1 Introduction

Herodotus’ Histories is a record of his inquiry into the origins of the Greco-Persian Wars (between 499 BC and 449 BC), as he states himself in the first sentence of the work.¹ Throughout the work, we find uses of the accusative-and-infinitive (AcI) construction that are familiar to classicists, but quite remarkable from a linguistic perspective. An example is given in (1):²

Example (1) is a speech report with an AcI construction as its complement. Interestingly, whereas the AcI construction in speech reports is usually directly dependent on a verb of saying, here it extends to the relative clause (τῷ oὔνoμα ɛἶναι Παντίτην ‘whose name.ACC to-be.INF Pantites.ACC’).³ I single out the relevant part in (2) and use brackets to indicate the structure:

The AcI in (1) also extends to the continuation of the report (νoστήσαντα δὲ τoῦτoν ἐς Σπάρτην … ἀπάγξασθαι ‘after having returned to Sparta this.ACC to-hang-himself.INF’). While both usages of the AcI are well known to classicists (see e.g. Smyth [1916, sections 1598–1600, and 1920, sections 2630–4], who sees the infinitive in subordinate clauses as attraction), they are puzzling from a linguistic point of view. Let us focus on the latter for now, the uses in continuations. While in this very example, one may argue that the continuation is still part of one and the same sentence, we will see that there are other cases where this is impossible. These cases show that we have to treat the continuations as independent main clauses. This is already interesting in itself for syntactic reasons since it means that we have non-finite main clauses, but it also raises the question central in this article: what is the relation between the infinitive and the previously mentioned verb of saying, if not one of syntactic dependency? Not only the AcI, but also the so-called oblique use of the optative mood is found in such peculiar positions. We will see examples in section 3.

This article aims at a better understanding of these peculiar uses of the AcI and optative. In order to achieve this, I take a combined linguistic-narratological perspective. At the linguistic side, I present an analysis in pragmatic/semantic terms of the oblique uses of both markings in general from which the peculiar uses are then shown to come out as natural consequences. In this way, we also arrive at a better understanding of the linguistic relation between the oblique morphology in the complement and the embedding verb in the matrix clause. Next, I explain why, even though it is tempting to classify some of the peculiar uses as Free Indirect Discourse, it is fundamentally different from what we see in modern novels. The pragma-semantic analysis is then used to explain what these uses are useful for. As we will see, they facilitate a faithful rendering of original discourse relations without losing the information that it is a report. This helps us to explain why Herodotus is the most prominent user of these constructions (and according to Cooper [1974: 25] for some uses even the only one). I argue that part of the answer can be found in Herodotus’ narrative style: a combination of three traits of the Herodotean narrator makes this construction particularly suited for him.

At a higher level, the article provides a case study of how authors use the inventory of their language to find a midway between speaking in their own voice and representing the speeches of others. As such, this study may well shed some light on the use of similar tools in other languages, such as morphological evidentials in languages like Quechua (Faller, 2002) and Cheyenne (Murray, 2014), the reportative Konjunktiv (subjunctive) in German (Fabricius-Hansen and Sæbø, 2004), but also, for example, evidential uses of English seem and must (von Fintel and Gillies, 2010).

Throughout the article I will focus on Herodotus’ narrative of the battle of Thermopylae (Histories 7.207–233), one of the most famous battles in Ancient Greek history. For readers not familiar with the language, I will first say a few words about speech reports in Ancient Greek in the next section (section 2). Then, in section 3, I will present the data and explain in more detail why certain alternative accounts could not work. In section 4, I will develop the analysis in terms of presupposition-triggers, followed by the comparison with Free Indirect Discourse in section 5. Section 6 discusses the usefulness of the construction and why Herodotus is the most prominent user.

2 Three Report Constructions

In Ancient Greek indirect discourse, a verb of saying can take as its complement:

1. a regular indicative finite that-clause (Ancient Greek complementizers: ὡς, ὅτι);

2. a finite that-clause with the verb in a special mood, called the optative, formed by a suffix inserted between verb stem and inflection (only possible when the matrix verb is in past tense); and

3. an infinitival clause.

