Finiteness is a central concept in many linguistic theories, yet it is poorly understood. In this squib, we provide new data that must be incorporated into current research on finiteness: the Latin infinitival structure known as the “accusative and infinitive” (AcI), which has properties that are typical of canonical nonfinite clauses, can be syntactically unembedded.
It is clear that this is unexpected. A common view—found in one variant or another in Hornstein 1990:115–117, 146–154, Klein 1994, Rizzi 1997, Bianchi 2003, Adger 2007, and Giorgi 2010—is that finiteness is responsible for anchoring the clause to the actual utterance (e.g., for the interpretation of tense). Since a root clause must be temporally anchored to the utterance time, one would not expect nonfinite clauses to be root clauses.
Finiteness has morphological, syntactic, and semantic dimensions, which do not always align. An example from Latin is clauses with historical infinitives, which are morphologically nonfinite but syntactically unembedded and semantically like finite forms in having deictic time reference and speaker assertion semantics. What makes AcIs different from these and similar structures is that they behave like nonfinite clauses both morphologically and semantically, yet are demonstrably syntactically unembedded in certain situations.
2 The AcI Construction
The AcI canonically consists of an infinitive and a nominal subject that is in the accusative rather than the nominative. Nonsubject arguments and adjuncts are realized as in a finite clause. AcIs are common as complements of report predicates, like fateor ‘confess’ in (1), but also occur in longer passages of indirect discourse as in (2).
ego [me amare hanc] fateor.
I. NOM me.ACC love.PRS.INF her.ACC confess.PRS.1SG
‘I confess that I love her.’
(Terence, Andria 896)
[reddi captivos]1 negavit
return.PRS.PASS.INF prisoners.ACC deny.PFV.PST.3SG
[esse utile]1; [illos enim adulescentes esse
be.PRS.INF useful they.ACC for young.ACC be.PRS.INF
et bonos duces]2, [se iam
and good.ACC officers.ACC himself.ACC already
‘He denied that it would be expedient to return the prisoners; for, he said, they are young and good officers, while he was already consumed with age.’
(Cicero, De officiis 3.100)
(2) contains three reported clauses, each of which takes the form of an AcI.1 The two instances of the infinitive esse ‘be’ belong to the first and second AcIs; the infinitive in the third AcI has been elided.2 The report predicate negavit ‘denied’ is sandwiched inside the first AcI, and the second and third AcIs follow without any additional overt report predicate.
Two explanations for the lack of an additional report predicate immediately spring to mind. First, it could be that the three AcIs are asyndetically coordinated and syntactically embedded as a whole under negavit. This analysis is ruled out by the context. The overt report predicate is a negative utterance verb, and the first AcI expresses what is being denied. The next two AcIs, on the other hand, are positively asserted, not denied, and therefore cannot be embedded on a par with the first AcI.
A second possible explanation is that there is an implicit affirmative utterance verb in the structures of the second and third AcIs. This is compatible with the interpretation we expect from the context, but in section 4 we show that such a null verb leads to empirically incorrect predictions.
We therefore defend a third analysis under which the first AcI is syntactically embedded under negavit while the second and third AcIs are syntactically unembedded. Combine this with the claim that AcIs are nonfinite, which we defend below, and we have counterexamples to the idea that nonfinite clauses cannot be root clauses.
3 Latin AcIs Are Nonfinite
There is no agreed-upon set of features to identify a clause as (non)finite. Nikolaeva (2007b, 2010) identifies morphological, syntactic, and semantic features of finiteness. Here, we look at deficient tense, aspect, and mood (TAM) categories, agreement, and subject case assignment. While none of these features can be taken as defining nonfiniteness crosslinguistically, we do show that AcIs pattern with what is expected from nonfinite forms on just about any test that has been proposed and that is relevant for Latin.
3.1 Deficient Tense, Aspect, and Mood
Infinitives cannot express tense relative to the utterance time (Hornstein 1990:147). Infinitives instead either have bound tense,3 which means that they are tenseless and use the matrix tense as their own, or have dependent tense and express time relative to (but not necessarily identical to) the matrix. For aspect and modality, the typical situation crosslinguistically is that these categories can be expressed but to a more limited degree than in finite clauses (Nikolaeva 2010: 1180).
