IV.3 Discourse segmentation

§1. The present chapter focuses on how we segment Herodotean and Thucydidean discourse in reading, and on the relevance of particles to that. “We,” “segment,” and “discourse” require some glosses. “We” refers to the community reading the two Histories nowadays. This reading community does not apply a uniform practice in reading and/or teaching and/or researching those texts. However, the task we share of attempting to comprehend the texts does entail certain overarching aims and general constraints.

§2. Segmentation is a part of reading. While reading, we process stretches of text in a piecemeal progression. The size of such stretches may be bigger in the case of silent reading (as our eyes can span more words), and smaller in the case of reading aloud (eyes and ears focus on fewer words at a time), but the fundamental cognitive and intellectual experience is the same: we receive verbal input in pieces. [1] Usually written pages provide larger divisions of discourse (indentation, for example), and we may glance at them in a cursory manner as we start reading a clause; still, what we can follow word by word is a relatively small portion of discourse. [2] If we listen to stretches of text, segmentation happens in a more radical way, as the absence of written pages compels us to focus just on the verbal segments we receive sequentially.

§3. Finally, I take discourse as a comprehensive and dynamic notion that embraces various linguistic components of texts (morpho-lexical, syntactic, semantic, and pragmatic), as well as paralinguistic and extralinguistic components (punctuation, prosody, paragraphing, links to the contextual setting, short and long-term memory). [3] Furthermore, “discourse” evokes a distinctive feature of discourse analysis as a subfield of linguistics, that is, the consideration of discourse units, especially above the sentence level. [4] In light of that, segmentation encompasses micro and macro divisions, ranging from the way in which a vocative expression is set off to what made editors of the two Histories assign chapter numbers.

§4. Our piecemeal processing of discourse relies on several components that we make use of in order to segment texts. These components include the syntactic arrangements of words. We process signs of integration across words, and signs of boundaries between them. For instance, we associate a sense of completeness with main clauses, and we perceive syntactic integration in the constituents of a subclause. Likewise, we sense boundaries between list entries, and we isolate anacoluthic constructions from the surrounding co-text. All this is to say that we see syntactic characteristics of the text such as the presence of subordinating conjunctions, and the occurrence of finite or nonfinite verbs, as important cues in discourse segmentation.

§5. Further input comes from the spatial organization of the written text. If I visualize the OCT editions of Herodotus and Thucydides, my mind’s eye recalls section and chapter numbers on the margins, and a larger space between words corresponding to the beginning of new sections. I also recall a series of commas, full stops, high stops, semicolons, and, from time to time, parentheses. Indentation is very sporadic. No capital letters, no exclamation points, no dashes occur. [5] On the larger scale, chapter and section numbers trigger the expectation that some more or less major boundary occurs—a topic shift, or a scene-shift, or a change of grammatical subject. As a matter of fact they affect paragraphing. [6] On the smaller scale, section numbers trigger expectations about discourse boundaries that are less strong than those of chapter numbers. Their lengths may vary, but the relation to main and subclauses tends to be constant: a section is made of multiple main and subclauses, supposedly showing some coherence. Sometimes sections coincide with períodoi, [7] especially in the editions of Thucydides, but this is not a rule. On the whole, visual divisions above the sentence level trigger certain expectations about discourse boundaries in the reader, and the visualization of written pages influences the way we come to grips with discourse segmentation.

§6. Interpretation of how texts divide into segments, then, depends on linguistic as well as paralinguistic signs. Paralinguistic signs support verbal expressions without being words, and have a meaning besides or beyond the words to which they are attached. In written code, generally speaking, paragraph boundaries, indentation, bold, italics, underline, and all punctuation marks are paralinguistic signs. For spoken code we can consider timbre, tone of voice, pace, and all the basic components of prosody—intonation, length, rhythm, and intensity. [8] Punctuation marks are particularly relevant to our processing of Greek texts (which lack features such as bold and italics). Punctuation is semiotically complex, and in fact it represents meaning components that are akin to particles. Therefore, punctuation marks deserve special consideration.

§7. The next sections discuss the main factors (modern as well as ancient) that contributed in the past, and contribute in the present, to determine the segmentation of ancient Greek texts, as well as some labels that have been attached to various units of discourse. Only in the light of those factors it is possible to establish matches and mismatches concerning the punctuating and segmenting roles of particles. The chapter, then, proposes a “holistic” principle of segmentation, which relies on two broad communicative notions, that is, “acts” and “moves.” [9] Particles contribute to discourse segmentation by marking acts and moves together with other co-occurring features.

§8. Let us start with contemporary conventions about the most standard signals of segmentation in written texts, that is, punctuation. For the purposes of the present chapter, I am going to single out a few points that account for conventional and cross-linguistic functions of punctuation marks. These points, in turn, will allow me to spell out some basic assumptions that modern readers either activate almost unconsciously, or tend to take for granted when approaching excerpts of ancient Greek prose.

§9. A recent monograph on the guiding functions of punctuation provides the following definition: “… punctuation is a nonverbal medium-dependent graphic feature that is transmitted via the visual channel and which, for the most part, communicates supplementary paralinguistic information.” [10] Such a view stresses the nonverbal character of the signs, along with the nonverbal communication they effect. They are about words, but are not words per se (hence “para-linguistic”); they communicate—at a subsidiary level—interpretations about the flow of discourse [11] and its segments. With our eyes we naturally process this paralinguistic input together with the linguistic information, and the joint processing of the two types of information guides our comprehension of texts.

§10. But what, exactly, is the paralinguistic function of punctuation? That is to say, how does punctuation contribute to the meaning of words? The scholarly literature acknowledges a broad duality: some functions hinge on the general idea of revealing the grammatical structure, while some other functions hinge on the general idea of revealing elocution. Some works strongly favor what Skelton (1949:165) calls “the grammatical principle”: Some other works state the function of clarifying syntax, but at the same time they take the elocutionary/rhetorical functions of punctuation on a level of complementarity: A similar view is expressed by Truss about two distinct functions of commas: An example of a grammatical “job” that punctuation marks might perform is when it is necessary to introduce a pause in order to make sense of a string of words. Consider Unless we punctuate after “talked” (“King Charles walked and talked; half an hour after his head was cut off”), continuous reading would yield a paradoxical image. An example of the rhetorical “job” might involve the use of single quotation marks: In some punctuating systems single quotation marks signpost a particular inflection of voice, which provides the word “exact” with the implied meaning “as they call it,” or something equivalent to that.

§11. Even though both the grammatical and the elocutionary functions of punctuation nowadays tend to be seen as complementary and necessary, there is an area of significant imbalance that I would like to stress. The elocutionary side (often called “rhetorical” [12] ) regards emphasis and prosodic qualities attached to words or word-groups. Elocution-based punctuation does not indicate the structure and the segmentation of thoughts; rather, it conveys a “mode of speech.” [13] Conversely, the grammatical side is explicitly related to text structure and segmentation. “Grammatical” punctuation shows where boundaries are, and how segments relate to one another (Kirkman, (t3)); it even represents “an independent attempt” (i.e. independent of elocution) to represent structure (Skelton, (t2)).

§12. Over the past decades studies on the prosodic values of punctuation have increased, which explore the relation of interdependence between discourse and intonation. [14] This is in contrast to the theoretical view expressed by the authors quoted thus far: the analysis of the prosodic qualities we attach to words is not confined to a more or less subjective “mode of speech” (Skelton) that hints at “literary qualities as rhythm, direction, pitch, tone and flow” (Truss, (t4)). Conversely, the emphasis on specific words reflects the informational status speakers assign to them. [15] As the next paragraphs will show, prosodic continuity and discontinuity directly convey the segmentation of discourse. Therefore, the duality in the functions of punctuation that I mentioned at the beginning of this discussion can be re-interpreted in terms of deep interlacing rather than opposition, and in terms of co-construction of boundaries, structure, and meaning.

§13. On this point Chafe 1988b is a salient article. It postulates that written language has its own prosody, albeit covert. Punctuation units (which are defined as “stretches of language between punctuation marks” [16] ) reveal the prosody of the writer’s inner voice. They do so in an approximate manner, as this example clarifies: The same question mark in these cases stands for two intonational realizations of questions that in fact are quite different: the intonation of the former question has a rising contour (the end being on a higher pitch than the start), whereas the intonation of the latter has a falling contour (the end being on a lower pitch than the start). [17] Punctuation units, Chafe shows, may correspond to “spokenlike” units imagined by oral readers (engaged in reading the written text aloud), or to a mix of “spokenlike” and “unspokenlike” units imagined by readers engaged in silent reading. [18] A series of experiments demonstrates that punctuation units in written styles tend to be longer than intonation units in spoken styles. [19] According to the author the reasons for this difference are two. First, “writers and (silent) readers are able to process larger chunks of information at a time;” second, “written language provides syntactic as well as punctuational clues to prosodic boundaries” (Chafe 1988b:424).

§14. Both points are very relevant to the present chapter. Chafe’s first point (“writers and (silent) readers are able to process larger chunks of information at a time”) holds true for writers and readers of texts that are meant to be read. In the case of many ancient Greek texts, however, the language was composed to be performed (in songs, on a stage, or through public delivery). The performance-constraint implies that listeners/spectators were supposed to receive smaller chunks of information at a time. This should constitute a caveat when we modern readers approach the modern punctuation of ancient Greek texts. His second point (“written language provides syntactic as well as punctuational clues to prosodic boundaries”) is crucially relevant to discourse segmentation and the role of conjunctions, sentential adverbs, and particles in it. Conjunctions, for example, may signal prosodic breaks without being preceded by commas (Chafe 1988:415 and 419). Chafe’s point about conjunctions is intuitively unchallenging; yet he suggests that in written language the syntactic articulation of thoughts helps in segmenting discourse whether or not punctuation marks stress syntax. There is a furher factor that determines punctuation in written language, Chafe continues. This factor is the inner voice of writers, whether it (deliberately) creates exceptionally short or exceptionally long units. [20] This is one of the reasons why linguists are developing more and more research on the relation between syntax, prosody, and discourse. [21]

§15. From the preceding discussion it emerges that syntactic arrangements of words and our individual voicing of texts jointly influence punctuation choices. Our research on ancient Greek particles confirms the latter point: while searching for functions of particles, it is more productive to consider the interlacing between grammatical and prosodic or elocutionary factors in discourse segmentation, rather than to split them (or, even more problematic, to see only grammar as structural). Particles show how grammar and prosody are integrated. Their position can tell us about prosodic boundaries and about discourse segmentation by complementing rather than contrasting syntactic information. Particles reveal grammar and elocution without a clear-cut division of labor.

§16. In an article on the absence of punctuation at the end of speeches in Latin poetry, Feeney aligns with others in stressing the gap between the textual display ancient readers had at their disposal, and the modern layout of printed editions. [22] He points out the problem of punctuation, in particular, by reporting Heyworth’s statement: “The punctuation of a classical text involves the imposition of modern conventions upon an alien environment.” [23] Similarly, Denniston (1950:430) notes: “The punctuation and accentuation of our MSS are not to be trusted over-implicitly, and frequent changes should probably be made. Editors have been rather haphazard in this matter.” Dover (1997b:27-28) quotes examples from Thucydides and Isaeus where modern punctuation differs across different editions; he concludes that the disagreement of editors is ultimately about the length of sentences, which is determined by colons, commas, and full stops. Sansone (1990:174) simply states: “(…) it is unscientific to assume that one can rely on the punctuation in a modern text of a classical author to provide evidence for the stylistic characteristics of that author.” In a critical edition of Plato’s Ion, Rijksbaron (2007:68-69) acknowledges the shaky reliability of punctuation in printed editions of the dialogue (since 1578, the date of Estienne’s edition); consequently, he warns the reader that in several points of his own edition he prefers to rely on the alternative punctuation offered by Byzantine manuscripts. [24]

§17. These remarks confirm that in modern editions of ancient Greek works punctuation is not a stable feature of the text. It is less original than the words, [25] and it is the result of editorial decisions taken at several stages. To some extent modern punctuation does reflect pre-print punctuation, [26] but correspondences are not one to one, quantitatively and qualitatively. In addition, modern punctuation lends itself to non-univocal interpretations; for example, opting for a semicolon instead of a full stop may be a decision made independent of syntactic and semantic reasons. [27] Moreover, translations only partly reflect the segmentation suggested by Greek words and by modern punctuation: however literal, they naturally adopt the punctuation habits of the target language, which adds to the instability in interpreting segments. [28]

§18. My argument focuses on particles, punctuation marks, and periodic style, but also on larger-scale divisions, specifically the relationship between particles and chapter or section numbers as signposts for indentation and paragraphing. I shall consider three major realities that presumably influence the modern punctuation and segmentation of ancient Greek works. The first is that the “covert prosody” of written texts (to recall Chafe’s terminology) gives editors a range of choice in determining where and how to segment discourse. The editorial decisions based on the inferences of their readings may be independent of decisions about whether or not to reproduce the punctuation of earlier editions (see the previous section). The second reality is that we have inherited from early print books a set of punctuation marks that were established during advances in orthography made in this era (see the upcoming paragraphs in this section). The third reality is the presence of a parallel inheritance, namely the multifarious punctuation and division systems elaborated throughout centuries of manuscript transmission (see the next few sections). In other words, I see segmentation practices as resulting from the combination of these three aspects: processing syntactic and semantic information along with imagining the author’s inner voice, mirroring previous print editions, and mirroring pre-print editions.

§19. Let us consider the second reality, that is, punctuation since the start of print editions. The very first appearance of a standard version of commas, colons, semicolons, and full stops (with the shape and the graphical relation to words we know and still use) date back to the Aldine edition of the treatise De Aetna of Pietro Bembo (1496). In a later sizable handbook on orthography (Orthographiae ratio, 1566), the section Interpungendi ratio by Aldus Manutius the Younger (1547-1597) describes, in no more than ten pages, the punctuation marks selected for printed texts: commas, semicolons, colons, full stops, question marks, and parentheses. [29] The main unit of segmentation is unequivocally the sentence: punctuation illuminates sentences otherwise obscure (“equidem usu didici, obscuras saepe sententias, si recte distinguantur, illustrari”); a single full stop means the end of a sentence (“restat unicum punctum, quo sententia concluditur, ac terminatur”); commas are applied to nouns and verbs rather than to whole or incomplete sentences (“[virgula]: ea vero non integram sententiam, nec sententiae partes terminat, sed nomina, vel verba distinguit ….” [30] The set of punctuation marks selected by Manutius the Younger excludes exclamation markers and dashes, which are added at a later point. [31] This does not mean that the effects potentially signaled by exclamation marks and dashes were not considered in the Renaissance; it only means that the selected set of marks was felt sufficient for clarifying discourse segmentation.

§20. Chafe (1988b:401-402) stresses that 18th century handbooks privileged a “spokenlike” way of punctuating (in English), but that later works focused on grammatical conventions [32] (see also above, (t2)—(t4)). Along the same lines, I find that modern and contemporary editions of ancient Greek prose privilege the grammatical principle of punctuation over the elocutionary principle; at any rate, grammatical considerations are the decisive factor in punctuating texts. At some point (perhaps near the publication of 19th century grammars of ancient Greek) philology embraced a line of thinking that teaching practices of the time probably enhanced: punctuation clarifies the syntax of sentences. This was (and still is) an important service to readers of texts in dead languages, for different reasons: partly the lack of any spoken counterpart, partly the indirectness of indications about delivery in ancient times, make the task of comprehension more effortful. Therefore, scholars came to regard syntactic articulation as the substantial component of texts, which punctuation should partially illustrate and clarify.

