SIBLING TERMINOLOGY IN HOMER: PROBLEMS WITH ΚΑΣΙΓΝΗΤΟΣ AND ΑΔΕΛΦΕΟΣ*

Homeric Greek has two sibling terms: (1) κασίγνητος, a general term for any sibling; (2) ἀδeλφeός, a male ego’s male sibling. Existing scholarship has tried to draw various kinds of distinctions between the two terms, mostly linguistic. The restriction of ἀδeλφeός to male ego -- a new observation -- poses anthropological problems, but the solution proposed here is once again linguistic. In a nutshell, my solution is based on Perpillou's argument that ἀδeλφeός is a relic of a lost collocation *φράτηρ ἀδeλφeός "uterine brother", which must have dropped out of hexameter in the early Dark Age when it became metrically intolerable. Other issues in the relationship between the two terms are to be explained by the fact that ἀδeλφeός is an archaism, and not a neologism as one might think.

* I am indebted to David Rosenbloom and an anonymous reader for CQ for many useful comments and recommendations on an earlier version of this paper; and to Margalit Finkelberg for prompting me to investigate the topic in the first place. All faults are my own.

THE PROBLEM
In early Greek hexameter κασίγνητος/-η may refer to any sibling of anyone; but ἀδελφεός only ever refers to a male sibling of a male ego. 2 'Ego' in this context refers to the person from whose perspective the relationship is expressed: if the Iliad refers to Agamemnon as the brother of Menelaus, Menelaus is 'ego'. Henceforth the terms are abbreviated as κ. and ἀ., and relationships between specific individuals are expressed as 'ego-sibling', for example 'Menelaus-Agamemnon' means Menelaus = ego, Agamemnon = sibling. Table 1 shows occurrences of both terms in the Iliad and Odyssey. Occurrences of the compound αὐτοκασίγνητος 'sibling sharing both parents' are indicated in round parentheses, so for example there are 57 occurrences of κ., plus 7 occurrences of αὐτοκ., for a total of 64. No compounds of ἀ. are attested. The distribution for κ. is fairly consistent with what we would expect, based on the proportion of male to female characters in Homer: the ratio of brothers to sisters is 4.82:1 for κ. (or 6.64:1 for κ. and ἀ. combined), and the ratio of male ego to female ego is 3.00:1 for κ. (4.25:1 for both terms). But ἀ., although it appears 20 times, appears in only one of the four possible combinations.
As Table 1 indicates, in Homeric Greek κ. is the more commonly attested sibling term. In Classical Attic, however, ἀ. comes to be the sole term for all siblings, varying in grammatical gender as κ. does in Homer; by the Classical Period κ. appears only in poetic and archaic contexts (see Part 3,below).
Duplication of terms is not in itself unusual in the Homeric kinship system. Homeric Greek has a much greater variety of terms for lineal and co-lineal kin than Classical Attic, as is illustrated by the standard existing survey of Homeric kinship terminology, that of Gates. 3 This is in spite of the fact that Attic kinship terms are much better attested. In Homer, multiple terms for son and daughter also exist: alongside υἱός 'son' and θυγάτηρ 'daughter' we find κοῦρος, πάϊς, τέκος and τέκνον, all 'child'. 4 The most striking cases of duplication are the terms for systems it is customary to adopt a system of abbreviations to avoid confusion: the abbreviations followed here are B = 'male sibling', Z = 'female sibling', Si = 'sibling of either sex'; mB = 'male ego's brother', foSi = 'female ego's older sibling of either sex', ayZ = 'any ego's younger sister'; and so on. Some languages make no distinction between siblings at all: that is to say, they have only one sibling term, perhaps varying in grammatical gender. Examples of this are Greek of the Classical Period (ἀδελφός/-ή) and Spanish (hermano/-a). Both terms are adjectival in origin. Many European languages employ one distinction, [I] gender of sibling, and so have terms for B and Z (Eng. brother, sister; Rus. брат, сестра; Rom. frate, soră). Other languages use two distinctions, thereby having four sibling terms. Hungarian distinguishes by [I] gender of sibling and [III] relative age, with terms for oB (báty), oZ (növér), yB (öcs) and yZ (hug); Basque by [I] gender of sibling and [II] relative gender, with terms for mB (anaia), mZ (arreba), fB (neba) and fZ (ahizpa). Some languages use two or more of these criteria but do not have a separate term for every possible combination. For example, Turkish distinguishes by [III] age of sibling but, while older siblings are further distinguished by [I] gender of sibling (oB ağabey, oZ abla), there is only one term for all younger siblings (ySi kardeş). In Polynesian languages all three criteria are used, with [II] relative gender as the primary distinction; if the sibling is the same gender as ego, there is a secondary distinction by [III] relative age, but if the sibling is the opposite sex the secondary distinction is [I] gender; thus the terms available are moB/foZ (for example, Māori tuākana), myB/fyZ (tāina), fB (tungāne) and mZ (tuāhine).
