Syntax, Scope and Prosody: Ancient Greek as a Human Language (Draft Paper)

Syntax, Scope, and Prosody: Ancient Greek as a Human Language Mark Hale Concordia University, Montr´eal

1 Preliminaries 1.1

The Problem

This paper concerns two ideas regarding archaic Indo-European languages which I like to refer to as ‘zombie’ ideas, because they simply will not die.1 In very rough form, these ideas are: 1a) Archaic IE languages like Greek and Sanskrit are ‘non-configurational’, or ‘free word order’, languages. Or ‘predominantly’ so, whatever that might mean. 1b) The word order properties of ancient metrical texts are determined by the constraints of the meter. Or ‘predominantly’ so, whatever that might mean. Neither of these ideas is universally held, but both, usually in some ‘hedged’ form (as in the ‘predominantly’ characterizations above), receive broad support in the scholarly literature. In my view neither of these ideas has much of a chance of providing a useful orientation for the scientific investigation of the syntax of archaic Indo-European languages—indeed, the fact that they are broadly believed in some weak form has, in my view, been a major impediment to our making progress on the study of the syntax of these languages. In this paper, I will attempt to define with relative precision the concepts involved in these assertions and argue that, when considered in a less informal manner, the proposals have little merit. The two ideas are often invoked together (though as we shall see below the logic of this is somewhat awkward). We can see this if we unpack the reasoning seen in Diggle’s treatment of a textual issue in Euripides, to which we will now, by way of introduction to the problem, briefly turn. In Diggle (1981:83) we find a detailed discussion of IT 507, which Diggle prints as: χάριν δὲ δοῦναι τήνδε κωλύει τί ϲε; “what prevents you giving this as a favor?” The issue Diggle addresses, based on the various opinions of previous editors, is whether the text should read τί ϲε or ϲε τί. Several methodological issues arise in the course of his discussion, which is thus worth citing more or less in full. He begins with a consideration of the relative ordering of enclitic pronominals and the enclitic indefinite τιϲ/τι in the last foot of the iambic trimeter: 1And there is some evidence that they eat brains.

Consider first the placement of the enclitic forms. So far as concerns the enclitic τι and the accusative pronoun, either order (ϲέ τι or τί ϲε) is metrically possible and both are found: ϲέ τι A. PV 835, S. OT 644, Men. EPitr. 433 (257 Koerte), 574 (398 K.), τί με (ϲε) S. fr. 176 P, Ar. Nub. 38, Lys. 656, Plut. 22 (and in a trochaic tetrameter at Pax 330), Men. Sam. 702, Dysk. 107, Phas. 53. Whether the words are the last in the sentence or appear in the middle of the sentence is not a factor which determines the choice of order. When τιϲ is used with the accusative pronoun, or when τι is used with the pronoun in the genitive or dative, the order is dictated by metre: μέ (ϲέ) τιϲ Hcld. 248, El. 559, Hel. 1615, Ba. 649, IA 849, Ar. Plut. 1014, Men. Dysk. 142, τί μοι (ϲοι, ϲου) Cycl. 175, Hec. 992, IT 582, Ba. 658, IA 1188, A. Eum. 442, S. El. 902, Ph. 761, OC 1414, GLP 32.4 Page, Ar. Au. 51, Lys. 861, Ran. 44, 175 (and in a trochaic tetrameter at A. Pe. 705), Men. Asp. 173 (coni.), 420, Dis Ex. 62, Dysk. 299 (suppl.), 877 (suppl.), Epitr. 264 (88 Koerte), 1070 (712 K.), Perik. 485 (235 K.), fr. 97.1. He then turns to a consideration of the relative ordering of the interrogatives τίϲ/τί and pronominal clitics. Now consider the placing of the interrogative forms τίϲ and τί. I can find no example of τί με (ϲε). But this is of no significance, since I am equally unable to find an example of με (ϲε) τίϲ (τί). Metrical necessity dictates the order at Ion 433 ἀτὰρ θυγατρὸϲ τῆϲ Ἐρεχθέωϲ τί μοι / μέλει; and at Ba. 832 τὸ δεύτερον δὲ ϲχῆμα τοῦ κόϲμου τί μοι; cf. Men. Mis. 216 (18 Koerte), if Handley’s supplement is right. And so these passages tell us nothing. And it would be doubly irrelevant to adduce in evidence passages where the interrogative is the opening word of its clause, since here again the alternative order is excluded: Andr. 1104-5 ὦ νεανία, τί ϲοι / θεῶι κατευξώμεϲθα; A. PV 83, S. El. 352, 887, Ph. 559, 753, OC 1420, Men. Mis. 308. I have therefore found no evidence either in tragedy or in Aristophanes or Menander to justify the assumption that, when the interrogative form is used, a trimeter will more naturally end with ϲε τί than with τί ϲε. As mentioned above, several issues arise concerning Diggle’s approach to the textual issue before him,2 but here I will focus only those most relevant to the questions at hand. Note the first conclusion drawn by Diggle: when either order between indefinite τιϲ and the enclitic pronoun is possible (e.g., when we are dealing with the neuter τι and an accusative clitic pronoun), both orders are attested. What does this mean? As far as I can see, there are two possibilities a priori: either which order one gets is based on considerations of the grammar (i.e., one order is triggered in one set of syntactic configurations, the other in another) or the order is ‘free’—i.e., the 2For example, why would one believe that the ‘last foot of the iambic trimeter’ line has any properties which make it worthy of separate investigation for the question at hand? It seems very likely that if certain ordering principles hold (or fail to hold) of Euripides’ trimeters generally, they will equally well hold (or fail to hold) for the final foot of the iambic trimeter. Is there something special about this foot that is supposed to make interrogatives act differently, with respect to word order, than other positions in the iambic trimeter line?

2

grammar does not dictate any specific ordering between the elements.3 Under the first assumption, Euripides’ text, since it can show either order in these cases, presumably attests the order which has the appropriate interpretation (since structural differences entail interpretational ones, see below) for Euripides’ communicative intent. Under the second assumption, since the interpretation is independent of the order (the grammar ordering elements, regardless of their structural and semantic relationship, in ‘either order’) again it is a safe assumption that Euripides has used an order which has an appropriate interpretation, given his communicative intent (since either ordering would, this is obviously a very safe assumption). Thus, under either possible scenario, Euripides has placed the elements in a sequence which gives, given the workings of his grammatical system, the interpretation he desired to communicate. Note that in all cases he has found a way to do this within the constraints of the metrical system—quite a simple task in this instance, in which either ordering scans! It seems likely that if Diggle were adopting the first approach to this issue—that the ordering exists to express some meaning, encoded by the grammar—the logical next task would seem to be to (1) determine what meaning difference is expressed by the two orderings, (2) determine which meaning is relevant to the IT 507 passage under discussion, and (3) edit the passage with the ordering that expresses the meaning appropriate to the passage. Of course under the ‘free word order’ assumption there is no point in doing any of that work—each of the two orderings express all possible meaning relations, there being no relationship between order and interpretation. From the fact that he does not explore possible meaning differences, it seems likely that he leans in the direction of the ‘free word order’ hypothesis (though he is not explicit). When the ordering of the elements has metrical implications (i.e., the two elements cannot be ‘swapped’ while still respecting the constraints of the meter), Diggle says that ‘the order is dictated by metre’. Note that, as in the ‘swappable’ case, both orders are attested. While we will consider below in some detail what it might mean to assert that ‘the order is dictated by metre’, for the present let us approach the matter relatively superficially. Again, we need to deal with two different ‘working hypotheses’ about the relationship in Attic Greek between word order and interpretation, to which, however, we would now add a third possibility: (a) The syntax of Attic Greek determines the observed word order and differences in word order are related to differences in interpretation (Attic Greek is ‘configurational’). (b) The syntax of Attic Greek determines the observed word order, but differences in word order are unrelated to differences in interpretation (Attic Greek is ‘non-configurational’). (c) The syntax of Attic Greek does not determine the observed word order: that order is determined by the metrical system. The interpretation is thus not dependent upon the observed ordering. 3One might well wonder, as a general consideration, whether the underlying assumption in such a case is that the causal structure of the world, which seems to hold of everything from clusters of galaxies to the Higg’s boson, is broken at the level of Attic Greek word order, or, if not, just what factors are responsible for the actual observed order in each instance, if they are not related to the grammar.