§21. This view is related to the idea that in ancient Greek prose syntactic segmentation, and, in particular, hypotactical segmentation, is primary. [33] The function of the full stop punctuation mark at the end of periods illustrates this idea well. [34] Modern scholars have established a conceptual match between periods in the sense of períodoi, and periods in the sense of full stops. Modern and contemporary readings of ancient Greek prose take periods as períodoi (that is, complex sentences) bounded at each end by full stops; they are the grammatical and stylistic measure of complete segments in written discourse. [35] Decoding a prose text from full stop to full stop or períodos by períodos is the safest way to follow how discourse is segmented. Sentences and períodoi are so much seen as the building blocks of discourse that scholars count them for statistical or heuristic purposes. For example, Müller (1980:25) compares the length of sentences in large excerpts of Herodotus and Thucydides by counting the segments between full stops. Yaginuma (1995:133) considers the high word count of Thucydidean sentences and “the extraordinary size of his periods” as decisive factors supporting the idea that Thucydides wrote for readers rather than for listeners. [36]

§22. But what exactly makes a períodos complete, and divisions across periods the safest? An answer that would take into account the comprehensiveness of discourse would be less obvious than one would think, as the criteria for “completeness” and “division” may differ. [37] What most interests me at this point is the answer provided by scholars of Herodotus and Thucydides (undoubtedly grounded on pre-modern literature on rhetoric). 19th and 20th century descriptions of the prose composition of the two historians include assessments about their periodic style. What makes a períodos a complete unit to those scholars is the syntactic articulation of thoughts—in particular, the way in which the mechanism of hypotaxis is exploited. What spans two consequent full stops is a construction of main and subordinate pieces that overall express cohesion and integratedness. [38] The more complex the arrangement of syntactic hierarchies within a períodos, the more elaborate is the periodic style. Periodic complexity means syntactic complexity.

§23. In addition to the criterion of syntactic complexity scholars apply a perfection scale, based on diachronic evolution, to assess períodoi: what is later and more complex is better; what is less complex and earlier is less good. Thucydides, according to Classen (1862:LXXIX) adopts a mature periodic style; the development is completed, and multi-clause periods are most common (“Was (…) die Satzbildung betrifft, so ist zwar die Periode, welche sich durch Vorder-, Nach- und Zwischensätze gegliedert, bei Th.[ukydides] in vollständiger Ausbildung und geläufigstem Gebrauche”). A few decades later Blass (1887:223-224) explicitly disagrees with Classen: Thucydides’ work is ranked somewhere between the ancient “additive” and the later fully periodic style; Classen has overestimated it (“Thukydides’ Schreibart steht (…) in der Mitte zwischen der locker anfügenden alten und der späteren periodischen, und Classen behauptet viel zu viel, wenn er die gegliederte Periode dei Thukydides in vollständiger Ausbildung und geläufigstem Gebrauche sein lässt”).

§24. The sense of periodic perfection emerges from some converging ideas posited by Blass (1887) and by Müller (1980). When subclauses precede the main clause, Blass originally posits (1887:224), the resulting períodos is “ascending” (“aufsteigend”) and perfect. Conversely, when the main clause comes first, and subclauses follow, the period is “descending” (“absteigend”) and not perfect (or not even a proper period). The latter case instantiates “die Art des Anreihens” (“the stringing style,” 1887:224). [39]

§25. About a century later, Müller retrieves Blass’ notion and graphically represents the trend of perfect “ascending” períodoi as follows (Müller 1980:32): Müller calls this “arc of tension” (“Spannungsbogen,” 1980:32), which applies to the “perfect” configuration of periodoi. The subclauses that precede the main clause create an ascending trend that culminates in the crest, after which a shorter descending segment follows (“Es entsteht ein ungleicher Spannungsbogen, der im vorderen, längeren und mehrgliedrigen Teil des Satzes allmählich ansteigt, in der Pause einen ruhenden Gipfelpunkt hat und im Schlußkolon [40] schnell abfällt”). [41] Müller sees Herodotus’ style as an intermediate but still imperfect step towards a flawless periodic style. [42] Sometimes the shape of his períodoi is not a nice arc; in that case we find “verunglückte Periode” (1980:10). [43] He compares a fragment of Hecataeus, which he judges artless (“kunstlos”), not exciting (“spannungslos”), clumsy (“holperig”), and monotone (“eintönig,”) [44] to Herodotus 1.8 (the beginning of Candaules’ episode). He notes that Herodotus’ harmony in the varied sequence of kôla replaces the abrupt row of kôla in Hecataeus (“Die länger gewordenen Kola stehen nicht mehr abrupt nebeneinander, sondern sind schon viel harmonischer aneinandergefügt. Die Verknüpfungen sind abwechslungsreicher und komplizierter, sie klammern die Kola enger aneinander” (1980:7). [45] Therefore, Herodotus’ style constitutes an improvement in the use of “strung” diction (“Zweifelsohne ist ein Fortschritt im Gebrauch der λ. ε. [λέξις εἰρομένη] zu erkennen,” 1980:7).

§26. Now let us turn to a textual example that illustrates how modern punctuation (at least in Thucydides) is used to clarify the syntax of períodoi. This períodos (drawn from Stuart Jones’ edition) is nearly coextensive with chapter 22—in fact this chapter has no internal sections. The speaker is Nicias, who explains all the equipment and the resources that are required in order to confront the Sicilian contingents. The length speaks for itself: the idea that Thucydides uses long periods depends on the fact that modern punctuation places full stops after a long series of subclauses and a main clause in between. [46] But this does not mean that Thucydides conceived the passage as one períodos, and that he “uses” or even prefers long períodoi. [47] When we face the OCT page of this excerpt, our eyes automatically glance through a large stretch of text spanning the initial chapter number to the full stop that concludes the first períodos. This reflexive action—perhaps along with Thucydides’ reputation—makes us assume complexity even before we begin to decode words. But we also presume thematic coherence: we are confident that once main clauses are detected, the remaining parts will depend on them, and meaning will be accommodated according to the hierarchies established by syntax. Syntactic complexity is certainly there, and thematic coherence is undeniable. Nevertheless, to embrace such a sizable stretch of text as one periodic unit prevents us from appreciating the individual components in the order in which they are presented. The presumption of periodic unity obscures internal major boundaries signaled by particles; it blurs Thucydides’ meticulous construction of discourse acts that illustrate Nicias’ points. I refer the reader to the suggestions I make in the detailed analysis of this passage in IV.5.2. There I take particles, conjunctions, and other features as signposts for discourse acts whose syntactic ranking does not exhaust their pragmatic force. Such a pragmatic and rhetorical segmentation goes unnoticed if we exclusively orient our reading to answering questions such as “where is the main clause?” and “which subclauses/participles occur?”.

§27. In conclusion: I took the example of períodoi to make a few general connections. First, I connect modern punctuation marks to the set of symbols derived from and used in pre-print and early print versions. Second, I connect grammatical principles to the elocutionary principles guiding punctuation. Third, I connect the way Greek texts are punctuated in modern editions to scholarly efforts to clarify grammatical structure. I show that scholarly appreciation of elaborate períodoi and qualitative judgments about periodic perfection have resulted in syntactic idealism. Finally and most important, I draw a connection between periodic complexity and hypotaxis. It emerges from all the connections I make that readerly segmentation is based on syntactic phenomena of subordination and coordination, as well as on the punctuation marks that are intended to clarify them. Let us see now how input from pre-print strategies of punctuation and from ancient notions of períodos enriches the perspective of discourse segmentation.

§28. Even though names of marks, graphic shapes, and conventions vary chronologically as well as geographically, it is commonly thought that before the first printed editions of classical texts, signs in Western punctuation mainly served performative needs—essentially where and how to pause, while reading aloud or performing. Ong indicates that pre-print punctuation was based on performative criteria, and that clarifying syntax was secondary. [48] The ancient Greek habit of employing scriptio (or scriptura) continua (no word separation)—adopted by the Greeks from the oldest times, and by the Romans only in the 1st-2nd century CE, to imitate the Greek texts [49] —indicates the crucial role that aural decoding played in understanding writing. [50] In medieval times the term denoting a group of signs “added in the classroom to guide the inexperienced reader in the correct pronunciation of text” was prosodiae (Saenger 1997:53). All this commentary attests to widespread belief that there is an inherent connection between the segmentation of discourse and the way in which a written text is realized with one’s voice. [51]

§29. In a recent article, Vatri (2012) demonstrates that ancient Greek reading did not have to be oral for subjects to be able to decode scriptio continua. [52] In fact, the author shows, Thai readers—who read unspaced texts—visually process stretches of language “in the same way as in Western readers of spaced alphabetic scripts” (640). Vatri concludes that cues to decoding words and segmentation depend more on specific combinations of letters which readers recognize rather than on whether reading is silent or not. To this I would add that in light of what Chafe (1988b; see above §13) postulates, the segmentation of discourse is influenced by two phenomena: certain arrangements of words, including syntactic arrangements, and the covert (or inner) prosody of written language. By “written” I mean both “written to be read silently,” and written transcriptions of performed language. They differ, of course, but the idea of prosodic imagery remains in either case, holding both for the implicit interaction of writer-reader as well as for the explicit interaction performer-listener. Thus, in my opinion, it is irrelevant whether silent reading was practiced or not at the time of scriptio continua. Discourse segmentation was applied to texts and processed in any case, in accordance with both word arrangements and inner prosody. I also add that a considerable amount of texts in archaic and classical Greece were designed to be publicly performed, which made the practice of reading aloud a necessary step.

§30. Among the ancient punctuation marks that are adopted in manuscripts of ancient Greek texts, I single out paragraphḗ (παραγραφή, later παράγραφος) and stigmḗ (στιγμή) or distinctio. Paragraphḗ is one of the oldest punctuation marks, and it corresponds to a stroke placed at the left margin of texts or near the start of a line. The word is first attested in Isocrates (Antidosis 59). [53] In manuscripts of plays a paragraphḗ may mark a change of speaker. In lyric texts it separates metrical groups of verses. The function in literary prose often is to signpost the end of sentences or sections, or to mark transitional points. [54] Johnson (1994:68) hypothesizes that such strokes were “practical aids to reading aloud,” and were visually convenient whenever readers had to look back at the texts after possible breaks of any kind. For the purposes of the present chapter I observe that, on the one hand, a paragraphḗ can be compared to section and chapter numbers in modern editions of Herodotus and Thudydides. [55] On the other hand, its function in marking discontinuity in the source of utterances can be compared to modern quotation marks. In both cases paragraphḗ boundaries occur between units of discourse that are larger than simple sentences; this reflects a need (or a wish) to indicate segmentation on a larger scale than individual clauses and sentences.

§31. Stigmaí (later distinctiones) are dots placed at different heights to signal pauses. Higher, middle, and lower dots signal minor, medial, and major pauses respectively. [56] Pfeiffer (1968:177-180) remarks that the adoption of τελεία στιγμή, μέση στιγμή and ὑποστιγμή dates back to Aristophanes of Byzantium (3rd-2nd century BCE). [57] Medieval Byzantine manuscripts of the 9th to 10th century BCE still employ these different dots. [58] They show evidence, however, that the lowest dot (τελεία στιγμή, analogous to our full stop) does not always occur where we would put a full stop, and is, on average, relatively rare. As for middle-high dots (μέση στιγμή), it is often not clear which kind of boundary they stand for: sometimes they delimit segments in the middle of a clause, which evidently do not coincide with clause boundaries. [59] Along similar lines Perria (1991:203-204) remarks that there is an extraordinarily rich spectrum of punctuation marks to be found in a particular collection of 9th century manuscripts. Besides στιγμαί, a few διαστολαί (similar to commas), and a few question marks, various strokes or “nails” (in the author’s Italian text “chiodi”) appear not only between main and subclauses but also after partitive constructions, or between subject and verb, or before and after parenthetical expressions. The author associates the abundance of these signs with the marking of vocal inflections and intonation.

§32. In spite of the relative scarcity of information about ancient punctuation theories, a particularly rich and old source stands out. This is the work of Nicanor, a Greek philologist who lived in the 1st century CE. He is one of the four commentators mentioned in the scholia of Venetus A of the Iliad, and was known as “the punctuator” (ὁ Στιγματίας). [60] His contribution is summarized and discussed in an article by David Blank, quoted above (n57). Blank suggests that the eight punctuation marks established by Nicanor (which I shall shortly introduce) are based on the Stoic distinction between complete and incomplete units of sense. What is strikingly relevant to the topics of this chapter, however, is that Nicanor connects the occurrence of different particles to different lengths of pauses, and therefore to different punctuation marks.

§33. According to Blank (1983:49-50), Nicanor’s system of punctuation, devised for Homer, is as follows. Five types of stigmaí (“full stops” in Blank’s translation) mark completeness (αὐτοτελεῖς λόγοι), while two hypostigmaí (corresponding to high dots) and one hypodiastolḗ (a kind of comma) mark incompleteness (ἐλλεῖπον). The five full stops indicate four to one moments of silence (χρόνοι) ranging from a stronger to a weaker degree of disconnection across two segments of discourse. [61] The deepest separation, which corresponds to a pause of four chrónoi (and corresponds, graphically, to a teleía stigmḗ), occurs between asyndetic clauses, and after vocative or exclamatory expressions. A less deep separation (three chrónoi, represented by a hypoteleía stigmḗ) is for segments introduced by δέ or by another súndesmos, [62] or as Nicanor puts it, τῶν ἰσοδυναμούντων τῷ δέ (“among those whose force is equal to that of δέ”), such as γάρ, ἀλλά, and αὐτάρ. Then comes a pause of two chrónoi (Nicanor’s πρῶτη ἄνω στιγμή) for segments connected by correlatives, such as μέν … δέ; ἤ … ἤ; οὐκ … ἀλλά. Α pause of one chrónos, equivalent to a δεύτερα ἄνω στιγμή, separates segments introduced by καί. Finally, the τρίτη ἄνω στιγμή stands for one chrónos of pause preceding segments introduced by τε. All these full stops mark lógoi that are complete in themselves (αὐτοτελεῖς).

§34. The remaining marks separate incomplete segments; all of them correspond to one chrónos of pause. Two ὑποστιγμαί distinguish between “dramatic” vs. “undramatic” (ἐνυπόκριτος vs. ἀνυπόκριτος) renderings of segments, which remarkably evoke intonation patterns; [63] the “dramatic” ones separate protases from their subsequent apodoses, the “undramatic” isolate τὰ διὰ μέσου, parenthetical segments. Finally, the weakest disconnection is represented by a ὑποδιαστολή, which separates for the sake of clarity (πρὸς τὸ σαφέστερον διασταλτέον); it also separates relative pronouns from their antecedents, incomplete segments sharing a verbal argument, and complete segments sharing elements κατὰ κοινόν.