In (3) we see (constructed) examples of each, all reporting the utterance γράφω ‘I am writing’:

The use of the optative in speech reports is called the oblique optative. It resembles the reportative subjunctive (Konjunktiv) in German (see e.g. Fabricius-Hansen and Sæbø, 2004). I extend the use of the word oblique to infinitival constructions used in speech reports, as in (3c).⁴

In such infinitival constructions, Ancient Greek does not express the subject of the complement’s verb if it is co-referential with that of the matrix verb, as in (3c). If it is not co-referential, the subject of the infinitival clause is marked with an accusative case. So, literally, ‘he-said her-ACC to-write’ translates as ‘he said that she was writing’. This construction is called the accusative-and-infinitive construction, AcI (Accusativus cum Infinitivo) for short.

As a final remark, I mention that Ancient Greek is a non-Sequence-of-Tense language (Bary, 2012). Simply put, this means that the tense from the original utterance is retained in the report without modification. We see this in (3a) where the present tense from the original is retained in the Greek report, while in the English translation we use a past tense (the present tense being only possible in specific cases, see e.g. Abusch, 1997; Bary and Altshuler, 2015; Bary et al., forthcoming).

3 Uses of the Reportative Markers

In this section, I will introduce some more terminologies and present a short overview of the various oblique uses of the optative and AcI. We find the Ancient Greek oblique optative usually in combination with a verb of saying or thinking on which the clause with the optative depends syntactically, as in (3b). The same holds for the oblique AcI, as in (3c). Examples from Herodotus are given in (4) and (5), respectively:

I will use the term reportative markers as a cover term for the oblique uses of the AcI and optative, and I will refer to their use exemplified in (4) and (5), where they occur in clauses that depend syntactically on a verb of saying, as the normal use. As we have already seen in the introduction, apart from this normal use, the reportative markers can also be used in more peculiar positions. I distinguish four positions and label them (i) to (iv). In (1) the AcI is used (i) in a subordinate clause within a report. In addition, the markers can also be used (ii) in continued indirect discourse, stretching over more than one sentence. (1) could be a case in point, but here we could also argue that it is still one and the same sentence. With other instances, this seems rather implausible, however. Consider (6), in which we have a say-construction (λέγoντɛς δι᾿ ἀγγέλων ‘saying through messengers’), followed by four optatives, after which Herodotus switches to AcIs.

Here the length and, as we will see, in particular the presense of the particle γάρ, makes it unnatural to consider this as belonging to one and the same sentence.

Moreover, we find the reportative markers (iii) in cases where the verb of saying is mentioned only parenthetically, as in (7):

In (7), we have a parenthetical say-construction (ὡς λέγɛται ὑπὸ τῶν ταύτῃ oἰκημένων ‘as said by those who dwell there’), followed by a series of AcIs. Although these infinitives do indicate that we have to do with a report, syntactically they do not depend on the verb of saying.

Lastly, the reportative markers can also be used (iv) without explicit mentioning of a verb of saying or thinking.⁵ This is exemplified in (8):

Here, no verb of saying is present. In the middle of the story, Herodotus suddenly starts using infinitives, probably to indicate that he is reporting what he has heard from others.⁶

These more peculiar uses (i) to (iv) form a challenge for a linguistic analysis of these reportative markers. Had we only the normal use of the reportative markers (the use in which they occur in a clause that syntactically depends on a verb of saying), we could try to develop an analysis along syntactic lines, especially in view of the fact that there is no clear contribution to the meaning in those cases (I will come back to this later).⁷ Then the reportative markers could be considered a case of agreement (or maybe concord) with the verb of saying, without introducing a meaning element themselves (compare the third person inflection -s in English that agrees with the subject, but does not contribute to the overall meaning itself). In view of the more peculiar cases described in this section, however, this is untenable. For one thing, in these cases the markers do contribute information that we would otherwise not have had, namely that the content is said. Moreover, there is nothing in the sentence that the reportative marker can depend on syntactically.

Let us work this out in some more detail. Why could it not be a case of syntactic dependency on a verb of saying? If it were, we would either have to treat the whole report as one long sentence depending on an initial verb of saying or we would have to say that it’s more than one sentence and stipulate that there are implicit verbs of saying in the post-initial sentences. To begin with the first option, examples like (7) and (8) already show that this is not possible as a general solution, since here there is no verb of saying that it could possibly stand in a relation of syntactic dependency to. But for (6) as well, this is untenable since it has the particle γάρ in the reported speech, a particle that, as a rule, introduces main clauses, which means that here a new sentence starts. A final argument against this option comes from examples like (9), where the initial explicit verb of saying is one expressing a command, while the continuation, without explicit verb, is interpreted to be just said (and not commanded), showing that the reportative markers cannot be said to still depend on the explicit verb:

This leaves us with the second option: we have to do with separate main clauses, but with implicit verbs of saying. As Haug, Jøhndal & Solberg (2017) argue for similar cases in Latin, it is again the particles that show that this option is untenable. They discuss the Latin particle enim ‘for’, which is the rough functional equivalent of Ancient Greek γάρ. In short, their argument is that the discourse relation the particle expresses (simply put, one of causality) can be seen to be a relation at the level of the content of the reported speech. Consider (7), for example, which contains γάρ in the sentence in parentheses. γάρ, like enim, always scope over the whole of its sentence. If there were an implicit verb in this sentence, we would expect γάρ to scope over that as well, resulting in the interpretation that saying that there happened to be a festival was a cause that the people were dancing. The actual interpretation is, of course, that a causal relation holds between the (events expressed by the) AcI clauses directly: γάρ marks there being a festival as the cause of the dancing. This means that the stipulation of a null verb would give the wrong predictions.

In light of these problems that an analysis in terms of a syntactic dependency will inevitably have, I will opt for a different route and analyze the various uses of the reportative markers along pragma-semantic lines, following Bary and Maier (2014). The main idea is that reportative markers are presupposition triggers. As will be explained in the next section, this allows for a uniform treatment of both the normal and the peculiar uses of the reportative markers.⁸

From a narratological perspective, this article aims to explain the use and usefulness of these constructions, and, in particular, why Herodotus is the most prominent user of them. According to Cooper, Herodotus seems to be even the sole author where we find uses (iii) (with a parenthetical speech verb) and (iv) (absolutely free) with the AcI (Cooper, 1974: 25, 5n). Categories (i) and (ii) (the uses in subordinate clauses and continued reports, respectively) are more wide-spread,⁹ although the use of the infinitive for very long passages again seem to be a peculiarity of Herodotus (De Bakker [2007: 33–4] mentions the story of pharaoh Rhampsinitus and the thief with its 947 words as the longest uninterrupted instance). If the Ancient Greek language apparently allows for these uses, we need to understand why it is Herodotus in particular who uses them. A tentative answer to this question will be provided in section 5.

4 Reportative Markers as Presupposition Triggers

As we have seen in the previous section, the reportative markers show the following behaviour. If they are not embedded under a verb of saying, they clearly contribute something to the meaning of the sentence as a whole, namely that what is expressed is a report of an utterance by someone else. In the normal case, however, if they are embedded under a verb of saying, there is no clear contribution, since the embedding verb already tells us that the complement is reported. In particular, in the latter case, we do not get a reduplication of reports (it is said that it is said that …). This means that we have the following desiderata for the semantics of reportative markers: whatever their semantics are, they should turn a clause into a report if the clause in question is not overtly embedded, but dissolve if it is. As we will see, this is exactly the behaviour of presupposition triggers.

Presuppositions can be characterized as information that is taken for granted by the participants in a conversation (see Van der Sandt [2015] for a good introduction to the topic of presuppositions). This information has a different status from information that is presented as new. Consider (10), where the presupposition is written in a smaller size:

By uttering (10), a speaker presupposes that the Netherlands has a king and conveys as new information that he likes to swim. The distinction between information that is presupposed and information that is presented as new is encoded in our language. In (10), it is the use of the definite description the king of the Netherlands that induces or triggers the presupposition. But the class of presupposition triggers is much broader, and also includes, for example, verbs like to stop or to know. If someone utters (11), we infer from that that Peter used to smoke.

The information that Peter used to smoke is a presupposition and it is triggered by the use of the verb to stop. Similarly, in (12) we infer that Beijing was formerly romanized as Peking, a presupposition triggered by the use of the verb to know, the presupposition trigger.

There are tests to determine whether a linguistic element is a presupposition trigger. These diagnostics are based on the fact that presuppositional information tends to emerge as inferences in environments where standard inferences do not survive (Van der Sandt, 2015: 330). One test is constancy under negation. Applied to (9), this gives us:

In (13), the presuppositional information that the Netherlands has a king is preserved (and hence passes the test) and only the non-presuppositional part (i.e. that he likes to swim) is negated.