3

Interestingly, although Diggle asserts that it is the meter that is dictating the observed order, nowhere does he demonstrate, or even acknowledge that such demonstration is required, that the observed order fails to directly express, through the workings of Euripides’ grammar, precisely the meaning Euripides intended for the line in question. If he is assuming ‘free word order’, such a demonstration is of course not possible: if any ordering can express any meaning, then the intended meaning can be grammatically expressed by the observed order. The lines scan of course (as all lines must), but the ordering is dictated by the grammar, and the meter is merely a filter on whether the grammar-produced sentence, with its associated interpretation, can be used at this point in the play. Under assumption (a) above, Diggle would need to show, in order to invoke the meter as determinative as to observed order, that in those lines in which the indefinite pronoun τιϲ/τι precedes the enclitic pronominal this ordering cannot be attributed to the effects of the grammar directly, and likewise for those lines in which the order is the inverse. No such demonstration is forthcoming. The relative power of the grammar and the meter to determine observed word order comes up again in Diggle’s reasoning when he asserts, concerning the relative ordering of interrogative τίϲ/τί and the pronominal enclitics, that it would be irrelevant ‘to adduce in evidence passages where the interrogative is the opening word of its clause, since here again the alternative order is excluded’. But a sentence-initial sequence such as με τί is not excluded by the meter, since the two elements are generally ‘swappable’. Obviously, Diggle is referencing the fact that the grammar of Attic Greek does not allow sentence-initial enclitics, thus any sentence which begins *με τί... can be excluded a priori. But this is true as a general proposition (holding both of ‘swappable’ and non-swappable enclitic-interrogative sequences) only if the grammar, not the meter is determinative as to observed order. But which is it? Is the metrical system in the “driver’s seat”, or is the grammar? To work towards an answer to this question is one of the goals of the present paper.4 4As to the specific passage under discussion by Diggle, the matter is quite complicated. In general, Attic Greek respects the requirement, spelled out for the similarly-structured syntax of Vedic Sanskrit in Hale (1987), that pronominal clitics do not appear to the left of a WH-moved interrogative, which would strongly favor the reading adopted by Diggle, seemingly without explicit motivation, of τί ϲε. There is, however, a slight complication: as has been known since Wackernagel’s seminal article on the matter in 1897, Attic Greek shows more ‘violations’ of the processes which give rise to second-position placement of clitics than do languages such as Vedic Sanskrit, or Homeric Greek. One of the structures which give rise to these ‘violations’ involve cases in which the pronominal clitic remains in its constituent of origin (part of the process which gives rise to Wackernagel effects is the movement of pronominal clitics to some relatively high position in the tree—see Hale, forthcoming). If the pronominal clitic remains within, e.g., the Verb Phrase, and that Verb Phrase were to undergo fronting around the WH-element— such fronting being well-attested in both Vedic and Greek—the pronominal clitic would end up to the left of the interrogative. This seems to me to be attested in three instances (of the hundreds and hundreds of interrogatives in the Euripidean corpus): e.g., Ion 1350 ἔχει δέ μοι τί κέρδοϲ ἢ τίνα βλάβην; ‘what benefit or what harm does it hold for me?’. The material to the left of the interrogative in IT 507 is the complex VP χάριν δὲ δοῦναι τήνδε κωλύει ‘prevents from giving this as a favor’—if ϲε had remained within this VP (rather than moving to some high, but under the landing site for WH-movement, position in the tree), it could have fronted with that VP around τί. One may assume that if we understood the order within such fronted VPs better, we could rule out or in the observed (very late in the VP) placement of ϲε. At present, I, at least, can make no determination.

4

1.2

Grounding Assumptions

It will be worthwhile to spell out in some detail the assumptions which I bring to the task of evaluating these proposals. Different assumptions may, of course, lead to a different assessment. While I do not think that any of the assumptions I am making are particularly idiosyncratic, considerable confusion can be avoided if we work within a relatively explicit terminological and conceptual framework. The first central assumption that I will make is that all aspects of linguistic structure are constructed. This is largely definition—i.e., I will consider as a ‘linguistic structure’ such objects as are created in the course of a derivation in a grammar. There may be many other kinds of information that can be extracted from an utterance than just what was encoded in its structure: if I say to you “maybe you should come to dinner next week; my cousin will be there” you may conclude that I am trying to set you up with my cousin, or that, because we both know you despise my cousin, my invitation is intended as a joke. Nothing in the pair of utterances encodes either of these inferences: they represent conclusions you have drawn from the linguistic evidence actually provided to you by these utterances, together, of course, with other sources of information you have access to. We will be concerning ourselves in this discussion only with linguistic forms thus defined. As a general model, it has been fairly standard for some time to assume that the syntax computation mediates between sound and meaning, i.e., between phonological and semantic interpretation. Both phonological interpretation and semantic interpretation get as their input (i.e., as the entity they will be interpreting) a syntactic representation (though not necessarily the same syntactic representation). This model is often conceived of as something like this: Semantics scope interp.

Syntactic Computation

M1

M2

M3

...

Mn

Phonology prosodic phrasing

Additionally, I believe that, as with theory construction in general, there is an ontological commitment in our linguistic theorizing. This means that if we claim that in a passage such as the one quoted below, Greek τε is conjoining NPs, there must be NPs. If conjunction is a linguistic operation, the NPs in question must be aspects of linguistic structure—i.e., they must have been constructed in the course of the derivation. We will say, using standard terminology for operators like conjunctions, that they ‘scope over’ what they take as an argument. Thus in the example below the first τε ‘scopes over’ τὰϲ Νηρέωϲ κόραϲ ‘the daughters of Nereus’, the second over ἰχθύων πᾶν γένοϲ ‘the whole tribe of fishes’. 2) μὰ τὴν Καλυψὼ [τάϲ τε Νηρέωϲ κόραϲ], 5