§35. Blank also records the dismissive attitude of nineteenth-century philologists towards the criteria of “the Punctuator” (1983:48, with nn. 5-8). One of the points on which Friedländer (first editor of the fragments of Nicanor’s punctuation of the Iliad) “disagrees vehemently” with Nicanor concerns the relatively strong pause that Nicanor prescribes before καί and τε clauses. [64] Friedländer’s characterization of Nicanor as articulating an excessively meticulous system of punctuation reveals an interesting discrepancy between the two men’s attitudes toward the proper segmentation of language. A view of discourse that stresses the written aspect does not allow and segments to start after a full stop. However, they are acceptable from the perspective of an oral delivery, especially when “and” starts a new intonation unit, and coincides with the beginning of a separate discourse act or move. [65]

§36. On the whole, Nicanor’s system confirms that punctuation marks and performative “instructions” were still deeply linked in the 1st century CE. The clearest sign of this intertwining seems to me the suggestion of a full stop (τελεία στιγμή, corresponding to the longest pause) after vocative and exclamatory expressions. From this I infer that Nicanor must have associated vocative and exclamatory expressions not only with separate but also with special intonational contours. Analogously, the term ἐνυπόκριτος (“dramatic” [66] ) captures the prosodic relevance of a mark between a subclause and the related main clause.

§37. We should also note from the evidence of Nicanor that punctuation is seen as relating to a mix of lexical and syntactic clues emerging from texts, several of them being the conjunctions and adverbs that we call particles. At least for Nicanor, then, particles can determine the location of punctuation marks. We suppose that also in modern times particles can do that. There are differences, however: while in Nicanor the link is explicit and consistent, in printed editions the association is suggested implicitly and often inconsistently (let us think of full stops or commas before X δέ); also, Nicanor gives importance to the lexical item per se, whereas modern editors presumably combine the evidence of particles with their reconstruction of períodoi (e.g. καί preferably is not preceded by a full stop).

§38. The inquiry into the theoretical and diachronical complexity of these themes continues with the retrieval of a few ancient ideas related to discourse segmentation. The first are the ideas of completeness and incompleteness, which date back to the Stoics. According to Blank (1983:59-60) completion and incompletion are central to Stoic thoughts on grammar and rhetoric. They divide units of saying, or utterances, into two categories, one which they call λεκτά, are αὐτοτελῆ (“self-contained”) if the expression is complete (ἀπαρτισμένην ἔχοντα τὴν ἐκφοράν), and the other ἐλλιπῆ if the expression is incomplete (ἀναπάρτιστον ἔχοντα τὴν ἐκφοράν). [67] Note that such ideas do not directly refer to a specific configuration of syntactic segments. If we think of intonational completion and incompletion, certain intonational profiles may match the delivery of certain syntactic units; but here there is no link between complete segments and specific syntactic segments, not even through intonation.

§39. Aristotle’s definition of períodos does not clarify the syntactic status of what is λεκτά or ἐκφορά for the Stoics. Still, some English translations impose syntactic readings of the terms used by Aristotle. [68] The philosopher famously defines períodos as “a way of saying or diction (λέξις) that has a beginning and an end, and whose size is easy to take in at a glance” (λέγω δὲ περίοδον λέξιν ἔχουσαν ἀρχὴν καὶ τελευτὴν αὐτὴν καθ’ αὑτὴν καὶ μέγεθος εὐσύνοπτον, Rhetoric 3.1409a). What primarily characterizes a λέξις that is not constructed from περίοδοι, but is εἰρομένη (“strung-on”), is the lack of τέλος. [69] Since the example he provides of λέξις εἰρομένη is a version of the incipit of Herodotus’ Histories, which consists of a complete unit of sense (“This (is) the exposition of the investigation of Herodotus of Thurii”), the lack of τέλος must reside elsewhere. It is clear through his reference to round racetracks that Aristotle intends τέλος to mean “rounding out, fulfillment, coming full circle” rather than a linear end. [70] A few centuries later Cicero renders períodos as ambitus verborum. [71]

§40. This “circular” conceptualization of completeness was influenced by different practices in antiquity. [72] While commenting on the periodic style of Thucydides, Lamb (1914:105) explicitly connects the notion of períodos with one of them, that is, intonation. He writes: “in its widest and most literal sense it [period] merely means the rounding or circuit made by the rise and fall of the voice in anything beyond the simplest statement of fact; and this general meaning is fairly well given by the English word ‘compass’.” I align with Lamb. From the point of view of prosody, the starting and the concluding points of a períodos imply various inflections that the voice goes through between two similar prosodic moments. [73]

§41. Another important term in ancient discussion of discourse segmentation is κῶλον, “member.” [74] Where do kôla come into the picture, in Aristotle’s thoughts? Ιn Rhetoric 3.1409b a períodos is said to be either “in kôla” or “simple” (περίοδος δὲ ἡ μὲν ἐν κώλοις ἡ δ’ ἀφελής). A λέξις “ἐν κώλοις” has to be complete, with distinct parts, [75] and easy to repeat in a single breath; it does not include division, but is a whole (ἔστιν δ’ ἐν κώλοις μὲν λέξις ἡ τετελειωμένη τε καὶ διῃρημένη καὶ εὐανάπνευστος, μὴ ἐν τῇ διαιρέσει … ἀλλ’ ὅλη, Rhetoric 3.1409b). [76] In a later passage Aristotle further distinguishes diction-in-kôla into either one “with distinct parts” or “contrasting” (τῆς δὲ ἐν κώλοις λέξεως ἡ μὲν διῃρημένη ἐστὶν ἡ δὲ ἀντικειμένη, 1409b). To illustrate these two types, Aristotle adduces examples from Isocrates’ Panegyricus, from which we can infer the syntactic status of the kôla he refers to. Interestingly, the quotations show a mix of main clauses, elliptical clauses, subclauses, participial and infinitive constructions, verbal phrases and noun phrases. [77] Aristotle sees the expression of contrast “in each of the two members,” ἑκατέρῳ τῷ κώλῳ (ἀντικειμένη δὲ ἐν ᾗ ἑκατέρῳ τῷ κώλῳ ἢ πρὸς ἐναντίῳ ἐναντίον σύγκειται ἢ ταὐτὸ ἐπέζευκται τοῖς ἐναντίοις, 1409b-1410a), but “members” may be, for example, substantivized participles (see n77). Therefore, kôla are not only full clauses; the category must be syntactically inclusive rather than exclusive.

§42. Along the same lines, when Aristotle discusses παρίσωσις (assonance at the beginning of words) and παρομοίωσις (assonance at the end of words, or homoiotéleuton), the similarity of elements is considered across kôla: παρίσωσις δ’ ἐὰν ἴσα τὰ κῶλα, παρομοίωσις δὲ ἐὰν ὅμοια τὰ ἔσχατα ἔχῃ ἑκάτερον τὸ κῶλον· (1410a). Freese (1939) translates, “equality of clauses is parisôsis; the similarity of the final syllables of each clause paromoiôsis. This must take place at the beginning or end of the clauses.” However, taking kôlon as “clause” does not do justice to the Greek (or to Aristotle). The first example that is offered (of assonance at the beginning) Ἀγρὸν γὰρ ἔλαβεν ἀργὸν παρ᾽ αὐτοῦ, “for he received from him land untilled” (Aristophanes fr. 649, reported in 1410a), where the assonance is between Ἀγρὸν and ἀργὸν within the same clause, and not across clauses. Analogously, the second example of assonance at the end is ἐν πλείσταις δὲ φροντίσι καὶ ἐν ἐλαχίσταις ἐλπίσιν (reported still in 1410a, and unattributed by scholars); again, the homoiotéleuton regards φροντίσι and ἐλπίσιν, which belong to different phrases, not to different clauses. Clauses are not the only syntactic construction Aristotle takes into account when he considers kôla. Finally, when Aristotle prescribes that kôla and perídoi should not be either too short or too long (δεῖ δὲ καὶ τὰ κῶλα καὶ τὰς περιόδους μήτε μυούρους εἶναι μήτε μακράς, 1409b), he uses for his criteria the rhythm of discourse and the listeners’ convenience. His remark further supports the interpretation of kôlon as a unit that serves performative functions but is syntactically unspecified.

§43. Further important notes about kôla come from the author of On Style (περὶ ἑρμηνείας). [78] As Schenkeveld puts it (1964:30), for Demetrius “the existence of kôla is essential.” The kôlon is the basic unit; already at the beginning of the treatise kôlon is defined as something that “aims at putting a thought together,” (βούλεται διάνοιαν ἀπαρτίζειν, par. 2). [79] Its crucial function as a building block becomes clear when Demetrius compares kôla in prose to metrical units in poetry (par. 1). A kôlon is either a complete thought, or a complete part of a thought (par. 2). [80] Furthermore, Demetrius gives a name to short kôla, that is, κόμματα: ἡ δὲ τοιαύτη βραχύτης κατὰ τὴν σύνθεσιν κόμμα ὀνομάζεται (“such a brevity in composition is called kómma”; par. 9). [81] Períodoi are made of kôla and kómmata, but while períodoi depend on kôla, kôla exist independently of períodoi. What makes a mix of kôla and kómmata a períodos, then? Demetrius’ definition of períodos is: ἔστιν γὰρ ἡ περίοδος σύστημα κῶλων ἢ κομμάτων εὐκαταστρόφως πρὸς τὴν διάνοιαν τὴν ὑποκειμένην ἀπηρτισμένον (“the period is a system of kôla or kómmata that completes the underlying thought in a well-turned way,” par. 10). [82] The shortest períodos is of two kôla, the largest is of four kôla (however, the size of kôla is not specified). The periodic style is called ἑρμηνεία κατεστραμμένη, the non-periodic style ἑρμηνεία διῃρημένη. [83] The latter is “loosened” as regards kôla: ἡ εἰς κῶλα λελυμένη, οὐ μάλα ἀλλήλοις συνηρτημένα “[the separating style] is loosened with respect to kôla, they are not quite linked to each other” (par. 12). All in all, Demetrius does not seem to offer definitions of kôla that associate them exclusively with clauses. “Kómmata” is introduced as a notion relative to kôla, but syntactic descriptions are absent. Note that if the term kôlon hints at parts of wholes, the etymology of kómma is straightforwardly related to segmentation: from κόπτω, κόμμα literally is something “cut off” (just as English “segment” comes from Latin secare, which also means “cut off”).

§44. In the Roman era, Cicero provides further clues on ancient notions of kôla and kómmata; he calls them membra and incisa respectively. According to Schenkeveld Cicero belongs to the group of post-Aristotelean writers who couple period with rhythm rather than with a unifying thought (διάνοια). Orator ad M. Brutum 66-67 contains a remarkably extended section on the use, in forensic speeches, of membra and incisa, explicitly associated with kôla and kómmata. The períodos (ambitus) is said to consist of four members (constat enim ille ambitus et plena comprehension e quattuor fere partibus, quae membra dicimus), [84] which pleases ears (auris impleat) and is neither too short nor too long. These four members are similar to verses (in a spurious addition they are said to approximately correspond to an hexameter), and what links them are “bonds of continuation” (quasi nodi apparent continuationis, quos in ambitu coniungimus). Conversely, if we want to express ourselves member by member (membratim volumus dicere), we pause (insistimus). Examples of incisa (kómmata) quoted by Cicero are very short syntacticly independent but semantically correlated clauses (missos faciant patronos followed by ipsi prodeant; furthermore, domus tibi deerat? followed by at habebas), while examples of membra (kôla) are two interrogative clauses (cur clandestinis consiliis nos oppugnant? followed by cur de perfugis nostris copias comparant contra nos?). [85]

§45. Quintilian also uses the terms incisum, membrum, and ambitus. [86] He recalls that according to Cicero a period should encompass a single breath (Institutio oratoria 9.4.125). [87] St. Jerome (4th century CE) translates the Bible by using a system of punctuation that he calls per cola et commata. [88] Still later in the Etymologiae of Isidorus of Sevilla (7th century CE) texts are divided into períodoi, kôla and kómmata, and there are indications of where one’s voice should rest. [89] The distinction between membrum and incisum is kept also in Old English psalmody, where the period coincides with the verse, and each segment corresponds to a “breath-group.” [90]

§46. The evidence I have collected in the latter section has an interesting counterpart in modern studies on prose colometry. “Colometry” literally means “measurement in kôla”; the concept per se is neutral with respect to poetic meter and to prose. [91] Recall that for Demetrius kôla are in prose what meter is for poetry (On Style, par. 1; see above, §43). Dionysius of Halicarnassus in De compositione verborum devotes significant attention to the stylistic features that make prose poetic; those features primarily include word placement, rhythm, and sound. Kôla and kómmata are important units in this process. [92] In the 20th century Hellenists as well as Latinists have employed the notion of prose kôla in different types of discussions. I shall delineate some of them.

§47. Eduard Fraenkel was the first to contribute new ideas to the ancient notion of kôla in prose (1932, 1933). [93] His divisions into kôla (marked by vertical bars) are “retrospectively” inferred from postpositive elements (such as ἄν), which signal a boundary before the word they are phonologically attached to. This applies to participial phrases as well. [94] On the level of lexical cues, Fraenkel includes negative markers in the items that may indicate kôlon boundaries, [95] even though he does not specify how to distinguish between negations with phrasal scope and negations with clausal scope.

§48. Robbins 1979 applies the notion of colometry in prose to analyze Cicero’s Third Oration Against Catilina. He identifies periods of four kôla each, and each kôlon is comparable, in length, to a dactylic hexameter. Kôla are clauses or long phrases, whereas kómmata are short phrases or single adverbs, and even single conjunctions that connect two parts of a period (e.g. ut, 1979:58). The conclusion is a bold statement: “It is … an injustice to classical prose writers to print them straight out, using only grammatical punctuation, in disregard of the carefully composed cola and periods. What a tremendous help it would be to students in learning and to teachers and scholars in explaining the classics if we were to write them out, in Saint Jerome’s words, per cola et commata” (1979:62).

§49. In a pedagogically oriented article, Harrison (2007) proposes that students arrange Latin (and Greek) texts by indenting clauses and phrases on separate lines. She equates this practice with reading per cola et commata, thus assimilating cola to clauses and commata to phrases (2007:291). The segments thus obtained facilitate comprehension: “The line-divisions break the text into meaningful chunks and help students see which words do (and do not) go together” (2007:292). Such an approach is seen as “… opposed to more common translation methods, such as translating individual words, trying to intuit the sentence whole, or hunting and picking to reconstruct the sentence in English sentence form” (2007:293). The graphic representation of the segmentation Harrison devises is simple: “Main clauses generally begin flush left. Sub-units (e.g. long prepositional phrases) and subordinate clauses and their sub-units are indented. Parallel items, such as correlatives and compound phrases or clauses, are lined up under one another” (2007:294).

