1. Introduction
At least since Perlmutter’s (Reference Perlmutter, Jacobs and Rosenbaum1970) influential work, the grammar of aspectual verbs such as begin has inspired a range of opposing studies cross-linguistically. One line of research argues that aspectual verbs and their infinitival complements involve raised subjects, related in form and meaning to the non-finite verb but not the aspectual verb. Another line of research in the spirit of Perlmutter’s analysis states that aspectual verbs and their infinitival complements are ambiguous between a raising and a control structure, emphasizing the similarities to infinitival constructions including verbs like threaten or promise.
Studies arguing in favor of a raising analysis for aspectuals, on the one hand, refer to classical diagnostics for raising structures (Kiss Reference Kiss1995, Rochette Reference Rochette, Johnson and Roberts1999, Huddleston & Pullum Reference Huddleston and Pullum2002, Léger Reference Léger2019), where the non-finite verb determines whether the subject is a noun phrase, clause, or expletive, and also determines the thematic role of the raised subject, ranging from subject to clause to expletive (1).
Studies highlighting the ambiguous nature of aspectual verbs (Perlmutter Reference Perlmutter, Jacobs and Rosenbaum1970, Rizzi Reference Rizzi and Jay Keyser1978, Schroten Reference Schroten1986, Lamiroy Reference Lamiroy1987, Gunkel Reference Gunkel, Rolf Thieroff, Fuhrhop and Teuber2000), on the other hand, stress diagnostics for the presence of an external argument, among them the possibility of deverbal nominalizations and the imperative formation of aspectuals – all patterns require a suppressed external argument involving agentivity. Data such as (2) are therefore used to support a control rather than a raising analysis of aspectual verbs and their infinitival complements.
While argument-structural evidence for the analysis of aspectual verbs is drawn from a range of different languages, only a few languages allow word order properties as additional evidence for a raising and/or control analysis of aspectual verbs. As regards modern German, not only raising structures but also control structures are attested in restructuring contexts; examples of both types of structures may consequently show characteristics of a monoclausal structure such as pronoun fronting and third construction (Haider Reference Haider2010). Only control constructions, however, can occur in a biclausal structure. Thus, if aspectual verbs can be found with word order patterns that indicate a non-restructuring context, this would support an analysis of aspectual verbs as ambiguous between a raising and a control structure. Such evidence is in fact shown in (3). Since Bech’s influential study on the grammar of infinitival constructions in German (Bech Reference Bech1983:117), extraposition of the infinitival complement has been regarded as indicative of the clausal status of the non-finite complement and thus the biclausal nature of the infinitival pattern.
Since the pioneering work by Bech (Reference Bech1983), linearization patterns of infinitival complements in modern German are well understood (Grewendorf Reference Grewendorf1988, von Stechow & Sternefeld Reference Stechow and Sternefeld1988, Haider Reference Haider1993, Wurmbrand Reference Wurmbrand2001). This applies only to a limited extent to the history of the German language. As studies of Old High German (Demske Reference Demske and Eythórsson2008), Middle High German (Askedal Reference Askedal and Ole Askedal1998) and Early New High German (Maché & Abraham Reference Maché, Abraham, Lobenstein-Reichmann and Reichmann2011, Demske Reference Demske2015) have shown, there is no systematic relationship between raising and control constructions and word order properties in historical stages of German. In a recent quantitative study, De Cesare (Reference De Cesare2021) presents evidence that infinitival complements of raising and control predicates behave alike with respect to their word order properties until the mid eighteenth century. Up to this time, raising and control constructions behave in the same way, insofar as, on the one hand, intraposition of the infinitive is the preferred positional variant for both types. On the other hand, extraposition of the infinitival complement is recorded in the historical data not only for control verbs, but also for raising verbs, as illustrated below with the raising verb pflegen ‘use to’, using an example from the seventeenth century.
In the present article, we wish to contribute from a diachronic perspective to the still open question of whether aspectual verbs are in fact raising verbs or are rather ambiguous between a raising and a control structure. Using new data from German language history, we will use historical changes affecting word order to reassess the status of aspectuals in modern German. The article is organized as follows: In section 2 we briefly review the literature on aspectual verbs in German including historical accounts on word order properties of infinitival complements. In section 3, we set the stage for our empirical study and section 4 presents our study on the diachronic development of aspectual verbs focusing on word order properties. Section 5 concludes.
2. Aspectual verbs in German
2.1. Aspectual verbs between raising and control
The defining feature of raising verbs is that they do not select an external argument, so the grammatical subject in raising constructions is assumed to originate in the infinitival domain and to have raised to the governing predicate. It is thus the non-finite verb that imposes selectional restrictions on the respective subject. In this regard, aspectual verbs pattern with classical raising verbs like scheinen ‘seem’ in that they allow a number of phenomena that are only compatible with a raising analysis. For example, aspectual verbs may combine with weather verbs selecting a non-thematic subject, as illustrated in (5), or with process verbs requiring a theme argument as subject, as in (6).
