In this paper I argue that the unembedded Accusativus cum Infinitivo in Ancient Greek is a case of hearer-induced grammaticalization. AcI embedded under verba dicendi or δοκεῖν in deontic contexts are ambiguous: while the speaker intends the deontic reading as a side meaning of the embedding verb, it can also be attributed to the AcI. If the hearer opts for the latter analysis, a new function of the AcI emerges, which ultimately leads to its de-embedding. I show that this grammaticalization process is parallel to similar reanalyses in sound change. In a short outlook the analysis is extended to the emergence of absolute constructions.
In recent years there has been a growing interest in de-embedding or insubordination, to use the more recent term coined by Evans (2007). The phenomenon is long familiar to students of the syntax of the older Indo-European languages. Thus, it is not the aim of this paper to add further examples to the ongoing discussion. Nor is it my intent to assess the role de-embedding plays in grammaticalization theory (on both see Evans 2007 and the papers in Evans and Watanabe 2016). Rather, what I am interested in is to give a possible scenario for the actuation of a remarkable case of de-embedding, the Ancient Greek unembedded AcI.
The Greek infinitive is attested in a variety of uses, most of which are unsurprising given the fact that it originated from an adjunct noun phrase which is headed by an event noun and denotes the purpose of the event depicted in the embedding sentence.1 One use, however, deviates strikingly: the unembedded Accusativus cum Infinitivo (AcI), which is not only a case of insubordination, but rather surprisingly inherits the unusual case marking of its subject from its embedded antecedent. The origin of this construction is undisputed, as is the grammaticalization path. What has not been dealt with, however, are the mechanisms behind this type of grammaticalization. In this paper I argue that it is hearer-based and triggered by the exact same factors as hearer-based sound change. The paper is part of a project aiming at giving a unified account of sound change and grammaticalization.
Throughout this paper, I use the term grammaticalization to characterize the change from embedded to unembedded AcI. I follow Traugott (2008: 220) in understanding grammaticalization broadly “as the output of processes of language use that lead to systematic changes in morphosyntactic form and meaning.” Under this definition, the terms grammaticalization and grammatical constructionalization (Traugott and Trousdale 2013) are grosso modo interchangeable.2 But even under a narrow definition of the term, de-embedded AcI qualify as grammaticalizations, or, more accurately, as secondary grammaticalizations, i.e. changes the source of which is itself the product of a grammaticalization process (for the term see Traugott 2010). Grammaticalization as a process has been described as a “drift […] toward tighter structures, toward less freedom” (Haspelmath 1998). Such a drift can indeed be observed in the development of the de-embedded AcI, both in semantics and in syntax. As will be illustrated below, embedded infinitives may have various readings which depend exclusively on the embedding context. The unembedded AcI, on the other hand, has exactly one reading: it is necessarily deontic.3 This semantics, then, has become a feature of the construction. De-embedding also results in an increase in grammatical structure compared to embedded AcI. In the latter, the accusative of the subject is licensed by the embedding verb, a phenomenon described in the literature as exceptional case marking (ECM) or raising to object (Goldstein 2016: 269). In unembedded AcI, the case is licensed from within the construction itself. This change comes with an important corollary: raising to object implies that the AcI be a VP (or TP, depending on one’s premises). It cannot be an S (or CP), since the latter is a barrier for ECM or raising (see again Goldstein 2016: 269, 272 for the difference between VP- and S-infinitives in Ancient Greek). Unembedded AcI serve as full-fledged independent sentences: they are S/CP in their own right. Note also that the unembedded AcI, like other grammaticalizations, is an instance of what Himmelmann (2004) calls expansion, since it is used in a context in which infinitives did not occur before the change.
In the following section a short overview of the prehistory and the uses of the Greek infinitive is given. In section 3 I present a picture of the unembedded AcI, while section 4 introduces the notion of actuation. In the fifth section I give an account of the emergence of the unembedded AcI in terms of the theory developed in section 4. Section 6 introduces absolute constructions as a further example of a construction evolving from the same type of hearer-based change. A summary is given in section 7.
2. The infinitive in PIE and early Greek
2.1. A short prehistory of the Greek infinitive: morphology
All types of Greek infinitives are based on deverbal event nouns in oblique cases. The oldest marker, which is attested from Mycenean onwards and is probably of PIE age, is formed by reanalyzing the locative of an s-stem. Thus, Greek ἔχειν, Myc. e-ke-e can be transposed to a PIE form *seg̑h-es-en, the ending of which is notably similar to the one presupposed by Latin -ere < *-es-i and Vedic -sáni < *-s-én-i. An exact reconstruction of a PIE infinitive marker, however, remains impossible as the attested forms differ both in case marker and stem formation: while the Latin infinitive continues an *i-locative, Greek has *-en, most likely also a locative marker (Stüber 2000: 139). Vedic combines both, -sáni being an *en-locative hypercharacterized by an additional *-i (Stüber 2000: 159). In addition, Vedic presupposes a zero grade affix not attested elsewhere in deverbal neutral s-stems (Debrunner 1954: 233; Stüber 2000: 159 takes this to be an archaism), while both Greek and Latin show full grade. Still, the continuation of an *en-locative both in Vedic and in Greek is remarkable and points to a common PIE ancestor, since in both languages *en-locatives of s-stems or other event nouns are otherwise not attested.4
In the case of IE languages, locatives of event nouns are a surprising starting point for a grammaticalization path leading to infinitives. For one thing, locative complements are not attested with any of the verbs embedding infinitive complements in the early IE languages. Locative complements can thus be excluded as a possible source of the infinitive. For another, the semantics of adjunct infinitives is purposive and quite similar to the dativus finalis which seems to be a much more likely source for the infinitive—many of the infinitive markers attested in early Vedic actually derive from datives. The IE locative does not typically denote purpose, though it may be used as a Goal-case. At first glance, this might indeed be the missing link, since Goal-cases are attested throughout the world as a source for purposive infinitives (Haspelmath 1989; Heine and Kuteva 2002). However, the situation in the IE languages is slightly more complicated. As shown by Vedic, which still distinguishes neatly between the dative and the locative, Goal-locatives are restricted to nouns denoting places or, more generally, objects in the world. With abstract entities such as events, Goal is always denoted by the dative (Keydana 2013: 329). If Vedic faithfully continues the late PIE system, Goal cannot be the origin of the infinitive.
