Variation versus Change Clausal Clitics between Homer and Herodotus

Enclitic distribution in Greek (and archaic Indo-European generally) is governed by a set of generalizations known as wackernagel’s law, according to which enclitics occur in “second position.” As has long been known, surface exceptions to Wackernagel’s Law in Homer are uncommon, but in Herodotus are far more frequent. Wackernagel himself attributed this difference to syntactic change: in Homer a single mechanism is responsible for second-position clitic distribution, while in Herodotus the old second-position rule competes with new placement rules. Although the nature of theseinnovativemechanismshasneverbeenexplicated,philologistshaveadoptedthis view with apparent unanimity. The central claim of this paper is that the alleged syntactic change is an illusion. What Wackernagel and others have observed in Homer and Herodotus is a difference in usage , not grammar. Specifically, Herodotus uses constructions that yield non-canonical surface patterns (i.e., the clitic is not “second” in its clause) more often than Homer. As the same generalizations capture the distribution of clitics in both Homer and Herodotus, there is no validity to the claim that Wackernagel’s Law is weaker in the classical period than in the archaic, or that there are new distributional rules at work.


Introduction
wackerangel's law (wl) is a generalization about the surface position of enclitics (Wackernagel 1892, Hale 1987a, 1987b, Goldstein 2010, 2014, 2016a, which predicts that they occur in "second" position (2p). The following examples illustrate canonical second-position behavior ('-' marks the host-clitic relationship; the relevant clitics are in bold):1 (1) Canonical 2p distribution i. Κροῖσος-δέ-μιν ἐκάθηρε. 'Croesus purified him.' (Hdt. 1.35.2) ii. (ἀπὸ ταύτης) ω -γάρ-σφι τῆς μάχης, Ἀθηναίων θυσίας ἀναγόντων ἐς τὰς πανηγύριας τὰς ἐν τῆισι πεντετηρίσι γινομένας, κατεύχεται ὁ κῆρυξ ὁ Ἀθηναῖος ἅμα τε Ἀθηναίοισι λέγων γίνεσθαι τὰ ἀγαθὰ καὶ Πλαταιεῦσι. 'Since this battle, the Athenian herald prays that good things befall the Athenians and Plataeans together, when the Athenians conduct their sacrifices at the festivals that occur every four years. ' (Hdt. 6.111.2) In example (1i), both the discourse particle δέ and the pronominal enclitic μιν are hosted by the first word of the clause, Κροῖσος. Example (1ii) shows that "second position" has to include reference to prosodic constituents. The clitics γάρ and σφι are hosted not by the first morphosyntactic word of the clause, which is the preposition ἀπό. If we assume that ἀπό is proclitic here, then it forms a prosodic word (indicated with a subscript ω) with ταύτης, which in turn hosts γάρ and σφι. As robust as the pattern in (1) is, it is not without its counterexamples: (2) Noncanonical Distribution (ἀπὸ Βαβυλῶνος) ω -δὲ καὶ τῆς λοιπῆς Ἀσσυρίης χίλιά-οἱ προσήιε τάλαντα ἀργυρίου καὶ παῖδες ἐκτομίαι πεντακόσιοι. 'From Babylon and the rest of Assyria, a thousand talents of silver came to him and five-hundred castrated boys. ' (Hdt. 3.92.2) Here the discourse particle δέ is arguably where we expect it, but the same cannot be said for the pronominal clitic οἱ. It is not hosted by the first prosodic word of the clause, but by the first word after the prepositional phrase ἀπὸ Indo-European Linguistics 4 (2016) 53-97 Βαβυλῶνος καὶ τῆς λοιπῆς Ἀσσυρίης. It has been repeatedly observed that the frequency of counterexamples such as this increases between Homer and Herodotus. This increase in non-canonical examples between the archaic and classical periods has been attributed to syntactic change.
The central claim of this paper is that the generalizations governing the distribution of clausal clitics in Homer and Herodotus do not in fact differ. Despite appearances to the contrary, there is actually no difference in the position of the clitic in examples (1) and (2). In the latter, a prepositional phrase has simply been adjoined to the clause (more specifically, the s constituent). The difference in the rate of canonical second-position behavior between Homer and Herodotus is due to the frequency with which constructions that yield noncanonical surface patterns, such as we have in example (2), are used. In other words, it is due to variation in usage. The difference is not due to a change in grammar (cf. Hale 2007: 38 with regard to Maori passives). No new secondposition rules develop between Homer and Herodotus, and the Homeric mechanisms of clitic distribution are not weaker in Herodotus.
I am not, however, claiming there is no syntactic difference between the grammars of Homer and Herodotus. In fact, Homeric and Herodotean syntax do appear to differ in the structure of the left periphery of the clause. In Herodotus, there is evidence for a focus preposing construction that leads to non-canonical surface patterns such as we have in example (2). My corpus offers no evidence for such a construction in Homer.
The remainder of this paper is organized as follows. Section 2 offers background on the Greek clitic lexicon and Wackernagel's Law. This background serves as the foundation for section 3, which presents the standard claim that new mechanisms of clitic distribution arise in post-Homeric Greek. Section 4 examines in more detail the quantitative data on which this claim is based. Here I argue that when we include authors from the Hellenistic period, the claim of syntactic change is harder to maintain. Section 5 then argues that the distribution of clausal clitics in Homer and Herodotus can be handled with the same distributional generalizations. Four constructions are identified that lead to the non-canonical surface distribution of clitics: wide-scope adverbials, contrastive topicalization, focus preposing, and participial clauses. Section 6 compares the use of the four constructions in Homer and Herodotus, which reveals that the frequency of counterexamples in Herodotus is actually negligible. Section 7 brings the paper to a close with summary remarks as well as a conspectus of new research questions that this study raises. The Appendix contains two sections. Section 8 contains the specifics of what was considered canonical second-position behavior. Section 9 provides a sample of the recalcitrant data.