Moreover, aspectual verbs can combine with verbs of propositional attitude, selecting a clausal subject, as in example (7). A further diagnostic for a raising construction is the occurrence of impersonal verbs, resulting in impersonal constructions, licensing either a dative (8a) or an accusative object (8b).
However, aspectual verbs are also attested in environments where there is evidence that they select an external argument. For example, aspectual verbs such as anfangen ‘begin’ allow for deverbal er-nominalizations (9a), a nominalization pattern creating names for logical subjects and thereby deleting the external argument throughout the word formation process. As pointed out by Fukuda (Reference Fukuda2007) for the English counterpart, the German deverbal noun Anfänger is likewise lexicalized, deriving from the Middle High German antvâher ‘instigator’. With other aspectual verbs, nominalizations using the er-suffix are not productive (9b).Footnote 2
Another argument for a control analysis of aspectual verbs is supposed to come from imperative formation: According to Perlmutter (Reference Perlmutter, Jacobs and Rosenbaum1970), English aspectuals allow the formation of an imperative, suggesting that aspectuals may have an external argument in contrast to pure raising verbs such as scheinen ‘seem’. This observation obviously carries over to German:
Following Rochette (Reference Rochette, Johnson and Roberts1999:149), we assume that the grammaticality contrast in (10) is due to the different semantic contributions of aspectual verbs like anfangen ‘begin’ and scheinen ‘seem’: While anfangen is an aspectual modifier stressing the initial phase of a process, scheinen is an indirect evidential marker, referring to the reliability of a proposition (Diewald & Smirnova Reference Diewald and Smirnova2010:180). Note that imperative formation with aspectuals is restricted to processes including an agent argument.
A further major argument in favor of a control analysis is provided by aspectual verbs attested with a nominal but no verbal complement (Perlmutter Reference Perlmutter, Jacobs and Rosenbaum1970). In the following example beginnen ‘begin’ seems to be the only candidate for the main predicate and hence the only possible source for theta role assignment in the clause. Note however that subjects may have different thematic roles: They may be agent or theme arguments, suggesting that it is not the verb beginnen that assigns a thematic role to the subject argument.
According to Pylkkänen & McElroy (Reference Pylkkänen, McElroy, Traxler and Gernsbacher2006), examples like the one in (11) involve a non-syntactic process of complement coercion in which the entity-denoting object is transformed into a salient event description that satisfies the selectional requirements of the aspectual verb. According to Rochette (Reference Rochette, Johnson and Roberts1999), the selected semantic category is ‘process’; see also Jackendoff (Reference Jackendoff1997) and Pustejovsky (Reference Pustejovsky1995).Footnote 3 Depending on the accomodated non-finite verb, the grammatical subject receives a ± agentive reading. An approach in terms of complement coercion hence shows that data like (11) do not necessarily provide an argument for an analysis of aspectuals as control predicates.Footnote 4
The brief overview of the ambiguous nature of aspectual verbs heading infinitival constructions in German provides suggestive evidence in favor of a raising analysis. A control analysis, however, may not be refuted altogether. In the following section, we briefly review what has been said so far about the nature of the subject of aspectual verbs and their role for the structure of the respective infinitival constructions.
2.2. Previous accounts: the nature of subjects
Most studies of aspectual verbs in German assume that they come in two variants, a raising and a control variant, just as the verbs drohen ‘threaten’ and versprechen ‘promise’. What all verbs supposedly have in common is that the thematic role of the grammatical subject can either come from the embedded infinitive or the verb heading the infinitival construction. Undoubtedly, the status of the subject plays a prominent role in any analysis of aspectual verbs and verbs of the drohen-type. Kiss (Reference Kiss1995) argues, for example, that there are no selectional restrictions on the surface subject of aspectuals and subscribes to a uniform raising analysis. He maintains that the limited acceptability that arises when aspectual verbs are combined with impersonal predicates can be attributed to the fact that subjects of aspectual verbs are obligatory. Note that our corpus shows that such constructions are actually used; see (8). Yet another position is taken by Haider (Reference Haider1993:245) who contends that aspectual verbs are a marked class of control verbs, selecting a barely constrained subject argument. The common assumption, however, is that aspectual verbs occur as both raising and control verbs, highlighting the parallel behavior of aspectual verbs and the verbs drohen ‘threaten’ and versprechen ‘promise’ as regards classical diagnostics concerning the nature of the subject (Gunkel Reference Gunkel, Rolf Thieroff, Fuhrhop and Teuber2000, Reis Reference Reis and d’Avis2005).