Latin infinitives conform to this picture. They never express absolute tense, only dependent tense (in AcIs) or bound tense (in many control structures). (3) illustrates the three possible “tense” forms of the Latin infinitive.4 (3a) shows them under a present tense report predicate where they are interpreted relative to the utterance time. In (3b), the report predicate is in the past, and the same infinitival forms are now interpreted relative to some past time.
Marcum bonum esse/fuisse/futurum esse
Mark.ACC good.ACC be.PRS/PRF/FUT.INF
‘I say that Mark is/has been/will be good.’
Marcum bonum esse/fuisse/futurum esse
Mark.ACC good.ACC be.PRS/PRF/FUT.INF
‘I said that Mark was/had been/would be good.’
We conclude that Latin AcI structures never express deictic tense, only dependent tense.
Aspect is obligatorily expressed in the indicative, but cannot be expressed in the infinitive (Pinkster 2015:384). Therefore, while (4a) can have both an imperfective and a perfective verb, both will be reported as shown in (4b) with an aspect-neutral infinitive.
‘Mark was eating/ate.’
Marcum cenavisse dico.
Mark.ACC eat.PRF.INF say.PRS.1SG
‘I say that Mark ate/was eating.’
Mood is also deficient, as there is no subjunctive of the infinitive.5 One sees this in conditional structures, which make a distinction between ordinary conditionals in the indicative and “potential” conditionals in the subjunctive, as shown in (5a).6 Both would be reported as in (5b), and since the infinitive cannot express mood, the contrast has been neutralized in the consequent (Menge 2000:825).7
si hoc dicis/dicas,
if this.ACC say.PRS.IND/SBJV.2SG
‘If you say/should say this, you are/would be wrong.’
puto te, si hoc dicas,
think.PRS.1SG you.ACC if this.ACC say.PRS.SBJV.2SG
‘I think that if you say/should say this, you are/would be wrong.’
3.2 Deficient Agreement
Agreement, particularly in person, is also typically deficient in nonfinite forms. Here too, there is crosslinguistic variation: several varieties of Romance have “inflected infinitives,” which agree even in person (Raposo 1987, Ledgeway 1998, 2000, Vincent 1998), and of course there are languages where even finite forms show no agreement in person (e.g., Mandarin Chinese).
Latin presents the canonical picture in which finite forms display rich agreement while infinitives do not agree. Indicatives, subjunctives, and imperatives all agree in person and number with their subjects in Latin. In contrast, participles and gerundives agree with their subject in gender, number, and case, while infinitives and gerunds do not agree at all. The exception is periphrastic forms consisting of an infinitive and a participle, but even here the infinitival part of the periphrasis never agrees, as shown in table 1.
|Active .||Passive .|
|Active .||Passive .|
3.3 No Nominative Assignment
A third property that is often used to distinguish nonfinite forms from finite ones is the absence of nominative case assignment (Nikolaeva 2007b:4). This too is not a universal property of nonfinite forms, as demonstrated, for example, by “personal infinitives” in various Romance languages (Ledgeway 2000, Mensching 2000),8 but again the situation in Latin is (almost) the canonical one. Finite verbs invariably assign nominative case to their subjects,9 while nonfinite forms (participles, infinitives, gerunds, and gerundives) typically do not. An exception is the ill-understood “historical infinitive” that is used in narrative contexts (6).
Graecus primo distinguere et dividere,
Graecus.NOM firstly distinguish.PRS.INF and divide.PRS.INF
quemadmodum illa dicerentur.
as they.NOM.PL say.PRS.SBJV.PASS.3PL
‘The Greek [i.e., the philosopher Philodemus] first drew distinctions and split hairs about how they [i.e., those doctrines] were meant.’
(Cicero, In Pisonem 69)
We briefly return to this construction in section 5. This case apart, the subject of an infinitive appears in the accusative, even when there is no evidence for raising to object, as in (7), where the infinitival clause seems to be the subject of the matrix verb.
. . . si eos hoc nomine appellari fas
if them.ACC that.ABL name.ABL call.PRS.PASS.INF right
est . . .
‘if it is right to call them by that name’
(Cicero, Pro Murena 80)
Of course, the existence of such structures does not in principle rule out that other types of AcI constructions involve raising to object. We cannot go into the details of the Latin-specific literature (see Jøhndal 2012:68–76 for a summary), but two views dominate: either the accusative is a default case (Ferraresi and Goldbach 2003, Goldbach 2003, Calboli 2005) or the infinitive assigns accusative case either directly or mediated by a null complementizer (Cecchetto and Oniga 2002, Melazzo 2005, Jøhndal 2012). Either way, it is clear that the infinitive is unable to assign nominative case and therefore behaves as traditionally expected of a canonical nonfinite form.