Presuppositional information is often given explicitly in the preceding discourse, as in (14):

On a Van der Sandtian (1992) account of presuppositions, in which presuppositions are treated as anaphora (in the discourse semantic sense of the word, i.e. elements that need to be resolved in the textual context), the presupposed information in this case binds to this preceding material (‘binding’, too, understood in the discourse semantic sense and not as in e.g. Binding Theory). However, presuppositions can also be used to make shortcuts, as Karttunen remarked:

People do make leaps and shortcuts by using sentences whose presuppositions are not satisfied in the conversational context. This is the rule rather than the exception … If the current conversational context does not suffice, the listener is entitled and expected to extend it as required. He must determine for himself what context he is supposed to be in on the basis of what was said and, if he is willing to go along with it, make the same tacit extension that his interlocutor appears to have made. (Karttunen, 1974: 191)

Example (15) illustrates this:

It presupposes that there is a post office, a street going down to it and at least three brick houses there, and the speaker asserts that John lives in the third of them. Still, even if the presuppositional information is not already part of the common ground of the participants of the conversation, (15) can be uttered felicitously. This is also possible with the examples in (10), (11), and (12). The presuppositional information is then said to be accommodated by the hearer, a term introduced by Lewis (1979b), to deal with the non-presuppositional part of the utterance.

Following Bary and Maier (2014) and building on Fabricius-Hansen and Sæbø (2004) and Schwager (2010) for the German reportative subjunctive and Tagalog daw, respectively, I claim that the reportative markers of Greek are also presupposition triggers. The presupposition that they trigger is that someone said the content expressed by the clause that contains this marker.

In the following examples, I will abbreviate this presupposition as SAID, as in (16), where this presupposition is triggered by the optative:

These presupposition triggers show the same behaviour as the more familiar ones we have seen earlier. The presupposed information may be given explicitly, as in (17) (the normal use) and (18) (use (ii), in continuations), and then the presupposed information binds to that, in a way analogous to what we have seen in (14):

The difference between (17) and (18) is that in the former the information is given in the sentence itself, whereas in the latter it is given in the previous discourse. The occurrence of reportative markers in subordinate clauses, use (i) exemplified in (1), and the one with a parenthetical say construction, (iii) exemplified in (7), are just special cases of the former.

As we would expect of presupposition triggers, the presuppositional information can also be only presupposed and not given before. In that case it has to be accommodated by the hearer. This is the case with use (iv), the use without any verb of saying, as in (19):

Note that what first seemed remarkable uses of the reportative markers are actually—once they are seen as presupposition-triggers—natural consequences of one and the same meaning.¹⁰

One might wonder whether it is correct to assign one and the same meaning to the optative and AcI in their functions as reportative markers. Indeed, there are certain differences. For one thing, very long continued reports are always in the AcI. In addition, if the report shifts from one kind of complementation to the other, it is usually in one direction: from the optative to the AcI and not the other way around. Still, this does not legitimise a different kind of analysis. Example (20), for instance, shows that a syntactic analysis is untenable even if we restrict ourselves to the optative.

Let me repeat the argument given in section 3. First, the particle γάρ marks the start of a new sentence, which means it cannot be a case of coordination under the explicit verb. Furthermore, the discourse relation that comes with γάρ shows that there is no hidden verb of saying. Rather than treating the two reportative markers differently, a pragmatic explanation for the above two differences suggests itself: due to its other usages (the potential use and that in wishes), the optative may be avoided in favor of the AcI for reasons of ambiguity.

Since we knew already that the passages discussed in this section are to be interpreted as a report of someone’s words, it is legitimate to ask what we have gained with the analysis provided in this section. The benefit is to be found in our understanding of the mechanism behind these uses: the way in which the optative gives us the information that someone said the content of the words is very similar to the way in which the use of the verb to stop tells us that the subject used to do the activity denoted by the verb’s complement. This then explains that only in some cases does a reportative marker make a clear contribution to the interpretation of a sentence as a whole.¹¹ Cooper has convincingly argued that traditional teaching does not recognize that infinitives need not ‘stand in a grammatical relationship to some verb of speaking or thinking if they are to reveal their oblique potential’ (Cooper, 1974: 76). This article offers an alternative analysis of the relation between verbs of speaking and associated oblique infinitives taking into account the difficulty Cooper observes with the more traditional picture in terms of a syntactic dependency.¹² In section 6, I will show that the presuppositional nature of the reportative markers also helps us understand why it is useful to have this device, both in general and for Herodotus in particular. Preparations for that discussion will be made in the next section.