A tentative explanation for the locative lies in the formal similarity of the -i-locative and the dative in *-ei̯. It is conceivable that both markers continue one and the same desinence which originally covered both the dative and the locative function. This assumption is further corroborated by Hittite, where both cases are marked by -i—at least if we consider Hittite to be archaic in this respect. Concerning the locative infinitive, this suggests that the grammaticalization took place at a point in time when both cases were not yet distinct, viz. before late PIE.
The other Greek infinitive markers, -έμεν and -εναι / -μεναι, are likewise based on locatives of deverbal nouns. However, these formations are of later origin. And since the locative and the dative collapsed into a single category in Greek, it is impossible to decide if they developed out of true locatives or rather out of formal locatives in dative use.5
2.2. A short prehistory: types of embedding
Diachronic morphology reveals that infinitives were originally used as adjuncts. Adjunct infinitives come in two types, as rationale clauses and as purpose clauses.6 In both types, the subject of the infinitive phrase is latent. With rationale clauses, it is controlled by the subject of the embedding sentence. Purpose clauses have a more complicated structure. They always contain at least one latent argument which is controlled by the object of the embedding sentence. This may be the subject of the infinitive phrase, but it can also be any other argument. In the latter case, the reference of the latent subject is free, meaning that it is determined by semantics and pragmatics.7 Example (1) illustrates a rationale clause, (2) a purpose clause:
Both examples clearly illustrate the purposive semantics of the adjunct infinitive sketched above in fn. 3. The infinitive phrase denotes an event which the agent intends to result from the action depicted by the embedding verb. The difference lies in the control relation, subject control in (1) versus object control in (2). In the oldest language, adjuncts never occurred with the complementizer ὥσθε. Quite surprisingly, its use originated in complements and spread to adjuncts only later.8 In the context of this paper, the most important fact about both types of adjunct infinitives is that they always have a modal, purposive reading.
Complements, both modal and nonmodal, are attested already in Mycenean. This is quite remarkable since in Vedic, another IE language with early attestation, complements are a very recent phenomenon and nonmodal infinitives are absent altogether.9 For Mycenean see examples (3) and (4):10
In example (3) the infinitive depends on the participle ophēlontes. Hence it has a deontic reading. In (4), the infinitive phrase, an AcI, depends on the speech act verb phāsi. It denotes the content of an utterance, i.e. something like “that q”.11
We may thus conclude that the development of the infinitive had already reached a very advanced state in early Greek, since modal semantics are no longer obligatory even in the oldest attestations.
2.3. A short prehistory: matrix infinitives
From the earliest attestations of alphabetic Greek, the language makes use of unembedded infinitives used as main verbs. Two types can be discerned. The first, which may well be of PIE origin since it has parallels in other IE languages such as Vedic, licenses a subject in the nominative:
In this example, the infinitive follows an optative, λύσαιτε, to which it apparently stands in free distribution. As with λύσαιτε, the subject, a 2.pl., remains unexpressed. Still, the latent subject bears nominative case. This is evident from the appositive participle ἁζόμενοι, whose phi-features are licensed through agreement with it. Overt subjects in the nominative are attested in the 3.pers., see Schwyzer (1950: 382). This type of matrix infinitive probably developed out of the purposive use of the infinitive via reanalysis, since both share the buletic reading and the surface syntax, most notably the absence of an overt subject (Keydana 2013: 74 fn. 37). Though still frequent in Homeric Greek, it vanishes in the classical language (Schwyzer 1950: 380). However, matrix infinitives are not ousted from the system altogether. Rather, in a clearly inner-Greek development, a new type arises: the unembedded AcI, used mainly in requests and orders.12
3. The unembedded AcI
Like the matrix infinitive with a nominative subject, the unembedded AcI is attested from Homeric Greek onwards:
(6) is a typical example of the unembedded AcI.13 As in (5), the infinitive parallels a finite verb in the preceding clause, in this case the imperative ἐχέτω. The accusative subject is overt, and there is evidently no embedding verb. Note that the context is some sort of treaty—the typical genre for the unembedded AcI.
Unembedded AcI are also attested in other literary genres. The following example from Herodotus is the climax of a message from Mardonius, the Persian military commander, to the Lacedaemonians. Mardonius challenges his enemies and offers the following deal:
Thukydides uses the unembedded AcI in an account of a treaty between the Athenians and the Lacedaemonians:
The affinity of the unembedded AcI for legal texts is confirmed by its frequent use in inscriptions. As already observed by Schwyzer (1950: 380), AcI “herrscht in den Inschriften der archaischen Zeit fast allein und wird erst später mehr und mehr durch den ursprünglich nicht ganz gleichwertigen τω-Imperativ ersetzt.”
The most ancient epigraphic attestation can be found in a decree from Dreros, after 650 BC:14
It is highly unlikely (and has, to my knowledge, never been proposed) that in this early Cretan example the AcI depends on ἔϝαδε (as it does in ex. 14 below), mainly because of the cataphoric pronoun ἇδε, which, together with the dative πόλι, saturates the subcategorization frame of (ϝ)ἁνδάνω. The assumption of an independent AcI is further strengthened by the fact that the actual decree opens with a subordinate ἐπεί-clause.