Looking at aspectual verbs in French, Lamiroy (Reference Lamiroy1987) and Ruwet (Reference Ruwet1983) state that animate subjects support a control reading, while their inanimate counterparts promote an understanding as a raising construction. This same generalization seems to apply to German as well:
But as pointed out by Gunkel (Reference Gunkel, Rolf Thieroff, Fuhrhop and Teuber2000) and Reis (Reference Reis and d’Avis2005), animate subjects are also attested in unambiguous raising structures such as (13), suggesting that there is no unique relationship between the agentivity of the subject and a particular infinitival structure.
Consequently, predicates ambiguous between a control and a raising interpretation are taken to get either reading in cases where an animate subject is used: In the following example, Alfons may be either announcing that he will break up the reception (= control variant) or there may be direct evidence that he will do so in the near future (= raising variant).Footnote 5
Highlighting the parallel behavior of aspectual verbs and drohen-type verbs, Gunkel (Reference Gunkel, Rolf Thieroff, Fuhrhop and Teuber2000) and Reis (Reference Reis and d’Avis2005) maintain their view that aspectual verbs have not only a raising but also a control variant.Footnote 6 In particular, Gunkel (Reference Gunkel, Rolf Thieroff, Fuhrhop and Teuber2000) argues that aspectual verbs allow for a raising construction only when they select an infinitival complement including a non-finite verb with an inanimate subject (15a) and (15b). Animate subjects are compatible with an interpretation as raising verbs if they are non-agentive (15c). Gunkel (Reference Gunkel, Rolf Thieroff, Fuhrhop and Teuber2000) further states that the embedding of passive constructions is ungrammatical whenever the infinitival complement contains a subject that is both animate and agentive and thus triggers a control structure, while the grammatical subject simultaneously requires a raising structure (15d).Footnote 7
In our view, however, agentivity of the subject does not rule out the embedding of passive constructions under an aspectual predicate, as shown by the impersonal passive construction in (16). Because tanzen ‘dance’ is an intransitive verb, there is no internal argument to be raised and the agent argument is blocked due to passive.
Furthermore, the ambiguity between a control and a raising interpretation observed for drohen-type verbs cannot be substantiated by means of corresponding examples including aspectual verbs. Unlike for the verb drohen ‘threaten’, we assume that the aspectual verb beginnen ‘begin’ is a predicate modifier, never heading an event apart from the one denoted by the non-finite verb. Therefore, only the non-finite verb rezitieren ‘recite’ assigns a thematic role to the animate subject in (17), which excludes an ambiguity in the sense of (14).
As for the correlation between animacy and agentivity, on the one hand, and the nature of complement structure, on the other, Gunkel’s (Reference Gunkel, Rolf Thieroff, Fuhrhop and Teuber2000) approach for German undoubtedly extends beyond an approach like Lamiroy (Reference Lamiroy1987) for French. Nevertheless, the problem remains that there is considerable doubt about the assumption that aspectual verbs like beginnen ‘begin’ have a full verb variant in German, as is the case with drohen-type verbs. Moreover, the historical record for both verb types apparently supports their different infinitival grammar: While there is ample evidence that the raising variant of drohen ‘threaten’ and versprechen ‘promise’ evolved from a control structure (Reis Reference Reis and d’Avis2005), there is no such evidence for aspectual verbs in German (von Monsterberg-Münckenau Reference Monsterberg-Münckenau1885, Jedrzejowski Reference Jedrzejowski2015); see section 3 for more details. Before turning to the historical data, we will briefly sketch word order properties of infinitival complements, which are known to be a strong diagnostic in German for distinguishing between raising and control constructions.
2.3. Word order diagnostics and restructuring
In present-day German, infinitival complements in raising and control constructions display systematic differences regarding their placement: While infinitival com-plements in raising constructions are generally restricted to a position preceding the governing verb in final position (18), their placement is less restricted in control constructions. Infinitival complements of control verbs allow both intraposition (19a) as well as extraposition (19b). A subclass of control verbs admits a further word order pattern called “third construction,” which includes a discontinuous infinitival complement (20).
This word order distinction further correlates with the restructuring properties of the infinitive embedding verb: Classical raising verbs like scheinen ‘seem’ and control verbs of the vergessen-type all exhibit typical diagnostics for a monoclausal structure such as pronoun fronting (21a), while only biclausal structures allow pied piping (21b); see Haider (Reference Haider2010:311–313) for a comprehensive list of diagnostics.
Classical raising verbs like scheinen ‘seem’ and control verbs of the vergessen-type are taken to belong to the set of so-called restructuring verbs in present-day German. Differences between the two classes regarding argument structure may be captured by distinguishing between functional (classical raising verbs) and lexical (control verbs of the vergessen-type) restructuring verbs (Wurmbrand Reference Wurmbrand2001, Reference Wurmbrand2004): The latter predicates are taken to occupy the verbal head, thus establishing thematic relations with their internal and external arguments (22), while functional predicates occupy typical positions of functional heads (23) like auxiliary and modal verbs do.