5 Unembedded Indirect Discourse and Free Indirect Discourse

In the previous section, I analyzed the reportative markers as triggering the presupposition that someone said the content expressed by the clause that contains this marker. One might wonder whether content is a sufficiently fine-grained notion here. For one thing, it would make the Greek phenomena that we have discussed so far quite different from the narratological device called Free Indirect Discourse (FID), where, as the term is commonly used (e.g. Schlenker, 2004; Maier, 2015), the utterances or attitudes reported are to a large extent presented from the character’s (that is the reported speaker’s) perspective. In this section, I will argue that, despite the clear similarities between the two devices (both are reportative constructions without syntactic embedding), there is indeed a difference in that the constructions in Herodotus do not carry the same suggestion, which will lead me to conclude that content is indeed a sufficiently fine-grained notion for the analysis of the reportative markers in Herodotus.

Let me start with a short discussion of FID, a report construction that has attracted considerable attention, first mainly from narratologists and more recently also from linguists, and has led to an immense literature (to give only a tiny selection: McHale, 1978; Banfield, 1973, 1982; Ginsburg, 1982; Fludernik, 1993; Vandelanotte, 2009; Schlenker, 2004; Eckardt, 2015; and Maier, 2015). It goes without saying that the short discussion that follows can never do justice to all of this. I focus on the aspects relevant for a comparison with what we find in Herodotus.

In the passage in (21) from Austen’s novel Emma, FID is marked with italics:

Here, we see the defining characteristics of FID: (i) The thoughts expressed are attributed not to the narrator, but to Emma, a character in the story. (ii) Interestingly, this is achieved without any embedding under a verb of thinking. (iii) An additional feature of FID is that it gives the suggestion to be quite a faithful report of the original thought or utterance.

Although the sentences in italics are not direct reports (the impression that the narrator wants to give us is not that Emma thinks it was a wretched business but instead it is a wretched business), the impression is that it comes close to that. (I deliberately use suggestion and impression since these constructions are typically used in fiction. Also, when I speak of a report of thoughts, I am aware of the fact that we may not (always) think in natural language in reality, but in literature we clearly pretend that we do.) We see this (fictitious) faithfulness to the original utterance (or to the first-person perspective of a character, if one likes) in various aspects of the language (e.g. Banfield, 1973, 1982; Maier, 2015):

1. indexicality/deixis: all indexical elements, except for tense and person features, are to be interpreted from the character’s perspective. This can be illustrated with the by now canonical example (22) (cited in e.g. Banfield, 1982; Doron, 1991; Schlenker, 2004):

   (22)	Tomorrow was Monday, Monday, the beginning of another school week. (Lawrence, 1971: 185)

   Here, the suggestion given is that the character thinks Tomorrow is Monday and we see that the indexical adverb tomorrow, but not the present tense of is is retained in the FID report.

2. syntax: all sorts of elements (e.g. interjections, direct questions, exclamatives) that cannot occur in indirect discourse can occur in FID. As for exclamatives, (20) provides examples of their occurrences in FID. In indirect discourse, this is impossible, as (23) shows (on the reading of that as a complementizer):

   (23)	#Emma thought that such an overthrow of everything she had been wishing for!

3. word choice (e.g. definite descriptions): words are interpreted as the character’s formulations. Consider first the indirect discourse example (24):

   (24)	Oedipus believed that his mother was not his mother.

   This attitude report has in principle two readings. On one reading, the definite description his mother is interpreted character-oriented and Oedipus believes something like My mother is not my mother, which is a contradiction. However, the sentence has a second reading in which it is the narrator who refers to this person as his mother (Oedipus himself may think of her in terms of Iocaste, for example), a reading in which Oedipus does not believe in a contradiction.¹³ FID reports do not have two readings in this respect. Only a character-oriented (and here contradictory) reading of definite descriptions is available:

   (25)	His mother was not his mother, Oedipus believed.

A few cautionary remarks may be in place here. I do not wish to present either indirect discourse or FID as fixed constructions. As for indirect discourse, it is well known that there is considerable freedom: a reporter can choose to stay close to the original words or rephrase the utterance in his own words to a large extent. However in FID too, it has been observed that the narrator can, for example, insert a proper name when the referent is not salient enough for the narratees to be picked out by the use of a third person pronoun (Fludernik, 1993: 136; Maier, 2015).