μὰ θαἰερὰ κύματ᾿ [ἰχθύων τε πᾶν γένοϲ], ‘by Calypso and Nereus’ daughters, by the holy sea swell and the whole tribe of fishes’ (Eur. Cycl. 262-263) Note that, because τε occurs in ‘second position’ (a matter to be discussed in much greater detail below) in the entity being conjoined, quite subtle aspects of the clause structure of Greek can be determined by examining its placement. For instance, in (3) we have conjoined verbs (or ‘verb phrases’—I’ve bracketed the two conjuncts): 3) καὶ πῶϲ ἂν αὑτὸϲ [κατθάνοι τε] [καὶ βλέποι]; ‘and how can the same (person) be both dead and be alive?’ (Eur. Alc. 142) Unsurprisingly, the nominative subject αὑτὸϲ (from ὁ αὐτὸϲ with ‘crasis’ — i.e., phonological fusion, of the two elements) is outside the scope of the conjunctions (since it isn’t being conjoined to anything). The fact that τε does not immediately follow αὑτὸϲ is key to the proper interpretation, as anyone who is familiar with Attic Greek is readily aware. Linguistics is, of course, largely about making these implicit procedures we use to produce and understand linguistic strings explicit, so if it appears I am belaboring the obvious at various points along the way, that perception would not be completely inaccurate. We can turn to more complex examples, using what we have learned from these familiar ones, such as the sentences in (4ab). 4a) καὶ Φρύνιχοϲ ... // αὐτόϲ τε καλὸϲ ἦν, καὶ καλῶϲ ἠμπίϲχετο. ‘and Phrynichus...was himself beautiful, and dressed beautifully’ (Ar. Thesm. 164) 4b) γυναῖκεϲ, ἥκω πόλλ᾿ ἔχων λέγειν φίλα, αὐτόϲ τε ϲωθείϲ (...// ...// ...) vίκην τε Θηϲέωϲ ἀγγελῶν, ... ‘Women! I have come with much welcome news— I myself am being saved...and am bearing news of the victory of Theseus.’ (Eur. Supp. 634) Given the position of τε, the first cannot mean ‘Phrynichus himself was beautiful and (Phrynichus himself) dressed beautifully’. Phrynichus, the subject, is outside the scope of the τε and αὐτόϲ is inside its scope (and thus not to be interpreted with the second conjunct). It means, then, that ‘Phyrnichus was himself beautiful, and (Phrynichus) dressed beautifully.’ By the same reasoning, the second example does not mean ‘(I) myself am being saved, and (I myself) am bringing news of the victory of Theseus.’ Given the placement of τε relative to αὐτόϲ it must rather mean ‘(I) am being saved myself, and (I) am bringing news of the victory of Theseus.’ Is this supported by the normal interpretation of the text? I believe it is (we will return to the matter later in the paper). However, at this juncture we come back to the general point: our theoretical constructs entail ontological commitments. 6

Thus, if a ‘Kolon’ (which I take to be a prosodic structure, as has become common) is built out of a non-prolative infinitive clause (but not, e.g., out of prolative ones), then, because the building of prosodic structure is a linguistic operation, non-prolative infinitive clauses must (a) exist5 and (b) have been constructed in the course of the linguistic derivation. If they differ in their prosodic structuring (as well as in their semantics, of course) vis-`a-vis prolative infinitive clauses, then, since the phonological system does not access semantic interpretation (and vice-versa), in order for these two linguistic structures to be treated differently by the phonology, they must differ in their structural properties. These assumptions provide us with two clear diagnostics for linguistic structure: scope relations (established via semantic computation) and prosodic phrasing (constructed in the course of phonological computation). The primary mechanism for determining scope relations is our interpretation of the text. This can be a challenge, but for some operators (coordinators and subordinators, for example) it is not generally controversial. There are several ways to determine prosodic structure, including (1) overt indications in the text (punctuation, word spacing, and the like—though these are often editorial!), (2) evidence provided by the meter (caesura, lineend, limits on resolution), and (3) evidence provided by the interaction between prosodic operations (e.g., clitic ‘shifting’ appears to be sensitive to aspects of previously-constructed prosodic structure, as we will see below).

2

Some Recent Approaches to Attic Greek Syntax

Attic Greek is often touted as an example of a “predominantly free word order”, or “basically non-configurational” language. There is a strong tradition that treats this property as the fading residue of an alleged even less restrictive word order syntax in the Indo-European protolanguage. Sometimes this is expressed rather romantically, as in Dover’s (1960:68) assertion that “[i]f the Greeks had not possessed so intense a degree of artistic self-consciousness, it may be thought likely that syntactical patterns would have established themselves much earlier and much more firmly.” And that Greek literature “maintained a resistance to that drift towards syntactic uniformity which has been the fate of other languages...”. Sometimes, more technically, as in Devine & Stephens (2000:149-150) statement that “it is not unreasonable to think that Greek preserves a progressively weakening residue of (presumably early Indo-European) nonconfigurational features... The more nonconfigurational a language is, the leaner its nuclear phrase structure...” In addition, it has become popular in some recent work to attribute aspects of the clitic syntax of archaic Indo-European languages (including Greek) to the workings of prosody alone (Keydana 2011) or those of some “prosody-dominant” mechanism (Goldstein, 2010), diminishing the role of syntax in accounting for several crucial aspects of clitic syntax in archaic IE languages. Such approaches appear to take prosody as a syntactically-independent given, and seem to envision a system in which the clitics are simply left out of the derivation until such time as prosodic structure has already been constructed, and then inserted into an appropriate position in that 5Note that it may be that ‘non-prolative infinitive clauses’ turn out to be reducable to some other structural entity or configuration that the Kolon-forming phonological computation operates over—this is of no general significance in the present context.

7

structure. It is hard to figure out what the architecture of such a system might be. The authors seem almost determined not to inform us as to their architectural assumptions. Such a derivation is impossible in the kind of system sketched above, since the clitics could not, if inserted during phonological computation, contribute to semantic interpretation, which they clearly do. One of the primary reasons for the sense that Attic Greek is a non-configurational language, which following Devine and Stephens, themselves following the only serious attempt to formally characterize such a language (K. Hale 1981 and related work), I will refer to as a ‘flat phrase structure’ system, seems to be the phenomenon of discontinuous constituency. It is definitely worth one’s while thinking about discontinuous constituency, but we will not do that for the most part in this paper. Before putting it aside altogether, however, imagine two languages, both of which want to contrastive focus, as English does in (5a). The other language, call it Hsilgne, conveys that same semantics as in (5b). Similarly for wh-clauses in (6ab). 5a) Sanskrit data I’ve got; Greek data I really need! 5b) Sanskrit I’ve got data; Greek I really need data. 6a) In which book did you read that? 6b) Which did you read that in book ? Imagine that the fronting that we see in the two languages is the result of need to place the contrastively-focussed element in the domain of a “contrastive-focus head” (in 5ab) or the whelement into the domain of a wh-head (in 6ab). In English, even though ‘data’ is not part of what is being contrasted (since it is the same in both contrasting NPs), it fronts in (5a). And even though ‘in’ and ‘book’ are not wh-elements, they front in (6a). By contrast, in Hsilgne, only the element which needs to enter into the relationship with the higher head is ‘moved’ to the relevant position. That is, there is no ‘pied piping’. Which language is doing something weird? What is the technical mechanism that insures ‘pied piping’, and how ‘minimalist’ is it? Attic Greek is not Hsilgne, as we will see abundantly below. But its apparent ‘discontinuous constituency’ arises (in those cases in which it is actually possible) from a failure to ‘pied pipe’. Since I consider ‘pied piping’ pretty inexplicable and odd, this makes Attic Greek a very natural human language (and English, and English-like languages, a little weird).

3

Constituency

Languages with ‘flat phrase structure’ do not have constituents.6 Their syntax looks like this, where W1 , W2 , W3 ,...Wn are words, and S is a sentence: S W1 W2 W3 ... Wn 6Actually, in my view natural languages with ‘flat phrase structure’ do not exist, and thus don’t have much of anything.