In later inscriptions, unambiguous examples abound. Thus, the AcI is the very first sentence of an Arcadian inscription from Tegea, ca 390 BC:
Substantial additional evidence for independence comes from alternations of finite verb forms and infinitives, which occur frequently in inscriptions from various dialects. Cf. the following example from an Athenian decree, ca. 440–432 BC:
The decree begins with an imperative, παρασχόντον, followed by an infinitive, hελέσθαι, another imperative, νεμάντον, and finally the infinitive καταστε̃σαι. The same pattern is attested in the famous Gortyn law, ca. 480–450 BC, cf. Baunack and Baunack (1885: 76). In the following paragraph we find an imperative, καταδικαδδέτο̄, followed by the infinitive κρίνεν:
In both examples, no embedding verb intervenes between imperative and infinitive. Cases like these are broadly attested and lead Schwyzer to the obvious conclusion that the unembedded AcI and the imperative stand in free distribution (Schwyzer 1950: 383): “Häufig wechselt in Inschriften der a.c.i. mit τω-Formen ohne ersichtlichen Grund.”15
The origin of the matrix AcI is, again according to Schwyzer, “Ellipse des Verbum regens” (1950: 383). The term ellipsis, however, is slightly misleading, because ellipsis is usually understood as a synchronic process. The matrix AcI does not fit this picture. Ellipsis is typically triggered by a preceding conjunct. In the examples given here, as in many others, this condition is clearly not fulfilled. In (10), for example, the matrix AcI is the first sentence of the inscription. In (11) and (12), the AcI alternates freely with the imperative. We thus have to conclude that with unembedded AcI Schwyzer’s ellipsis is to be understood as a diachronic process. It is what Evans (2007) calls a conventionalized ellipsis.
The grammaticalization path seems obvious. In the first stage, AcI in the context of legal texts or more generally any type of request were embedded under some verbum regens. Examples of such embedded AcI are attested in Greek inscriptions, though clear examples are surprisingly hard to come by. This is due to the fact that in most decrees potentially embedding verbs are used as a formula independent of the actual decree itself (as e.g. in ex. 9). Still, embedding under the very same verbs is attested. The embedding verbs denote judicial acts (this type is quite rare, but see ex. 13) or psychological states (ex. 14). Also attested are simple verba dicendi (ex. 15). It is noteworthy that in the context of legal texts neither deontic modal verbs nor verbs of request are ever attested.
The earliest example of an embedding verb denoting a judicial act that I am aware of is from Cimolos, after 338 BC:
Especially in Cretan, ἔϝαδε is frequent; mostly, however, the verb is used as part of the introductory formula. The earliest of the rare examples of an AcI actually still embedded under ἁνδάνω is from Dreros, 7th c. BC:
(14) is a probable example of a dependent AcI mainly because of the absence of any cataphoric pronoun coreferent with the AcI (as in ex. 9). The preposed relative clause is a weaker counterargument than the ἐπεί-clause in (9), since it depends on the (latent) object of τίντεσθαι.
Introductory εἶπε is typical for early Attic decrees. The following is a well-preserved example from 424/3 BC:
δοκεῖν is yet another verb frequently attested in decrees. It is discussed below in section 5.1.
In the second stage of the grammaticalization process, the AcI became autonomous. At least part of this grammaticalization path is attested elsewhere: as shown by Heine and Kuteva (2002: 92), complements can turn into purpose clauses, and complementizers into purpose clause markers. For an example of de-embedding in German, which occurs in similar contexts, see below, ex. 17. As can be seen from the data given so far, the second stage did not supersede the first. Rather, embedded and unembedded AcI are attested side by side in alphabetic Greek, most prominently in inscriptions.
Both the source and the target of the change are undisputed. Still, some important questions remain unanswered. How did the AcI acquire a deontic reading? And why did it become de-embedded in the first place?
To answer these questions, we have to look into the actuation of the change and to identify possible factors by which it may have been triggered. Does it somehow emerge due to structural reasons? Or is it routed in usage? And if so, who is responsible: speaker, hearer, or both?
In recent years, actuation has been a topic of intense study in the field of phonology. And since both grammaticalization and phonological change take place in communication between speakers and hearers, it seems advisable to treat them alike.17
I do not consider the language system a plausible culprit for initiating language change. The language system does not exist outside the knowledge of a language acquired by every speaker. It emerges and changes during the course of every speaker’s life as a consequence of the fact that the speaker is constantly confronted with language use, both as a speaker and as a hearer (see, as an illustration, the impressive data in Harrington et al. 2000). Already acquired knowledge as well as cognitive predispositions, be they language-specific or not, may influence microchanges in each individual speaker, but they cannot actuate them.18 Rather, change emerges out of language use, and since use always presupposes two parties, the speaker and the listener, both are possible initiators.
Recent studies in phonological change have shown that change typically emerges from the interplay of variation produced by the speaker and categorization made by the listener. Lautabsichten, to borrow a term from Trubetzkoy (1958), are never pronounced in isolation. Thus, production always suffers from coarticulation effects and various degrees of undershoot, the result being substantial variation. This is a challenge to hearers who have to cope with it by forming hypotheses as to the underlying Lautabsicht. This categorization is determined by prior experiences (known types of deviation, knowledge of the lexicon etc.). Thus, hearers confronted with new or infrequent patterns are especially prone to form ‘wrong’ hypotheses. John Ohala was the first to study this mechanism in greater detail. He famously distinguishes two types of change, hyper- and hypocorrection (Ohala 2012). Hypocorrection shall be discussed here in some detail because of its similarity to what we propose happened in the development of the unembedded AcI. Ohala’s now classic example of hypocorrection is a scenario in which a speaker utters the string /ut/ as [u̟t]. This slightly deviant pronunciation is due to the fact that any speaker prepares the coronal gesture of the [t] by fronting the tongue body during the articulation of the preceding vowel. As a result, the transitions the hearer is confronted with in [u̟t] are ambiguous.19 The hearer perceives these transitions as cues for the following coronal, but at the same time also as cues for /y/. Depending on his perception grammar formed by prior experience, this ambiguity may lead to different results. If the cue for coronals outweighs that for a front vowel, nothing changes. But if the latter cue outweighs that for coronals, the hearer will misparse [u̟t] as /yt/, the result being a microchange, subphonemic at first and originally restricted to this very hearer.20 A similar change is perceptual metathesis which occurs frequently with rhotics (Blevins and Garrett 2004). E.g. in Vedic, a deverbal noun /karʂʈu-/ ‘plowing’ existed, which was (probably) pronounced as [kʌrʂʈu-]. Rhotics, however, are a challenge for perception since their F3, the most salient cue for a rhotic, is elongated. This makes it difficult for hearers to decide on the exact place of the rhotic within a given string. Again, ambiguity arises: hearers may perceive the word as /karʂʈu-/ or as /kraʂʈu-/. If the latter happens, metathesis takes place.
