguous and ambiguous categories is consequential: The work uses a hypothesized correlation between syntax and information structure to disambiguate sequences it has identified as ambiguous, such as XP-XP-Verb, in which the verb is both verb-third and verb-final (pp. 144–147). The argument runs the risk of circularity if one does not first establish that a correlation exists in the unambiguous clauses. Without this first crucial step, one only has asserted associations that are then used to sort ambiguous data, a process that can create the appearance of a pattern where none is present. This sort of argumentation is present in Schlachter’s analysis of the verb-third/-final ambiguous clauses: She asserts, but does not demonstrate, an association between focus-initial discourse structures and verb-third, and a negative association between this information structure and verb-final. These assumed associations then provide the means of categorization: Any ambiguous clauses with initial focused constituents must be verb-third and cannot be verb-final. The work does not state exactly how many of the 24 (or 22) ambiguous clauses are disambiguated in this way, but the process and its results must be viewed with caution.Footnote ²⁹

Beyond these issues, there is little evidence that Schlachter’s (Reference Schlachter2012) information structurally defined syntactic categories are relevant for Otfrid. Beginning with the verb-third pattern (XP-XP-Verb-XP), recall that Schlachter identifies these as mostly copula constructions with a preverbal adverb that is focused or “focuses” the other preverbal constituent, while being itself thematic.Footnote ³⁰ Of the 45 asyndetic verb-late main clauses in my dataset, 10 fall into Schlachter’s verb-third category.Footnote ³¹ None of the 10 is a copula construction; only two exhibit a preverbal adverb (II 8 8 and II 8 44). Turning to the contention that verb-third is associated with focus-initial clauses, this does not appear to be true for Otfrid, whose clauses with the XP-XP-Verb-XP sequence exhibit no unanimity in their initial constituents: Five of the ten clauses have initial pronouns, which, Lötscher’s analysis notwithstanding, are generally understood as thematic.Footnote ³² The remaining five begin with full DP/NPs or prepositional phrases, prosodically heavy constituents that are more likely to be focused.Footnote ³³ None appear to have any relevant antecedents. In sum, I find no evidence of an information structurally defined verb-third clause in Otfrid.

The same can be said for the verb-final (XP-XP-XP-Verb) pattern, which Schlachter (Reference Schlachter2012) associates with (initial) anaphoric constituents; these clauses should only convey familiar information rather than new information. There are 22 main clauses in my dataset that exhibit the unambiguous verb-final pattern.Footnote ³⁴ Five (23%) contain only thematic or anaphoric preverbal constituents.Footnote ³⁵ Preverbal constituents here include subject, object, and indefinite, as well as anaphoric adverbials, such as thar ‘there’. In contrast, 10 (45%) of the 22 clauses contain preverbal constituents with no apparent antecedents whatever in their chapters.Footnote ³⁶

1. (24)

   

In this chapter, as shown in 24a, Jesus asks his disciples who people think he is. They respond that people have mistaken him for John, among others. This reply, and the constituent John in particular, represents new information within this context. Example 24b illustrates that constituents with definite determiners, Schlachter’s “d-pronouns,” are not automa-tically discourse-given. Thaz méz has no earlier reference in the chapter—this discussion of measurements and the dimensions of the jugs constitutes new information. Instead, the definiteness of the noun empha-sizes this one particular measure to which Otfrid introduces the reader. In colloquial English one might say something like: There is this measurement we call a sester. In my dataset, only five tokens begin with a d-pronoun, all of which are full DP/NPs.Footnote ³⁷