8

Languages with constituency have hierarchical structure: constituents are elements dominated by a shared node. In the tree below, B and W1 are constituents of A, and W2 and W3 are constituents of B. A B

W1

W2 W3 Pied piping makes possible many excellent arguments for constituency. Indeed, it is hard to imagine any other reason for the fronting of ‘in which book’ in (6a) except that these elements form some kind of morphosyntactic object (a constituent) and that it is that object which is being ‘moved’ by the syntax. There are no languages, even allegedly ‘flat’ ones, which move three random unrelated words to clause-initial position to form a question. The idea that these elements form a constituent is also strongly supported by our semantic intuitions, as well as by a variety of ‘substitution tests’, which are, however, difficult to make extensive use of with dead people, who tend to be not very responsive to direct inquiry. So, if, for example, Attic Greek were to show evidence of ‘pied piping’ of the type seen in (6a), we could conclude that Attic Greek must have PPs. If it has PPs, its phrase structure is not flat. In addition, since wh-movement is ‘movement into a relationship with a wh-head’, if Attic Greek pied pipes PPs in interrogative and relative structures, it needs to have the relevant CP-domain structure to support a wh-head and landing site for movement to that head. Attic Greek pied pipes [+WH] PP’s: 7a) [ ἐπὶ τί ] γάρ μ᾿ ἐκεῖθεν ἦγεϲ; ‘(Because) for what did you bring me from there?’ (Ar. Av. 340) 7b) [ ἀφ᾿ οὗ ] γὰρ ὁ θεὸϲ οὗτοϲ ἤρξατο βλέπειν, ἀπόλωλ᾿ ὑπὸ λιμοῦ. ... ‘Because from which (time) the god regained his sight, I’m perishing from hunger.’ (Ar. Plut. 1173) Now, I suppose, given this fact, one could propose that Attic Greek phrase structure that is ‘basically flat’ except for PP’s (and NP’s, presumably), something like this:

9

CP C′

SPEC

S

+wh

... Wn

PP

W1 W2 P

NP W3 W4

Indeed, perhaps some such conception of phrase structure lies behind Devine & Stephen’s notion of ‘lean phrase structure’ or others proposal of ‘relatively’ or ‘predominantly’ free word order— since the advocates of these positions have not spelled their proposal out in any detail, it is impossible to speculate. But the tree is pretty hideous, and the mechanics of its construction are going to be pretty hard to state, since whenever we have a PP we need to build one, but otherwise we just put the words in ‘any order’. And, for the record, can you build a machine (or write a computer program, or have a computational mind) that will put things in ‘any order’? As far as I can tell, it is trivial to generate a given order by leveraging syntactically irrelevant factors (using the numerical value of a milliseconds counter, e.g.), such that it will appear random to someone attending the system’s actual output, but the output will be systematically related to that irrelevant factor. Do we really think anyone talks by doing this? And, of course, we have not exhausted the evidence for constituency in Attic Greek. A very standard test for constituency is scope relations. Natural language provides with a number of operators which take scope over constituents. Attic Greek has many such. Generally, these operators take as their complement the element over which they scope. Conjunction, for example, is a trivial operator, as is disjunction. Greek has a tonic coordinator, καί ‘and’, and a tonic disjunction, ἤ ‘or’. As is well-known, it is normal for these elements to immediately precede the material over which they scope. In addition, Attic Greek has numerous subordination operators, such as ἐπεί ‘when, because’, which also typically takes as its complement the element over which it has scope. 8) ἀλλ’ Ἄργοϲ ἔλθω; πῶϲ, ἐπεὶ [φεύγω πάτραν]; ‘But do I go to Argos? How (can I), since I am an exile from my home country?’ (Eur. Her. 1285) It seems pretty clear that the morphosyntactic objects over which these elements scope are not random subsets of the words in a clause. Indeed, imagine an English-like language that had a conjunction operator and a subordination operator that could conjoin/subordinate random sets of words from within a clause. Start with the clauses: 10

9a) I buy books. 9b) I visit Cambridge. and now use our ‘conjunction operator’ (which I’ll call AND) to conjoin the bolded words below to the matrix clause, or our subordination operator (which I’ll call WHEN) to subordinate the bolded words to the matrix clause. I’ve left the word order unchanged, but ‘visit’ is not within the scope of the coordination in (10a), and Cambridge is not within the scope of WHEN in (10b). 10a) I buy books AND I visit Cambridge. 10b) I go to Widener WHEN I visit Cambridge. No sensible meaning emerges from such operators, and, as I said, no language does these horrible things. Thus anything that can be conjoined in Attic Greek (as in human language generally) must be a constituent, and anything that can be subordinated must be a constituent. The existence of many types of morphosyntactic objects in Attic Greek follow from these simple facts: NPs, PPs, AdjPs, VPs and a variety of higher-order constituences (clauses, including infinitival and participial clauses) all show up conjoined or subordinated, and thus all must exist. And everyone has always assumed they did. The analyses of syntax presented in the traditional literature have always been build up around the notion that Attic Greek has constituents (not using that term, of course). That is, hierarchical phrase structure has been assumed. No one, even those who say that the language has flat or ‘relatively flat’ phrase structure, fails to discuss Greek syntax in terms that directly entail that it does not have that kind of structure. It’s very odd.

4

The Point, Finally

All of this, I am quite confident, you all already know. I now want to turn to some challenges that we face when trying to understand the nature of the Greek clause (I’ll stop saying ‘Attic Greek’ from now on), assuming that we can now all agree that we are not allowed to put aside those challenges by pretending that Greek is ‘non-configurational’, ‘kinda non-configurational’ or ‘pretty much free word order’. Instead, we will assume that Greek has structures of the type seen in other natural human languages, and computes over those structures in the manner of natural human languages. Note that this is a good idea even if we turn out to be wrong. The most restrictive hypothesis must be excluded before more promiscuous ones may be entertained, and it is hard to imagine a less restrictive theory of Greek syntax than ‘the words of any given clause may come in any order’. To make the challenges more exciting, we’ll deal almost exclusively with Attic Greek drama; in fact almost exclusively with Euripides.7 That is, with poetic texts, from which even those with 7Generally, all the trimeter lines from all more or less intact plays have been considered, at least a little, for the discussion to follow, excluding Rhesus (whose authorship is disputed). The project is a work in progress, of course. Much more careful scrutiny is still required.

11

relatively sane ideas about Greek phrase structure feel a strong sense of un-ease when extracting data for syntactic argumentation. I will try to show that this anxiety is misplaced, and that the texts offer much useful data to us which prose texts cannot. Much of this has been discussed, but still has not been systematically investigated by someone who refuses to entertain notions of flat phrase structure, and thus might be able to actually make some progress.

5

Scope

Let’s start with something simple. We’ve seen that in general Greek, like other languages, uses an Operator[argument] type structure with καί ‘and’ and ἐπεί ‘when, since’. Doubtless you can all expand this small set of operators quite extensively off the top of your heads (εἰ ‘if’, ὥϲτε ‘so that’, etc.). These elements quite trivially scope over the material to their right.8 There are, as you also know, some instances of operators of this general semantic type which, however, seem not to take their argument to their right, or at least not entirely so: the conjoining elements τε and δέ, for example, or the subordinator γάρ ‘because’. So, unlike καί, the conjunction τε (e.g., in examples (2), (3) and (4) above) appears inside the element over which it has scope, rather than to its left. In these instances, as I said, the operator appears to be inside the element over which it has scope, rather than taking it as a complement. As Wackernagel rather famously noted (though not using this data), this includes placement directly inside what would otherwise be perfectly well-formed constituents. The position of these operators is syntactically difficult, given that elements within an NP or PP are, as a matter of principle, to be construed with the head of that morphosyntactic object. ‘Flat’ phrase structure would solve the problem in the sense that there would be no ‘constituents’ in the language, and thus no ‘interrupted constituents’ in the examples above. But we have seen that this idea fails on general principles, and thus is not available to us as a way out of this difficulty. More strikingly, it appears that these very same elements create havoc for our otherwise well-behaved operators such as ἐπεί and εἰ. 12a) ἐπεὶ γὰρ ἤιϲθεθ᾿ ἡμέραν τὴν κυρίαν ἥκουϲαν, ὕδαϲι ποταμίοιϲ λευκὸν χρόα ἐλούϲατ᾿, ... ‘(but what she did within the house you will be amazed to hear:) because when she learned that the fated day had come, she bathed her fair skin in fresh water...’ (Eur. Alc. 157) 12b) ...εἰ γὰρ ἔϲτι Λοξίαϲ, ‘because if Loxias lives,...’ (Eur. Tro. 356) 8Some of them allow fronting, usually of a very limited type, of morphosyntactic objects to their left. We will ignore that for a moment.