Such microchanges are determined by two types of parameters. The first belong to the realm of channel bias (Garrett and Johnson 2013). Channel bias subsumes coarticulation effects (like in [u̟t] above) or the limits of perception, e.g. elongated cues (as in the perceptual metathesis discussed above). Channel bias is complemented by analytic bias, i.e. the knowledge of a language against which perception takes place. Analytic bias determines the sensitivity to variation (Moreton 2008) and the direction of cue interpretation (Hamann 2009). In our Vedic example, channel bias would lead to a chance distribution of ‘correct’ and metathesized perceptions. It is analytic bias, in this case the knowledge that Vedic prefers sonorants in complex onsets to sonorants in complex codas (Keydana 2008), which pushes hearers in the direction of the metathesized form.21
Actuation outside of phonology is based on similar preconditions. In phonology, undershoot and coarticulation obscure the Lautabsicht and make the hearer’s task challenging. In semantics and pragmatics, it is the attempt to be economical and effective which leads the speaker to make utterances which are hardly ever fully explicit. Speakers rely heavily on a shared common ground, they use anaphors, and sometimes they even are innovative. They may use new implicatures, thus inviting inferences (Traugott and Dasher 2002). As in phonology, listeners have to cope with deficient or ambiguous information. They have to reconstruct the meaning of an utterance, to resolve anaphors, to accomodate presuppositions, and even to learn new generalized invited inferences (GIINs). At the same time, they avoid pragmatic overload (Eckardt 2009),22 thus constraining the creativity of the speakers (Schwenter and Waltereit 2010).
5. A scenario for the unembedded AcI
How would an actuation scenario for the unembedded AcI look like?
As mentioned above, it is quite rare that AcI in legal contexts are embedded under verbs denoting judicial speech acts as in example (13). Verbs denoting requests do not occur at all. Rather, the embedding verb is typically a simple verbum dicendi. With AcI, these verbs are used in two different contexts. One, exemplified by (4) above, is reported speech. The other is a request, or to put it slightly more formally, a deontic speech act. Cf. (15) above and the following example from Cyrene, 4th c. BC:
(16) contains a decree cast in the words of Damis, son of Bathuklēs. Its content is introduced by ἦιπε. The core lexical meaning of the construction εἰπεῖν + AcI is simply “say [that q] AcI ”—deviations from this semantics are always fully predictable from the context. In the following, I will refer to this lexical meaning as type 1. (16), however, is a decree. It is not the intention of the speaker to inform the hearer of some utterance made by Damis. Rather, the whole point of the inscription is that the content of this utterance is legally binding, meaning that the speaker invites the hearer to infer that the speech act expressed by ἦιπε is actually deontic. This conveyed meaning of εἰπεῖν + AcI, “request [that q] AcI ”, will be referred to as type 2. Note that this invited inference does not change the lexical meaning, since it is restricted exclusively to well-defined contexts in which the speaker can rely on the ability of the hearer to derive the intended side meaning felicitously. Note also that this innovation clearly targets the embedding verb—the semantics of the AcI remains unchanged. This is evident from the fact that the same type of inference occurs with complements other than the AcI, e.g. in constructions like εἰπεῖν τάδε (cf. also ἅδ᾽ ἔϝαδε in ex. 9 above). Here, too, the conveyed meaning is clearly deontic.
Cases like (16) are crucial for the scenario developed here because they form what Heine (2002) calls a bridging context: even in a decree, a construction ‘speech act verb (like εἰπεῖν) + AcI’ is ambiguous and may be analyzed either as type 1 or as type 2. It is up to the hearer to infer which use is intended.
As I argued above, the speaker probably intends the utterance as type 2. Why he avoids a verb of request remains unknown to us. But as a result of this choice, he invites an inference, namely that with the use of the speech act verb a deontic meaning is implied. In the given context, this is hardly a bold move. The speaker can rely on the fact that when hearing or reading a decree the hearer cannot fail to grasp the conveyed meaning of the utterance. To borrow a term from Du Bois (2007: 163), the hearer aligns to the situation forced upon him by the speaker, and there is no doubt that the discourse succeeds as intended by the speaker. Still, assuming that hearers and speakers parse utterances on the premise that meaning is essentially compositional,23 the hearer has to decide what part of the utterance the deontic reading is tied to. And since (16) is a bridging context, he has two options. He may analyze the sentence exactly as intended and infer a modal side meaning of the verb (type 2). In this case, nothing changes. But he may also classify the verb as type 1. The deontic modality, then, must be associated with the AcI. Surprisingly, this alternative reading is pragmatically equivalent to the intended one, since “say [that necessarily q] AcI ”, the reanalysis made by the hearer, can be used in exactly the same contexts as “request [that q] AcI ”. As a consequence, the communication remains felicitous and the reanalysis goes unnoticed. As the hearer has no reason to reassess his ‘wrong’ interpretation, he will eventually turn this new Particularized Conversational Implicature (PCI) into the Generalized Conversational Implicature (GCI) that the AcI may express a request (thus returning a modal reading to the infinitive once again).
This miscategorization, then, accounts for the deontic modality of the unembedded AcI. But how did it turn into a matrix infinitive? The answer lies with the hearer turned speaker.