4 AcIs Can Be Root Clauses
Despite the nonfinite nature of AcIs, we will argue that AcIs can be used as root clauses (i.e., without any syntactic embedding) in so-called unembedded indirect discourse (UID).10Bary and Maier (2014), who introduced the term UID for contexts like (2), discuss the Ancient Greek discourse particle gar in such contexts, the rough functional equivalent of Latin enim seen in (2). These particles unambiguously indicate the start of a new sentence and regularly11 occupy the second position of their sentences, strongly suggesting a sentence break at the orthographic semicolon in (2) and militating against the coordination view that we dismissed on other grounds in section 1. However, on its own the presence of enim does not tell us whether the sentence it introduces contains an embedding null report verb. We will show that this analysis can be refuted on the basis of both the scope of enim (section 4.1) and evidence from rhetorical questions (section 4.2).
4.1 The Discourse Particle enim in AcIs
iam eum, ut puto, videbo;
now him.ACC as think.PRS.1SG see.FUT.1SG
misit enim puerum . . .
send.PFV.PST.3SG for boy.ACC
‘I will see him, I think. For he has sent a boy . . . ’
(Cicero, Epistulae ad Atticum 10.6.5)
We can represent the logical form (LF) of the second sentence in (8) as in (9). For concreteness, we assume that enim is somewhere in the CP domain. DEIC-PAST relates the event to the utterance time and marks it as prior to it.
(9) [CP enim [TP DEIC-PAST [VP send a boy . . . ]]]
Enim scopes over the whole of its sentence and marks the past sending of a boy as standing in a causal relation to the directly preceding discourse segment. Therefore, we infer a temporal relationship between the two sentences: the sending of a boy precedes the seeing. Moreover, misit puerum ‘send a boy’ is marked with deictic past, meaning that it is past relative to the utterance time.12
Let us now look at two examples with enim in the vicinity of an AcI. In (10), an AcI is explicitly embedded under the report verb dixit ‘said’.
periucundus mihi Cincius fuit
welcome.NOM me.DAT Cincius.NOM be.PFV.PST.3SG
ante diem iii Kal Febr ante lucem; dixit
28 January before light.ACC say.PFV.PST.3SG
enim mihi te esse in Italia . . .
for me.DAT you.ACC be.PRS.INF in Italy.ABL
‘Cincius was a very welcome arrival (before daybreak, 28 January), for he told me that you [= Atticus] were in Italy . . . ’
(Cicero, Epistulae ad Atticum 4.4.1)
In the first sentence, Cicero (writing a letter to his friend Atticus) expresses delight at Cincius’s arrival. In the second sentence, enim signals an explanation of the first sentence. The LF is as in (11). Notice that REL-PRES here denotes bound time (i.e., time relative to some t that is bound by the matrix tense), so that the interpretation is simultaneity with t, which—as indicated by the matrix DEIC-PAST—is prior to the utterance time.
[CP enim [TP DEIC-PAST [VP say [CP [TP REL-PRES
[VP you are in Italy]]]]]]
Enim thus relates Cincius’s past saying—as expressed by dixit ‘said’—to the previous discourse. By contrast, the content of Cincius’s saying, the fact that Atticus is in Italy, cannot be a reason for Cincius’s arrival to be very welcome. We clearly do not get the default interpretation of temporal sequence, namely, that Cincius first was a welcome arrival and then told me that you are in Italy. Rather, the second sentence is the cause for the arrival being welcome.
The PROIEL corpus (Haug and Jøhndal 2008) contains 36 examples such as (10), where enim occurs in a sentence with the verb dico ‘say’; in all those cases, it scopes over the entire utterance. More generally, enim occurs in 186 sentences containing a complement clause (of any kind of verb). Only in 12 of those sentences13 does it surface inside the complement clause, and in all those cases, it still scopes semantically over the whole sentence. None of the 12 examples involves a verb of saying, but (12) shows how enim attaches semantically to the matrix verb despite its position.
cuius enim scelere impulsi ac
whose for evil.deed.ABL egged.on.PRF.PTCP and
proditi simus iam profecto
betrayed.PRF.PTCP be.PRS.SBJV.1PL now truly
‘For you now truly see by whose evil deed we were egged on and betrayed.’