How do aspectual verbs and verbs of the drohen-type fit into this classification? As regards the latter class of verbs, they are in one of their manifestations well-established raising structures (Gunkel Reference Gunkel, Rolf Thieroff, Fuhrhop and Teuber2000). As Reis (Reference Reis and d’Avis2005) shows, drohen ‘threaten’ also allows for discontinuous and extraposed infinitival complements just as lexical restructuring verbs. Regardless of whether the infinitival complement is intraposed or not, however, the restructuring predicate has a temporal-aspectual meaning in both instances, characteristic for the raising variant, which suggests a classification as functional restructuring verb in Wurmbrand’s terminology. This means that discontinuous and extraposed infinitival complements are not restricted to a subclass of control verbs, but are compatible with the raising variants of drohen ‘threaten’ and versprechen ‘promise’. Note that the classical raising verb scheinen ‘seem’ does not allow the discontinuous or extraposed occurrence of its infinitival complement.Footnote 8 In its control reading, drohen ‘threaten’ passes all tests for a biclausal, hence non-restructuring configuration.
According to Wurmbrand (Reference Wurmbrand2001:15), aspectual verbs like beginnen ‘begin’ are always restructuring verbs, being ambiguous between functional and lexical restructuring. With respect to word order, aspectuals behave like drohen ‘threaten’, allowing for intraposition, third construction and extraposition. Remarkably, as noted above, third construction and extraposition are excluded for a classical functional restructuring predicate like scheinen ‘seem’.
Wurmbrand (Reference Wurmbrand2001:209) suggests that verbs like versprechen ‘promise’ and drohen ‘threaten’ only allow for the control verb reading as soon as the infinitival complement is extraposed.
In our view, there are no analogous differences in meaning when considering the positional variants of aspectual verbs: Following Wurmbrand (Reference Wurmbrand2001) a control reading would be expected when the infinitive is extraposed (25c), whereas intraposition and third construction are compatible with a raising interpretation. Such a difference, however, cannot be established from the following examples.
With respect to the question whether aspectual verbs are pure raising verbs or also include a control variant, the word order properties of their infinitival complements obviously do not provide conclusive evidence: If the extraposition of an infinitival complement provides a strong argument for a biclausal structure, then aspectual verbs do indeed encompass a control variant. Such an analysis seems problematic from the point of view of argument structure and the nature of subjects (see sections 2.1 and 2.2), which raises the question of the diagnostic value of the extraposition pattern. In the next section, we will report evidence that word order properties of infinitival complements in present-day German are in fact a recent development.
3. A diachronic perspective
The observed word order variation of infinitival complements of aspectual verbs in present-day German might give the impression that unlike other raising verbs such as scheinen and pflegen they are not fully grammaticalized and hence still exhibit word order properties that are usually associated with control verbs, namely, extraposition of the infinitive. While this hypothesis has been discussed in previous literature for the verbs drohen and versprechen, whose raising variant is claimed to have developed from the control variant only recently (see De Cesare Reference De Cesare2021:113-114 and references therein), it seems less convincing for the aspectual verbs beginnen and anfangen.
First, unlike other verbs, beginnen and anfangen have been shown to exhibit raising behavior already in older stages of German. Jedrzejowski (Reference Jedrzejowski2015) finds evidence for raising structures with beginnen already in Old High German, where beginnen can embed impersonal predicates (26) and weather verbs (27). In addition, while he could find clear evidence for raising, there was no unambiguous evidence for control in his Old High German data, which leads him to the hypothesis that the raising variant of beginnen did not develop from the control variant, as assumed for other raising verbs.
Similar evidence can be found in Middle High German, where the verb beginnen also clearly shows raising patterns, as exemplified in (28) and (29). As for the Old High German examples, beginnen still selects a bare infinitive in Middle High German.
As far as the verb anfangen is concerned, Ebert et al. (Reference Ebert, Reichmann, Solms and Wegera1993:394) report that it is attested with subjectless constructions and with weather verbs in Early New High German, which indicates that the raising variant is already present then. Like beginnen, anfangen selects a bare infinitive as well until about the sixteenth century (Ebert et al. Reference Ebert, Reichmann, Solms and Wegera1993:403). The status of the infinitive thus further suggests that aspectual verbs already yield monoclausal construals in older stages of German, although it is to be noted that the relation is not unique, since zu-infinitives can also show restructuring properties.
Second, an explanation based on a grammaticalization scenario is also not straightforward from the point of view of word order. As shown in the previous section, in present-day German, infinitival complements in raising and control constructions display systematic differences regarding their placement, which in turn is related to the restructuring properties of the infinitive embedding verb. There is, however, ample evidence that systematic word order differences between raising and control constructions, respectively functional restructuring and other (non-)restructuring predicates, are a fairly recent development in the history of the German language (Ebert Reference Ebert1976, Askedal Reference Askedal and Ole Askedal1998, Reis Reference Reis, Müller and Reis2001, Demske Reference Demske and Eythórsson2008, Maché & Abraham Reference Maché, Abraham, Lobenstein-Reichmann and Reichmann2011, De Cesare Reference De Cesare2021). For example, Demske (Reference Demske and Eythórsson2008, Reference Demske2015) points out that extraposition is attested across all types of verbs taking infinitival complements in Old High German, even for verbs already showing raising properties, including modal verbs (30).