The AcI started out as a device to encode subordination. By extracting a mistaken GIIN in the context of legal texts, hearers derived a new GCI for the AcI by associating it with deontic modality. This conventionalization then leads to grammaticalization, as the former GCI turns into the primary function of the AcI (in the genre of legal inscriptions and treaties in general). As a consequence, the old function, subordination, which is still very much alive outside legal contexts, is demoted. This demotion together with the newly established firm connection of the AcI with deontic modality finally opens up new vistas for syntax. Since subordination is no longer at the core of the construction, it can now be used outside of embedding. The result is a construction in direct competition with the imperative, i.e. exactly what we find in the examples from Homer onwards.
The actuation sketched here is a hypocorrection. The speaker invites an inference which can potentially be attached to the embedding verb and to the embedded structure alike. Thus, he generates ambiguity, but he does so involuntarily. The change is clearly not speaker-invited. Still, the structural ambiguity forces the hearer to decide between two possible analyses. The actual change occurs if the hearer opts for the ‘wrong’ analysis. Thus, it is hearer-based. This scenario resembles quite strongly the hypocorrection leading to perceptual metathesis. In both cases, the information offered by the speaker is correctly perceived by the hearer: he processes cues for the lowered F3 of rhotics, and likewise he infers deontic modality. In both cases, however, the exact location of the information is dubious. Lowered F3 is elongated, and it is unclear if it is associated with a segment in the coda or rather in the onset. Likewise, the implicature can be associated both with the embedding verb and the AcI without any difference in meaning. In both cases the change is essentially triggered by channel bias, although it has analytic overtones. With metathesis, it is the a priori knowledge that lowered F3 must be tied to one segment. With the AcI, it is the premise that semantics is essentially compositional: deontic modality has to be encoded somewhere.
The proposed shift from mere subordination to deontic modality and the subsequent de-embedding have a striking typological parallel in German sentences like (17):24
Like the Greek AcI, German subordinate sentences with dass simply denote embedding: “that q”. Sentences like (17), however, lack the embedding verb. Remarkably, they remain verb-final and thus still look like subordinate clauses. Like their Greek counterpart, they have a clear deontic reading which is not part of the original semantics of dass-embedding. Both observations clearly point to a scenario similar to the Greek one: the construction started out under embedding and acquired its semantics due to a hearer-induced reanalysis. The development is slightly less cogent than in the Greek case, since de-embedded deontic dass seems to be licensed only if a stressed particle like ja or bloß is added (while this type of particle is not licensed in subordinate sentences).25
5.1. δοκεῖν as a bridge verb
So far, the current proposal suggests embedding under verba dicendi as the point of origin for the development of unembedded AcI. However, another frequent introductory verb in legal inscriptions is δοκεῖν. The verb is mostly used as a formulaic opening syntactically independent of the rest of the inscription. But as with other verbs mostly attested in formulaic use, examples where the verb still clearly embeds an infinitive do exist. The most archaic is from the late 7th century part of the already quoted 4th century inscription from Cyrene, SEG 9:3:
The initial ἔδοξε is clearly formulaic. Remarkable, however, is the following δοκεῖ. On the one hand, it takes the following infinitive ἀποπέμπεν as complement. On the other, ὁριστόν marks the deontic modality explicitly—I am not aware of any other such case in the corpus.
An example without an explicit marker of the deontic reading can be found in a decree from Eretria, 411 BC. Though this analysis cannot be proven, it seems natural to assume that the AcI is still embedded under ἔδοξεν:
Due to the frequency of this verb in decrees, any account of the grammaticalization of unembedded AcI must reckon with embedding under δοκεῖν (rather than εἰπεῖν) as a possible starting point. Since the one example with lexically marked deontic modality is completely isolated, possible scenarios should start from a simple structure δοκεῖν + infinitive or δοκεῖν + AcI. As with εἰπεῖν, these structures form bridging contexts. Lexically, the verb denotes epistemic modality. Examples for an epistemic reading abound from the Iliad onwards. An original epistemic semantics is also confirmed by attestations of δοκεῖν with nominative subjects, which never show a deontic reading. Additional evidence comes from the fact that it is quite straightforward to derive the stronger deontic reading attested in (19) from the weaker epistemic one than vice versa. In the type exemplified by this example, it is again the context which makes it possible for the speaker to convey deontic modality as an additional side meaning. As with verba dicendi, the speaker invites the hearer to infer a semantics “it seems necessary [that q] AcI ”, while the hearer erroneously perceives “it seems [that necessarily q] AcI ”. δοκεῖν is thus as likely a starting point for the change as any verbum dicendi.
6. Extending the analysis: absolute constructions
Unembedded AcI are just one case of hearer-induced grammaticalization. Another striking example are absolute constructions. In the following, I sketch a possible scenario for the Greek genitive absolute, following the basic ideas put forward in Lowe (2015), which, although intended to account for the Vedic locative absolute, can be applied to the Greek data nicely. The Greek genitive absolute presupposes case syncretism, thus slightly obscuring the origins of the construction (Keydana 1997: 74–77). Nonetheless, a possible scenario can be given within the Greek system.
I follow Lowe (2015: 103) and Ruppel (2013: 211) in assuming that the construction originated in adnominal participles denoting natural phenomena like the rising of the sun or the advent of seasons. Since genitives have a temporal use in Greek, a form like g.sg. ἔαρος could mean “in spring”. An example for this use can be found in the 19th Homeric hymn:
Here, the adjective πολυανθέος modifies ἔαρος as an epitheton ornans. As such, it has a nonrestrictive reading. The whole NP thus refers to spring as one holistic entity. In the following example, however, the modifier is clearly restrictive:
In a nonabsolute reading, the restrictive participle is used to refer to a “particular manifestation” (Lowe 2015: 104) of spring—its incipient early phase as opposed to other such manifestations as e.g. late spring. The sentence, however, is ambiguous and lends itself to an alternative analysis. Instead of understanding ἔαρος here as a particular manifestation of spring, it can also be taken as referring to the unique holistic concept of spring as in (20) above. Under this reading, however, the participle can no longer be taken to be adnominal, since ἱσταμένου necessarily restricts the meaning of the noun it modifies, and restricting a unique reference does not make sense. Rather, it forces a reanalysis with a sentence-like structure where the participle heads the phrase. This, then, is the absolute construction.26
Ambiguous sentences like (21) are thus a probable source of the emergence of the genitive absolute. The scenario developed here parallels exactly the one proposed above for the unembedded AcI, where the starting point for the reanalysis is a bridging context. As with the AcI in deontic contexts, the speaker intends one reading, namely early spring as one manifestation of spring. The hearer, however, opts for another interpretation of ἔαρος, viz. that of spring as a unique and holistic concept. As a consequence, this forces him to assume a different syntactic representation. The reanalysis goes unnoticed, since the discourse remains felicitous under the new interpretation. The moment the hearer turns speaker, however, the new construction becomes manifest—at least if used in unambiguous contexts. Absolute constructions are thus another example of hearer-induced grammaticalization.