12

In these examples, γάρ appears within the scope of the ἐπεί and εἰ operators, but is to be construed, if our interpretation of the thousands of relevant text passages is correct, as rather having scope over them. That is, BECAUSE [ WHEN X ] and BECAUSE [ IF X ] are surfacing as WHEN [ BECAUSE X ] and IF [ BECAUSE X]. It is structurally impossible to build a syntactic tree over the strings in (12a) and (12b) in which the logical operators will dominate one another in an appropriate fashion. Since they must dominate one another in an appropriate fashion, or the relevant meanings would not be being expressed, we can only draw one conclusion: Not all aspects of the order of elements in Greek clauses involving operators such as γάρ is determined by the syntax. When we ask ourselves what properties the ‘operators such as γάρ’ share, in addition to triggering this mismatch between form and interpretation, it is readily apparent that they are all enclitics.9 Enclitics are prosodically deficient elements which must be ‘hosted’ on their left. If we rewrite sentences (12ab) with the operators in ‘expected’ position, given their scope interpretation, we get: 13a) γὰρ [ ἐπεὶ ἤιϲθεθ᾿ ἡμέραν τὴν κυρίαν // ἥκουϲαν ] 13b) γὰρ [ εἰ ἔϲτι Λοξίαϲ ] It is apparent that in these structures the enclitic operator has nothing to ‘lean on’ to its left, yet, because of its own prosodic deficiency, it must lean. The syntax, of course, has no idea that these elements are prosodically deficient,10 any more than it knows anything else about the phonological properties of the elements over which it computes. It thus happily builds the structures in (13ab) and passes them off to the phonology. The phonological computation, when building prosodic structure, has the power to reposition these elements. The most restrictive hypothesis—and the one strongly supported by the available evidence—is that this repositioning (a) is sensitive only to the prosodic, not syntactic, structures involved (thus allowing the deficient elements to seem to be placed ‘inside’ morphosyntactic constituents with whom they have no direct semantic relationship), and (b) is as local as possible within the domain of computation (i.e., if one is trying to prosodically phrase the BECAUSE clause, γάρ is moved as minimally as possible within that prosodic domain). The result of this is that the syntax constructs representations like those in (13ab). It is these representations that wend their way to the semantic interpreter. Obviously, the semantics will thus reflect the scopal relations the syntax has constructed. This representation is passed on to the phonology as well, which modifies it minimally, giving the surface strings in (12ab). The semantics never sees this representation, and thus the positioning of γάρ in this string is semantically inert. 9I won’t use the terminology ‘postpositive’ here, which seems like a needless complication—the status of the editorial accent marking is orthogonal to the prosodic deficiency of these elements, amply evidenced by their failure to appear in clause-initial position, or after the caesura or verse-end in trimeter verse. 10Note that prosodic deficiency is a phonological diacritic on the relevant lexical item; it does not equate with any specific accentual property.

13

6

A Note on Some Variable Particle Orderings

The system just posited, with elements arranged in the syntax in a manner in keeping with their scope relations, and enclitics subjected to a subsequent prosodic repositioning when not properly hosted within their domain, allows us to explain a number of otherwise puzzling facts regarding the ordering of enclitics in Attic Greek. Again, this is not the place for a comprehensive investigation of these matters—we will be limited to a few cursory remarks. There is a particle, γοῦν, in Attic Greek which is universally ascribed, in origin, to a sequence of the particles γε and οὖν, in that order, with the customary elision of the final vowel of γε. Indeed, some editors, and some manuscripts, render the particle as γ᾿ οὖν to reflect this origin. Interestingly, there are contexts in which the particles surface with οὖν earlier in the string than γε, contrary to what one might expect given the univerbated γοῦν. One such context, as Denniston emphasizes (1950:422), is after clause-initial negation, when we get the sequence οὔκουν... γε, which Denniston identifies (following in part earlier scholarship) as “the negative form of γοῦν.” But why should following the negation trigger a re-ordering of the particles? οὖν is a strongly second position enclitic, which may be assumed to take the clause it is within in its scope (with the same provisos as hold for γάρ above), and to undergo the requisite prosodic inversion. By contrast, γε is an emphatic particle, taking a constituent (or sometimes an individual word) as its scope. It is likewise expected, as an enclitic, to undergo prosodic inversion. Thus, in an example such as: προλάζυμαι γοῦν τῶι χρόνωι τῆϲ ἡδονῆϲ. ‘therefore I get in advance at this time the pleasure’ (Eur. Ion 1027) we would assume that the syntactic representation of the string was: [οὖν [γε προλάζυμαι] τῶι χρόνωι τῆϲ ἡδονῆϲ] with οὖν ‘therefore’ taking the entire clause as its scope, and γε emphasizing the verb προλάζυμαι ‘I receive in advance’. Neither οὖν nor γε is properly hosted on its left in this string, so we get a prosodic derivation in the phonology along the lines of: [οὖν [γε προλάζυμαι] τῶι χρόνωι τῆϲ ἡδονῆϲ] > [οὖν [προλάζυμαι=γε τῶι χρόνωι τῆϲ ἡδονῆϲ] > [προλάζυμαι=γε=οὖν τῶι χρόνωι τῆϲ ἡδονῆϲ] which, with the expected elision of γε, will yield the attested sentence. What do we expect to happen if such a structure interacts with what surfaces as sentence-initial negation? Here’s an example: οὔκουν θραϲεῖά γ’ αὐτὸν ἐλπὶϲ ἀμμένει. ‘In fact an encouraging prospect does not await him.’ (Eur. Andr. 444) οὔκουν is, of course, the univerbation of οὐ(κ) and οὖν. The input structure for this string would appear, given its interpretation, to be: 14