An Overview of Second-Position Clitics in Greek
The clitic lexicon of Ancient Greek is notoriously large. It comprises pronominal clitics, discourse particles, modal particles, certain verb forms, and a conjunction (cf. Smyth 1956: §181;Dik 1995: 32). Table 1 presents the clitic pronouns attested in Homer and Herodotus (for more detail, see Goldstein 2016a: 6-7). As Table 1 illustrates, the system of pronominal clitics is richer in the singular (the token frequency of singular forms is also higher than that of plural forms). Table 2 reveals that the inventory of non-pronominal clitics is even larger (clitics followed by the name of an author are found only in that corpus; the absence of a designation means that the clitic is found in both Homer and Herodotus).
The heading "Discourse Particle" is deliberately vague, as the function of most of these words is not well understood (cf. Spencer and Luís 2012: 34-36). The boundary between discourse particle and indefinite is not in fact as sharp as Table 2 suggests. The words that bear an accent are standardly classified as "postpositives" in the secondary literature, and not true enclitics. I depart from this practice here because there are no distributional properties that correlate with this distinction.
Since the pioneering work of Hale (1987a) and (1987b) on the distribution of enclitics in Indo-Iranian, it has been clear that there is no single "second position." In fact, at least three distributional categories need to be recognized, as Table 3  Domain membership is apparently determined by semantic scope (cf. Rice 2000 on affix ordering in Athabaskan). Sentential clitics mark intersentential relationships, while clausal clitics realize grammatical features of the clause itself, and phrasal clitics those of sub-clausal phrases (cf. Anderson 2005: 145). I abstract away from the variation that one encounters in the domain-internal ordering of enclitics (on which see Monro 1891: § 365, Ruijgh 1990, Goldstein 2016a. As far as surface distribution is concerned, the crucial difference between sentential clitics and clausal clitics is that the former occur in preposed phrases, whereas the latter do not (see also The bracketed constituents are in each case topicalized phrases (which are discussed in detail in section 5.1.2). The sentential clitic γάρ 'for' occurs inside the topicalized phrase, while the clausal clitic σφι 'for them' is hosted by the first prosodic word thereafter. Situations such as this, in which multiple secondposition clitics do not form a chain, I refer to as splaying. Whether a series of clitics is splayed or contiguous, their typical order in a sentence is as follows: phrasal clitics precede sentential clitics, which in turn precede clausal clitics. This yields the order μὲν-γάρ … σφι above (both μέν and δέ in this example are phrasal and not sentential, as each scopes over its dp). In this paper, I focus specifically on the distribution of clause-domain clitics. There are two reasons for this bias. The first is relevance: the claim of syntactic change between Homer and Herodotus has been based by and large on this class. The second is practical: the distributional properties of clausal clitics tell us more about clause structure than sentential or phrasal clitics.
As already illustrated in example (1ii), clause-domain clitics are hosted by the first prosodic word of their clause.8 This pattern is found in both Homer and Herodotus (example 1ii is repeated here for convenience): In each case a clausal clitic is hosted by a prosodic word that crucially does not form a syntactic constituent. In example (4), the negation οὐ and the pronoun σύ do not form a syntactic constituent, as here the negative scopes over the whole clause, and not just the pronoun. Since οὐ is proclitic here, οὐ σύ does, however, form a prosodic word, which serves as the host of both γε and the modal particle. The same pattern is also found with οὐδέ (e.g., Il. 3.368) and μηδέ (e.g., Hdt. 1.9.1). In example (5), a clausal clitic occurs inside a dp that is the complement of a pp. On the assumption that prepositions can be proclitic, the host is again a prosodic constituent (namely a prosodic word), but not a syntactic constituent. The following tree illustrates the mismatch between prosodic and syntactic constituency in example (5i): Indo-European Linguistics 4 (2016) 53-97 (6) Syntax-prosody mismatch Crucially, only a prosodic node dominates ἐπὶ κόσωι, namely the prosodic-word node ω. It is this prosodic constituent that ἄν selects as its host. The string ἐπὶ κόσωι does not form a syntactic constituent (as illustrated above, this string is not exclusively dominated by a single node), so the host-selection of clausal clitics has to make reference to prosodic constituency. Syntactic constituency is relevant to the extent that it constrains the eligible prosodic words that may serve as licit hosts.