It is undisputed that the embedded AcI in deontic contexts is the source of the unembedded AcI. In decrees and other unambiguous contexts, the message intended by the speaker can easily be conveyed without relying exclusively on lexical meaning. The speaker is thus allowed to invite an inference without risking pragmatic overload. The hearer, on the other hand, is supposed to infer the invited inference that the embedding verb conveys a deontic meaning in spite of its lexical semantics. However, he misparses the implicature: instead of attributing the deontic modality to the embedding verb, he associates it with the AcI. Although this interpretation deviates from the speaker’s intentions, it goes unnoticed because the communication succeeds either way. Nonetheless, the reanalysis has significant consequences the moment the hearer conventionalizes the implicature. As soon as the deontic modality is tied to the AcI, this construction can be used to denote just that and is no longer restricted to use in embedded constructions. Conventionalization thus leads to syntactic change. A similar type of hypocorrection lies at the heart of absolute constructions. Thus, both developments point in the same direction: at least some types of grammaticalization and at least some types of sound change work exactly alike. It remains to be seen if it is possible to develop a unified theory of actuation covering both phenomena.
Bach, E. (1982). Purpose clauses and control. In P. Jacobson and G.K. Pullum (Eds.), The Nature of Syntactic Representation, pp. 35–57. Dordrecht: Reidel.
Bary, C. and E. Maier (2014). Unembedded indirect discourse. In U. Etxeberria, A. Fălăuş, A. Irurtzun, and B. Leferman (Eds.), Proceedings of Sinn und Bedeutung 18, pp. 77–94.
Baunack, J. and T. Baunack (1885). Die Inschrift von Gortyn. Leipzig: Hirzel.
Bermúdez-Otero, R. and G. Trousdale (2012). Cycles and continua: on unidirectionality and gradualness in language change. In T. Nevalainen and E.C. Traugott (Eds.), Handbook on the history of English: rethinking and extending approaches and methods, pp. 691–720. Oxford: Oxford University Press.
Blevins, J. (2004). Evolutionary Phonology. The emergence of sound patterns. Cambridge: Cambridge University Press.
Blevins, J. and A. Garrett (2004). The evolution of metathesis. In B. Hayes, R. Kirchner, and D. Steriade (Eds.), Phonetically Based Phonology, pp. 117–156. Cambridge: Cambridge UP.
Debrunner, A. (1954). Altindische Grammatik, Volume II,2. Die Nominalsuffixe. Göttingen: Vandenhoeck & Ruprecht.
Du Bois, J.W. (2007). The stance triangle. In R. Englebretson (Ed.), Stancetaking in Discourse. Subjectivity, evaluation, interaction, pp. 139–182. Benjamins.
Duhoux, Y. (2008). Mycenean anthology. In Y. Duhoux and A. Morpurgo Davies (Eds.), A Companion to Linear B. Mycenean Greek Texts and their World. Vol. 1, Chapter 9, pp. 243–393. Louvain-La-Neuve: Peeters.
Eckardt, R. (2009). APO: Avoid pragmatic overload. In M.-B. Mosegaard Hansen and J. Visconti (Eds.), Current trends in diachronic semantics and pragmatics, pp. 21–41. Bingley: Emerald.
Eckardt, R. (2012). Grammaticalization and semantic reanalysis. In K. v. Heusinger, C. Maienborn, and P. Portner (Eds.), Handbook Semantics, Berlin, pp. 2675–2702. Mouton de Gruyter.
Effenterre, H. v. (1946). Inscriptions archaïques crétoises. Bulletin de correspondance hellénique 70, 588–606.
Effenterre, H. v. and F. Ruzé (1994). Nomima. Recueil d’ inscriptions politiques de’l archaisme grec, Volume 1. Roma: Ecole Française de Rome.
Evans, N. (2007). Insubordination and its uses. In I. Nikolaeva (Ed.), Finiteness: Theoretical and Empirical Foundations, pp. 366–431. Oxford: Oxford University Press.
Evans, N. and H. Watanabe (Eds.) (2016). Insubordination, Volume 115 of Typological Studies in Language. Amsterdam; Philadelphia: Benjamins.
Ferraresi, G. (2016). Zur Konjunktion dass in Exklamativsätzen und in Rekursionsstrukturen. In R. Lühr (Ed.), Idiosynkrasie. Neue Wege ihrer Beschreibung, Wiesbaden, pp. 81–100. Reichert.
Fortson, B.W. (2012). Latin -rier and its Indo-Iranian congeners. Indogermanische Forschungen 117, 75–118.
Garrett, A. and K. Johnson (2013). Phonetic bias in sound change. In A.C. Yu (Ed.), Origins of Sound Change. Approaches to Phonologization, pp. 51–97. Oxford: Oxford University Press.
Goldstein, D. (2016). Classical Greek Syntax. Wackernagel’s Law in Herodotus, Volume 16 of Brill’s Studies in Indo-European Languages and Linguistics. Leiden; Boston: Brill.
Grossman, E. and S. Polis (2014). On the pragmatics of subjectification. The grammaticalization of verbless allative futures. Acta Linguistica Hafniensia 46(1), 1–40.