(Cicero, Epistulae ad Atticum 3.8.4)
The generalization that emerges is that enim always scopes over the entire sentence, as outlined in (11). Contrast this with (13). Here, enim marks the start of a new sentence, which, as we have discussed, cannot be coordinated with the first AcI. This means that the second AcI cannot be syntactically dependent on the overt verb of saying, dixi.
dixi [hanc legem Publium Clodium
say.PFV.PST.1SG this.ACC law.ACC Publius Clodius.ACC
iam ante servasse]1; [pronuntiare enim
already before serve.PRF.INF promise.PRS.INF for
solitum esse et non dare]2.
be.in.the.habit.PRF.INF and not give.PRS.INF
‘I said that Publius Clodius had already complied with this law. For, I said, he had been in the habit of promising but not paying.’
(Cicero, Epistulae ad Atticum 1.16.13)
The context is a law being debated, which makes it legal to promise to pay for votes as long as the payment is not actually made. Cicero reports a joke he made in the Senate, with the second AcI delivering the punchline. Enim crucially marks a causal relationship between the first AcI and the second one, and this causal relationship is part of the report. Therefore, the LF in (14), which uses a phonetically null verb ‘say’, gives the wrong semantics.
[CP enim [TP DEIC-PAST [VP say [CP [TP REL-PAST
[VP he is in the habit of promising but not paying]]]]]]
One possibility would be to say that enim scopes under the null verb of saying. However, this involves the very unattractive assumption that enim can scope under null verbs of saying, whereas it always scopes above overt verbs of saying. A simpler generalization is that enim always scopes over all the syntactic material in the sentence in which it occurs. Given that enim in (13) clearly scopes over the content of the report only, this means that the LF must be as in (15).
(15) [CP enim [TP REL-PAST [VP he is in the habit of promising but not paying]]]
(15) is correct as far as it goes, in that it represents the assertoric contents of (13). However, a crucial part of the meaning of (15) is that this proposition is not asserted in the speech act corresponding to the utterance of (13), but is understood as the assertoric content of another speech event that is available in the context.14 We assume that the root infinitive gives rise to a presupposition that it is in the assertoric content of some speech event in the context. This modal contextual dependency is reminiscent of modal subordination (Roberts 1989), whereby a syntactically unembedded modal expression is interpreted within the scope of a modal operator in the previous discourse. Similar kinds of semantic subordination in report contexts are observed for the German report subjunctive (Fabricius-Hansen and Sæbø 2004), as well as for the German modal sollen (Faller 2017); see Solberg 2017 for a detailed semantic analysis of the Latin construction.
The argument from enim can be repeated with other discourse particles as well as with discourse relations that arise in the absence of overt markers. The details would take us too far afield. Let us just observe that in longer stretches of indirect discourse, we infer discourse relations between the propositions expressed in the AcIs, not between assertions of those propositions. Moreover, the inferred discourse relations are interpreted as parts of the report, not as parts of the main narration. That is, the narrator does not commit to these relations. An analysis in terms of implicit embedding verbs will therefore get the discourse structure systematically wrong.
4.2 Rhetorical Questions
Questions in indirect speech are generally in the subjunctive. Latin grammars observe, however, that indirect rhetorical questions are usually not subjunctives, but AcIs (see, e.g., Kühner and Stegmann 1914: 537–538). Such interrogative AcIs have the same relative tense interpretation as the noninterrogative AcIs discussed above. For example, the perfect infinitive in the reported rhetorical question in (16) is evaluated relative to the time of the patricians’ rage.
quod ubi est Romam nuntiatum . . .
this.ACC when be.PRS.3SG Rome.ACC announced
fremunt omnibus locis: . . . num etiam in deos
rage.PRS.3PL all.ABL places.ABLQ even in gods.ACC
immortales inauspicatam legem valuisse?
immortal.ACC ill-omened.ACC law.ACC prevail.PRF.INF
‘When the news arrives in Rome, [the patricians] rage in every corner: . . . Had the ill-omened law prevailed even against the immortal gods?’
However, the grammars also note (e.g., Kühner and Stegmann 1914: 541) that this pattern is found only in UID. Truly embedded (indirect) questions are in the subjunctive even when they are rhetorical (17).