3
The Standard Claim: New 2p Rules Wackernagel (1892) observed that fealty to the second-position "rule" that he was arguing for varied between Homer and Herodotus. Figures 1 and 2 and their accompanying tables compare the raw frequencies of canonical 2p behavior of the modal particle ἄν and the pronominal clitic σε in these two authors.9 Indo-European Linguistics 4 (2016) 53-97  The grey region of the bars represents the number of non-canonical examples, i.e., the number of cases in which the clitic is not hosted by the first prosodic word of its clause. The χ2 test of independence reveals that for both clitics the difference in the frequency of canonical 2p behavior between the two corpora is highly significant (p < .001, df = 1); the effect size is small (φ < .3).10 defined graphically: that is, that it referred to the second orthographic word after a major mark of punctuation. I have not used this definition of second position because it is not linguistically real, i.e., it is not based on linguistic constituents. For a detailed exposition of what I counted as canonical second-position behavior, see section 8 in the Appendix. 10 The p-value represents the probability of obtaining a value at least as extreme as the one  In Homer, there is a single second-position rule, which accounts for the uniform behavior of enclitics in this corpus. In Herodotus, by contrast, this older second-position mechanism is but one of several, as a result of which clitic distribution is less uniform than it is in the archaic period. The subsequent literature on clitic distribution has followed this view with apparent unanimity. Illustrative is the following remark of Dover (1960: 15;cf. 17): "The progressive tendency in post-Homeric Greek [is] to distribute [enclitics] within the clause, instead of concentrating them after the leading [prosodic word]." Howorth (1955: 93); Dunn (1989); Taylor (1990: 30, 131-133); Slings (1992);Fraser (2001: 164-166);and Taylor (2003) express similar views.
Wackernagel's claim suffers from at least four problems. First, it is based on a limited sample, namely Homer and Herodotus. As we will see in this section, when we cast a wider net, we see that the alleged weakening of wl in the classical period is an illusion.
Second, his analysis is incomplete. While he claims that new rules are in operation, he does not specify what the new rules do. Their only distinguishing feature seems to be that they cause enclitics to occur further into the clause than the old rule. In addition, Wackernagel says nothing as to how the alleged set of rules in Herodotus are supposed to interact. If the old rule is still partially active, when should we expect it to be operative and when to be ineffectual? In the end, his account does not make clear predictions, so it is impossible to evaluate it. It is unsettling that his view has commanded such wide assent, given that there is so little to agree or disagree with.11 Third, Wackernagel appears to have been working with a null-view of Greek clause structure, in which no assumptions were made about underlying syn- 11 Fraenkel ([1933] 1964: 94) offers the most explicit view of post-Homeric innovations in clitic distribution (cf. Marshall 1987: 15-16, 121, Devine andStephens 1994: 422). For the modal particle ἄν, he argued that it could occur in canonical second position or second in a Kolon (a prosodic constituent that seems to be roughly equivalent to an intonational phrase; see Goldstein 2010: 16-22 for more) or directly after the verb. While Fraenkel's supplement is certainly an improvement, it too fails to specify when we should expect these various distributional possibilities. As for the idea that clitics are already starting to lose their host promiscuity and select verbs for hosts in the classical period, there is not enough evidence to support this, beyond the cases that are mentioned in, e.g., section 9.2.
Indo-European Linguistics 4 (2016) 53-97 The rise of periphrastic do ( from Kroch 2001: 27;cf. Ellegård 1953) tactic structure. On this view, exceptions to wl amount to any cases in which a clitic is not the second orthographic word after a major mark of punctuation. As we will see below in section 5, if we make assumptions about the structure of the clause and where clitics appear within that structure, there turn out to be far fewer exceptions. And finally, Wackernagel and subsequent scholars have not given enough attention to the distinction between changes in textual output and actual syntactic change. As we will see, the usage patterns of texts from the classical and Hellenistic periods exhibit considerable variation. While this variation can lead to grammatical change, in and of itself it cannot be equated with change.

A Closer Look at the Quantitative Data
Before turning to the Greek data, I present two quantitive profiles that are widely believed to be representative of syntactic change. Figure 3 is a plot graph of the development of do-support in the history of English (the y-axis represents relative frequency).

figure 4 Relative frequency of καί and τε
A much-discussed property of this development is the s-shape of curve characterizing the increase. It is thought that this pattern of gradual increase followed by a sharp rise, which then plateaus, is said to be typical of syntactic change (Kroch 1989a(Kroch , 1989b. For our purposes, what is crucial is that the graph shows a continuous upward trend over time, even though there are dips.
For syntactic loss, we can turn to the history of conjunction in Greek, where we can observe head-initial καί gradually eclipse the postpositive conjunction τε, as illustrated in Figure 4.12 After Homer, the difference in frequency between καί and τε widens. Even though the absolute frequency of the individual conjunctions varies over time, the gap between them steadily increases. Figure 5 visualizes this increase.13 When we turn to the data from second-position clitics in Greek, we find stark differences. Table 6 tallies the number of tokens in which ἄν and σε occur 12 Figure 4 presents counts of the morphemes τε and καί without regard to their function. Thus tokens of adverbial καί and "epic τε" are also included. This does not, however, affect the point being made here. Frequency counts are based on the editions of the Thesaurus linguae graecae (www.stephanus.tlg.uci.edu). 13 One could interpret the loss of τε as a case of lexical change, and not syntactic change per se. While the lexicon is of course affected, I argue in Goldstein (2016b) that the rise of καί is part of a shift toward head-initial syntax in the archaic ie languages generally. Head-initial conjunction also ousts enclitic conjunction in Indo-Iranian, Latin, and Germanic.
Indo-European Linguistics 4 (2016) 53-97 figure 5 Difference in relative frequency of καί and τε after the first prosodic word of their clause from Homer in the 8th (?) century through Chariton in the first or second century ce. In other words, this is the number of canonical examples of wl in each corpus.14 Figure 6 presents this data visually. This graph differs strikingly from those considered earlier in this section. While the rate of canonical 2p behavior does dip between Homer and Herodotus, later corpora do not continue this downward trend. The absence of a clear diachronic trend is the first indication that what we are looking at with Homer and Herodotus is not syntactic change, but rather variation in usage. 14 The numbers for σε come exclusively from forms that have no graphic accent. It is of course possible for enclitics to receive in certain prosodic contexts. When an enclitic is morphologically distinct (e.g., μιν), its prosodic status is clear. In cases like σε, where the tonic and enclitic forms are not morphologically distinct, there are forms with a graphic accent that can be interpreted as enclitic or tonic. On account of this ambiguity, the graphically-accented cases have been excluded from consideration. (Circularity is potentially lurking behind some subset of the data, if textual editors themselves are relying on the surface position of a pronoun to determine whether or not it should bear a graphic accent.) Frequency counts are again based on the editions used by the Thesaurus linguae graecae.