Extraposition is still possible with raising structures at least until the Early New High German period; see (31) and (4) above. At the same time, however, there is a preference for the intraposition of infinitival complements regardless of verb type throughout the fourteenth and fifteenth centuries (including infinitival complements of verbs of the non-restructuring type in present-day German), as illustrated in (32).
Conducting a large quantitative study on the placement of infinitival complements since the fifteenth century, De Cesare (Reference De Cesare2021) provides conclusive evidence that raising verbs occur predominantly with intraposed infinitives since about 1750, which becomes the obligatory pattern in present-day German. Control verbs on the other hand, continue showing variable word order behavior with an increasing preference for extraposed infinitival complements appearing in the youngest investigated period, present-day German (figure 1).Footnote 9 De Cesare (Reference De Cesare2021) argues that word order patterns undergo a process of specialization over time, and that different word order patterns specialize for different types of infinitives, namely monoclausal and biclausal infinitives, in order to avoid structural ambiguity: Raising structures involving a monoclausal construal specialize for intraposition, while control constructions involving a biclausal structure specialize for extraposition. Note that for lexical restructuring verbs, extraposition is still the preferred pattern in present-day German, even though restructuring is an option (De Cesare Reference De Cesare2021). It is still not clear, however, how aspectual verbs fit in this diachronic scenario.
Figure 1. Raising and control versus word order (De Cesare Reference De Cesare2021:105).
In the present article we look into word order properties of aspectual verbs over time and evaluate them against the emergence of a systematic distinction of raising and control verbs in the recent history of German, aiming at a deeper understanding of the syntactic behavior of aspectual verbs in present-day German. Since, however, word order properties are also strongly correlated with restructuring properties of the infinitive embedding verb, developments in word order properties can also be indicative of the type of restructuring of the constructions under discussion, rather than their raising versus control status. Given the mixed evidence for the raising versus control status of the aspectual verbs beginnen and anfangen, this further distinction might indeed be needed to explain the mixed behavior of aspectual verbs.
In light of the diachronic evidence for prototypical raising verbs, viz. functional restructuring verbs, however, it is to be noted that an increase in intraposed patterns is not necessarily indicative of a change from lexical to functional, since this correlation between restructuring properties and word order was not present in older stages of the language. Instead, word order patterns might have been conventionalized over time depending on the predicate type, as suggested in De Cesare (Reference De Cesare2021).
4. The present study
In view of the generalizations presented so far, the main goal of the present study is to investigate whether the diachronic development of infinitival complements’ linearization patterns with aspectual verbs can bring additional evidence to the discussion about the status of aspectual verbs. As discussed in the previous sections, the present-day word order distribution does not provide conclusive evidence for either analysis. Unlike typical raising verbs yielding functional restructuring like scheinen ‘seem’, aspectual verbs allow word order patterns other than intraposition in present-day German, which goes against a unified raising analysis. Taking the diachronic perspective into consideration, however, has shown that a raising interpretation cannot be excluded on the basis of word order, since raising structures did allow for variable word order in the past and are only recently increasingly attested with the intrapostion pattern. By looking at what linearization patterns are attested with aspectual verbs over time, and by comparing the trajectory of the development with that of typical raising and control verbs, we might thus be able to shed light on the question as to whether both a raising and a control variant are available and whether the aspectual verbs beginnen and anfangen are ambiguous with respect to their restructuring status. Assuming that the proposed selectional restrictions hold for aspectual verbs, we thus take properties of the subject to be the distinctive factor and, in accordance with the criteria proposed in previous studies (see section 2.2), we investigate whether animacy and/or agentivity influence the frequency of intraposed infinitives.
Based on the development scenario described in De Cesare (Reference De Cesare2021), in which word order patterns have specialized for different structure types, there are two possible scenarios for aspectual verbs. If both a raising and a control variant are available, word order properties of aspectual verbs should develop in opposite directions, that is an increase in intraposed infinitival complements is expected only for the putative raising variant, while the putative control variant is expected to show an increasing preference for extraposition, along the lines of more prototypical raising and control verbs. If, however, only raising is a structural option, an overall increase of intraposition should be observed, independent of the properties of the subject. Considering the restructuring properties of aspectual verbs opens yet another possibility, however. An increase in intraposed infinitives is also expected if aspectual verbs yield functional restructuring, whereas an increasing preference for extraposition would be indicative of lexical restructuring or no restructuring at all.