§50. Habinek (1985:127), building on E. Fraenkel’s work, provides a useful list of units that serve as “rhythmical and rhetorical cola” in Latin texts (1985:128): “any weighted or expanded substantive, regardless of case; any weighted modifier, be it participial, adjectival, or adverbial in character; all prepositional phrases; phrases, such as infinitives, that substitute for substantives; and verb phrases.” In addition, the author discusses single-word constituents (128-132); items in enumeration (132); brachylogy [96] (133); correlative constructions (134-135); and vocatives (135-136). Thus, Habinek’s examination of kôla is based on various syntactic constructions; note that full clauses are hardly considered in his list. [97]

§51. Finally, Scheppers 2011 is a monograph entirely devoted to “the colon hypothesis” of the title. His basic hypothesis is twofold: as the author puts it, “the colon is the unit to which Greek word order rules … are applicable,” and “the colon is the elementary discourse unit, in other words: discourse essentially comes in cola” (17; italics in the text). Towards the beginning of the work (2011:15), Scheppers revises E. Fraenkel’s kôlon typology and lists his own four categories of kôla: “autonomous verb clauses” (including “main finite clauses,” “subordinate finite clauses,” “participle clauses,” and “infinitival clauses”); “parallel members” (such as οὔτε … οὔτε …); “syntactically non-integrated constituents” (such as parenthetical clauses); and finally, “fronted constituents” (including “topic noun phrases,” “fronted prepositional phrases,” “fronted markers,” and “common grounds for complex structures”). He also suggests several specific criteria for identifying kôlon boundaries in his textual analyses of Lysias and Plato’s excerpts. “Fronted negatives,” for example, are one of them (243 and 258). [98]

§52. As Scheppers’ monograph is sizable, and the majority of texts under examination are prose texts, this is a thorough contribution on Greek kôla. The scholar’s application of his kôlon criteria in the textual analyses sometimes are questionable, and theoretical issues related to word order perhaps remain unsolved, but the fundamental idea is valuable in itself: if we posit the existence of kôla as segments of prose discourse, in accordance with the accounts (however general) of ancient scholars, an entire world of linguistic details emerge. Such details, which regular hypotactical analyses obscure, give us an enriched view of discourse segmentation, and can affect our comprehension of content.

§53. At least one more feature of modern readings of ancient Greek prose texts is influenced by earlier practices, besides modern punctuation and modern acknowledgment of prose colometry. This is the paralinguistic segmentation realized by chapter and section numbers, indentation, and extra space between words in the OCT editions. [99]

§54. Hemmerdinger (1981:54) tells us that Jungermann (1608) was the scholar who introduced chapter division (τρήματα sive sectiones) in Herodotus’ Histories, whereas Hudson (1696) introduced chapter divisions in Thucydides. [100] With the exception of ms D of Herodotus (10th century), which, according to Hemminger (1981:54) shows a few traces of an ancient chapter division by means of Ionian numbers, medieval manuscripts do not include any such divisions. Therefore, our identification of loci in both works relies on a quite modern macro-segmentation of the texts. The criteria that brought scholars to divide chapters across centuries can generally be inferred by the texts themselves: different chapters straddle different settings or individual setting components, [101] different topics, or different (parts of) speeches. For section divisions uniform criteria are less easy to infer, as internal divisions within stretches of discourse that share the same communicative goals—moves, in our work—can be established on the basis of different assumptions. However, a crucial observation is that the texts display linguistic features from which more segmentation emerges than that suggested by numbers, or a different one. [102]

§55. In addition to chapter and section numbers, the OCT editions report two further paralinguistic signs: indentation and extra space between words. Both can co-occur with chapter and section numbers, but also occur independently. As for indentation, a sample of 11 chapters from book 1, 5, and 9 for Herodotus’ OCT edition (Hude), and book 1, 5, and 8 from Thucydides’ OCT (Stuart Jones) shows that no consistency (neither in form nor in content) underlies chapter starts that are indented and chapter starts that are not. Therefore some degree of arbitrariness has to be assumed, which we could at least lessen if we rely on actual linguistic features. In a very few cases indentation corresponds to section starts instead of chapter starts, and in one case it has no corresponding number. [103]

§56. As for extra space between words, the same sample of chapters gives a more interesting result: while chapter and section starts that are not indented are consistently marked by extra space between words (unless the new unit begins a new OCT line), there are uncountable extra spaces between words without any corresponding numbers, in both Histories. [104] The interest lies in the consistency behind this unnumbered segmentation, which derives from the occurrence of particles and particle combinations. Extra spaces are found, for instance, before X γάρ and X μὲν δή segments. Even though there is no full coincidence between extra spaces in the OCT editions and extra spaces in the two old manuscripts I inspected, as far as the sample of chapters abovementioned are concerned, [105] I interpret all these extra spaces as recorded on the basis of a collation of pre-print editions. [106] The striking match between paralinguistic extra space between words and the linguistic features that in my reading start moves [107] confirms that particles and other co-occurrent features are crucial devices for discourse segmentation.

§57. How should we assess the role of particles with respect to all the factors and views that I have delineated so far? First of all, particles constitute a device for marking discourse segmentation, but other linguistic devices may do the job as well, in combination with particles or independently: these are syntactic devices (e.g. hyperbata, subordinating conjunctions) or lexico-semantic devices (e.g. negations). Therefore, our interpretation of how to chunk ancient Greek discourse has to rely on multiple features. However, particles stand out, because of a quality that makes several of them similar to punctuation and other kinds of paralinguistic segmentation. This quality is the capacity of suggesting when and how discourse proceeds, or when and how discourse stops. A genitive absolute cannot do that, but δέ or μέν can.

§58. As far as the relationship between particles and modern punctuation is concerned, a few considerations are in order. According to Parkes (1992:1) the primary function of punctuation is “to resolve structural uncertainties in a text, and to signal nuances of semantic significance which might otherwise not be conveyed at all.” This applies to particles as well. Punctuation and particles, in other words, both function as road signs. “Punctuation marks are the traffic signals of language: they tell us to slow down, notice this, take a detour, and stop” (Truss 2003:7). Likewise, as Denniston himself puts it (1954:xli-xlii), “The reader or listeners, when he has reached a certain point, meets a particle which looks back to the road he has traversed, and beckons him on in a certain direction.” [108] The resonance between the metaphors is significant. Punctuation and particles are both multifunctional. For example, the same graphic sign “comma” can be used to mark incompleteness of thoughts across contiguous segments in quite different ways, for example to join clauses, to set off list-entries, and to attach details. [109] Analogously, δέ in different constructions can signpost different relations across discourse segments. [110]

§59. However, it would be unfair to analogize particles to punctuation in modern editions of ancient Greek texts. [111] Particles represent more, and give more information, than modern punctuation does. Although they are subject to some variation in textual transmission, particles are more original than both modern punctuation and “posthumous” chapter and section divisions. [112] Furthermore, unlike punctuation marks, they are lexically differentiated, and frequently grouped in combinations. It is up to us modern readers to take on the challenge of grasping the hints that particles offer about segmenting and articulating discourse, by improving our sensitivity to their implications and to metacommunication.

§60. I do not deny that modern punctuation is a helpful tool for decoding ancient Greek texts. As I already noted, to some extent modern punctuation may even be inspired or motivated by the occurrence of particles, especially peninitial-position particles. [113] My point, however, is that full stops and commas do not simply aid our reading. On at least the visual and cognitive levels they signify much more. They paralinguistically suggest stronger and weaker discourse boundaries, and they shape the length of segments. Full stops cause us to presuppose that what occurs between them is a unit of some sort. Parentheses cause us to interpret the information contained within them as secondary information. The absence of exclamation marks causes us to assume that exclamatory acts are unlikely to occur. The list may continue. In other words, modern punctuation marks suggest a certain reading of segments, units, and grammar, or a reading of them in a certain direction. This may or may not harmonize with what particles suggest. My analyses will show matches and mismatches between segments suggested by modern punctuation and segments suggested by particles. The mismatches comprise the more interesting evidence, as they motivate us to observe more closely what particles convey that modern punctuation cannot.

§61. A further comparison will be made between the occurrence of particles and chapter and section numbers as we find them in print editions. In those cases the discourse boundaries relate more to multiple sentences and multiple períodoi, in a word to segmentation on a larger scale. Also at this level matches and mismatches are observable. In some current editions we may even find indentations at chapter starts, and, as I already said earlier, extra spaces between words. But indentation does not consistently accompany chapter starts, and extra space between words does not accompany only section and chapter starts, which makes consistency hard to find. [114] However, extra spaces do almost consistently show a match with particles: as I will point out, extra spaces largely coincide with move boundaries signaled by particles and particle combinations. My general claim is that the occurrence of linguistic features, especially particles and particle combinations, is more reliable and more precise than modern segmentation also above the sentence level, at least in terms of lexical consistency and variation.

§62. Let us now briefly recall considerations about labels for segments. We have seen that the discussion of periodic styles in modern times privileges syntactic ranking (hypotaxis) and associated trends (subclauses following or preceding main clauses). The consideration of kôla and kommata in periods (períodoi) unfortunately is absent. Ancient views of segmentation, conversely, pay close attention to kôla and kómmata that represent smaller but no less relevant building blocks of discourse than períodoi (kómmata, in turn, represent smaller elements than kôla).

§63. The fact that the length and, most of all, the grammatical form of kôla and kómmata are left unspecified is, paradoxically, a very important cue. It means that no specific syntactic characterizations can be associated with them. Despite the tendency to translate kôlon as “clause,” kôla do not necessarily coincide with clauses. Ancient punctuation practices (for all their diachronic and typological variety) signal the same: marks occur beyond clause boundaries. Both ancient segmentation and ancient punctuation apply a broader perspective than periodic or sentence-based articulation. Both consistently show that grammar and eloquence do not conflict. Modern studies in prose colometry recall and reinforce these ideas, which in ancient times were integral to the composition and interpretation of prose texts.

§64. Overall, ancient views about segmentation, pre-print punctuation, and the modern retrieval of the notion of prose colometry show that segmentation is inspired by several phenomena of language production and language reception. Criteria and signs may differ. Priorities in establishing the criteria may change across time and reading (or teaching) habits. However, all these systems or frameworks start from words, from the linguistic material available. The distribution and the combination of linguistic features play a paramount role; although they may have been affected by textual transmission, they chronologically precede any ancient punctuation and paralinguistic marking. Now it is time to clarify how we can process language in order to infer discourse segmentation, where “discourse” is informed by syntax, semantics, and pragmatics.

§65. In order to exemplify the perspective that is adopted in the textual analyses (see below 3.10-3.11, and IV.5.2-5), let us consider what the full stop in modern punctuation and the chapter number do in the following passage, drawn from the end of Thucydides’ account of the plague that devastated Athens over the first years of the war (Book 2.47-54). The full stop invites us to consider ταῦτα μὲν … and οἱ δὲ … as quite separate segments. A further modern paralinguistic sign is the chapter number, which precedes οἱ δὲ Πελοποννήσιοι. It warns us that a major discourse boundary occurs. My point is that the language encodes this segmentation regardless of the presence of a full stop and a chapter number.

§66. Let me expand on how Thucydides’ language provides all the information necessary for processing the two segments without the need of an intervening full stop. In order to understand this excerpt we need to process ταῦτα first. The pronoun refers to all the facts and events connected to “the disease,” which we infer from τὰ κατὰ τὴν νόσον γενόμενα. Then we need to realize that what follows (οἱ δὲ Πελοποννήσιοι) introduces a totally different subject matter. How do we interpret the role of μέν and δέ in this context? The adjacent μέν and δέ clauses cannot be a manifestation of the μέν – δέ construction expressing semantic contrast. Rather, they juxtapose what the discourse is about: first “these (events),” and then “the Peloponnesians.” On a larger scale, the μέν act rounds off a long section (the seven chapters that Thucydides devotes to the plague). At the same time it anticipates that the account will continue. This is the function that μέν performs in this utterance. [115] The δέ statement, conversely, opens a separate and new narrative section (starting with a new δέ segment). We may even “hear” the inner voice of the historian and imagine a specific intonation matching the μέν statement (resembling the tone of an interim conclusion [116] ), and a different intonation matching the δέ statement (launching a new topical spin).

§67. We would not be able to comprehend how discourse is flowing [117] at this point in Thucydides’ account if we did not process all this information, which regards the local co-text as well as a broader co-text and context. These steps sound natural; nonetheless, they require a combination of certain operations and skills: the activation of the short-term discourse memory (what was said in the previous units), anaphora processing (to retrieve a relevant referent for the pronoun), competence in syntax (to combine τά with γενόμενα, for example), competence in semantics (to connect “the disease” with the plague, for example), and encyclopedic and historiographical knowledge (to infer that anything to be said about the Peloponnesians at that point does not have the Athenians as its main subject). The full stop and the chapter number in modern editions adumbrate only a small part of this entire process.

§68. This chapter calls for paying attention to discourse as an organic whole made of several linguistic components. The approach we advocate implies a fundamental assumption: just as with any other kind of textual interpretation, interpretation of particles relies on Apollonius’ and Heliodorus’ idea that particles “co-signify” (συσσημαίνουσι) with other words. [118] The discourse analyses of Thucydides and Herodotus undertaken in the next part of the chapter and in IV.5 are based on a holistic principle. They consider the fundamental dimensions of language influencing each other (syntax, semantics, pragmatics), on local as well as global levels of discourse—that is, from phrases to paragraphs, from acts to genre. [119] Considering the pragmatic dimension, in particular, means taking account of the relation between language and dedicated communicative settings (or situations), and between co-text and paralinguistic context.

§69. We contend that this holistic view aligns with the equally holistic view of language that ancient rhetors expressed in several works. The pivotal concept for them was that of performance: for most of the genres and periods under discussion, written compositions ended up as discourse embodied through speakers or singers – a type of discourse that was voiced and received aurally. This practice informed the description, by ancient writers, of forms, patterns, and stylistic rules of language. [120] We moderns have to content ourselves with only the written part of texts designed for performance. [121] However, words provide excellent clues on several sophisticated levels of communication. For example, they may reveal discourse segmentation, cognitive and/or syntactic projections, rhetorical figures, discourse patterns, phonological resonance, and context-dependent references. [122] Among the various components of literary language that help us detect all of this, particles stand out. They typically signal discourse procedures, communicative actions, and attitudes that reveal discourse strategies.

§70. Two notions that we use in the light of this holistic principle are those of act and move. [123] Both reflect a discourse perspective that includes but also reaches beyond syntax and periodic divisions. Although classifications of act and move types are neither proposed nor wished for, acts and moves postulate a simple idea: people perform them to achieve discrete communicative goals and broader communicative goals respectively. In Herodotus and Thucydides, acts and moves are performed by the two authors; through authorial mimesis they are meant to be performed by speaking characters.

§71. Let us first concentrate on discourse acts, and on the role of particles in marking them As II.2 §§24-25 recalls, Bakker (1997b:146-155) compares hexametrical kôla in Homer to intonation units. While discussing Dionysius of Halicarnassus’ view of sentences, Viljamaa (2003:173-176) also compares the ancient notion of prose kôla to the contemporary notion of intonation units. In his work on hyperbata, Markovic (2006:127) assumes the same. [124] Scheppers (2011:18-35) starts his investigation by illustrating the centrality of intonation units, and shows that particles can mark kôlon boundaries. Certainly, particles may be compared to prosodic marking, and prosodic units may or may not coincide with syntactic boundaries. Nicanor indicates as much when he assigns longer and shorter pauses to stretches of text starting with certain particles (see above, §§ 33-34). To the extent that peninitial-position particles emphasize the group of words to which they are attached, they may signal prosodic boundaries and indicate ad hoc intonational contours—it is conceivable to think of a range of μέν clause intonational contours, δέ clause intonational contours, and so on. Goldstein’s thorough research (2010, 2013b) on the prosodic relevance of the distribution of clitics in Classical Greek—including several particles—is a further landmark for us: it demonstrates that intonation has a crucial bearing on the construction of semantic and pragmatic meaning. In our terms, intonation has a crucial bearing on discourse segmentation.