I find the remaining seven clauses difficult to assess in a way that is consistent with Schlachter’s (Reference Schlachter2012) analysis.Footnote ³⁸ This difficulty arises in part from the nature of the clauses themselves: Their preverbal constituents include some obviously discourse-given, anaphoric elements, but also full DP/NPs, whose status remains unclear to me. Rigorous analysis of these clauses is hindered by the fact that Schlachter’s work does not adequately lay bare its methodology for assessing whether constituents are truly thematic. As discussed, isolating initial d-pronouns is not a satisfactory means for identifying thematic constituents. Schlachter’s section 4.4.2 presents one possibility for assessing whether a clause continues an argument or narrative sequence. Namely, it looks at whether verb-final main clauses occur at the beginning, middle or end of text sections, presumably building on the assumption that new information is conveyed in section-initial—not in section-internal/-final—clauses. This cannot be a satisfactory method for Otfrid and its much longer chapters. Even if I assumed that all seven clauses were consistent with Schlachter’s discourse analysis—which seems unlikely to me—the data would then have 12 clauses that have the predicted information structure, 10 that do not. There is no evidence that syntactic pattern correlates with information structure.

4.2. Light and Heavy Verbs in Otfrid

The previous section demonstrated that existing information structural accounts of asyndetic verb-late clauses in OHG do not elucidate the current data. Yet it is difficult to relinquish Behaghel’s notion that prosody influenced constituent ordering in the early Germanic clause, more so than more modern conceptions of the clause might allow for. In this section, I present evidence of a correlation between prosody and syntax by focusing on the finite verb. Hopper (Reference Hopper1975:52–57) provides the context for this analysis. The goal of his work is to reconstruct the basic clausal patterns of Proto-Germanic by comparing the strikingly similar surface patterning of the early Germanic languages (pp. 13–14).

In Hopper’s Germanic clause, there is no assumption of an implicational relationship between verbal syntax and clause type—main or dependent. Instead, he asserts that clauses may be independent of, or integrated into, their preceding discourse, and he connects each discourse status to different syntactic configurations and verb weight. Verb-second and verb-final may occur with either discourse type, and neither is crucially defined as main or dependent:

1. (25)

   

The only predicted correlation here is between verbal syntax and verb weight: Light verbs tend to occur in clause-second position, heavy verbs—in clause-late; verbal syntax does not particularly correlate with the main/dependent distinction. That is, it is the weight of the verb that matters, and the attraction of light verbs to second position—an expression of Wackernagel’s Law—can occur in either context. Heavy verbs, in contrast, tend to remain in clause-late position, unless the pragmatic context is emphatic, in which instance they can appear in clause-first position.

Because the distinction between light and heavy verbs is important, let me define them more precisely. According to Hopper (Reference Hopper1975:56), verbs “of light weight” comprise one or two syllables and have little or no lexical content. Semantically the light verb contributes little to the clause. I understand prosodic lightness as following from the lack of lexical content in that verbs that are lexically empty do not require a lot of phonetic material to distinguish them from other lexical items. Thus, light verbs are defined primarily by their lexical lightness. A copula verb is the classic example of this class: Its various conjugations are short with prosodically light syllables, and it is so devoid of lexical content that some linguistic varieties, like African American Vernacular English, drop it entirely. Hopper identifies other types of verbs as being of low lexical yield, including auxiliaries and “certain unemphatic verbs” (p. 19), such as “verbs of making, writing and dedicating in inscriptions and verbs of saying in colloquy” (p. 56); these verbs tend to occur in the unstressed, enclitic clause-second position. Hopper does not explain why auxiliaries and verbs such as to make belong in the same light verb category. Thus, to avoid a circular argument (verbs of making are light because they occur in second position, and verbs of making occur in second position because they are light), I consider why some apparently lexical verbs might actually be light.

Auxiliary verbs are a good starting point because they demonstrate that lexical content can be spread across a complex predicate. Unlike auxiliaries in Modern German, finite verbs in early periphrastic perfect and passive constructions were not semantically empty bearers of inflection, as evidenced by inflecting nonfinite verbs attested in early dialects. Yet they still required a noun-like participle to convey the full meaning of the predicate. The finite verb itself is light, compared to the lexical nonfinite verb. The idea that predication can extend beyond the finite verb to include other substantives and need not be confined to the verbal complex is not a new one. Jespersen (Reference Jespersen1942) first coined the term light verb to describe English verbs with little lexical content, such as to make, get, give, and take, that would combine with substantive complements and whose predication contribution was not the same as that of a main verb. An example of this sort of construction from English is to give an answer, where the verb is light and does not predicate as fully as the simplex variant to give money (Sundquist Reference Sundquist2018:261). Jespersen’s definition of light verb is different from the one I work with: Jespersen refers to a distinct class of verbs, different from auxiliaries in that they form complex predicates with noun phrases, prepositional phrases, and adjective phrases. In fact, Butt & Lahiri (Reference Butt and Lahiri2013) argue explicitly that light verbs are not auxiliaries. However, these light verbs and the verbs in auxiliary+participle constructions all exhibit a complex predicate in which the main verb, to one extent or another, is bleached of lexical content and relies on some additional constituent to fully convey meaning.