In order to address these questions, we conducted a diachronic corpus study investigating the word order behavior of two ingressive aspectuals, that is of two aspectual verbs focusing the beginning of an action (beginnen and anfangen ‘begin, start’).
4.1. Data and method
For our investigation we consulted online available corpora that together cover the period from around 1350 to present-day German. For the historical record, data were extracted from the Baumbank.UP,Footnote 10 a treebank of Early New High German texts, the Deutsches Textarchiv (DTA),Footnote 11 and the GerManC Corpus,Footnote 12 which contain texts ranging respectively from the sixteenth to the nineteenth century and from 1650 to 1800. For the twentieth and twenty-first centuries we searched the DWDS Kernkorpus Footnote 13 and DWDS Kernkorpus 21,Footnote 14 respectively. Additional observations come from two corpora that have been built up as the empirical base of two other projects: a corpus of early prose novels (fifteenth and sixteenth centuries)Footnote 15 and an offline corpus of newspaper texts (seventeenth to nineteenth centuries).Footnote 16 For the corpora that are POS- and Lemma-annotated (DTA, DWDS Kernkorpus, and DWDS Kernkorpus 21) we formulated search queries that target our items of interest. In particular, the search targeted sentences containing the following elements:
-
1. A subordinating conjunction (KOUS);
-
2. The matrix verb of interest (anfangen or beginnen);
-
3. A zu-infinitive (PTKZU, VVINF).
Since the corpora are not syntactically parsed, the queries are based on linear order only. The search query for intraposition is given in (33a): Here the matrix verb of interest appears in last position and is directly preceded by the infinitive. Example (33b) illustrates the query that was used to find both extraposition and third construction cases: In this case the infinitive follows the governing verb, but potential discontinuous elements of the infinitive can precede the verb as any other word was allowed to appear before the matrix verb and between the governing verb and the infinitive. Both queries also include the specification to order the hits chronologically:
Once the corresponding hits were obtained, an additional manual check was performed in order to exclude false hits. These include, for example, sentences like the one in (34), which were selected by the search query for extraposition and third construction. Since it was not possible to specify hierarchical relations, the search query also finds those sentences where the aspectual verb does not belong to the subordinate clause itself but simply follows in a declarative V2 sentence.
For the remaining corpora, observations were extracted by searching for the above-described target sentences manually, with the help of a lemma-search targeting either the matrix verbs or the infinitive marker zu ‘to’.
In total we obtained 4,738 observations. For each observation we annotated the following variables: the position of the infinitive (intraposed, extraposed, third construction), the aspectual verb (anfangen or beginnen), the animacy (animate or inanimate) and agentivity (agentive or non-agentive) of the subject and the infinitive verb. In the following sections we look at the attested word order patterns over time both from a qualitative and from a quantitative perspective.
4.2. Qualitative results
Looking at the availability of word order patterns in present-day German, our data show, as previously observed, that infinitival complements of aspectual verbs are not restricted to an intraposed position. Extraposed infinitival complements are also attested, which is not the case for the prototypical raising verbs scheinen and pflegen (De Cesare Reference De Cesare2021). While extraposition of the infinitive is taken to be evidence for its clausal behavior in present-day German, our data show that this cannot be the case with aspectual verbs. Extraposition is in fact not only attested in the putative control cases (e.g. as in (35)), but also in unambiguous raising contexts, namely sentences with an expletive subject as in (36a–b).
Similarly, intraposition is attested with both animate and agentive subjects (37), namely in the putative control cases, and with expletive subjects (38a) or inanimate and non-agentive subjects (38b), namely in the putative raising cases.
This type of distribution is not restricted to the present-day German data but also holds over time, that is both intraposition and extraposition are attested in our older data with no particular restriction concerning the subject type. In both examples in (39) the subject is inanimate and non-agentive, but in (39a) the infinitive is intraposed while in (39b) it is extraposed.
In (40) the subject in both sentences is animate and agentive, but again both word order options are attested. The infinitive in (40a) is extraposed, while in (40b) it is intraposed.
The attested word order distribution confirms the observation of previous literature that extraposition is not restricted to control contexts in older stages of the language. On the contrary, typical raising contexts also show extraposed infinitives. These results also suggest that any changes in the word order behavior of aspectual verbs are unlikely to be motivated by a shift from control to raising, since the raising variant is attested already in Old High German (see Jedrzejowski Reference Jedrzejowski2015). An increase in intraposed infinitives might instead be due to increasing conventionalization of word order patterns, as was the case for prototypical raising verbs such as scheinen and pflegen. Unlike the latter group, however, the aspectual verbs beginnen and anfangen still seem to show extraposed infinitives in present-day German.