§72. Our work intends to draw connections to studies in prose colometry as well as to research on intonation units/“intonational phrases” (Goldstein 2010:42). As our focus is on discourse strategies, we explore the discourse relevance of what is behind the idea of intonation unit/intonational phrase, rather than using the terms themselves. Likewise, we propose to consider what may be seen behind the term kôlon, rather than applying the term itself. Therefore, while constantly keeping in mind links between particles and prosody, on the one hand, and the links between particles and kôla, on the other, we call the basic segments of verbal communication discourse acts. Their value is more theoretical than philological—that is to say, the aim is not to reconstruct “original” discourse acts—; what we identify throughout our volumes is potential discourse acts. [125] In modern languages discourse acts tend to be realized by intonation units in spoken varieties, and by punctuation units in written varieties. [126] They represent small steps, but these steps are strategic in that they let the speaker make points. They are acts because they perform communicative actions together with, and beyond the words uttered. [127]

§73. The following subsections will exemplify discourse segmentation in acts. Since their grammatical form is not what defines them, the analysis focuses on acts in continuous texts rather than on selected acts from different excerpts. In this way the very process of segmentation is in focus as well. The goal is to show how chunking discourse into acts changes the reading and the interpretation of the text, and how particles elucidate discourse acts. Given the relatively small scale of acts, comparisons will be made especially with modern as well as pre-print punctuation.

§74. Different is going to be the case of moves, that is, units typically made up of multiple acts. Comparisons in that case will include also paralinguistic segmentation above the sentence level. Also, the discussion of moves will require more general considerations, and special attention to the recurrence of particle combinations and other constructions. I remind the reader that chapter IV.5 blends the analysis of acts and that of moves, and clarifies the relation between them.

§75. Earlier in the chapter (§§ 24-25) I mentioned Müller’s idea of successful periods, that is, syntactically complex wholes displaying a series of subclauses that culminate in a main clause. Müller adopts for these periods the notion, originally coined by Blass (1887), of “aufsteigende Periode” (“ascending period”). Discussing the periodic style of Herodotus, Müller suggests that the historian, in failing to sustain such “ascending periods,” falls short of a perfection that later writes come to practice. The Histories of Herodotus from time to time includes what he calls “unsuccessful” (“verunglückte”) periods. Müller lists some examples, among which is a passage from the episode of the Cnidians in book 1 (1.174.2-3). Müller is not the only scholar to noticed the peculiar way in which the narrative is articulated in that passage: How and Wells (1928:134) calls 174.3 “a model of confusion.” This view is a longstanding one and seems to involve the issue of syntactic hierarchies: long before Müller and How and Wells, Bekker (whose edition of Herodotus dates back to 1845) decided to add a relative pronoun at the beginning of section 2, even though the reading is not attested in any manuscript.

§76. Let us first report the text as it appears in the OCT edition. [128] This way, we can begin by interpreting the excerpt according to its articulation in periods: from full stop to full stop, with intermediate commas, with section numbers as the only signals of higher level segmentation, and with an instance of indentation at the start of the chapter. Godley’s translation directly follows, featuring in turn full stops and commas together with syntactic divisions appropriate to the flow of discourse in English. [129] Section 1 coincides with a period, with the people of Caria as the subject. The full stop occurs before οἰκέουσι δέ, where section 2 starts. The next full stop occurs only at the end of section 3; the subject of this second period is the Cnidians, one of the peoples of Caria. Overall this print edition segments the text into two periods including nine commas and a parenthesis (at the beginning of section 3). The punctuation yields a total of 13 complete and incomplete segments.

§77. Now, how would we process the content differently if we segment by discourse acts? Let us visualize the segmentation that results from using occurrences of particles, and features such as clauses, negations, and absolute constructions, to draw our boundaries. [131] The following translation attempts to reflect the sequence and the pace of acts. Each segment (which I number for convenience) represents a discourse act. For the sake of comparison, the modern punctuation of the OCT edition is retained (including the parenthesis at the beginning of section 3). Overall we obtain 19 discourse acts, 13 of which end with a modern punctuation mark. The 13 acts regularly coincide with dependent or independent clauses. [132] In the rest of the 6 cases our criteria were as follows: the negation οὐδέν separates act 2 from act 1 as οὔτε separates act 5 from 4. Two verbs belonging to two different clauses separate acts 2 and 3. The noun phrase ὁ Κεραμεικὸς κόλπος separates act 12 from 11 by analogy with the syntactically parallel noun phrase (with hyperbaton) ἡ κατὰ Σύμην τε καὶ Ῥόδον θάλασσα, which separates act 14 from 13. Acts 17 and 18 are different clauses, even though no comma divides them. This brief survey of act boundaries shows how many different linguistic features contribute to identifying them.

§78. Every act is a strategic step in communication, regardless of whether or not it includes a verb form, and whether the verb form is finite or non-finite. Act divisions level out syntactic hierarchies, in fact: participial clauses stand on an equal footing with main clauses, just as vocatives stand equal to both. Hence the genitive absolute in act 7 is an act per se, just as 9 and 10 are. No relative pronoun is needed. Only modern readers who artificially impose a syntactic convergence of multiple clauses upon ὤρυσσον οἱ Κνίδιοι (act 17) would find the pronoun wanting, and still be left with a very imperfect period.

§79. Particles signal act divisions just as decisively as other linguistic features. Notice the act-peninitial position of δή in 8 and τε in 10. [133] τε in 10 marks also the coordination between two genitives absolute. Further act-peninitial particles are μὲν γάρ in 11 and ὦν δή in 15. They are combinations that introduce moves, as later paragraphs will clarify. As lexical markers, μὲν γάρ and ὦν δή are more precise segmenting devices than the parentheses, which are technically redundant; the former marks the opening of unframed discourse, while the latter conveys frame recall. [134] In fact, it is likely that these particles’ occurrence was what originally motivated editors to insert the parentheses. [135] Finally, in act 1 the combination μέν νυν complements the priming function of the act as a whole (οἱ μέν νυν Κᾶρες): the Carians are cognitively “primed” as the focus of attention over the next few acts, and the μέν component announces projection. [136] Then a δέ act (act 6) shifts the attention to the Cnidians.

§80. Act divisions show how the pragmatic and cognitive gist of the passage unfolds in sequence. After the Carians have been dealt with in acts 1-5, the Cnidians come to the fore in act 6 as the next focus of attention—the pinning-down function of the construction καὶ ἄλλοι καί is notable. [137] Once this new focus has been established, acts 7 to 16 further narrow the focus with a series of geographical points that culminate in mention of a certain piece of land (τὸ … ὀλίγον τοῦτο, act 15) that becomes the next crucial cognitive signpost. In fact, acts 15 and 16 together pave the way for the pivotal action of the entire episode, ὤρυσσον οἱ Κνίδιοι (“the Cnidians were digging,” 17). [138] Κνίδιοι is retrieved, and echoes Κνίδιοι in act 6. The syntactic prominence of ὤρυσσον οἱ Κνίδιοι as a main clause matches its cognitive prominence as the pivotal action. However, in the flow of discourse ὤρυσσον οἱ Κνίδιοι is not the climax of an imperfect period; rather, it is the device that Herodotus exploits to fix the image of “digging and digging.” [139] For soon after ὤρυσσον οἱ Κνίδιοι appears another act, not quoted in (t12), that contains the same verb form ὤρυσσον (174.3). It is this repetition that fixes the image, and is instrumental to the narration of the wonder of the inexplicable wounds (introduced by καὶ δή, 174.4). [140]

§81. A more systematic comparison of modern punctuation, act segmentation, and medieval punctuation will be discussed in IV.5 using four longer excerpts. For now, let us complement these observations on 1.174.1-3 with the evidence from the oldest extant manuscript of Herodotus. [141] Several punctuation marks confirm the act boundaries of my interpretation. Those occurring in the manuscript where no modern punctuation marks appear are a comma separating acts 2 and 3; a high dot separating 4 and 5; a mid dot separating 11 and 12, and 13 and 14. No parenthesis occurs. Finally, there is no consistent match between high dots, mid dots, modern commas, and full stops.

§82. In summary, we see that the pragmatic organization of the piece fits the cognitive needs of listeners and readers well, even though the syntactic organization in main and subclauses in this passage is unsatisfactory to Müller, How and Wells, and Bekker. Particles help see the boundaries of discourse acts, and are more precise than modern punctuation in signaling the directions of discourse. Finally, a sample of pre-print punctuation on the passage reveals that medieval readers established boundaries equal to but also smaller than clauses. Also, the distribution of medieval punctuation marks does not necessarily coincide with the placement of modern commas and full stops. All of this confirms the non-absolute value of paralinguistic segmentation.

§83. Now let us consider a sample from Thucydides. In this case too we are dealing with a stretch of discourse that has been judged as ineffective. The passage 2.53.4, is from the chapters on the plague. It famously illustrates the shameless lack of respect for gods and human laws by the people affected by the desperate situation. Yaginuma (1995:137) calls this complex period “absteigend,” “descending,” referencing Blass 1887 (see above, §23): the main clause occurs at the very start, and is followed by a series of subclauses. Keep in mind that this type of period is considered less felicitous than the ascending type, as the movement from independent to dependent elements is seen as lowering the tension. Blass assimilates this type of periods to parataxis. Yaginuma, moreover, points out that the passage proves that Thucydides wrote for readers instead of listeners: “several sentences are forcedly made into an ‘absteigende Periode,’ and so would be rather hard to follow in hearing” (1995:137).

§84. Let us first report the text, starting on a new line after every comma, to emphasize how we normally process the segmentation offered by print editions. My analysis of acts in Herodotus showed that some modern punctuation seems to reflect the occurrence of peninitial particles. This is clearly the case here: four out of five segments ending with modern punctuation marks start with μέν or δέ in peninitial position. Hammond’s translation reflects this five-part division: the first two segments correspond to two simple sentences (“No fear of god or human law was any constraint.” and “Pious or impious made no difference in their view, when they could see all dying without distinction.”). The third segment extend to the colon (“As for offences against the law, no one expected to live long enough to be brought to justice and pay the penalty:”); the fourth and fifth are separated by a comma (“they thought that a much heavier sentence had already been passed and was hanging over them, so they might as well have some enjoyment of life before it fell.”). The modern commas, however, have two adverse effects on interpretation that are not supported by the text. One is the implication that the sense remains incomplete through all the segments until the last. The other is the implication that punctuation marks are not warranted for any other boundary.

§85. The inadequacy of modern punctuation may be grasped by comparing it against the medieval punctuation found in an 11th century manuscript of Thucydides. [142] These marks are recorded in italics and curved brackets in the version of the passage I present below (t14), which represents my segmentation in acts. My translation in acts follows. First the medieval punctuation. At 6 and 11 the high dot and extra space before the next word indicate a strong boundary, a reading that is not conveyed by the use of modern commas. The evidence of pre-print punctuation thus challenges the implication of incompleteness suggested by commas. As for the implication that no other intervening boundary requires punctuation, consider acts 2, 4, 7, and 10. The medieval comma in act 2 after οὐδείς probably suggests that οὐδείς grammatically and prosodically relates to φόβος as well as to νόμος. The high dot at the end of act 4 marks the boundary between the infinitive construction and the substantivized infinitive, which is a regular clause boundary. The comma at the end of act 10 is particularly interesting, as it reflects, as we will see, a specific discourse strategy.

§86. Several acts in my segmentation have no corresponding punctuation mark, either print or pre-print, at their end (see 1, 3, 5, 8, 9, 12, 14, 15). Most of the boundaries are clause boundaries (3/4; 5/6; 8/9; 14/15; 15/16); therefore, acts coincide with clauses. Mismatches between acts and clauses deserve ad hoc comments. But before entering into details it makes sense to illustrate the general difference between the interpretation deriving from a canonical segmentation and the interpretation resulting from the segmentation in acts.

§87. The discourse analysis I offer based on my segmentation does not challenge the syntactic articulation of the passage or the semantic prominence of the main clause that opens the period (θεῶν δὲ φόβος ἢ ἀνθρώπων νόμος οὐδεὶς ἀπεῖργε). Rather, it challenges the idea that “descending” syntax (main clause followed by a series of subclauses) reflects decreasing communicative interests. By looking instead at the passages’ pragmatic articulation in acts, we see that there are significant intertextual connections that have been previously overlooked. The discourse acts turn out to be carefully sequenced and juxtaposed to convey the people’s progressive introspection. Indeed, the acts that follow the main clause even reach a climax and denouement, totally in contrast to the view that the discourse “descends.”

§88. Act 1 establishes the two parallel realms that constitute the subject of the upcoming discourse, fear of gods and human law. Act 2, which begins with οὐδείς, [143] negates both categories. The act-initial position of οὐδείς emphasizes the attitude of denial, renounciation, and failure that underlies other negative expressions in the Plague chapters. [144] Act 3, τὸ μὲν κρίνοντες, fulfills two pragmatic functions. τὸ μέν [145] projects a multiple-act stretch of discourse. The participle κρίνοντες introduces into the audience’s minds the concrete image of a plurality of individuals thinking about and evaluating the situation. While it is remarkable, from a grammatical point of view, that the text should lack a pronoun to clarify who these individuals might be, in cognitive terms the omission simply means that the referents are contextually accessible. [146] Acts 4 and 5 are variations of infinitive constructions: 4 is an infinitive clause and 5 features an infinitive as substantive. [147] Act 6 coincides with a participial phrase that alliterates with acts 4 and 5 (ἐν, ἐκ, ἐν). [148] Act 7 (τῶν δὲ ἁμαρτημάτων “of the wrongdoings”) [149] has a crucial pragmatic and cognitive value that explains why it grammatically precedes the nouns it refers to. It is a δέ priming act that projects further acts on the topic “wrongdoings,” that is, the topic exemplifying the lack of respect for law. [150] In act 8 (οὐδεὶς ἐλπίζων) οὐδείς resonates with the negation in act 2 (οὐδεὶς ἀπεῖργε); ἐλπίζων, a participle that refers to mental activity, resonates with κρίνοντες in act 3. Act 10 (βιοὺς ἄν) is a separate act because it is semantically ambivalent: it could be attached both to act 9 (μέχρι τοῦ δίκην γενέσθαι, “alive until the (moment of the) trial”) and act 11 (τὴν τιμωρίαν ἀντιδοῦναι, “alive to pay the penalty”). [151]

§89. Act 12 (πολὺ δὲ μείζω) begins the climax of the indirect thought, [152] and it features a discontinuity that the modern comma at the end of 11 obscures, but which the pre-print punctuation acknowledges (high dot and relatively large space before πολύ). [153] Rusten (1989:192) takes πολὺ δὲ μείζω to be dependent on ἐλπίζων (8), with an emphatic predicative position. Syntactically and semantically it may also depend on κρίνοντες (3). As for the noun that the comparative μείζω modifies, Rusten thinks it is τιμωρίαν to be connected to the subsequent κατεψηφισμένην (“pronounced against”) in act 13; [154] Rhodes, on the other hand, (1988:103) reads πολὺ δὲ μείζω as a generic adverbial phrase: “people tended much more to think that a sentence already decided was hanging over them.” Act 13, the longest in this excerpt (τὴν ἤδη κατεψηφισμένην σφῶν ἐπικρεμασθῆναι) represents the peak, because it expresses the impending doom from the internal point of view of the characters involved. [155] Finally, the denouement of the micropiece constituted by the reported thought (“before ruin fell, it was appropriate to enjoy a bit of life”) features disentanglement on more levels. Semantically, clauses progress from the impending doom to some enjoyment, even though the chronological order of the moments is the opposite. Performatively, whereas acts 13 and 14 include more consonants, 15 and 16 employ more vowels—ἀπολαῦσαι alone includes two vowels and two diphthongs. It is as if the sequence of sounds reflects in performance the loosening of life—an ultimate form of empathy with the peoples’ feelings.