[οὖν [οὐ(κ) [γε θραϲεῖα] αὐτὸν ἐλπὶϲ ἀμμένει]] Which will go through step-wise prosodic derivation of the form: [οὖν [οὐ(κ) [[γε θραϲεῖα] αὐτὸν ἐλπὶϲ ἀμμένει]]] > [οὖν [οὐ(κ) [[θραϲεῖά=γε αὐτὸν ἐλπὶϲ ἀμμένει]]] > [οὐκ=οὖν [θραϲεῖά=γε αὐτὸν ἐλπὶϲ ἀμμένει]]] to which additional accentual rules apply to produce the attested string. The ordering behavior of οὖν and γε is thus not the result of explicit stipulation, with a special condition holding in the case of initial negation, as the traditional account has it, but rather follows directly from general principles. In fact, the only reason the negation appears to be playing a special role here in triggering οὖν ... γε rather than γ’ οὖν order arises from its frequent clauseinitial placement. We would expect, then, that other frequently clause-initial elements would trigger the same switching of order between the two particles. And this in fact appears to be the case. Thus, it is not unusual to find clause-initial X=μὲν (where X, or its constituent, is usually contrastively focussed with the δέ-marked element of a following clause). If we find emphasis (let’s say of Y) by γε and an οὖν in such a clause, we do not get *X=μὲν Y=γοῦν, as the following example and its derivation show. ὠχρὸν μὲν οὖν οἶμαί γε καὶ κακοδαίμονα. ‘In fact I think (he’ll be) pale and ill-fated.’ (Ar. Nub. 1112) οὖν [[μὲν ὠχρὸν] [γε οἶμαι] καὶ κακοδαίμονα]] > οὖν [[μὲν ὠχρὸν] [οἶμαί=γε καὶ κακοδαίμονα]] > οὖν [ὠχρὸν=μὲν οἶμαί=γε καὶ κακοδαίμονα] > ὠχρὸν=μὲν=οὖν οἶμαί=γε καὶ κακοδαίμονα Other particle complexes show similar systematic behavior. Thus οὔκουν and μένουν, if some other element takes initial position, will show up as X οὖν οὐ(κ)...and X οὖν Y μὲν. This is common with certain usages of ἀλλὰ, as well as with clause-initial WH-elements, as the following examples show: φιλεῖϲ; τί οὖν οὐ κατεκλίνηϲ ὦ Μύρριον; (< οὖν [τί οὐ κατεκλίνηϲ]) ‘Are you in love? Then why don’t you lie down, Myrrhinekins?’ (Ar. Lys. 906) ἀλλ᾽ οὖν ἐγὼ μὲν δαίμοναϲ μαρτύρομαι || ὡϲ... (< οὖν [ἀλλὰ [μὲν ἐγὼ] δαίμοναϲ μαρτύρομαι ὡϲ...] ‘Then nevertheless I call the gods to witness that...’ (Eur. Med. 619) None of this is surprising, under the assumptions we are pursuing here, nor does it require overt stipulation.

15

7

The Role of Prosody

That the inversion operation we have just been discussing is prosodic in nature is clearly indicated by a number of phenomena, with respect to which it will now be useful to talk about what the phonological system does with the syntactic representation it gets handed. It will also be useful at this juncture to talk about the role of metrical texts in allowing us to recover relevant prosodic information. We will consider today only the spoken (rather than sung) Iambic Trimeter of Attic drama. This has the basic structure (a × symbol represents an anceps syllable—either quantity is permitted at that point in the line): ×

×

×

¯˘¯ ¯˘¯ ¯˘¯

There are three well-established structural properties of immediate interest in the trimeter. First, we have the line-end, indicated by the symbol above. Second, though not indicated above, almost every trimeter line has a caesura after either the fifth (more common) or seventh (less common) element:

˙˙¯˘¯ ¯˘¯ ˙ or ¯˘˙˙˙¯ ¯˘¯

×

×

×

×

×

×

¯˘¯ ¯˘¯

Finally, the metrical system allows resolution of a long as two shorts at certain positions in the line. Each of these metrical phenomena has prosodic implications, the line-end implying some degree of prosodic distance between the elements that straddle it (recall that prosodic structure is built out of syntactic structure), the caesura allowing (in my view) somewhat less prosodic distance to straddle, though still precluding, e.g., host + clitic sequences. The prosodic implications of resolution will be our immediate concern. Although a long syllable may be realized as two shorts in resolution positions, the two shorts must be in close prosodic nexus, and thus cannot normally be separated by a word boundary. Note that the caesura and line-end thus tell us that the elements which straddle them have some necessary degree of prosodic independence, while the resolution context tells us that the elements which straddle the resolution must have some degree of prosodic integration. The attested resolutions do include instances of words which we typically separate with a space in our modern editions straddling the resolution domain. Here are some instances (I will indicate the foot resolved across a ‘word’ boundary with in the schema, and mark the relevant ¯˘¯˘ text): 14a) καλόν γε :: τὸ::: γέραϲ τῶι ξένωι δίδωϲ, Κύκλωψ.

˘¯˘˘¯˘¯ ¯˙˙˙¯˘¯ ˘¯˘¯

‘Fine indeed the gift you have given the guest, O Cyclops!’ (Eur. Cycl. 551)

¯¯˘¯ ¯˙˙˙˘˘˘˘¯˘¯ ˘¯˘¯

14b) οἴμοι· τόδ᾿ ἤδη δόλιον ἐϲ ἐμὲ μηχανᾶι :::

‘Ah! This now is some trick you’re trying to pull on me.’ (Eur. Ba. 805) 16

14c) ... ˙:ὅ::: ϲε φυλάξαϲθαι χρεών.

˙˙¯˘¯˘˘¯ ¯¯˘¯ ˙

˙˙ ‘... you must avoid this’ (Eur. IA 989)

These passages (they do not exhaust the types of data) tell us that an article + noun sequence can be11 treated, as in (14a), as sufficiently close, prosodically, that the restrictions on resolution are satisfied. Similarly, a preposition + object sequence may do so (the pronominal status of the object in 14b is not relevant—note that it is the tonic pronoun in that example). Not surprisingly, clitics leaning on a host to their left also form a prosodically integrated whole (as in 14c). Note that resolution contexts provide evidence for these structures, but it would be a mistake to believe that these prosodic relations only existed when we had direct evidence. The genitive plural of the definite article, being a heavy syllable, cannot appear in contexts which give us ‘resolution’ evidence, but if definite articles can get phrased together with their nouns when we can use metrical evidence, we should assume that they can do so when we cannot (unless there is evidence to the contrary). Since the caesura and line-end require prosodic separation, it will come as no surprise that we do not normally find a definite article separated from its noun by these boundaries, nor a preposition from its object. And clitics are not found after the caesura or line-end. It is also not surprising, then that we find structures such as the following: 15) ἐν ἴϲωι γὰρ ἦν τόδ᾿, ... ::: ‘because that would be in a fair (way),...’ (Eur. IA 1199)

... ¯˘¯˘¯˘¯ ˘˙˙˙

The subordinator γὰρ, of course, has scope over this entire clause, and yet, as we have come to expect, it appears, for prosodic reasons, inside rather than to the left of the clause. The syntax built γὰρ [ἐν ἴϲωι ἦν τόδε], the prosodic computation created the requisite cohesion between the preposition and its object, thus γὰρ [ἐν=ἴϲωι ἦν τόδε], and when γὰρ sought out the nearest host in its domain, it found the clause-initial ‘phonological word’ ἐν=ἴϲωι. Note the order of the two prosodic computations—this will be of interest later. In many instances in which there is no evidence from resolution domains, the evidence from this type of prosodic ‘repositioning’ of clitics still reveals that prosodic integration has taken place. 16a) ἔξεϲτ᾿, ἐρώτα· τοῦ χρόνου γὰρ οὐ φθονῶ. ‘It’s allowed, ask, since I won’t begrudge you the time.’ (Eur. Hec. 238) 16b) ἐν ὧι γὰρ ἦν μοι πάντα, γιγνώϲκω καλῶϲ, κάκιϲτοϲ ἀνδρῶν ἐκβέβηχ᾿ οὑμὸϲ πόϲιϲ. ‘because my husband, in whom everything was for me, I know well, has proved the basest of men’ (Eur. Med. 228) 11It seems clear from data we will not address here that if there is focussing stress on the article or the noun, the prosodic relationship may be disturbed. The matter requires a serious study.