In sum, Hopper’s (Reference Hopper1975) clausal types indicate that the weight of the finite verb shapes verbal syntax in crucial ways and provide one with testable research questions. Are light verbs more likely to occur in clause-second position (in main and dependent contexts)? Are heavy verbs more likely to occur in clause-late or -first position (in main and dependent contexts)? Light and heavy verbs can be distinguished on lexical grounds—heavy verbs predicate fully on their own, whereas light verbs do not—but these sorts of distinctions can be difficult to draw based on historical data, because such an analysis involves clarifying semantic nuances inaccessible to the modern non-native speaker. This statement applies particularly to any search for complex predicates built outside of the verbal complex.

Thus, I define light and heavy verbs along syntactic lines: I treat the finite verbs in periphrastic formulations—that is, auxiliary + participle and preterit/optative-present + infinitive constructions—as light.Footnote ³⁹ The nonperiphrastic but obviously lexically empty copula verbs also belong in this category. All remaining nonperiphrastic lexical verbs count as heavy, with three exceptions: I treat separately the verb duan ‘to do’ and the two nonperiphrastic preterit-present verbs eig ‘to have’ and weiz ‘to know’ in the hopes of minimizing the potentially confounding impact of idiosyncratic data on correlations. Eig and weiz present an interesting case in that they have the stative semantics of other preterit/optative-presents (and the copula verb) and could be considered of low lexical yield. However, Jespersen (Reference Jespersen1942) would treat them as light. The case of duan in 26 illustrates the difficulties in identifying light verbs with complex predicates built outside of the verbal complex, for example, with a noun phrase.

1. (26)

   

The degree to which duan predicates on its own varies across the examples. In 26a, duan could be taken the most literally in that dátun thaz dúam can translate as ‘did another glorious deed’, though a somewhat less literal meaning is also possible: ‘effected great renown’. The more literal reading is not possible for 26b, however, in which datut zúhti ‘effected an education’ must refer more globally to the set of actions undertaken by the bishop that effected Otfrid’s education. The duan in 26c is semantically the least literal, most generic of the three; it must combine with rédina and is best translated simply as ‘said’. Also attested in the work is the simplex verb counterpart, redinôn ‘to state, explain’.

The nuances in the expression of just one potentially light verb, duan, should demonstrate how difficult it can be to identify lexically light verbs without the aid of syntactic correlates, such as periphrastic constructions. I am relatively confident that 26c contains a light verb construction, but it is less clear whether the predicates in 26a,b are light or heavy. Pursuing this issue requires a fine-grained analysis that lies beyond the scope of this article. Thus, for the present analysis I divide the data into two main groups: Clauses with periphrastic or copular predicates, which are unproblematically light, are separated from nonperiphrastic/copular predicates. Also taken out of the provisionally heavy group of predicates are clauses with nonperiphrastic preterit-present verbs and duan, the latter of which has real potential for a generic use in Jespersen-style light verb constructions. Despite these precautions, there is the possibility that other “lighter” nonperiphrastic verbs, such as duan, found their way into the “heavy” category. However, in defining the “light” category along syntactic lines, I ensure that no heavy verb was tagged as light. This methodology could inflate the rate of apparently heavy verbs in clause-second position erroneously because of the presence of unidentified light verbs in the heavy category. Given these limitations, it is notable that the correlation between heavy verbs and verb-final syntax, as I show below, is still significant.