(Sophocles’s sons try to get control of his property on the grounds of imbecility. In court, he points to his latest work, Oedipus at Colonus.)
tum senex dicitur . . . recitasse
then old.man.NOM say.PRS.3SG.PASS recite.PRF.INF
iudicibus quaesisse-que num illud carmen
judges.DAT ask.PRF.INF-and whether that poem.NOM
‘The old man is said to have read [his poem] to the judges and to have asked whether that poem seemed like the work of an imbecile.’
(Cicero, Cato Maior de Senectute 22)
We have shown that UID in Latin does in fact involve syntactically unembedded infinitives—that is, nonfinite root clauses—that are semantically interpreted through a pronominal-like dependent tense linked to a contextually available speech event.
Many languages have clauses that may be root clauses even though they have formal features characteristic of subordination (Evans 2007), and there are parallels—both semantic and syntactic—to the structure we have discussed here. Semantically, we have already noted the similarities with modal subordination—in particular, in German reportive constructions. But these involve forms where there is no evidence for nonfiniteness.
On the syntactic side, root infinitives are not unheard of elsewhere (see, e.g., Grohmann and Etxepare 2003). However, they are associated with a range of nonassertoric meanings such as elliptical answers, jussives/hortatives, rhetorical questions, and counterfactuals (Rizzi 1993). By contrast, the Latin construction does express an assertion, although typically not one of the external speaker.
The interest of our construction is clear if we contrast it with the above-mentioned historical infinitive (6). As pointed out by Nikolaeva (2007a:159), the infinitive in these structures licenses a nominative subject, expresses a speaker assertion, and has deictic past tense reference. We can add that it is a syntactically independent root clause. In short, “the only obvious indicator of nonfiniteness is morphology [viz., the lack of agreement]” (Nikolaeva 2007a:159).
The situation with unembedded AcIs is exactly the opposite. These are also syntactically independent root clauses but, unlike historical infinitives, they do not license nominative subjects, do not express speaker assertions, and do not have deictic time reference. In short, they are truly nonfinite root clauses. This is a finding that must be incorporated into current research on finiteness.
1 Notice that the subject of the first AcI is itself an AcI: what we translate as ‘to return the prisoners’ is really a passive AcI ‘for the prisoners to be returned’. This is orthogonal to our concern.
2 Elision of certain forms of the verb esse ‘be’ is a general phenomenon in Latin and not restricted to AcIs.
3 This is sometimes called anaphoric tense, but we will avoid that term here, since it risks confusion with discourse anaphoric tense (tense referring to a time introduced in the previous textual discourse), which is proper to finite verbs.
4 We use constructed examples here because attested examples do not come in minimal pairs and we are illustrating an uncontroversial point about the verbal system of Latin (Pinkster 2015:521).
5 AcIs can however express counterfactuality, in the form of a future participle of the lexical verb and the perfect infinitive of the auxiliary (Menge 2000:826). Arguably, the counterfactual interpretation arises not from the infinitive per se, but from the future in the past-construction as a whole.
6 Note that despite the subjunctive mood, this is not a “subjunctive conditional” in the sense of a counterfactual. Counterfactual conditionals are also in the subjunctive, but in addition require a shift to the past.
7 In fact, since all dependent clauses in reports take subjunctive mood, there is no contrast in the antecedent clause either.
8 But note that the personal infinitive appears only in embedded contexts, so it is not relevant for this squib.
9Barðal et al. (2012) discuss what they call the “dative subject construction” but, as they admit, they have no evidence that the dative arguments in question are actually subjects in Latin.
10 UID should not be confused with free indirect discourse. In free indirect discourse, pronouns and tenses are shifted but the discourse is otherwise interpreted from the perspective of the reported speaker. For example, there are no de re readings of definite descriptions.
11 But not invariably; see Spevak 2010:16–17 for details on the position of enim.
12 In this particular example, that is enough to warrant the conclusion that the second sentence temporally precedes the first, since the first sentence has a deictic future, but that is immaterial to the point.
13 Caesar, Gallic Wars, 3.14.4; Cicero, Epistulae ad Atticum, 3.8.4, 3.18.2, 4.6.2, 4.12.1, 4.16.4, 5.5.1, 5.11.5, 5.21; Vulgate, Luke 9.44, Acts 8.23, 2 Cor. 5.10.
14 In (13), the agent of the two speech events is the same, as this is a first person report (‘I said . . . ’), but there are still two distinct events.