17

The latter example reveals in a striking way the fact that the prosodic repositioning cares little for the semantics of the structures it is manipulating. γὰρ ends up inside the relative clause (introduced with the ‘pied piped’ PP ἐν ὧι). The preceding clause ends with ‘I want to die’, but of course the relative clause doesn’t provide the reason for the desire to die—that comes from the matrix clause (‘my husband has proved the basest of men’). Because the relative clause is preposed to that matrix clause, γὰρ ends up ‘inside’ the relative clause. But the semantic weirdness that would be entailed by such a positioning is irrelevant: the semantic interpreter never knows that γὰρ ended up there! Note that this prosodic explanation, building on the resolution facts and the distribution of other enclitic elements, is vastly superior to the more tradition notion, as expressed, for example, by Cooper (2002: §2.69.14.9[10]): The position of γάρ is usually the second place in the clause. But if several words form one conception γάρ may follow this whole thought, and this kind of postponement occurs even in prose... The notion ‘one conception’ is not available to the syntactic computation, nor to the phonological one, and is hopelessly ill-defined. The examples explicitly cited by Cooper in support of his analysis include the following: 17) ἐν τοῖϲ Ἀλεξάνδρου γὰρ ˙ ὑβρίζεϲ δόμοιϲ

˙˙

‘because you indulged your haughtiness in Alexandros’ palace’ (Eur. Tro. 1020) The evidence supports the idea that a P + Art sequence can form a prosodic bond, and of course P + N may. Here we see the interesting sequence P=Art=NGEN sequence forming a prosodic unit. How one could conceive of this set of words as representing ‘one conception’— ‘in the of Alexandros’ —is beyond me. The interesting thing is not that Cooper is clearly wrong—we are all wrong about lots of things in the course of our scholarly lives—but that the architecture of the model demands that he be wrong, and an examination of the data shows that he is. Attic Greek doesn’t do what Cooper says it does, and no human language can. Such considerations reveal the tremendous advantage of having a clearly-defined set of principles for discussion of syntactic phenomena in Greek. The assumption that phrase structure is ‘flat’ or ‘sorta flat’, or that the meter ‘determines’ or ‘kinda determines’ word order, does not generate the set of principles needed to support serious scholarly discussion of the issues. The unhedged claims are false prima facie, and the hedged ones are free of coherent substance without explicit spelling out of what the hedges actually mean or entail.

8 Meter-Shmeter Examine the following passage:

18

ὁ δ᾿ οὐ θέλων τε καὶ θέλων οἴκτωι κόρηϲ τέμνει ϲιδήρωι πνεύματοϲ διαρροάϲ· “And he, for pity of the girl both reluctant and willing, cut the breath’s passageway with his sword,” (Eur. Hec. 566-7) By the arguments regarding scope and τε which we gave earlier, this should mean “NOT [both willing and willing]”, which is absurd, and certainly not what Euripides wanted his audience to think he meant. Clearly, the intended meaning is “BOTH [not willing and willing]”. Notice that if we ‘fix’ the order to give us the desired interpretation, we get: ὁ δ᾿ οὐ τε θέλων καὶ θέλων οἴκτωι κόρηϲ, which has an unmetrical short fourth element. So, some would have us believe, Euripides couldn’t write the line he wanted to, and had to say something absurd metri causa, figuring, I suppose, that his audience would say to themselves: “he can’t have meant that absurd thing, he must have meant X, which is almost what he said, anyway.” Under this conception of the relationship between the poet and his/her meter, the poet tries to construct a sentence which conveys the content of what he would like to say, but the meter often compels him to say something else instead. And rather than discard the line as not expressing the proposition s/he would like to express, s/he says it anyway. This strikes me as a denigration of the careful work of the poet, a failure to appreciate likely audience reaction to ungrammatical (or ‘wrong meaning’) strings, and, as a scientific proposition, entirely circular (since all lines scan, claiming that the need to scan determines order is self-confirming). Imagine a contrary scenario, in which the poet’s task—one which s/he takes very seriously— is to construct metrical lines that do express the content s/he desires at that particular juncture. In the case of the Rigveda (where similar issues arise), it is hard to believe anything else: would you really write a hymn to a god whose actions determine your fate that didn’t carefully and precisely express what you wanted to say? In the case of Attic drama it is worth noting that the dramatic context was a religious one, but it is probably sufficient to note that the audience would not appreciate a play which constantly misled them (by giving the wrong order of meaningful elements, or a screwy prosodification) as to the meaning of the utterances of the characters. Thus, rather than the rampant speculation the metri causa position imposes on us, how about this: the ordering must be legitmate, in linguistic terms, and must express a sensible meaning. If follows, as near as I can see, that οὐ must be proclitic to the following participle, creating the phonological word [οὐ=θέλων], so when τε ‘flips in’, the first hosting position it sees is after that phonological word. δὲ [ὁ

9

[τε οὐ=θέλων]

[καὶ θέλων] ...

What could word order metri causa mean?

Imagine that in our texts both the order ‘X Y’ and the order ‘Y X’ of the pair {X,Y} is attested. In addition, the two orderings scan differently, such that in a given metrical context, only one of the orderings is possible (and in another context, the other order is possible, of course). Since 19

under a sufficiently restrictive theory of syntax, distinct ordering entails a difference in interpretation, the orders can be assumed to express distinct meanings. I can imagine four basic types of scenario: Scenario I (no metri causa explanation licensed): the grammar generates: XY YX

with the meaning: A B

the text shows: XY YX

with the meaning: A B

Under this scenario, the poet has a linguistic system which generates strings with the sequence X Y with meaning A, and strings with the sequence Y X with meaning B. The poet uses these strings to express the meanings that they express. The composed lines scan, as all should, but the observed order is clearly grammar-driven, rather than meter-driven. Scenario II (minimal metri causa phenomena): the grammar generates: XY YX YX

with the meaning: A B B

the text shows: XY YX YX

with the meaning: A B A

!

In this scenario, the poet uses the Y X order both as the grammar would require (to express meaning B), and to express the meaning of the order X Y—a meaning that that sequencing is not assigned by the actual linguistic system. If the poet consistently does this in contexts in which Y X scans, but X Y would not, it would appear to be appropriate to attribute the ‘incorrect’ meaning for the Y X order to the effects of the meter. Scenario III (promiscuous metri causa phenomena): the grammar generates: XY XY YX YX

with the meaning: A A B B

the text shows: XY XY YX YX

with the meaning: A B B A

! !

In this scenario, regardless of which order the pair {X,Y} surface, the interpretations that the grammar generates for each ordering is available to the other. That is, the order appears to be being determined by the metrical requirements of the composed lines, while the meanings which the strings seem to bear appear independent of the observed order. Scenario IV (ungrammaticals surface, metri causa) the grammar generates: XY *Y X

with the meaning: A

the text shows: XY YX

with the meaning: A A

!

Under this scenario, one of the two orderings (Y X) cannot be generated by the grammar, regardless of intended interpretation. It is ‘ungrammatical’. Being an ungrammatical string, it has no linguistic interpretation. Nevertheless, this order is attested in the text, and it has the meaning that the XY order has. The orderings match the requirements of the meter, but the interpretations do not match the observed ordering.

20

In Scenarios II, III, and IV it would be appropriate to speak of metrically-conditioned word order. Note that, crucially, such an explanation seems licit only when there is a mismatch between the interpretation and the observed order. This makes sense, of course: when the order is appropriate to the interpretation, it would be silly to attribute that order to the meter (since any other order would fail to have the desired interpretation). Note additionally that it is impossible to advocate both ‘free word order’ and ‘metrically conditioned word order’. If Greek phrase structure were flat, any string of a given set of elements can be interpreted the same as any other string of those elements—indeed, it is the alleged observation of such a global mismatch between order and any specific interpretation (in, e.g., Warlpiri) that trigged the notion of ‘non-configurational’ languages in the first instance. Since the ‘free word order’ analysis says that in a Greek sentence it is only necessary that the words come in some order (and the order is not tied to interpretation), it is hard to see how you could prove that the meter was responsible for the observed order. In essence, every ordering has every interpretation, so the observed ordering does, and the poet could simply have used the grammar to generate the string in that order.