Left-Periphery Constructions in Homer and Herodotus
As mentioned above, Wackernagel was working with a null view of Greek clause structure. That is, any clitic that was not the second orthographic word in a clause was considered an "exception." However, there is ample evidence that second position should not be defined as the second orthographic word, but rather as a combination of domain-and host-selection properties. This section argues that if we make use of syntactic and prosodic constituency in the definition of second position, we can capture the distribution of clausal clitics in Homer and Herodotus with the same set of generalizations.
The following left periphery allows constituents to precede the host of clausal clitics (Goldstein 2016a: 20-27  Clausal clitics are hosted by the leftmost prosodic word of the highest available projection (that is, either cp or s), excluding any preposed projections. As the position of clausal clitics is conditioned by the available projections, they do not have a fixed position in the above tree. So if a wh-phrase occupies Spec,cp, it is predicted that the first prosodic word of the wh-phrase will host the clitic. But material that is adjoined to cp will precede the host of a clausal clitic. Furthermore, if any material is adjoined to s, both this material and any material occupying higher nodes is unavailable to host a clausal clitic.15 Indo-European Linguistics 4 (2016) 53-97 As discussed below in section 5.2, it is not yet clear if Homer has the nonmonotonic focus projection. 16 What the tree in (7) does is to expand the space of what constitutes canonical 2p behavior by acknowledging that certain types of constituents and subconstituents can occur before the host of a clausal clitic. As a result, many non-canonical surface patterns will turn out not to be exceptional: they merely reflect the presence of morphosyntactic material before the host of the clausal clitic. In the following sections, I identify four such constructions that license morphosyntactic material before the host of a clausal clitic: adverbials that scope over clauses (section 5.1.1); topicalization (section 5.1.2); non-monotonic focus preposing (section 5.2); and participial clauses (section 5.3).

5.1
Delimitation I use the term delimitation (borrowed from Krifka 2008) to refer to two constructions that yield non-canonical surface distribution of clitics, sentential adverbials and topicalization. These two constructions, while distinct, sometimes prove difficult to distinguish, which is why I have grouped them together under the heading of delimitation.