Also, I do not wish to say that no elements in indirect discourse are to be interpreted from the character’s perspective. Attitudinal particles, for example, can be interpreted in this way, as witnessed by the German (26) and the Greek (27):

As Döring (2013: 105) notes, wohl, expressing uncertainty, is to be interpreted from Reif’s perspective (who does not know the outcome of the match at the time of his utterance rather than the reporter’s (Kohl) (who does know the outcome)). And similarly, in (27), as Wakker argues, it does make sense for the reported speaker, but not for the narrator to stress the truth of the reported speech by the use of the particle combination ἦ μέν (Wakker, 1997: 216).

Notwithstanding this freedom with both constructions, the aspects that I have discussed under (1) to (3) (summarized in Table 1) seem to set FID apart from indirect discourse, and, hence, are good indicators to determine whether what we find in Herodotus qualifies as indirect discourse (without an embedding verb), or shares with FID the third characteristic mentioned and presents the character’s thoughts or utterances from the character’s perspective (to a larger extent than is possible in indirect discourse). This then, on the one hand, answers the question whether content is a sufficiently fine-grained notion for the Greek constructions and, on the other hand, provides us insight in Herodotus’ use of these constructions, a topic I will discuss in the next section.

Two comments on Table 1. First, I leave out tense, since this, Ancient Greek being a non-Sequence-of-Tense language (see section 2), behaves the same in direct and indirect discourse (and hence probably also in FID, if it exists). Second, although indexicals like tomorrow are by and large interpreted from the narrator’s perspective in indirect discourse, we also find character-oriented instances e.g. (26):

The implication of the analysis provided in section 4 seems correct.¹⁴ There are no indicators that the Greek speech reports featuring what we called the remarkable instances of reportative markers are more faithful to the original wording (or first-person perspective of the reported speakers) than normal indirect discourse.¹⁵ For example, we do not find exclamatives or non-pronominal indexicals like ἐχθές ‘yesterday’ to be interpreted from the reported speaker’s perspective. There is also more positive evidence: we do find elements that show that these constructions lack the faithfulness that we find with FID. We find, for example, definite descriptions that are to be interpreted with respect to the narrator. (29) is a case in point:

Here we have a continued AcI report of what the Greek said and within this report, Herodotus writes τὴν Ἕλληνɛς λέγoυσι Ἐρύθɛιαν νῆσoν ‘the island that the Greeks call Erytheia’. This phrase originates from Herodotus, not from the reported Greeks, and as we have seen such a rephrasing of definite descriptions in the narrator’s own words is impossible in FID.

To conclude this discussion, despite the similarities between the Greek constructions and what is usually called FID (both are reportative and syntactically unembedded), the discussion in this section has shown that the two are in fact fundamentally different: with the latter, but not the former, words are interpreted with respect to the reported speaker’s perspective. For that reason, following Bary and Maier (2014), I use the term Unembedded Indirect Discourse, and not FID, for what we find in Herodotus, signalling that it is basically just indirect discourse, be it indirect discourse that is syntactically unembedded. Alternatively, one might propose to stretch the term FID as to include these cases (and to give up the faithfulness to the reported speaker’s perspective as a defining characteristic). Although this is in principle a legitimate move, I deliberately refrain from doing so (following in this respect De Bakker [2007: 33]), believing that the two are fundamentally different. In the next section, I will discuss the usefulness of these constructions and why Herodotus is their most prominent user, focusing on his narrative of the battle of Thermopylae.

6 Unembedded Indirect Discourse and the Herodotean narrator

With FID in modern literature, the reader has the illusion of ‘being within a character’s consciousness’ (to borrow a phrase from Fludernik [1993: 325]). Although the exact nature of this baffling phenomenon is still unclear,¹⁶ it seems safe to assume that this is at least partly due to the fact that we interpret the reported words from the character’s perspective. Another factor at play here seems to be the fact that FID is often a report of thoughts, rather than utterances. Both elements are missing in what we called the remarkable uses of the reportative markers in Herodotus: there are no indicators that words are to be evaluated from the reported speaker’s perspective (above the level of indirect discourse) and the reports are usually reports of what is said rather than thought. However, if both the elements that contribute to the aforementioned interpretational effect of FID are missing in Herodotus, the question arises what the usefulness of Unembedded Indirect Discourse is. The presuppositional analysis proposed in section 4 suggests an answer here: even if the narrator does not intend to capture the original perspective, he might want to give a faithful report of the original discourse structure. To see that Unembedded Indirect Discourse is a useful device for this, consider first the constructed English example in (30):