§90. To sum up: act-segmentation brings attention to a number of syntactic, semantic, and prosodic details in this excerpt that directly contradicts the characterization of this períodos as hard to follow. One of these details is visible only after we take into account the discourse segmentation that I propose: 11 out of 16 acts end with verbs, and in particular two acts (11 and 16) end with 4-syllable verbs (ἀντιδοῦναι and ἀπολαῦσαι respectively). [156] In fact, act-analysis shows that there are more major discontinuities in this passage than previously acknowledged (see e.g. 7 and 12, given weak commas in modern punctuation), which challenges the interpretation of it as just one unit of discourse. [157] In addition, even though this excerpt fits the definition of “descending” períodos according to Blass, act-analysis reveals a progression from impersonal to personal elements of content that culminate in the vivid image of doom hanging over the heads of the people.

§91. In the segmentation discussed so far I have adopted a method of identifying acts based on particles and other linguistic features found in situ. [158] I have avoided general descriptions that attempt to use content to identify typical kinds of acts—such as “descriptive act,” or “exhortative act.” This is because we believe that there are no content-types that are independent of the actual stretches of discourse in which acts occur. The same holds for moves.

§92. Moves are discourse units above the act level. They consist of either several acts that share the same overarching communicative goal, or one act that reveals a major discontinuity with what precedes it. [159]

§93. Just as in the case of acts, I study moves in situ instead of classifying content-types of moves in advance. A Herodotean judgment about whether a certain explanation of a phenomenon is credible or not can easily be described as a move whose point is to express an assessment. But the description of a plain before a battle can be interpreted as the initial part of a long narrative move, as well as a discrete move with a digressive character. In the pragmatic perspective we propose, we are not so much aiming at classifying fixed types of narrative actions, but rather at identifying what a move performs in its communicative setting. For example, I do not view the abstract narrative task “description of a plain before a battle” as existing in some predetermined slot in historiographical discourse; the historian can choose to preface a battle narrative with whatever introductory information he considers essential to the ongoing discourse. This information may involve knowledge necessary for enabling the reader to follow the development of the battle; it may illuminate the conditions (striking or not, ominous or not) under which the battle takes place; it may build suspense; it may compete with images of the same plain known to the audience, etc. It is one or more of these ad hoc purposes that defines what the move does, that is, what its communicative goal is.

§94. From the actual linguistic realizations of moves, and especially from the occurrence of particles and other specific linguistic features, some general pragmatic characteristics of moves can be inferred nevertheless. The following paragraphs, which precede the textual analyses, outline them. Points are made about what distinguishes moves from acts, and how move boundaries can be detected. I also compare the term “move” to other labels that have been used for multi-sentence units.

§95. I started this section by saying that a move is a discourse unit made of more acts, or just one act that reveals a major discontinuity with what precedes it. This suggests that discontinuity is a decisive factor, whereas the number of acts –in other words, the move size—is less relevant. Discourse discontinuity is central to the identification of moves as well as acts; [160] if so, what makes the discontinuity across moves “major”? The relevant element that suggests a move start is some reorientation of the discourse. [161] Examples of reorientation include a change in subject, time, or place; [162] a change in verbal tense or mood; the occurrence of a postpositive particle with a projecting function; the occurrence of a performative verb (such as “I will explain”); and the occurrence of a marked demonstrative pronoun. All these features may appear alone, but more often they appear in combination. [163] Until the next reorientation, the discourse can be taken to unfold acts belonging to the same move. A one-act move results only from a combination of reorienting features.

§96. Moves are in part based on the cognitive idea of contextual frames. [164] Each move evokes, for the reader, a mental space that is the receptacle of items such as events, characters and arguments. The features that make events, characters, and arguments part of the same frame are what makes a move coherent. However, the term move adds to the cognitive idea of frame the interactional idea of communicative purpose. Often the beginning of a move encodes its communicative purpose, more or less straightforwardly. For example, a speaker who switches, within his speech, from the communicative goal “persuading the audience to attack X” to the goal “listing the operations that are necessary to implement the attack” may underscore the different moves that he performs by using words or by inserting discourse acts that suggest to the addressees the respective overarching intentions. [165]

§97. Major discontinuities can be underscored not only linguistically but also paralinguistically. In modern writing and speech, moves can be detected also through paragraphs and paratones (speech paragraphs). The boundaries of paragraphs are marked visually by indentation; the boundaries of paratones are marked through intonational discontinuities by relatively higher or lower pitches, and by relatively longer pauses. [166] Those are the paralinguistic signs that accompany a major discourse discontinuity.

§98. From the seventeenth century onward (see §54), section and chapter numbers became the canonical paralinguistic segmentation of Herodotus’ and Thucydides’ Histories above the sentence level. However, sections or chapters do not always coincide with what pragmatically is a move. That is to say, segmentation numbers in modern editions sometimes do not match the occurrence of the linguistic discontinuities I mentioned earlier. This is why I associate moves directly with the occurrence of different linguistic features, including particles and particle combinations, which indicate discourse reorientation. [167]

§99. So far I established the following: discontinuity is more relevant than size, and it effects reorientation in the discourse; major discontinuities identify move starts, and are suggested by different linguistic features; paralinguistic means in modern texts are associated with major discontinuities (indentation, change of pitch, long pauses); for Herodotus and Thucydides, however, section and chapter numbers reflect them only partially. The next pragmatic characteristic concerns predictability. A couple of observations are in order.

§100. First, even though the exact size of a move cannot be predicted by its start, a relative sense of move length—that is to say, a move at least longer than one act—can be inferred from certain move starts. I refer to the acts that we call priming acts. [168] In Herodotus and Thucydides priming acts project several related acts, or at least more than one. Therefore, from their occurrence we can predict that the move is going to be relatively long. Subsection 3.11.1 will discuss these particular move starts.

§101. The second observation concerns moves that are predictable because they are expected. They are expected due to genre conventions. Historiographers need or want to enact certain moves so that their discourse conforms to the genre. In that case moves are macro-moves: they are units of multiple moves that share a broad communicative goal, such as “providing a brief historical excursus about people X” or “narrating the reactions of X to the outcome of a battle.” [169] Broad communicative goals are generally expected by modern readers [170] in the historiographical texts of Herodotus and Thucydides; these goals contribute to the recognition of the genre as such.

§102. Perhaps the clearest example of macro-moves in historiography is the insertion of speeches supposedly delivered by historical figures. Clarity about their boundaries comes from the linguistic codification of their beginning and end, and from their pragmatic consistency. Language can tell us that a speaker is going to take the floor by means of a set of possibilities. For instance, in the phrase ἔλεγε Ξέρξης τάδε (Herodotus 7.8) the name Xerxes and the verbum dicendi provide semantic clues, and the forward-looking pronoun provides a pragmatic clue; τάδε that tell us (without the aid of modern punctuation) to expect a direct speech by Xerxes. The speech forms a macro-move made up of several internal moves (each one in turn made up of several acts). Its end is signalled through ταῦτα εἴπας ἐπαύετο (7.8δ), with a symmetrical verbum dicendi and another pronoun that we now process as backward-looking. References to “I”, “we”, and “you” (singular or plural) through personal pronouns, verb forms, and vocatives provide the speech with pragmatic consistency.

§103. I prefer “macro-move” to two possibly parallel notions. One is the notion of episode. Episodes, which originally designated the parts between two choral songs in drama, are particularly at home in historiographical narratives. Their constitutive elements are a place, a time, and one or more characters. [171] Across episodes, at least one of these constitutive elements changes. However, while episodes can represent macro-moves, macro-moves include more than episodes: for instance, elaborated arguments supporting a historical claim; ethnographic surveys regarding one people; speeches as well as accounts of indirect speeches; and several kinds of “digressions.” [172]

§104. Another label used for general segments of written texts (of different size) is the term “discourse mode.” [173] Scholars have classified discourse modes into the following subtypes: narration, description, reporting (that is, interrupting narration to report on the narrator’s present, past, or future), and registering (that is, taking stock of the events being narrated). Each mode is viewed as characterized particularly by the use of certain tenses. Works such as Allan 2013 discuss text samples featuring overlaps between two modes, but the model in itself focuses on prototype-categories, on abstract types/styles of communication. In terms of discourse modes, Herodotus and Thucydides regularly blend narration and description, registering and description, reporting and narration. For the analysis of actual texts, I regard the concept of move as more neutral: by being not linked to a priori classifications of subtypes, it lets the reading of overarching communicative goals emerge from the ‘local’ arrangement of its constitutive discourse acts.

§105. In general macro-move boundaries are easier to infer than act and move boundaries, due to relatively straightforward semantic discontinuities and content closures (such as “and the winter ended” or “this is what he said”). [174] The next subsections analyze co-signifying features by means of which move boundaries can be detected, which are less easy to process. The reason is that the features indicating move boundaries usually exceed semantic cues. The features analyzed constitute patterns, as they recur. This is why I privilege ad hoc passages over the examination of continuous excerpts—which I carried out in 3.10 for discourse acts. IV.5 complements this part by detecting moves in continuous texts.

§106. For each of the passages segmented and commented, I will indicate matches and mismatches with modern paralinguistic segmentation, that is, punctuation as well as chapter and section numbers, indentation, and extra space between words as they appear in the OCT editions.

§107. I start the analysis of particles in moves by examining an understudied pattern: the use of particles in short acts that start moves. II.2 discusses short δέ acts in Homer that reorient the attention of performer and audience to a different setting or character. These are called “priming acts.” Short δέ acts in Herodotus and Thucydides perform a similar priming function, but—perhaps not surprisingly, regarding the different genre—the spectrum of reorientation appears to be broader. For example, they may introduce new frames that include not only different individuals or a different spatio-temporal setting, but also a different scenario within an elaborate argument or debate. In this sense priming acts in Herodotus and Thucydides are closer to priming acts in drama. [175] The crucial point of commonality of all priming acts is projection: they foreshadow the unfolding of several elements, not just one. In other words, they introduce moves. They do so by constituting either the first act of a move, or a hinge act across two moves. Both Herodotus and Thucydides use them remarkably often. [176] A further characteristic is their shortness, which makes me compare them to kómmata, or short kôla. [177] The particles primarily involved in this pattern are γάρ, καί, and δέ. [178]

§108. Let us start with γάρ. [179] In genres where narratives are predominant (epic and historiography, within our corpus), γάρ can mark discourse discontinuity in several ways. γάρ segments can embed separate narratives, offer syntactically parenthetical thoughts, supply “external” elements that help comprehension, insert metanarrative comments, and expand on a piece of information previously mentioned. However short a γάρ segment may be, it always indicates a move boundary, because of the discontinuity it expresses. [180] In cognitive terms, γάρ lends itself particularly well to the marking of unframed discourse. This is suggested in II.4.2.1: for Homer and Pindar, instances of unframed discourse are storyworld information external to the frames of the ongoing narrative; the main narrator’s intrusions into a character’s feelings; and gnomic assessments contributing general views to particular events. The fact that parentheses in modern editions of Herodotus and Thucydides frequently bracket γάρ clauses is a further sign that the particle is compatible with discourse discontinuity. A linguistic feature that often accompanies γάρ moves is a tense shift; [181] especially at the beginning of “digressions,” [182] these shifts encode temporal discontinuity.

§109. In a short monograph on γάρ in Herodotus (De γάρ particulae usu Herodoteo, 1882) Broschmann remarks that γάρ may be used to “induce something new” (42), be it a new character, a new place, or a new event that is going to constitute the cause of a subsequent event. Broschmann’s idea of new content is consistent with our view of unframed discourse and discontinuity. In several cases the sense of inducing something new conforms to the pragmatic goal of expanding, which in our terms means projecting multiple pieces of information. γάρ acts that introduce embedded narratives in both Histories can be full γάρ clauses. [183] In Herodotus ἦν γάρ clauses typically illustrate this function. [184] Short acts with γάρ simply amplify the discontinuity effect, because of their strong forward-oriented function.

§110. The form of a short act comprising γάρ (or other eligible particle) derives from the fact that the following segment begins in a manner consistent with act starts, typically a constituent followed by a postpositive particle. In the following example, μέν is the signal that marks the division after Σικελίας γάρ; other postpositive particles, and elements such as a negation or a subordinating conjunction, can initiate an act as well. [185] The long geo-political excursus on Sicily at the beginning of book 6 (6.1.2-5.3) starts with the short segment Σικελίας γάρ. [187] The oblique case of the noun “Sicily” confers an almost accidental character to the piece of information. Although morphology suggests a weak function, Σικελίας γάρ in fact has an important strategic and cognitive value: it primes a geopolitical reference to Sicily, and it foreshadows an expansion on it. [188] In this case the OCT paralinguistic segmentation (the full stop, number 2, the extra space between words) and my discourse segmentation match.

§111. καί represents the shortest possible of the projecting acts, as the following examples show. This passage includes the peak and closure of the narrative about the Egestans’ trick (6.46.3-5). [189] The Egestans were at war with the city of Selinous, and they had claimed that they had enough money to support the Athenians in their travel to Sicily to aid them against Selinous (6.6). In the second part of chapter 46, when the Athenians discover that no funds are forthcoming from Egesta, Thucydides retrospectively (οἱ δὲ Ἐγεσταῖοι … ἐξετεχνήσαντο τότε, 46.3) narrates the trick that the Egestans had enacted in order to pretend to be wealthy. They had deceived the Athenian envoys by showing them many precious objects and offering dinner parties. At that moment the historian uses καί to introduce the turning point: the same envoys belatedly realize the truth, and the soldiers blame them. καί often introduces narrative peaks, especially in Thucydides. [190] Here, however, what is more relevant is the pattern of the short act: at the beginning of section 5, καί is followed by several acts (οἱ μὲν ...; ἐπειδὴ …; ὅτι οὐκ …; οἱ δὲ …) which detail the unpleasant aftermath of a pleasant experience. The very same pattern occurs a few words later: || καὶ | Νικίου μέν helps us process the discourse as a general addition that introduces details on the situation (see καί, which starts a move) followed by a first element of the move (μέν accompanying the mention of Nicias’ opinion). My translation attempts to render the force of these καί by means of enriched “ands,” and by the colon (“And so:,” “And in particular:”). Note that also in these two cases full stops, section number (46.5), extra spaces, and move starts coincide.