5.1.1
Sentential Adverbials Adverbial expressions that scope over the entire clause (or utterance) precede the host of clausal clitics: tions that in no way wronged the Persians, because we wanted to increase our dominion, we were not to take vengeance on the Greeks who did perpetrate injustice. ' (Hdt. 7.9.2;cf. 7.184.3,9.113.2) In both of these examples, we have adverbial expressions (ἦ and καὶ γάρ) that affirm the truth value of the sentence that they introduce. The adverbs thus straightforwardly c-command their scope domains. Following the tree in example (7) above, we can analyze these adverbial phrases as adjoined to their clauses,17 with the crucial result that they lie outside of the domain over which second position for the clausal clitic is calculated. We also find temporal adverbials adjoined high in the clause: Like the discourse adverbials above, the temporal expressions here c-command their clauses because they scope over them. To take example (9) as illustrative, preposed ἔπειτα makes clear that the event of the speaker's heart commanding to sail for Egypt took place within the interval 'afterwards,' not that the interval labeled 'afterwards' was part of what his heart commanded.18 There have been various attempts to establish a universal hierarchy of adverbials that will predict their distribution in the clause (e.g., Jackendoff 1972, Bellert 1977, Cinque 1999, Ernst 2001, Frey 2003. I cite here that of Frey (2003): we think of information focus as a monotone increasing function (that is, information is only added to a discourse), non-monotonic focus differs in that it can both add new information and cause the loss of old information. 17 I assume for the sake of simplicity that adverbials are adjoined to phrasal projections; the point would remain the same if, e.g., they were in the specifier position of a devoted functional projection. 18 It may be the case that preposed temporal adverbials uniformly modify the reference time of the utterance. Full investigation of this possibility would take us too far afield here.
Indo-European Linguistics 4 (2016) 53-97 (10) Adverbial Hierarchy sentence adverbials > frame and domain adverbials > event-external adverbials (e.g., causals) > highest ranked argument > event-internal adverbials (e.g., locatives, instrumentals) > (internal arguments) > processrelated adverbials (e.g., manner) > verb While the details of these hierarchies present challenges (for an overview, see Maienborn and Schäfer 2011), for our purposes they are of use in offering a boundary between adverbial expressions that occur within s/cp and are adjoined to s/cp (or some higher projection). For Herodotus, it appears that adverbial expressions that belong to the first three categories (sentence adverbials, frame/domain adverbials, and event-external adverbials) adjoin to s/cp, and accordingly occur to the left of the host of a clausal clitic: The preposed adverbial expressions in each of these examples provides information about the circumstances of the event. In example (11), Cambyses is successful in spite of an earlier mistake (ἐπὶ τῆι ἀρχῆθεν γενομένηι ἁμαρτάδι).
In example (11ii), the adverbial expression πρὸς-ὦν-δὴ τοῦτο τὸ κήρυγμα offers an explanation for the event(s) described in the remainder of the clause, i.e., why no one talked to Periander's son or received him in their homes. While the exact class of adverbials that we expect to be preposed is hard to define, it is clear that preposing is motivated by scope. What did Olivia buy?
Contrastive topicalization is a linguistic construction that enables a speaker to move between sub-questions, such as between (12i) and (12ii): (13) I will tell you who bought what. Noa, she bought rice. And Olivia, she bought cough syrup.
There are sub-types of topicalization that are conditioned by the status of the question in the discourse (for a full account, see Goldstein 2016a: 121-173). For our purposes, the crucial fact is that contrastive topicalization involves a noncanonical surface position of the clitic and that we can motivate this position on the basis of the meaning of the construction. A particularly clear example of contrastive topicalization is found in Herodotus' ethnography of the Persian empire, where he records that upon ascending the throne Darius divided his kingdom into twenty satrapies (subscript 'f' abbreviates 'focus'): (14) qud: What did Darius do as king?
i. Sub-qud: How did Darius organize the empire? καταστήσας δὲ τὰς ἀρχὰς καὶ ἄρχοντας ἐπιστήσας ἐτάξατο φόρους οἱ προσιέναι κατὰ ἔθνεά τε καὶ πρὸς τοῖσι ἔθνεσι τοὺς πλησιοχώρους προστάσσων καὶ ὑπερβαίνων τοὺς προσεχέας τὰ ἑκαστέρω ἄλλοισι ἄλλα ἔθνεα νέμων. 'After arranging the provinces and setting up governors over them,  (14). This is the same type of hierarchical qud that we identified in (12), to which Herodotus offers a pair-list answer, as illustrated by the first entry: The sentence opens with a prepositional phrase identifying the satrapy, which is followed by the amount of the tribute. The particle μέν signals that this satrapy belongs to a set of satrapies that are under discussion. Thus we see that contrastive topicalization denotes a set of a set of propositions (the values of the tributes do not correspond to those in the actual examples above, but are merely for illustration): (17) {{Satrapy 1 contributed 1000 talents of silver, Satrapy 1 contributed 2000 talents of silver …}, {Satrapy 2 contributed 1500 talents of silver, Satrapy 2 contributed 1700 talents of silver …}, {Satrapy 3 contributed 500 talents of silver, Satrapy 3 contributed 2500 talents of silver …} …}.
It has to be noted that topicalization is not necessary to achieve this type of semantics, as it does not appear to be the case that Herodotus topicalizes the satrapy phrase in each entry. So the question under discussion would be 'Which political entity contributed what amount of revenue?' To answer this question exhaustively, Herodotus has to name an amount for each political entity. Herodotus uses topicalization to go from one satrapy to the next, as in (14), where topicalization of the pp from Babylon and the rest of Assyria signals a topic transition.
Topicalization is also attested in Homer: In example (18i), the question under discussion is 'How can one acquire gifts from the gods?' This is broken down into two sub-questions along the lines of 'Can one acquire gifts from the gods when they give them of their own accord?' and 'Can one acquire gifts from the gods by one's own will?' Topicalized ἑκὼν 'by one's own will' marks the transition to this second sub-question. Example (18ii) follows a similar pattern. Here the preposing of Ζηνός draws a contrast between Zeus and the rest of the gods in terms of whom the goddess Sleep would dare to lull to sleep. Frame adverbials, which provide the temporal or spatial setting of the event described in the rest of the clause, are often found in this construction: There are two possible motivations for the preposing of the np [πέμπτηι δὲ ἢ ἕκτηι ἡμέρηι ἀπὸ τούτων]. One is that the phrase is preposed on account of scope. Alternatively, this could be a case of contrastive topicalization in which 'the fifth or sixth day' from the anchoring event is being contrasted with other times. From context, the first analysis appears to be the better fit, but it does not seem possible to exclude the second. Examples of precisely this type are the motivation for the second-order category delimitation. Concerning the syntax of topicalized phrases, they precede interrogative pronouns:  Table 7 details the frequency of the delimitation constructions. As widescope adverbials and topicalization are present in both Homer and Herodotus, the differences above reflect differences in usage, but not grammar.