10 An Implication of the Foregoing Imagine that we have a language, like Attic Greek, in which we find in our texts both the order OV and the order VO. Presumably these orderings are linked to differences in interpretation. The metri causa explanation for the observed order centers on the presence of an interpretational mismatch between text order and grammar interpretation. To tell whether we are dealing with an effect of the meter, it is thus necessary to figure out what interpretation (or interpretations— because linear order can realize a variety of distinct hierarchical relations) is consistent with OV order, and what interpretation is found in the text. In the case of the contrast between OV and VO order in Greek I know of no one who believes they understand what the interpretational difference (or range of possible interpretational differences) is. Obviously our ignorance in this domain in no way entails that no interpretational differences are present: our analytical skills in the uncovering of subtle interpretational differences do not seem to me to be particularly good. But if we don’t know what kinds of meanings go with the observed orders, or if we believe that the orderings are unrelated to interpretation (as under the ‘flat’ phrase structure assumption), there can be no positing of metrical determination of order. Under non-flat assumptions, we need an interpretational mismatch to posit metri causa effects, but not knowing the interpretations in play, we cannot know whether there is a mismatch. Under ‘flat’ phrase structure assumptions, every order is perfectly fine, including the ones that scan, for any available construal of the elements in the clause. No metrical condition is thus required to account for the observed order. So, OV vs. VO, or any other dyadic pair, provides no insight in the effect of the meter on order unless we understand the interpretational contrast between the orderings. As we saw in our discussion of ‘scope’ relations, it does appear that Greek is quite sensitive to order when interpreting operators. Recall the following pair of sentences:

21

18a) καὶ πῶϲ ἂν αὑτὸϲ κατθάνοι τε καὶ βλέποι; ‘and how can the same (person) be both dead and be alive?’ (Eur. Alc. 142) 18b) καὶ Φρύνιχοϲ ... // αὐτόϲ τε καλὸϲ ἦν, καὶ καλῶϲ ἠμπίϲχετο. ‘and Phyrnichus...was himself beautiful, and dressed beautifully’ (Ar. Thesm. 164) In this pair, we see that the ordering relations between between αὐτόϲ (let’s not worry right now about the article in crasis in (18a)) and τε exactly match what I take to be the intended meaning. If we ‘flip’ κατθάνοι and τε, which of course could occur in either order (with differing interpretations), (18a) would mean ‘and how can both the same person be dead and s/he ((presumably not ‘the same person’) be alive?’ If we flip καλόϲ and τε in (18b), we would get the meaning ‘and Phrynichus himself was beautiful and (Phrynichus himself) dressed beautifully’. But the point of the sentence is not that Phrynicus emphatically had these two properties. It is that ‘being himself beautiful’ it follows that he dressed beautifully (and, as the text goes on to say, that his poetry was beautifully crafted). Moving the ‘himself’ out of the first conjunct destroys the interpretation of the passage, in my view. The matter is quite elaborate in some cases, and exceedingly consistent: 19) εἰ γὰρ ϲὺ μὲν παῖϲ ἦϲθ᾿, ἐγὼ δὲ ϲὸϲ πατήρ, ‘because if on the one hand you were the son, and (on the other) I your father [I would not have banished but killed you]’ (Hipp. 1024) We have several operators interacting here: εἰ, γὰρ, μὲν and μὲν’s correlate δὲ. The interpretation is completely clear. The input structure needs to have been that in (20) for this interpretation to hold: 20) γὰρ [ εἰ [ μὲν [ ϲὺ παῖϲ ἦϲθ᾿]] , [δὲ [ ἐγὼ ϲὸϲ πατήρ,] ] ] δὲ needs to be hosted on the left in its domain and is not, so it ‘flips in’ after ἐγὼ. μὲν needs to be hosted within its domain, and is not, so it flips in after ϲὺ. εἰ is not enclitic, so doesn’t move in the prosody, but once again γὰρ needs to be hosted on its left, is not, and thus must be flipped in to the position after εἰ. Although various orderings of the words in this clause (or words of the same grammatical categories) are attested in Euripides, any shifting of the order the operator elements would change the interpretation. Imagine, for example, that our clause instead said εἰ ϲὺ γὰρ παῖϲ μὲν ἦϲθ᾿, ἐγὼ δὲ ϲὸϲ πατήρ. This would mean ‘if, because you are one the one hand a child, and I on the other your father, [I would have not banished but killed you]’ Note that ‘if’ would go with the banished/killed clause, clearly not the intended interpretation. Imagine, rather, that it said εἰ μὲν ϲὺ γὰρ παῖϲ ἦϲθ᾿, ἐγὼ δὲ ϲὸϲ πατήρ: this would mean ‘on the one hand if, because you were a child, and on the other I was your father, [I would have not banished but killed you]’. English is a little unhelpful here, but the intended meaning requires that ‘if’ be inside the μὲν clause. Material (syntactically) inside the μὲν-clause is not to be interpreted 22

as part of the δὲ-clause as well (shared material comes outside the scope of μὲν, as with τε). So this ordering doesn’t mean ‘on the one hand if you were a child, and (if!) on the other, I was your father’. What it this ordering does actually mean is, I think, incoherent. In the structures I have examined, which should include all instances the placement of εἰ and γάρ in trimeter lines in Euripides, we find ‘Scenario I’ without exception. If the ordering were determined by the meter, or even highly influenced by it, this would be an astonishing result. If the ordering were ‘free’ and phrase structure flat, it would be nothing short of a miracle. The simple principle we should follow in such cases is this: since in all cases examined so far in which we can systematically test for interpretational mismatches between order and interpretation, there are none, in cases in which we cannot test, we should assume that distinct orders are tied in a grammatically-conditioned (rather than metrically-conditioned) way to distinct interpretations. Obviously, we should work to uncover what those distinct interpretations actually are, but, in the absence of a clear characterization, we are left with no evidence for metri causa effects, nor for ‘flat’ phrase structure.

References Cooper, Guy L. III. 2002. Greek syntax: Early Greek poetic and Herodotean syntax, Vol. 4. Ann Arbor, MI: University of Michigan Press. Denniston, J. D. 1950. The Greek particles. 2nd Ed. Oxford: Oxford University Press. Devine, A. M. and L. D. Stephens. 2000. Discontinuous syntax: Hyperbaton in Greek. New York: Oxford University Press. Diggle, James. 1981. Studies on the text of Euripides. Oxford: Oxford University Press. Dover, K. J. 1960. Greek word order. Cambridge: Cambridge University Press. Goldstein, David. 2011. Wackernagel’s Law in fifth-Century Greek. Doctoral dissertation, UCBerkeley. Hale, Ken. 1981. On the position of Warlpiri in a typology of the base. Bloomington, IN: IULC. Hale, Mark. forthcoming. Wackernagel’s Law: phonology and syntax in the Rigveda. . 1987. Studies in the comparative syntax of the oldest Indo-Iranian languages. Doctoral disseration, Harvard. Keydana, G¨otz. 2011. ‘Wackernagel in the language of the Rigveda: a reassessment.’ Historische Sprachforschung 124: 80-107. Wackernagel, Jakob. 1892. ‘Über ein Gesetz der indogermanischen Wortstellung,’ IF 1:333-436.

23