Hamann, S. (2009). The learner of a perception grammar as a source of sound change. In P. Boersma and S. Hamann (Eds.), Phonology in Perception, Volume 15 of Phonology and Phonetics, pp. 111–149. Berlin; New York: Mouton de Gruyter.
Harrington, J., S. Palethorpe, and C. Watson (2000). Monophthongal vowel changes in Received Pronunciation: an acoustic analysis of the Queen’s Christmas broadcasts. Journal of the International Phonetic Association 30(1/2), 63–78.
Haspelmath, M. (1989). From purposive to infinitive—a universal path of grammaticization. Folia Linguistica Historica 10, 287–310.
Haspelmath, M. (1998). Does grammaticalization need reanalysis? Studies in Language 22, 315–351.
Heine, B. (2002). On the role of context in grammaticalization. In I. Wischer and G. Diewald (Eds.), New reflections on grammaticalization, pp. 83–101. Amsterdam; Philadelphia: Benjamins.
Heine, B. and T. Kuteva (2002). World Lexicon of Grammaticalization. Cambridge, Mass. and London: Cambridge University Press.
Himmelmann, N. (2004). Lexicalization and grammaticalization: Opposite or orthogonal? In W. Bisang, N.P. Himmelmann, and B. Wiemer (Eds.), What Makes Grammaticalization—A Look from its Fringes and its Components, pp. 21–42. Berlin; New York: Mouton de Gruyter.
Jacobsthal, H. (1907). Der Gebrauch der Tempora und Modi in den kretischen Dialektinschriften. Strassburg: Trübner.
Johnston, M.J. (1998). A syntax and semantics for purposive adjuncts in HPSG. http://www.cse.ogo.edu/~johnston/.
Keydana, G. (1997). Absolute Konstruktionen in altindogermanischen Sprachen, Volume 40 of Historische Sprachforschung, Ergänzungsheft. Göttingen: Vandenhoeck & Ruprecht.
Keydana, G. (2008). Hohe Sonorität in der Koda: Indogermanische Lösungen für ein phonetisches Problem. Historische Sprachforschung 121, 54–64.
Keydana, G. (2013). Infinitive im R̥gveda: Formen, Funktion, Diachronie, Volume 9 of Brill Studies in Indo-European Languages and Linguistics. Leiden: Brill.
Koerner, R. (1993). Inschriftliche Gesetzestexte der frühgriechischen Polis. Aus dem Nachlaß von Reinhard Koerner herausgegeben von Klaus Hallof. Köln: Böhlau.
Kratzer, A. (2002). The notional category of modality. In P. Portner and B.H. Partee (Eds.), Formal Semantics. The Essential Readings, pp. 289–323. Blackwell.
Lowe, J.J. (2015). Participles in Rigvedic Sanskrit. Oxford Studies in Diachronic and Historical Linguistics; 17. Oxford: Oxford University Press.
Moreton, E. (2008). Analytic bias and phonological typology. Phonology 25(1), 83–127.
Noël, D. (2007). Diachronic construction grammar and grammaticalization theory. Functions of language 14(2), 177–202.
Noël, D. and T. Colleman (2009). The nominative and infinitive in English and Dutch: An exercise in contrastive diachronic construction grammar. Languages in Contrast 9(1), 143–180.
Ohala, J. (2012). The listener as a source of sound change: an update. In M.-J. Solé and D. Recasens (Eds.), The Initiation of Sound Change. Perception, production, and social factors, pp. 21–35. Amsterdam: Benjamins.
Richards, M. (2008). Two kinds of variation in a minimalist system. Linguistische Arbeitsberichte 87, 133–162.
Ruppel, A. (2013). Absolute Constructions in Early Indo-European. Cambridge: Cambridge University Press.
Schwenter, S.A. and R. Waltereit (2010). Presupposition accommodation and language change. In K. Davidse, L. Vandelanotte, and H. Cuyckens (Eds.), Subjectification, Intersubjectification and Grammaticalization, pp. 75–102. Berlin; New York: Mouton de Gruyter.
Schwyzer, E. (1950). Griechische Grammatik. Zweiter Band. Syntax und syntaktische Stilistik. München: Beck.
Stüber, K. (2000). Zur Herkunft der altindischen Infinitive auf -sáni. Münchener Studien zur Sprachwissenschaft 60, 135–167.
Traugott, E.C. (2008). Grammaticalization, constructions and the incremental development of language: Suggestions from the development of degree modifiers in English. In R. Eckardt, G. Jäger, and T. Veenstra (Eds.), Variation, Selection, Development—Probing the Evolutionary Model of Language Change, pp. 219–250. Berlin; New York: Mouton de Gruyter.
Traugott, E.C. (2010). Grammaticalization. In S. Luraghi and V. Bubenik (Eds.), Continuum Companion to Historical Linguistics, pp. 269–283. London: Continuum Press.
Traugott, E.C. and R.B. Dasher (2002). Regularity in Semantic Change. Cambridge, Mass. and London: Cambridge University Press.
Traugott, E.C. and G. Trousdale (2013). Constructionalization and Constructional Changes. Oxford: Oxford University Press.
Trubetzkoy, N.S. (1958). Grundzüge der Phonologie. Göttingen: Vandenhoeck & Ruprecht.
Waltereit, R. (2011). Grammaticalization and discourse. In H. Narrog and B. Heine (Eds.), The Oxford Handbook of Grammaticalization, pp. 413–423. Oxford: Oxford University Press.
This paper grew out of talks given in Zurich and Vienna. It is my pleasure to thank the audiences for lively and stimulating discussions. I also want to thank two anonymous reviewers for their valuable comments.