5.2
Non-Monotonic Focus In the previous section, I presented a topicalization construction in which the preposed constituent was not the focus of the utterance. In this section, I present a construction in which the preposed (sub)constituent is the focus of its utterance.
On a view of discourse governed by questions under discussion, focus is what fills in a variable: 20 A tlg search of ἄν will yield 490 tokens, 10 of which I excluded on textual grounds, and 4 because because they are cases of iteration (Goldstein 2013), which left 476. The phrase my sister fills in the variable introduced by the interrogative pronoun who. It is clear that various types of focus have to be recognized (see, e.g., Gussenhoven 2007). Minimally, we need one that simply fills in requested information as in (22), and one that "overwrites" information that is already present (NMFoc abbreviates 'non-monotonic focus,' which is explained below): If we imagine information in discourse as steadily increasing (i.e., monotonic), the function of this construction is to overwrite, as it were, information in the discourse. So in example (22) The adjective ἐλάσσονος 'less' is preposed because it overwrites the expected value 'more.' Non-monotonic focus preposing differs from contrastive topicalization in that subconstituents can be preposed, as in example (23) These two examples raise a number of issues. First, in the focus preposing construction in Herodotus, the focused constituent contravenes information that is in some way present in the discourse. Examples (26i) and (26ii) do not fit this profile. In the first example, σαώτερος appears to be simply the informational focus of the utterance, that is, the information that answers the Indo-European Linguistics 4 (2016) 53-97 immediate question under discussion. And in the latter, it is not entirely clear that νῦν is even focused, given the presence of the scalar focus expression [καὶ Διὶ πατρὶ] 'even father Zeus.' Either this clause has two focus constituents or νῦν is preposed for some other reason. Furthermore, the position of the focused element in the focus preposing construction is at odds with what we find in Herodotus. As illustrated in example (26ii), the preposed focus element lies beneath c. In example (26i), however, the preposed constituent is in Spec,cp (or at any rate, above c). Given that these examples do not pattern with the Herodotean construction, they have not been classified as cases of non-monotonic focus preposing. Table 8 provides a quantitative overview of the focus preposing in Homer and Herodotus. If further investigation demonstrates the absence of the nonmonotonic focus construction in Homer, that would not affect the central claim of this paper, namely that there is no change in clitic distribution between Homer and Herodotus. The generalization would remain the same (clausal clitics are hosted by the first prosodic word in cp/s excluding any preposed constituents). The difference would lie in the number of preposed phrases from which clausal clitics are excluded.

5.3
Participial Clauses When a circumstantial participial phrase occurs sentence-initially, a clausal clitic either occurs second within that constituent or second within the finite clause:21 (27)  In example (27i), the modal particle ἄν occurs second within the participial phrase, while in (27ii) it occurs second within the finite clause. (Note, however, that the explanatory particle γάρ 'for' in (27ii) is not restricted to the finite clause: it appears second within the participial phrase.) I refer to the first type as vp-participial phrases and the second as participial clauses. Participial clauses are s constituents that form an independent domain not only for second-position clitics, but also negation (for further properties of participial clauses, see Goldstein 2016a: 221-259):   The final line of this example parallels the structure that we have above in example (27ii). The only problem with using ῥα as a diagnostic is that its meaning is not well understood. If the particle is in fact only a phrasal clitic (that is, it only scopes over πρός πλατάνιστον), then it is not telling us anything about the participial phrase. Most of the meanings that Cunliffe (1924: s.v. ἄρα) lists suggest that the particle has clausal scope, however.22 On this basis, I cautiously maintain that βωμοῦ ὑπαΐξας is in fact a participial clause. Homeric syntax thus allows participial clauses, but they are (for whatever reason) only rarely used.

Homer and Herodotus Compared
We are now in a position to reassess the issue of canonical 2p behavior in Homer and Herodotus. By acknowledging the four constructions presented in the previous section, we have drastically reduced the number of non-canonical (that is, exceptional) examples in both Homer and Herodotus. Figure 7 presents a new view of the behavior of clausal clitics in these two corpora. As detailed in Table 10, there are now 14 unclassified (or non-canonical) examples in Homer (out of 1090 total clauses), while in Herodotus there are 21 (out of 2896 total clauses). One could claim that what we are looking at here is syntactic change, but there is little appeal in such an analysis. First, as this class is not homogeneous, one would have to allege not one syntactic change, but several. Each would be then be based on slight evidence. Second, a χ2 with Yates' continuity correction yields a p-value of .1345, which is well above the standard .05 threshold for significance. In other words, there is no reason to reject the null hypothesis that the samples are drawn from the same distribution.

22
E.g., his first two entries, "Expressing consequence or sequence" and "Expressing explanation, indicating a reason or cause", suggest that the arguments of the particle are propositions.
Indo-European Linguistics 4 (2016) 53-97 figure 7 Decomposing non-canonical clitics What Wackernagel and others observed was not then the rise of new cliticdistribution rules, but rather the frequency with which these four constructions are used (cf. Probert 2015: 443, who argues that the differences in relative clauses between Homer and Demosthenes are more stylistic than syntactic). As a consequence, nothing supports the view that the placement rules of clausal clitics changed between Homer and Herodotus. I want to emphasize that the claim pursued here is not that the clause structure of Homer and Herodotus is identical. The point is rather that my study provides no evidence for a difference in the distribution of clausal clitics between Homer and Herodotus.