§112. A parallel example from Herodotus is of pragmatic as well as philological interest: || Καὶ | τὴν μὲν starts chapter 17—move start and chapter start overlap. Despite the fact that all manuscripts give καί, Reiske emended to κῃ, probably by analogy with the subsequent occurrence of ὧδέ κῃ. Perhaps the role of “and” at the beginning of a new chapter was not easy to explain. Lloyd (1989:247) remarks that “also” would be the appropriate translation, given that Herodotus here shares his opinion as he does elsewhere. Such an interpretation arises from the perception that καί in this case does not link conjuncts. I offer a pragmatic reading as an alternative. Τhe καί in question works as an “and” pragmatically enriched—“and so,” or “and now,”—which is intonationally isolated and stressed. The sense of continuity is expressed in relation to a relatively large portion of preceding discourse, not just to the preceding sentence. In fact, the Ionians’ opinion occupies the entirety of chapters 15 and 16. The occurrence of μέν after καί (καί τὴν μὲν Ἰώνων) suggests an act boundary after καί. As one would expect, what begins in 17.1 is an account articulated in more than one segment and more than one move—see || Αἴγυπτον μὲν; in fact it continues for the rest of the chapter (17.1-17.6). Therefore, καί works as a short and separate act introducing a multi-act unit of discourse.

§113. The final two short acts have δέ in common. [191] The following examples show three aspects of priming acts particularly well. Following the usual pattern, τεκμήριον δέ is a short act separated from a segment containing γάρ, and consequently projects a multi-act sequence (15.4 to 15.6) consistent with a move. The sense of projection, however, additionally derives from the semantics of τεκμήριον, which in Thucydides has a methodological meaning as well: specifying evidence of a certain fact is a crucial component of the scientific inquiry by the historian. The first aspect, then, is the idea that priming acts project a move by means of all their components, syntactic, semantic, and pragmatic. [192]

§114. The second aspect is that priming acts are sometimes hinges between moves. In (t18) τεκμήριον δέ can be said to start a move but also to precede a γάρ move (note the Greek semicolon after δέ, which suggests a relatively strong boundary). Thus τεκμήριον δέ represents a hinge between the historical information conveyed through the previous acts, and the elucidation of the evidence supporting that information.

§115. The third aspect has to do with the reorientation of attention at narrative turning points. The next passage occurs at the climax of the Prexaspes episode (Herodotus 3.74-75). [193] The Magi have made Prexaspes their friend, and instruct him in the official speech he has to deliver in order to dispel suspicion from the Magi. In a stroke of narrative genius Herodotus reports this plan without describing Prexaspes’ point of view at all, thereby allowing the tension of the moments preceding Prexaspes’ delivery of his speech to have its full impact on the reader or listener. Then, when the time comes, Herodotus writes: ὁ δέ reorients both the historian’s and the audience’s attention so that it falls on Prexaspes, who has been silent thus far. The short act also begins a move that gradually unfolds everything in an elaborate speech, and reveals the truth in total contradiction to the Magi’s expectations. This time there is a mismatch between moves and modern segmentation in sections and chapters. Ὁ δὲ || τῶν μὲν ... occurs between the beginning of chapter 75, about the Magi’s order for Prexaspes to reach the top of the tower, and the beginning of section 2, about the start of Prexaspes’ report. However, an extra space before Ὁ δὲ in the OCT edition does match my point about the move start.

§116. To sum up: all the examples I have given illustrate the same pattern: in certain co-texts particles indicate the occurrence of short acts. Content suggests a tight relation between these short acts and the multi-element units that follow. Except for the minimal case of short acts with καί, both the particle and the co-occurring word(s) foreshadow or project the subsequent elements. In the examined texts these co-occurring words can be pronouns (τοῦ, ὁ), names (Σικελίας), and nouns (τεκμήριον). This phenomenon is no less widespread in Thucydides than in Herodotus. Rather than judging it as a sign of spokenness from the perspective of syntactic incompleteness, we may regard it as a cognitively efficient means to guide the reader or hearer through the narrative. [194]

§117. An equally widespread phenomenon that is relevant to move boundaries concerns some usages of οὗτος in all its possible forms—adjective, pronoun, adverb. Herodotus provides copious evidence of that. Let us point out the context of this discourse strategy, first.

§118. Herodotus’ style includes the ability to dovetail discrete pieces of narration by balancing given with new information. [195] Cooper (2002:2705) summarizes the various devices that the historian uses to form a bridge with given information by saying: “endless new combinations of repeated proper noun, resumptive particle, resumptive participle, and demonstrative pronoun are taken up and recombined by our author.” [196] This kind of dovetailing technique facilitates the processing of discourse: the reader or listener is taken by the hand in her efforts to keep track of relevant actions and participants as the narrative progresses. Perhaps this is one of the features that famously led Cicero to compare Herodotus’ style to a calm river: sine ullis salebris quasi sedatus amnis fluit, “[he] flows like a calm river without any rapids,” Orator 12.39; such is the flow of his discourse.

§119. In this section I will shed light on how demonstrative pronouns and “resumptive particles,” to use Cooper’s terms, contribute to the calm flow of discourse in Herodotus. I will also assess their presence in Thucydides. Even though the phenomenon is less notable there, my analysis offers some partial counterargument to longstanding judgments concerning the asperity of Thucydides’ style, in particular the ancient opinion according to which his writing was δυσπαρακολούθητος, “hard to follow.” [197]

§120. Recently, a group of Spanish scholars has begun to focus on the discourse functions of the adverb οὕτως in different genres (prose and poetry). [198] Ruiz Yamuza in particular explores the consecutive usages of οὕτως. When οὕτως appears at the end of argument or fables, the idea of consequence relates to the conclusion(s) to be drawn; for those cases the author coins the term “epimítico,” “epimythic” (2012:228). [199] I take over the term and apply it to Herodotean and Thucydidean οὕτως and other οὗτος forms in acts or moves that wrap up historiographical accounts. “Epimythic” captures the metanrrative function of those forms: they are signposts for the conclusion(s) to be drawn once accounts are over. [200]

§121. This important role is played not only by the adverb οὕτως, but also by cognate pronominal and adjectival forms, especially the plural neuter form ταῦτα and the singular masculine nominative οὗτος. [201] The following examples show that different οὗτος forms can contribute (together with other semantic and pragmatic features) an epimythic function by signaling the conclusion of accounts of different size. In this passage the grammatical subjects involved are the Acarnanians and the Amphilochians, who waged war with the Ambraciots. The referents of ταῦτα are the terms of the truce agreed upon by both parties. Since the account of these terms occupies several narrative acts (114.3-4), ταῦτα signals its end, together with the rest of the act’s semantic content (ξυνθέμενοι, διέλυσαν, τὸν πόλεμον). [202] The act betweent the two extra spaces in the OCT edition (ταῦτα ξυνθέμενοι διέλυσαν τὸν πόλεμον) can be considered a separate move closing a multi-move section.

§122. The referent of the subsequent ταῦτα (μετὰ δὲ ταῦτα) is the whole of the events concerning the war between these parties. As frequently happens in historiography, μετὰ ταῦτα (with or without δέ [203] ) is a hinge phrase indicating that what follows is a new account, which implies a resetting of the temporal frame at least (with individuals and place unaltered). [204] In the light of that, ταῦτα works as a dovetailing device looking backward to a discrete group of events, as well as forward to the upcoming change of frame. As such it can be considered to open a new move, which harmonizes with the location of section number [4] in the OCT edition. Let us examine a similar case in Herodotus: In this passage Herodotus relates how Xerxes, finally persuaded by the oracles of Onomacritus, and by the advice of the Peisistratids and the Aleuades, decides to undertake the invasion of Greece. The king is the grammatical subject of the clauses preceding (t21), so there is no need to retrieve the referent with a demonstrative, especially in the nominative case, because the referent is already in focus. [205] What can justify, then, the use of a οὗτος form from a discourse perspective? The scene occurs at the end of chapter 6 in book 7. This means that at some point in the history of the segmentation of Herodotus’ Histories editors felt that these clauses determined a chapter boundary. The boundary is conveyed at the linguistic level by οὗτός τε δή. Besides οὗτος, τε can be connected to the two historians’ use of τε to achieve a coda-like effect in their texts. [206] In addition, δή regularly co-occurs with an epimythic οὗτος form (usually οὕτω δή and οὕτως … δή) to close a major narrative step. [207]

§123. Not only can οὕτως fulfill an epimythic role; it can also trigger the start of the subsequent move. Here is an example: The speaker is the Athenian envoy Euphemus, who is trying to convince the Sicilians of Camarina to secede from Syracuse and join the Athenians (6.82-87). Let us focus on ἔχει δὲ καὶ οὕτως. οὕτως here does not look backward to a referent in the preceding text, but rather forward, to the entire argument constituted by the discourse acts that will follow. In other words οὕτως signals that a new move is about to occur, which will expand (note γάρ at its beginning [208] ) on the point that Euphemus has just concisely made (“The Ionians from immemorial time are enemies of the Dorians”). Therefore οὕτως is a proleptic device. This function harmonizes with the idea that here ἔχει δὲ καὶ οὕτως is a hinge between the preceding and the following move. The extra space that precedes ἔχει (note that it does not coincide with any number) may reflect the discourse boundary occurring before the short hinge clause.

§124. Herodotus uses the masculine pronoun οὗτος in a similar way, that is, forward-oriented. Sometimes an individual who is going to play a major role in subsequent acts is recalled at the beginning by means of οὗτος, even when its referent, the proper name, appears as well. During the account of the conspiracy of the seven Magi, Herodotus tells us that Otanes was the first to suspect that the “Smerdis” on the Persian throne was not the son of Cyrus. At that point he explains who Otanes was—note the priming act Ὀτάνης ἦν, preceding the acts that start with Φαρνάσπεω μὲν παῖς and γένεϊ δὲ. By means of οὗτος ὁ Ὀτάνης the historian begins a relatively long move (68.1-69.3) entirely devoted to Otanes’ actions. [209] οὗτος, far from necessary from the perspective of anaphora resolution, becomes a device for projecting the topical salience of the subsequent series of acts. [210] οὗτος opens a move whose point is to focus on Otanes. [211]

§125. Let us now briefly focus on a pattern that involves constructions with epimythic οὗτος forms accompanied by μέν. This pattern serves to round off multi-move units (that is, macro-moves). The most recurrent construction is ταῦτα μέν. [212] The account of the facts about Pylos occupies several chapters (4.2-6; 8-23; 26-41), alternating with episodes that occur elsewhere. The act, then, also constitutes a move that rounds off a quite extendend multi-move unit. The extra space before ταῦτα μέν matches, once again, my point about this being a move.

§126. Further οὗτος forms are employed with μέν as well. [213] An exception is οὕτως, which usually does not co-occur with μέν. [214] The occurrence of οὗτος forms + μέν attests to the hinge function of acts between moves: while the demonstrative component looks back at previous discourse, μέν projects “more to come.”

§127. Let us now turn to οὗτος forms accompanied by δή. This construction is also to be found at move transitions, and it occurs especially in Herodotus, where δή is exceptionally frequent. [215] In all these cases δή occurs in act-peninitial position. In this example οὕτω δή works as an epimythic closure device, to mark the end of a story. [216] Note the extra space before the οὕτω δή move.

§128. Further οὗτος forms are used in combination with δή to achieve similar communicative goals, for example to summarize a point, as in this case: “For these reasons” + δή is an example of Herodotean dovetailing. Actually the construction summarizes the reasons already given, but at the same time prefaces the narration of the subsequent action. In other words, each act that signals a move end simultaneously represents a new step in the discourse. Paralinguistic segmentation in this case acknowledges the major discontinuity; τούτων δή μιν εἵνεκεν starts section 2 of chapter 74.

§129. We can also interpret the function of δή without οὗτος forms, whenever it occurs in act-peninitial position and/or near temporal expressions. [217] The general discourse function we assign to act-peninitial δή in Herodotus is that of marking major steps in narrative progression—just as in Homer. [218] Already Apollonius Dyscolus associates δή to the sense of temporary conclusion, before starting something else. [219] Let us exemplify the role of δή at move transitions by considering how Herodotus marks the main steps in the (reported) narration of the encounter between the Nasamonians and the Pygmies: This excerpt fixes in our visual imagination two main events, the Nasamonians by the trees picking fruit, and then the Pygmies carrying away the Nasamonians. Both steps in Greek include δή and other significant co-occurring features. First, we find the construction δή κοτε, which recalls ὅτε δή in Homer. [220] Second, τε δή notably occurs after an unframed δέ act, the unframed information concerning the two people’s mutual inability to understand each other. [221] τε δή retrieves the main frame of the account, and signals its end at the same time. [222] While the extra space and the start of section 7 do mark the move that starts with Ἄγειν τε δή, in the case of ἰδεῖν δή κοτε nothing on the paralinguistic level indicates any boundary (not even a modern comma). Yet, my interpretation of the moves that are relevant to major visual shifts springs from the two δή moments. All in all, unlike οὗτος forms + δή, which usually mark a move at the end of an account, δή without demonstrative pronouns can be found also at intermediate transitions of accounts. The most suitable combination fulfilling exactly this function is μὲν δή.

§130. Van Otterlo, H. Fränkel, and Immerwahr [223] are among the scholars who note that μέν and δέ can straddle two different accounts or themes, and mention μὲν δή and μέν νυν as variations of μέν in this function. [224] More specifically, they observe that μέν, μὲν δή, and μέν νυν are often used to close or recapitulate a series of events, while δέ starts a new account or theme. In our terms this equals to saying that μὲν δή and μέν νυν often close macro-moves.

§131. Let me anticipate the main points of this subsection. μὲν δή and μέν νυν occur almost exclusively in Herodotus. [225] I take them to mark moves that end or start macro-moves. The fact that in print editions about one third of μὲν δή and μέν νυν moves either conclude or start chapters [226] matches with this idea. [227] The oscillation between end and start can be explained if we consider the discourse function of the same combinations while macro-moves are unfolding, that is, when they occur in the middle of a complex narrative (or speech). The μέν component in this respect is revealing, because of the metanarrative idea “more to come” underlying its use.