For the grammaticalization path from purposive event noun to infinitive see Haspelmath (1989). It cannot be illustrated with data from Ancient Greek since event nouns in purposive constructions like the dativus finalis are not (or rather no longer) attested—the dativus finalis being generally very rare (but see ἔριδι in ex. 2). Final datives and homonymous infinitives still exist side by side in Early Vedic. Cf. RV 2.31.3 with an NP headed by an event noun (ánu nú sthāti […] rátham mahé sanáye … ‘he will stand beside our chariot for great gain …’) and RV 1.116.21 with an infinitive heading a VP (āvataṃ ráṇāya váśam […] sanáye sahásrā ‘you gave Vaśa the help for the battle, in order for him to gain thousands.’). On the Vedic data see Keydana (2013: 212).
The issue of the exact relation of grammaticalizations to constructionalizations, however, is still a matter of debate; see e.g. Noël (2007), Noël and Colleman (2009), the discussion in Waltereit (2011: 414–415), and Traugott and Trousdale (2013).
As deontic modality is mentioned here for the first time, a short note on modal semantics may be in order. Any proposition which is evaluated relative to possible worlds is modal. The set of worlds (each a set of propositions) accessed in the evaluation determines the nature of the modality. Thus, a purposive infinitive like the one in ex. 1 below is buletic because the worlds accessed conform to the intention of the intentional agent of the embedding sentence as presupposed by speaker and hearer. Deontic readings imply that only those worlds are accessed which accord with man-made rules or laws. Epistemic readings access those worlds which match the knowledge about the world of speakers and hearers. The worlds accessed are the modal base of the evaluation. For more details and for a formal account of modal semantics, see Kratzer (2002).
Another candidate for a PIE infinitive marker not continued in Greek is *-dhi̯ōi̯, on which see Fortson (2012).
Schwyzer (1950: 359) ponders the possibility that originally the infinitive in -έμεν had a locative meaning, whereas -εναι was a true purposive. βῆ δ᾽ ἴμεν would thus mean “schritt aus beim Gehen”, βῆ δ᾽ ἴεναι “schritt aus zum Gehen”. This hypothesis cannot be validated. It is, however, quite improbable, since as early as the Iliad clear cases of purposive -έμεν-infinitives are attested, whereas unequivocal cases of locative meanings remain to be found.
This distinction goes back to Bach (1982: 35–36). Bach uses the term in-order-to-clause instead of rationale clause. The latter designation, which is preferable for not being language-specific, goes back to Johnston (1998: 9).
For an explanation of this rather strange control pattern, see Keydana (2013: 98).
The only possible case of an adjunct with ὥσθε in Homer is I 42, which, however, is best analyzed as a complement depending on ἐπέσσυται.
They are, however, attested in Old Avestan (Keydana 2013: 76 fn.40). Hittite infinitive syntax deviates substantially from that of the other ancient IE languages. The patterns attested, especially those in serial verb constructions, are probably due to substrate influence. In any case, they are innovations.
Depending on one’s interpretation, up to nine infinitives are attested in Mycenean. All of them are complements, the most unambiguous being dependent either on ophēlo-, or (twice, on one and the same tablet) aphiēmi. The lack of attestation of adjunct infinitives is probably due to the genre of the Mycenean texts.
Two more AcI with verba dicendi are attested in Mycenean, both on Eb 297 (on which Duhoux 2008: 301–302). While the second one is parallel to Ep 704, the first one is peculiar since, if the interpretation is correct, it contains a latent subject: eukhetoi kwe e-to-ni-jo ekhehen thehōi ‘and she claims to hold an e-to-ni-jo for the deity’.
Latin, too, has unembedded AcI. However, the Latin unembedded AcI is restricted exclusively to exclamations and most probably originates in the more general use of the Latin accusative in such contexts. As will be seen in the following section, the domain of use of the Greek unembedded AcI is very different. I thus conclude that—apart from their superficial similarity—the Greek and the Latin unembedded AcI have nothing in common, neither in a diachronic, nor in a synchronic perspective.
Infinitives of both the present stem and the aorist stem are attested in unembedded AcI. See Jacobsthal (1907: 36 sqq.) for details.
The epigraphic data are printed as in the sources indicated.
Matrix AcI are also attested in unembedded indirect discourse (UID), see Bary and Maier (2014). This use of the AcI, however, differs substantially from the one discussed here. On the one hand, UIDs typically follow embedded AcI depending on speech act verbs; if they do not, a speech act verb can always be inferred from the context. On the other, UIDs are always interpreted reportatively.
The translation follows Effenterre (1946: 591); see also Koerner (1993: 338) and Effenterre and Ruz’é (1994: 271).
For an approach to grammaticalization similar to the one developed here, see Grossman and Polis (2014).
This is also acknowledged by scholars who treat syntactic change as a change in (micro)parameter settings. See e.g. Richards (2008).
This is the situation dubbed ‘chance’ by Blevins (2004).
For this version of Ohala’s scenario cf. Hamann (2009), for possible pathways from subphonemic to phonemic change cf. Bermúdez-Otero and Trousdale (2012). Hypercorrection is the opposite of hypocorrection, whereby e.g. a speaker utters [yt] for underlying /yt/, but the hearer presupposes coarticulation and perceives /ut/. See Ohala (2012) for details.
Hypercorrection is always triggered by analytic bias.
An anonymous reviewer raised the important question of how pragmatic overload can be measured. Eckardt herself does not raise this issue. The only possibility I see are psycholinguistic experiments with carefully operationalized pragmatic factors. I am not aware that any such research has been undertaken.
For the importance of this premise for studies in grammaticalization, see Eckardt (2012).
Further examples from a variety of languages can be found in Evans (2007).
For a detailed analysis of this sentence type see Ferraresi (2016), who observes that the use of particles is indeed more acceptable if the embedded sentence is topicalized (Ferraresi 2016: 93).
It is certainly true that “the temporal genitive is in itself relatively infrequent and unproductive” (Ruppel 2013: 34). However, other cases of potential ambiguity like the ones discussed by Ruppel (2013: 38–41) hardly fare better. It is quite conceivable that the absolute genitive continues an older construction, probably an absolute locative, which originated along the lines sketched here, and that the fact that this construction is marked by the genitive is a result of Greek case syncretism (Ruppel 2013: 220–221, Keydana 1997: 76–77).