6.1
Explaining the Usage Variation On my analysis, the difference between Homer and Herodotus is to be attributed to synchronic usage variation. The reason why the rate of 2p behavior for clausal clitics is lower in Herodotus than Homer is because the former uses the constructions presented in section 5 more than the latter. This claim raises the question of why. Why does Herodotus use these constructions more? This is a challenging question and one whose answer lies well beyond the scope of this paper, as it would require a detailed discussion of the discourse features of these two corpora. Nevertheless, I would like to at least sketch three possibilities, which are not mutually exclusive, for the usage differences that we observe between Homer and Herodotus: text-type, style, and syntactic nascency.
By text-type, I mean that the nature of the discourse itself is what is responsible for the varying usage frequencies. For instance, it could be the case that there is a greater need for contrastive topicalization in historiography as compared to epic. On this view, the frequency of the various constructions would fall out from the internal needs of the discourse. It is also imaginable that the pressures are external. That is, it could have simply been a part of the profile of Homeric epic that topicalization is not often used. Wackernagel (1892: 341) seems to have something along these lines in mind, when he writes that the second-position tendency of νιν is less robust in Pindar and the tragedians in comparison with Homer "wegen der grössern Künstlichkeit ihrer Wortstellung." If this analysis is on the right track, we should expect the frequencies of the constructions in section 5 to correlate with other textual properties.
A related possibility is that the usage ranges are due to an author's individual style. An adequate definition of style is notoriously difficult to come by, and here I conceive of it as the complement of the text-type properties mentioned in the preceding paragraph: these are features that exist solely at the level of the author and are not conditioned by the external or internal needs of the texttype. So on this view, a higher frequency of topicalization, for instance, would Indo-European Linguistics 4 (2016) 53-97 be a property of Herodotean prose that is not the product of the text-type in any way. It is simply how Herodotus builds a narrative. It is obviously no easy matter to tease apart stylistic properties from those conditioned by text-type.
The third possibility diverges from the preceding two. The differences that we observe may have nothing to do with text-type or style, but may be due rather to the nascent status of the constructions in section 5. That is, if we accept the view that syntactic innovation follows an s-shaped curve (as illustrated by Figure 3 in section 4), then what we observe between Homer and Herodotus is the gradual blooming of the various constructions. While this may well be the case, it does push the question back. What we would now like to know is why these constructions come to be used more in the interval between Homer and Herodotus.

Conclusion and Conspectus
By adopting a more precise definition of "second position" in which both canonical and non-canonical examples follow the same distributional generalizations, I have argued against the long-held view that new rules of secondposition clitic distribution arose between Homer and Herodotus. The distributional generalization for clausal clitics that I have proposed is just as robust in Herodotus as it is in Homer. Thus what Wackernagel and others have observed is a difference in usage between the two corpora.
It remains an open question as to what lies behind the usage differences between the two authors. As this analysis revamps the history of clitic distribution in Greek, it raises a number of new diachronic questions. Perhaps foremost among them is: When does the system of 2p distribution that we observe in Homer and Herodotus start to break down? How do we ultimately get to the head-adjacent clitics of modern Greek (see Kiparsky 2002, 2004)? And looking back in time, how far back can we project the left periphery argued for here-is this a Greek innovation, or is it an inheritance? While I have every confidence that we will achieve new insights into these questions and the history of archaic Indo-European syntax more generally, I hope to have made it clear that the challenges of investigating diachronic syntax on the basis of a restricted range of text types are not insignificant.

More on Second Position
For the quantitative profiles to have any meaning, it is essential to be explicit about what counts as exceptional. As established in section 2 above, the basic generalization is that the clausal clitic should be hosted by the first prosodic word of its clause. In the majority of cases, it is clear when an example violates this requirement. In some cases, however, it is not straightforward to apply. In the following sections I call attention to data that prima facie might seem exceptional but that were nevertheless counted as representing canonical 2p behavior for one reason or another.
One could analyze such a pattern as non-canonical, on the belief that the first element (or first two elements, in the case of expressions with three morphological words) of such strings should have been the host. For instance, since ἦ alone can host clitics, we would expect a clausal clitic to show up in between ἦ and μάλα, and not after them: (30) ὢ πόποι, ἦ μάλα-δή-με θεοὶ θάνατον δὲ κάλεσσαν: 'Alas, the gods have in fact called me to my death. ' (Hom. Il. 22.297) I counted examples such as this as canonical 2p behavior, because ἦ μάλα is either lexicalized (i.e., a single lexeme) or forms a single prosodic word (or both).

Complementizers as Hosts
The position of pronominal clitics in clauses with complementizers and relative pronouns exhibits variation that is not well understood. Typically the pronominal clitic is hosted by the complementizer or relative pronoun (and thus is second within cp), but at other times it is second within s: 23 Goldstein (2016a: 92-96) characterizes ἄν in such cases as a domain-widener. While it is true that the combination rel-ἄν is often rendered 'rel-ever' in this construction, it may not be entirely accurate to describe the modal particle itself as a domain-widener. According to Probert (2015: 83-97), the "indefinite construction" denotes generic quantification over occasions. In any scenario in which the occasions quantified over contain individuals, those individuals will be in the scope of the generic quantifier, from which the 'rel-ever' meaning results. On this view, the 'rel-ever' meaning is only a by-product of the construction, and perhaps has more to do with English -ever than the Greek construction. At any rate, an exact semantics of the "indefinite construction" is not necessary for our purposes. The crucial point is that the modal particle occurs after the complementizer in this construction, in contrast to example (32i 'He saw Hector lying on the plain. All around his comrades had sat down. He was gripped with strained wheezing, stunned in his mind, spitting up blood, since it was not the weakest of the Achaeans who had struck him.' (Il. 15.9-11) ii. ὡς-δὲ-ἄρα πάντες-οἱ ἐτετάχατο κατὰ ἔθνεα καὶ κατὰ τέλεα, ἐνθαῦτα τῆι δευτέρηι ἐθύοντο καὶ ἀμφότεροι. 'After they had all then been arrayed by him according to nation and battalion, both armies also offered sacrifice on the second day.' (Hdt. 9.33.1) It is not clear what conditions this variation. All that can be said at this point is that if a complementizer or relative pronoun is present in a clause, it is far more common for it to host the pronoun than not to. As the above examples show, both patterns are found in both Homer and Herodotus. Given that the alternation could be semantically or syntactically conditioned as with ἄν above, neither of the patterns in examples (33) and (34)  The inability of σε to intervene between ἵνα and μή in (35) is due either to the fact that that ἵνα μή is a complex lexical item occupying one syntactic node, or to the fact that together they form a prosodic word. Similarly, clitics do not intervene between a focus operator and its argument:25 (36)  See also Il. 8.478,9.379,9.385,9.445,19.372,22.220,22.439,22.351,23.346,Od. 14.140. Indo-European Linguistics 4 (2016) 53-97 Oceanus, who is the source for (them) all. Zeus son of Cronus, however, I would not approach or lull to sleep, unless he were to order (me) himself.' (Il. 14.243-248)