4.3. Quantitative analysis
To determine whether the distribution of word order patterns per context over time changes quantitatively we run a mixed-effects logistic regression evaluating the probability of intraposition as the dependent variable. Data were analyzed using the glmer function of the lme4 package (Bates et al. Reference Bates, Mächler, Bolker and Walker2015) in R (version 4.1.0, R Core Team 2021). The following predictors were included in the model: the continuous variable Time, the categorical factors Animacy (+animate vs. −animate), and Agentivity (+agentive vs. −agentive) of the subject as well as Aspectual Verb (anfangen vs. beginnen). In addition, to test whether a possible specialization scenario can be observed for the aspectual verbs beginnen and anfangen respectively, three-way interactions between Time, Animacy, Verb and Time, Agentivity and Verb were included in the model. Significant interactions of Animacy by Time or Agentivity by Time would indicate that the effect of the subject on the probability of intraposition changes over time, which could be indicative of a specialization process like the one described for typical raising and control verbs, hence supporting the hypothesis that aspectual verbs come in two variants. On the contrary, a general increase in intraposition, regardless of the subject, would support the hypothesis that the aspectual verbs beginnen and anfangen rather behave like raising verbs. A significant three-way interaction including the factor Verb would additionally show whether the two verbs behave differently in this regard. To reduce possible idiosyncratic effects due to preferences of individual authors or genre-specific preferences, random nested effects for Genre and individual Text were also included. The role of the embedded infinitive verb was also controlled for by adding it as a random effect.Footnote 17
In order to obtain main effects and their interaction, sum-coded contrasts were applied to the factors Animacy, Agentivity, and Aspectual Verb. The continuous variable Time was centered around the mean and log-transformed. Subsequent comparisons of interest were obtained by resetting the contrasts and/or releveling factors and refitting the model. The model results of main effects and their interaction are summarized in Table 1. The regression analysis confirms a main effect of Time: As the positive estimate shows, overall (across both aspectual verbs and subject conditions) intraposed infinitives become more frequent over time (p < .001). Significant main effects of Verb were also found, showing that anfangen yields overall less intraposition than beginnen (p < .001). Interestingly, main effects of animacy and agentivity also reached significance: Animate subjects yield significantly less intraposition (p < .01), and agentive subjects also decrease the probability of intraposition compared to the mean probability of intraposition (p < .05). However, while all main effects reached significance, most interactions failed to do so (see Table 1). The lack of an interaction between time and agentivity and time and animacy respectively, suggests that the effect of agentivity and animacy on the probability of intraposition does not change significantly over time, according to the model. In other words, no evidence for a specialization scenario could be found in our data. If a specialization process was underway, the effect of the subject on the probability of intraposition should have changed over time. Instead, the interaction including Agentivity is only significant when the factor verb is also included, suggesting that it is the individual verbs that behave differently over time. This is also indicated by the significant interaction between Time and Aspectual.
Table 1. Model results: main effects and their interaction
Successive comparisons between the two verbs show that although intraposition increases for both verbs over time, the effect is larger for beginnen than for anfangen (beginnen: β=21.47555, SE=2.24943, p=2e-16; anfangen: β=5.82121, SE=1.55962, p=0.00019). Looking at Figure 2, the developement observed for beginnen resembles that of typical raising verbs (cf. Figure 1), while anfangen shows a less steep increase. Overall, the probability of intraposition with the verb anfangen is relatively low even in present-day German (around 50 percent), as opposed to the near 100 percent of intraposition with beginnen.
Figure 2. Probability of intraposition over time per aspectual verb.
4.3 Summary and discussion
As far as the diachronic dimension is concerned, the development path of the different word order patterns does not correspond to the specialization scenario sketched for typical raising and control verbs. While animate and agentive subjects yield a lower probability of intraposition, no evidence for specialization could be found. In fact, unlike typical control verbs, which are increasingly found with extraposition over time, the aspectual verbs beginnen and anfangen rather show an increase in intraposition, regardless of the properties of the subject. This behavior over time rather matches that of prototypical raising verbs, while it does not provide strong evidence for prototypical control. Relating the word order behavior to restructuring properties, the trajectory of the development corresponds to that of functional restructuring verbs. Considering this additional factor might explain, on the one hand, the mixed behavior of aspectual verbs when it comes to thematic properties and, on the other hand, the increasing preference for intraposition typical of raising structures. Given the evidence from our diachronic study of word order properties, namely the overall increase in intraposition over time, we suggest that aspectual verbs are always functional restructuring verbs. However, as we cannot rule out the assignment of thematic relations in some cases, there might be a structural ambiguity involved. Following an idea originally proposed in Wurmbrand (Reference Wurmbrand2001:144-145), we assume that there is a structural position – head of vP or AspP – that is both functional and thematic, and that aspectual verbs might also occupy this position, in addition to a higher functional position. The control reading, that is the assignment of thematic relations, would then be possible if the aspectual verb is in the lower position, and not possible if the aspectual verb occupies the higher position. The ambiguity would thus be given by the two possible positions illustrated in (41). Importantly, as noted above, changes in word order are, in our view, not indicative of a structural change, that is from lexical to functional restructuring, but of changing preferences in language use due to conventionalization of the pattern for functional verbs. In other words, we assume aspectual verbs to be functional already in older stages of German.