Imagine that I write (30a) in an email to my colleagues, and that later that day someone at the meeting asks Does anyone know if Corien is coming? Then the reply can be (i), in which case we lose the information that the words in the post-initial sentences are a report of what I said. Or, alternatively, the reply could be (ii), in which case the repetition of embedding matrix clauses makes it difficult to sustain the original discourse relations. Unembedded Indirect Discourse, by contrast, gives us the best of both worlds: it has a marker of reportativity (the oblique optative or AcI), but one that, due to its presuppositional nature, does not break the original anaphoric chain. In the Thermopylae passage, we see this clearly in section 203 (repeated from (6)):

Had the continuation starting with oὐ γάρ ‘for not’ been interrupted by a repeated embedding matrix verb, the anaphoric link between the two parts would have been broken and more effort would have been required to interpret a causal relation between Xerxes not being a god and the soldiers not having to be afraid.

If we understand why Unembedded Indirect Discourse is a useful construction, this also helps us to explain why Herodotus is its most prominent user. I believe that part of the answer can be found in Herodotus’ narrative style. The combination of the following three traits of the Herodotean narrator make this construction particularly suited for him. First, as is well known, Herodotus, at least, pretends to provide to his narratees all versions of the events that he has heard.¹⁷ In book 2, he even explicitly states that this is the purpose of his history:

He even tells all versions, if it is clear to him, which one we should believe. In his narative of the battle of Thermopylae, we see this in 7.214:

Second, in addition to informant-speeches, Herodotus also recounts dialogues between characters in his story, so-called character-speeches. Example (33), again from the Thermopylae passage, is part of a dialogue between Xerxes and Demaratos:

The fact that Herodotus cannot possibly have known what was said does not prevent him from including such dialogues. This is probably to be seen in the light of a different conception of historiography in ancient times where a convincing (and entertaining) reconstruction of the past seems to have been more important than the discovery of what actually happened (De Jong, 2013).¹⁸

Taken together, informant-speeches and character-speeches take up a considerable part of the text: just under one-third.¹⁹ Still, this, in itself, does not explain the use of Unembedded Indirect Discourse. Even in combination with the length of these speeches (De Bakker [2007: 7] reports an average length of 26.9 words, the story of pharaoh Rhampsinitus and the thief being the longest example of what I call Unembedded Indirect Discourse, at 947 words), which clearly contributes to the need for continued report constructions, we do not yet have a full understanding. One might still wonder why Herodotus does not restrict himself to other report devices for such cases, such as direct discourse (or invent ways of reporting that come closer to modern FID). This can only be understood if we acknowledge a third trait of Herodotus’ narrative style, namely the fact that the Herodotean narrator, even when recounting other people’s words, usually remains present himself. He is an external narrator—he does not play a role in the events he recounts—but one that reveals himself clearly in the story (De Jong, 2013: 257, 263). In the Thermopylae narrative, he is sometimes present as a histōr, who weighs versions, as we have seen in (33), or as the organizer of the material, as in (35):

His presence makes the Herodotean narrator a dramatized narrator (De Jong, 2013: 263). The reader of the Histories never loses him out of sight. Strikingly, he is even present when he reads characters’ thoughts:

As an omniscient narrator, Herodotus has access to Xerxes’ thoughts and tells us that what the Lacedaemonians did seemed laughable to Xerxes. At the same time, he himself is present to tell us what was really the case. In general, I believe that Herodotus’ omnipresence precludes immersion into the story. Another example where we hear Herodotus’ own voice in attitude ascriptions is (37) (repeated from (4)):

As we have seen in section 5, definite descriptions in indirect discourse can, in principle, be interpreted from the actual speaker’s or the character’s perspective. The same holds for indirect attitude ascriptions, as in the first part of (37). In this case, it is clear, however, that we hear Herodotus’ own voice (and not that of the character Hydarnes) when he refers to the Phocians as such, since Hydarnes does not know what country the men are from.

Notably, Herodotus’ informant-speeches are never in direct discourse; indirect discourse is used almost exclusively (De Bakker, 2007: 161).²⁰ With character-speeches, on the other hand, we find both direct and indirect discourse. In addition to De Bakker’s interpretation of the alternation of the speech modes in terms of narrative pace and the effect of direct discourse of listening to a voice from the past, a factor that may also contribute to the difference between character-speeches and informant-speeches in this respect is the fact that the content of informant-speeches is much more easily confused with Herodotus’ own voice than the content of character-speeches. It is, for example, highly implausible that Herodotus would want to convey to his audience in his own voice the content of what he makes Xerxes say to Demaratus in (34), but for informant-speeches this confusion could easily arise. Hence, in informant speeches the report status has to be marked continuously throughout the report.