Variation with Disjunction and Conjunction
Conjunctions and disjunctions are typically assumed to occupy a phrase-structure position above that of the clause. Consequently, they should be unavailable as a clausal-clitic host: (38) Conjunction does not host pronominal clitic ἤδη γάρ ποτ᾽ ἐγὼ καὶ ἀρείοσιν ἠέ περ ὑμῖν ἀνδράσιν ὡμίλησα, καὶ οὔ-ποτέ-μ᾽ οἵ γ᾽ ἀθέριζον. 'Once I kept company with men even more fierce than you, and never did they despise me.' (Il. 1.260-261) As expected, the clitic pronoun μιν is hosted by the first prosodic word after the conjunction. Elsewhere, however, conjunction does host clausal clitics: (39) Conjunction hosts pronominal clitic i στῆ δὲ παρ᾽ Ἕκτορ᾽ ἰὼν καί-μιν πρὸς μῦθον ἔειπεν. 'He went and stood by Hector and said the following to him.' (Il. 7.46) ii. τοῦ μὲν ἄρ᾽ υἱὸς ἐπῆλθε, Θεοκλύμενος δ᾽ ὄνομ᾽ ἦεν, ὃς τότε Τηλεμάχου πέλας ἵστατο: τὸν δ᾽ ἐκίχανεν σπένδοντ᾽ εὐχόμενόν τε θοῆι παρὰ νηῒ μελαίνηι καί-μιν φωνήσας ἔπεα πτερόεντα προσηύδα. 'His son came up, Theoclymenus by name, who then stood near Telemachus. He found him pouring libations and praying by his swift black ship, and addressed him with winged words.  In example (39), we have two cases of the conjunction καί hosting the pronominal clitic μιν. Both are constructions that introduce speech. In all the examples of speech introduction that involve καί and μιν, the latter is always hosted by the former. Such consistent patterning raises the possibility that καί is a temporal adverb meaning 'then.' As an adverb it would occupy a node inside the clause (i.e., cp/s), and thus be able to serve as a host of clausal clitics. This alternation is also found with ἀλλά (e.g., Il. 1.508 and 23.523), αὐτάρ (e.g., Il. 6.157 and 23.95), and ἤ (e.g., Il. 8.13 and Od. 20.63).
Indo-European Linguistics 4 (2016) 53-97 In each example, one can make a case that the bracketed constituent is the focus of the utterance. But none of them seems to exhibit counterassertion, which is what one expects from the focus preposing construction. It may be the case that these examples provide as yet undeciphered clues to the structure of the left periphery.
'Accordingly, I did reason correctly in claiming (that the Greek record is) accurate up to Perseus.' (Hdt. 6.53.2) iii. τετραμμένωι γὰρ δὴ καὶ μετεγνωκότι ἐπιφοιτῶν ὄνειρον φαντάζεταί-μοι, οὐδαμῶς συνέπαινον ἐὸν ποιέειν με ταῦτα. 'For since I turned and changed my mind, a dream keeps coming and appears to me, and it does not at all agree that I do these things. ' (Hdt. 7.15.2) Although these examples have yet to receive an adequate analysis, it is remarkable that they are all hosted by the verb.

Embedded Participial and Infinitive Complements
In some embedded participial and infinitive phrases, a pronominal clitic occurs first in the embedded phrase. This phenomenon is referred to as ditropy, as it involves a split in the constituency of the clitic (Cysouw 2005, Spencer andLuís 2012: 66-67; the term goes back to Embick and Noyer 2001). It forms a syntactic constituent with the material to its right and a prosodic constituent with the material to its left: 27 I have translated the participial phrase as the complement of the verb κελοίμην with the infinitive Ἀργείοισιν ἀμυνέμεναι an adjunct expressing purpose. It is also possible to parse μῆνιν ἀπορρίψαντα as a cirumstantial participial phrase, in which case Ἀργείοισιν ἀμυνέμεναι would be the complement of κελοίμην. 28 These data do not lend support to the claim of Agbayani and Golston (2010) that "second position is first position." First, their claim is meant to characterize all clitics in all contexts, which is empirically untenable. Second, they register no awareness of the restricted distribution of 1p patterns. Nothing in their account would account for why it only shows up in embedded contexts. Third, a strictly 1p account, such as they advance, brings with it the problem of being unable to account for the far more frequent cases in which the clitic shows up second in the embedded constituent.