§132. Let us start the analyses with instances of μὲν δή and μέν νυν moves that summarize the events of sizable preceding discourse. The μὲν δή and μέν νυν acts in the respective passages are discontinuous with respect to what precedes and follows, so they constitute moves. [228] Their role is fundamentally metanarrative; note that in both passages the subsequent δέ acts make metanarration explicit (“From this point on my discourse inquires …,” and “I am going to prolong the discourse about Egypt …” respectively). In the case of (t28), the μὲν δή move concludes the account of the wars between Lydians and Persians, which starts at 1.46, and possibly even the entire Lydian account, which starts 46 chapters earlier (at 1.6). In the case of (t29) the μέν νυν move concludes the discussion of the Nile, which starts 24 chapters earlier (at 2.10). Note, by the way, the occurrence of τοσαῦτα in (t29), a variant of a οὗτος form. (t28) and (t29) close chapters in both Hude’s (OCT) and Legrand’s (TLG) editions. Also, in both editions full stops separate Λυδοὶ μὲν δή from Ἐπιδίζηται δέ. Additionally, the extra space before Λυδοὶ μὲν δή, which is unrelated to section numbers, paralleled by a rare indent in the TLG edition, match my reading of a move boundary. These moves frequently include οὗτος forms as well, and are usually followed by a δέ act whose content is totally different (as in (t28) and (t29). [229]

§133. Sometimes, in spite of the fact that modern segmentation at least partially acknowledges move boundaries, [230] modern punctuation obscures the closural force of μὲν δή and μέν νυν moves. Here are two examples showing that a simple comma does not match the subject-closing function; a full stop would fit better the major discontinuity in the discourse. μὲν δή is a metanarrative sign that the encounter between Croesus and Adrastus is over, for the moment at least—Croesus has met him, talked with him, and decided to host him (1.35-36). What follows is a new frame (note the tense shift in γίνεται), the beginning of the crucial account of the boar hunt, which will cause Adrastus to accidentally kill Croesus’ son. The comma of print editions does not do justice to this discontinuity. [231]

§134. Similarly, in the following passage modern punctuation and segmentation only partially represent the frame shifts that occur after μέν νυν: Both the Greek edition and the translation capture paralinguistically one of the major discontinuities in the discourse: the new topic begun at ἡ δὲ στρατηλασίη ἡ βασιλέος. The new chapter number (138) in the Greek edition corresponds to the indentation in de Sélincourt and Marincola. However, there is another discontinuity besides this one. Prior to the chapter-break, features that co-occur with the first δέ signal a new move. The first-person ἐπάνειμι is a performative verb (note the tense shift) that explicitly encodes and projects the historian’s narrative intervention. τὸν πρότερον λόγον (“the previous account”) clarifies the nature of this intervention, which is to resume the preceding, interrupted narration of the invasion of Greece. [232] Overall, this first δέ’s host clause constitutes a metanarrative move. It is likely that modern editions print a comma before ἐπάνειμι δὲ because editors construe μέν – δέ as a pair, which in Attic Greek can occur within the same thematic strand. But μέν νυν has a force different from μέν when it occurs alone: while single μέν is more flexible in function, μέν νυν always marks moves or even macro-moves. It is remarkable that in the OCT edition a very rare indentation occurs before ταῦτα μέν νυν, that is, an indentation that does not coincide with chapter starts. [233] Such a paralinguistic sign does fit the reading of a move boundary. The second major discontinuity occurs, as I have noted, at the chapter-break (with no indentation, by the way), where Herodotus shifts from metanarrative to actual narration; this second δέ act marks the beginning of a new macro-move.

§135. So far I have illustrated the appearance of μὲν δή and μέν νυν in moves that round off large units of discourse, and I have commented on two examples that show a mismatch between modern punctuation and move boundaries. Now two more examples follow, which illustrate mismatches directly on the level of chapter division. The discussion is central to the ambiguity of interpretation between ends and starts of relatively big discourse units.

§136. Let us consider the wording of a short move that in the TLG edition appears at the end of a chapter, whereas in the OCT edition it starts a new chapter: Legrand (TLG edition) places the utterance οὗτοι μὲν δὴ ἐν οὐδενὶ λόγῳ ἀπώλοντο at the end of chapter 69, whereas Hude (i.e. OCT edition, reported in (t32)) places it at the beginning of chapter 70. [234] The co-occurring elements that signal a major discourse discontinuity are the combination of a οὗτος form and μὲν δή. The comparison with other occurrences of οὗτος form and μὲν δή makes me align with Legrand rather than with Hude. Perhaps Hude did not want to separate μέν from δέ; however, in this case form and content clearly indicate a straddling of macro-moves, thematically and pragmatically.

§137. Legrand and Hude do agree on chapter segmentation in the following passage, but nevertheless the interpretation of an utterance starting a discourse unit can be challenged. The “first of the evil acts” is Cambyses’ order to kill his own brother Smerdis (3.30), whose murder has just been described. The οὗτος form τοῦτο in (t33) strengthens the backward-oriented discourse function of the act. In fact, the entire move resumes the preceding wording almost verbatim: 3.30 began with καὶ πρῶτα μὲν [τῶν κακῶν] ἐξεργάσατο τὸν ἀδελφεὸν Σμέρδιν, “First of all, he destroyed his brother Smerdis ….” πρῶτον μὲν δὴ λέγουσι Καμβύσῃ τῶν κακῶν ἄρξαι τοῦτο is therefore a standard case of a move that summarizes previous content. However, the same move also recalls the discourse structure of a list (“First …; second …”). The semantic clue “first … second” influences the discourse segmentation so that the first evil occupies the first spot instead of the last in the organization of chapters. In other words, there are reasons to think of the move in question as concluding but at the same time starting a multi-act discourse unit. How can we make sense of this apparent variability in positioning?

§138. The answer lies in the appreciation of intermediate or hinge narrative steps. In (t32) οὗτοι μὲν δὴ ἐν οὐδενὶ λόγῳ ἀπώλοντο (“So these [Megarians and Phliasians] perished without anyone talking about it”) in fact represents a step out of the account. It resembles an aside, and most likely a non-neutral one. [236] Let us consider now a similar μέν νυν move that occurs at an intermediate point within a long account: This passage occurs in the middle of Herodotus’ account of the cult of Heracles in Egypt (2.42.3-45.3). We have here a straightforward metanarrative move: the historian explicitly tells us that he is stepping out of the host account. μέν νυν signals such a move. [237]

§139. These last two examples show a key feature of these combinations. When the host moves conclude or start macro-moves, as well as when they signal different moments within the same account, μὲν δή and μέν νυν convey two important metanarrative meanings at the same time. One is that the speaker is either stepping out of the narration to summarize the events described, or is stepping out to comment on some general aspect of the narrative. [238] Such overviews are mainly realized a posteriori—they follow the detailed account—but every now and then can also be found a priori. The other metanarrative meaning is that this stepping out is, on a larger scale, intermediate. μὲν δή and μέν νυν acts never coincide with the absolute final part of accounts. They do not represent the definitive end, but they do prepare the way for it. [239] There is always either some further step that brings the account to a decisive close, or an entirely new account that begins immediately afterwards.

§140. The μέν component in these combinations gives a projecting function to these moves. It suggests that the speaker will go on with his discourse. [240] Earlier (§66) I associated this kind of μέν to modern nonfinal intonation at interim conclusions. As II.2 §52 reminds us, already Hoogeveen in the 18th century was pointing out μέν μεταβατικός (“transitional”), and even before him Devarius in the 16th century discussed μέν περιγραφικός, “summarizing.”

§141. Because of their intermediate role, these segments possess an instability that can be witnessed in divergences among editorial decisions. To resolve the ambiguity, I suggest that we consider these intermediate steps as hinge moves. They differ both from what precedes and what follows them, and they reveal a discourse strategy affecting multiple acts (backward and forward), not just the local one that hosts the μὲν δή or μέν νυν. [241]

§142. Not infrequently μὲν δή and μέν νυν appear immediately after the report of direct speech. [242] In these cases too the two combinations indicate that the speaker is stepping out of the previous discourse (the words uttered by another speaker) in order to proceed with the narration in the third person. On other functionally analogous occasions, μὲν δή and μέν νυν signal what in II.4 §18 is presented as “frame recall:” [243] the moves hosting them retrieve a discourse thread, and the cognitive frame associated with it, after some interruption. [244]

§143. The last remarks concern slight differences in usage between the two combinations. In general, both μὲν δή and μέν νυν work as a cluster (that is, the forces of the two individual particles are combined), but they can also work as a combination keeping the forces of the two particles separate. The latter emerges when the δέ act that follows centers on an element of content that pairs with the element attached to the μὲν δή or μέν νυν segment. [245] However, while in Herodotus δή alone is not rare, νυν occurs without μέν only 14 times (against 310 μέν νυν). [246] From this datum I infer that μέν νυν works as a cluster more frequently than μὲν δή. Another slight difference in use is that μὲν δή can occur in acts that describe exciting or vivid moments of accounts, while μέν νυν does not. [247]

§144. The combination μὲν οὖν appears more frequently in Thucydides than in Herodotus (85 occurrences against 7). LSJ (s.v. μέν) and Montanari (1329) take μέν νυν to be the Ionic variant of μὲν οὖν. The two combinations are similar in discourse function, from the usages I am going to comment. [248] Like μέν νυν, we find μὲν οὖν at closures together with οὗτος or τοιoῦτος/τοσαῦτος forms, immediately after the end of direct speech, and at the beginning of new accounts or at hinge points. [249] Unlike μέν νυν, in the TLG edition Thucydidean μὲν οὖν appears more frequently at the beginning of chapters and sections than at the end. [250]

§145. In order to illustrate the discourse functions of Thucydidean μὲν οὖν across different co-texts, let us look at four instances of μὲν οὖν within the same account, drawn from the narrative of the plague (2.47.3-54). Each of the μὲν οὖν segments occurs after a new section number, which reveals the editorial perception of a major discontinuity in discourse. In (t35) μὲν οὖν signals the beginning of a move introducing the opinion of doctors and individuals. The author for a moment steps back from the description of events: “Now, let us spend some words on how each doctor as well as individual knew about it [the illness] …” In (t36) μὲν οὖν signals a move that pulls back from the details, even when they seem important (like the information about dogs), and offers an overarching and recapitulating assessment: “As for the illness, then, … this was the general idea on the whole” (note the generic terms γιγνόμενον, ἐπὶ πᾶν). This use resembles several μέν νυν in Herodotus; τοιοῦτον echoes οὗτος forms. In (t37) μὲν οὖν indicates that a multi-act unit is ahead, which will deal with a new subject (ἐγένετο μὲν οὖν ἔρις), and which is related to the verse just reported. Therefore, μὲν οὖν in (t37) resembles that in (t35). In (t38) μὲν οὖν works as a hinge in a particularly clear way: it marks the ongoing move as resulting from the preceding account, and at the same time it introduces the notion of a comprehensive agreement between the events and the oracle (see the subsequent “The events, then, looked fitting the oracle”).

§146. These instances of μὲν οὖν, then, have the same metanarrative meanings I identified for μέν νυν in Herodotus: μὲν οὖν, like μέν νυν, marks a stepping out from the preceding flow of narration, and it indicates an intermediate assessment or its beginning. [252]

§147. This chapter developed out of the following basic questions: how do particles contribute to the segmentation of discourse in Herodotus and Thucydides? What is their relationship to ‘posthumous’ punctuation, and to section/chapter divisions? What is their relationship to discourse boundaries, such as syntactic boundaries and prose-kôla boundaries?

§148. The initial sections consider main segmentation cues separately. The starting point is modern punctuation and the duality often discussed in literature, that is, the “grammatical” and the “rhetorical/elocutionary” functions of punctuation marks. Next comes the attention of modern scholars for periodic style and hypotactical articulation of texts. A further crucial aspect is represented by pre-print modalities of punctuating and ancient descriptions of textual segments, which reveal a fundamental focus on delivery, and leave syntactic definitions unspecified. Then the focus shifts to the retrieval of prose colometry in modern times. The final cues concern the paralinguistic segmentation above the sentence level that we find in OCT editions.

§149. A discussion of matches and mismatches concerning particles in relation to these segmentation cues follows. Particles work as road signs (a metaphor used for punctuation); they may provide directions about single acts as well as to multi-act units. The consideration of the deep interlacing between prosody and grammar, the complex inheritance of pre-print paralinguistic signs, and the richness of particles in frequency and lexical variation make us opt for a holistic principle for segmentation. This principle takes into account syntax, semantics, and pragmatics. [253]

§150. The textual analyses illustrate the adoption of this principle. The two fundamental elements of discourse segmentation are taken to be acts and moves. Acts and moves are strategic communicative steps that show a different degree of discontinuity with what precedes and follows, and range in size from isolated phrases (priming acts) to macro-moves (expected by genre conventions). The general criteria used to identify act boundaries resemble those used to identify prose kôla. However, no content types of acts or moves are introduced. These segments are defined exclusively by the actual linguistic features that make them recognizable. So, patterns of co-occuring features lead us to identify, for instance, a γάρ move that in context X performs the communicative action Y; we do not deduce from γάρ that its host clause is a causal clause.

§151. Let me outline three general results emerging from the analyses. The first is that particles guide us to process discourse discontinuities on multiple levels. More specifically, particles, particle combinations, and other co-occurring linguistic features (such as οὗτος forms) signal discourse segmentation in a more authentic, more reliable, and more precise way than any kind of paralinguistic segmentation, old or recent.

§152. Sometimes the two communicative devices align—the paralinguistic one inspired by the linguistic one—but many other times there are mismatches. Mismatches should reorient our reading and make us privilege the linguistic over the (modern) paralinguistic signs. In fact, readers tend to infer the function of particles from (modern) paralinguistic segmentation, while actually it should be vice versa: the ad hoc function of particles should be stressed by means of ad hoc paralinguistic segmentation. For example, a modern comma that separates a μέν from a δέ construction might be totally inappropriate. If what co-occurs with that μέν and that δέ indicates a major discourse discontinuity, a full stop and an indent should be established between them. An exceptional match is constituted by the extra space between words that we find in the OCT editions, whose appearance seems to be especially guided by the presence of particles.

§153. The second general result is about two phenomena that particles reveal in the perspective of discourse segmentation, which no syntactic segmentation can show. The first relates to the idea of discourse hierarchy. Particles in priming acts that start moves have a superordinate role with respect to other particles that occur in subsequent acts. This superordinate role emerges from the fact that the priming acts pragmatically and semantically bind multiple segments that follow, in which other postpositive particles may occur. Such an organization of particle order reflects a discourse hierarchy. [254]

§154. The second phenomenon is the hinge function that especially particle combinations suggest in short moves between moves. What enables μὲν δή, μέν νυν moves (in Herodotus), and μὲν οὖν moves in Thucydides to work at the end of accounts but also at the beginning, and in the middle, is the following: most of the time the utterances that host those combinations represent intermediate closures either before a more comprehensive closure or before a parallel or contrasting move. For instance, they supply intermediate summaries, or they recall frames, or offer commentary on the ongoing narrative. They work as hinge moves, which qualify as stepping out of the preceding discourse, while foreshadowing further progress.

§155. The third and final result concerns parallel and distinct usages—inferable from the passages under investigation—in a general comparison between the two historians’ language. The parallel side is that acts in both authors emerge as comparably short segments. My analysis dismantles the idea that the size of periods in Thucydides is inherent in the historian’s language use.

§156. The distinct side is related to moves. Herodotus employs οὗτος forms and combinations such as μὲν δή to signal move boundaries more generously than Thucydides. I believe this to be part of Herodotus’ dovetailing technique, [255] which makes his flow of discourse as fluid as a river. [256]

§157. IV.5 complements these findings by merging the interpretation of acts and moves in situ, by drawing connections with other kinds of non-syntactic articulation (such as lexical resonance and sound effects), and by letting observations on coordination strategies (from IV.2) and on stance (from IV.4) converge in a unified discourse analysis.