This brings me back to the use of Unembedded Indirect Discourse. Given his narrative style, in which Herodotus always remains present himself, it is not surprising that if he recounts other persons’ words, he wants to mark them as such. As we have seen, the need to do this throughout the report is more pressing for informant-speeches. Here the content is more easily to be interpreted as Herodotus’ own voice. Moreover, they are generally also longer: an average of 43.2 against 23.1 words (De Bakker, 2007: 7). Since, as we have seen, a full matrix clause (he said that …) would disrupt the anaphoric chain of discourse relations of the often quite long utterances, the presuppositional reportative markers (oblique optative and AcI) come in extremely useful. This, I believe, makes Unembedded Indirect Discourse a tool particularly suited for Herodotus. In the battle of Thermopylae, we see this at work in 7.226:

Here, Herodotus provides us with an embedded report: we read what Dieneces is said to have heard, a character-speech embedded in an informant-speech. Dieneces had learned from a Trachinian that when the barbarians shot their missiles, they hid the sun by the multitude of arrows. As for the continuation τoσoῦτo <τὸ> πλῆθoς αὐτῶν ɛἶναι ‘such was the multitude of these’, Herodotus wants to mark that this is still the content of the embedded report. The presuppositional status of the reportative marker in the form of an AcI (in contrast to an embedding verb) makes it possible to do this without breaking the discourse link with the previous sentence.

One might wonder what form other Greek historians use for speech reports. Let us take a quick look at Thucydides, who also lived in the fifth century B.C. and who wrote about the Peloponnesian War. I will confine myself to a few remarks here without attempting to do justice to the immense literature on Thucydides’ speeches (e.g. Stadter, 1973; Scardino, 2007; Pavlou, 2013; Tsakmakis, 2017), most of which is from a historiographic/literary rather than linguistic perspective. We find continued indirect speeches in Thucydides as well (e.g. 6.49), but they are not as long as in Herodotus and they are rare. Part of the explanation is that Thucydides does not have informant-speeches to the same extent as Herodotus, and we saw that it is information-speeches that make the need for the syntactically free use of the reportative markers particularly pressing. As for character speeches, Thucydides is well-known for his long speeches in direct discourse (which Thucydides does not pretend to be verbatim reports of what was actually said, see Thucydides 1.22). Particularly famous among these are his very long political speeches, which function as breaks to reflect (Scardino, 2007: 717) upon what happens in the immediate situation and to relate this to universals concerning, for example, human nature. By choosing direct discourse, Thucydides is (seemingly) less present in the narrative, since the reported content is given in the character’s words. In addition to direct and indirect discourse, Thucydides also manipulates double voices or perspectives in more intricate ways (e.g. Bakker, 1997; Allan, forthcoming). It is probably this subtle manipulation of perspective that made him already in Antiquity known for his ability to produce in the mind of the readers the emotions of the characters (Plutarch, De Gloria: 3). The complex ways in which he did this deserve a study of their own. For now, I only refer to the corpus annotated for speech, perception and attitude reports (Bary et al., 2017), created in the Perspective Project, which is developed to help us understand how Thucydides manages to do this and what the exact differences with Herodotus are.²¹

7 Conclusion

What do you do when you want to report what someone else said and at the same time you want it to be clear throughout the report that you do not necessarily commit yourself to the content? In principle, you could add ‘she said’ in each and every clause. However, as I have argued, it will be hard, if not impossible, to sustain the discourse relations of the original utterance. The Ancient Greek reportative markers (oblique optative and accusative-and-infinitive) make it, due to their presuppositional nature, possible to mark the reportative status without breaking the original anaphoric chain of discourse relations. We have seen that a combination of three traits of the Herodotean narrator might explain why Herodotus is the most prominent user of these constructions. We have also seen some differences with FID, as we find it in modern literature. Since the presupposed information that they trigger is that the content of the clause is said, it lacks the suggestion of faithfulness to the first-person perspective of characters, an implication that seems correct.

A Glosses

I used the following abbreviations in the glosses:

For reasons of space and readability, I refrained from glossing gender, tense and aspect, except where it might be relevant for the reader. For the same reasons, I glossed mood only for non-indicative moods, voice only for (middle) passive voice, and number only with finite verbs (again, except where relevant).

Competing Interests

The author has no competing interests to declare.