What remains to be explained is the stable effect of the subject. Why do animate and agentive subjects yield less intraposition than inanimate and non-agentive subjects, if aspectual verbs are functional anyway? A possible answer to this question might involve a priming effect or a frequency effect: Since control verbs yielding lexical restructuring or non-restructuring always have an animate and agentive subject and since they are predominantly found with extraposed infinitives, language users might associate these properties of the subject with extraposition even when dealing with a functional restructuring structure, thus leading to the subject interfering in the word order choice.Footnote 18 Despite this effect of subject, the development of word order preferences for the aspectual verbs beginnen and anfangen follows that of typical functional restructuring constructions.
In addition, what seems to characterize our dataset is variation between the two verbs. Despite the overall increase of intraposition over time, anfangen is only found with an intraposed infinitive in about 50 percent of the cases in present-day German. Given the qualitative evaluation of the data, this cannot be indicative of the fact that anfangen shows a stronger control behavior than beginnen. Both verbs are in fact attested in unambiguous raising configurations, namely with weather predicates, non-agentive, and inanimate subjects, from Early New High German. Thus, in view of the similar syntactic behavior of both verbs, the differences in word order patterns preferences between the individual verbs suggest that additional factors might play a role in the linearization of infinitival complements.
The main difference between the two verbs is their morphological form. While anfangen belongs to the group of the so-called particle verbs, that is verbs with a separable prefix (i.e., an-), beginnen is considered a prefix verb, that is a verb with a non-separable prefix (i.e., be-). This morphological difference also has implications for their prosodic characteristics. Prosodically, particles are stressed, while non-separable prefixes are not (Eisenberg Reference Eisenberg2009). The presence of a separable prefix has been previously found to favor the reordering of the verbs in the right periphery, both in historical German (Sapp Reference Sapp2011) and modern Dutch (Den Besten & Broekhuis Reference Den Besten and Broekhuis1989). Furthermore, prosodic effects on the ordering preferences in verb clusters have been observed for spoken varieties of German (Schmid & Vogel Reference Schmid and Vogel2004). A similar effect might be involved in the explanation of the different word order behavior of beginnen and anfangen, but this proposal needs further investigation.
Moreover, as one reviewer suggests, anfangen, and particle verbs in general, is claimed to block restructuring properties, unlike beginnen, which might also explain why it yields less intraposition than beginnen. Looking at classical diagnostics however, we are not convinced that anfangen differs from beginnen with respect to restructuring, unlike the Dutch counterpart. While particle verbs are claimed to block verb raising in Dutch (Wurmbrand Reference Wurmbrand2001:97), for example, our data amply show that this is possible with anfangen in German (see section 4.2.1). In addition, we find pronoun fronting equally acceptable with beginnen and anfangen; see (42).
Based on our word order data, anfangen also shows an increase in intraposition over time, as shown above, which patterns with restructuring configurations. If anfangen was non-restructuring, an increasing preference for extraposition would be expected.
5. Conclusions
The aim of the present study is to contribute to a better understanding of the syntactic nature of aspectual verbs, given their mixed behavior with respect to raising versus control properties. Focusing on the two aspectual verbs anfangen and beginnen, both meaning ‘start, begin’, we compared the diachronic distribution of their linearization patterns to that of prototypical raising and control verbs. In addition, we evaluated their word order behavior from the perspective of restructuring.
Our quantitative analysis has shown that both verbs behave like typical raising verbs in this respect, showing an increasing preference for intraposed infinitives over time, while no indication for prototypical control behavior could be found in our diachronic data. In light of the evidence discussed in the literature, however, it cannot be entirely excluded that aspectual verbs establish thematic relations with the subject in some cases. Based on our novel diachronic word order data, we suggest that aspectual verbs are functional verbs, but that they can occupy different structural positions: If they occupy a higher functional position, that is a non-thematic position, they do not establish any thematic relations with their arguments, if they are in the lower functional position, that is a thematic position, they do.
In contrast to the prototypical functional verb scheinen, however, extraposition is still permitted in individual cases, which casts doubt on the status of extraposition as an unquestionable diagnostic, in particular when we keep in mind that the extraposition of infinitival complements as a diagnostic for the mono- and biclausal structure has only recently developed in the history of the German language.
Overall, the data also show a considerable difference between the individual lexical items under investigation. Since, however, structures compatible with raising are attested in the whole dataset for both verbs, we argued that this difference in word order preferences between anfangen and beginnen is due to additional extra-syntactic factors influencing the linearization of infinitival complements. In particular, based on previous findings on verb order in verb clusters, we advanced the hypothesis that the avoidance of intraposition for anfangen is motivated by prosodic factors following from the presence of a stressed separable prefix.
Open access