Consider the distribution of light and heavy verbs across the data, as shown in table 6. Verb-first and -second clauses are included as separate categories—due to their pragmatic differences and also to reflect the connection between verb-second clauses and light verbs (Hopper Reference Hopper1975). Not included in table 6 are 33 duan and 19 nonperiphrastic preterit-present tokens, which I treat separately, 13 ambiguous tokens comprising the discourse marking quad ‘said’ (see example 7a in section 1.2), and 16 coordinated clauses where I could not distinguish verb-first and verb-second.

Table 6. Light and heavy verb distribution across unambiguous clause types.

These initial numbers already seem consistent with the patterns outlined in Hopper Reference Hopper1975 in that almost half of light verbs surface in clause-second position (43.4%) and a majority of heavy verbs—in clause-late position (52%). Now consider the same numbers broken down according to verb placement only (table 7), leaving the ambiguous clauses to the side for the moment. There is a marked preference for heavy verbs in clause-first and -late position, which contrasts with the even distribution of light and heavy verbs in second position. The numbers continue to be consistent with Hopper’s hypothesis.

Table 7. Light and heavy verb distribution across unambiguous clause types, by verb placement only.

These distributions are statistically significant at p<.01 (chi-square = 33.98; df=2). Footnote ⁴⁰

Assessing the weight of the verb can also elucidate a set of ambiguous clauses. Recall from section 1.2 the 29 clauses in which a surface clause-second verb follows a complementizer. These clauses (see example 27) remained ambiguous in the typology because existing analyses disagree on whether their finite verbs are underlyingly clause-second or -final.

1. (27)

   

However, these clauses—like those in the unambiguous categories—also fit into the narrative outlined in Hopper Reference Hopper1975, in which the clause-second position prefers light verbs and disprefers heavy ones: 19 (65.5%) of these surface verb-second clauses have light verbs, and only 10 (34.5%) heavy.

Now I assess the distribution of light and heavy verbs across all aforementioned clause types, adding the disambiguated thaz+V2 clauses to the verb-second category:

Table 8. Light and heavy verb distribution across unambiguous clause types and disambiguated thaz+V2 clauses.

These distributions are significant at p<.01 (chi-square=42.38; df=2).

The final numbers under consideration are those for duan and the nonperiphrastic preterit-present verbs, eig and weiz.

Table 9. Light and heavy verb distribution across duan and the stand-alone preterit-present verbs.

For now, I only note that the distributions of duan and the nonperiphrastic preterit-present verbs, eig and weiz, resemble those of the lexically light periphrastics (see table 9), indicating that duan, eig, and weiz are probably also light.

These data indicate that verbal syntax in the Evangelienbuch is sensitive to verb weight: The finite verbs in clause-first and -late position are heavy most of the time, whereas the finite verbs in clause-second position are mostly light. These data provide empirical support for the patterns described in Hopper Reference Hopper1975 for Germanic, which are shaped by verbal weight, not just by whether the clause is main or dependent. Indeed, paying attention to the weight of the finite verb can elucidate seemingly noncanonical clausal patterns, such as asyndetic verb-late clauses, which result from a tendency for main verbs to occur in clause-late position, especially when they are heavy. This tendency holds in main and dependent contexts. The same is true for thaz+verb-second clauses, which exhibit more light verbs than heavy. Again, the clause’s status as dependent is not the defining factor of verbal syntax in these tokens; it is the verb’s weight. Both types of noncanonical clauses are but different manifestations of early Germanic’s sensitivity to prosody. Just as there is an undeniable correlation in Otfrid between verbal syntax and verb weight, there surely is one between verbal syntax and a clause being main or dependent. The fact that so many clauses in the Evangelienbuch seem to conform to modern expectations supports this point. It is easy to see how the stage is set for a reanalysis of this correlation between clause type and syntax as a central distributional principle of German syntax. However, reanalysis had not yet happened in Otfrid’s grammar, and it is the “noncanonical” clauses, anomalous only if one ignores verb weight, that signal this fact.