All of these systems are conjunctive. Furthermore, they are economical: there are few redundancies, so that for any given combination of criteria it is possible to know exactly which sibling term applies.
When redundancy occurs, it is an anomaly that requires explanation. For example, in Māori, in addition to the four terms outlined above, there is also an English loanword parata 'brother' (aB). In that case, the explanation is very easy to find: it is the dominance of the English language in New Zealand education and culture. In such cases conjunctivity tends to override economy. Kronenfeld notes another case in Ghana, among the Fanti, who have two overlapping terms: one for mZ, and another for aSi. The explanatory factor is again socio-cultural, in this case an inheritance mechanism from a male ego to his sister's children: the mechanism establishes a special relationship between a male ego and his sister. But as Kronenfeld points out, even that redundancy is not enough to push the Fanti to economize the system by having one term for mZ and another for fZ/mB/fB: the latter term would violate conjunctivity. 12 The Homeric case is similar to that of the Fanti, being conjunctive at the expense of economy. In a more economical division, where ἀ. = mB and κ. = fZ/ eight terms is possible; therefore, there is a total of 4140 possible typologies for distinguishing between these terms. They bring this figure down to just twelve, a figure that agrees well with their empirical data, by adopting a set of careful assumptions, among which are (1) the exclusion of disjunctive systems, on the principle that they are cognitively costly and therefore rare; (2) abandoning 'gender of ego' in favour of 'relative gender'. For example, Polynesian sibling terminology (outlined below) would be disjunctive if analysed in terms of 'gender of ego', but in terms of 'relative gender' it is neatly conjunctive. 12 D.B. Kronenfeld, 'Sibling typology: beyond Nerlove and Romney', American Ethnologist 1 (1974), 489-506, at 500-1. mZ/fB, κ. would be disjunctive. Unlike the Māori and Fanti cases, however, the explanatory factor for the redundancy is difficult to find. It is important to realize that this duplication is different in kind from the duplication we find in other Homeric kin terms. In a case such as the six terms for 'wife', the duplication is total: each term is entirely interchangeable with the others. If κ. and ἀ. were used in approximately similar ways in Homer, it would be easy to regard them as just another simple duplication. As it is, we have a problem. Table 2 presents a fuller account of sibling terminology in early Greek. The table does not list attestations of either term in prose texts. Very little prose from the Archaic period survives except in testimonia and paraphrases, which consistently use Attic/koinē and are therefore useless for our purposes. Of prose earlier than 500 that does survive, κ. appears only in two or three inscriptions in the Cypriot syllabary; and ἀ. in three prose inscriptions dating to the latter half of the sixth century. 13 Table 2 does not include occurrences of the compounds πατροκασίγνητος 'father's brother' and ματροκασίγνητος 'mother's brother', since these are not sibling terms; 14 and it does not list compounds of ἀ., as none are used in Archaic verse. Attestations of κ. in 'other hexameter' refer to attestations in both complete and fragmentary poetic texts, and in two Archaic inscriptions. 15 Attestations of κ. and ἀ. in 'elegy' refer to hexameter lines (there are no attestations in pentameters);  Table 2 is again consistent with what we would expect, based on the proportion of male to female characters in Greek verse. When we look at the figures for ἀ., however, some additional problems emerge. First is the problem with which we began, that in Homer ἀ. is only ever used for mB.
A second problem is this: why does ἀ. scarcely appear at all in the corpus of non-Homeric hexameter? We find one use of ἀ. in elegy, in the 'new Archilochus' fragment published in 2005, which matches Homeric usage and, being Archilochus, is very early; but that is the only non-Homeric use in hexameter. This cannot be explained by small sample size: non-Homeric hexameter is approximately a quarter the size of the Homeric corpus, and we do have 21 occurrences of κ. in the same corpus. The absence of ἀ. is especially counter-intuitive given that it went on to become the standard term for all siblings in later Greek. It shows, at any rate, that we cannot solve our problem by supposing that κ. is the older term and ἀ. the newer one, since we should then expect to find a smaller proportion of ἀ. in Homer, the older corpus. 17 Instead, we have a new problem.
Third: why is the use of ἀ. so clearly circumscribed in Homer (and Archilochus), but not in lyric? In lyric, ἀ. is attested for two of the four possible gender combinations, and presumably could be attested in all four combinations if more lyric verse survived. This Homeric-lyric divide suggests a metrical solution to our first problem, above. The metrical solution will be duly considered -and rejectedbelow.
The main distinction between the two terms, outside Homer, has always been that κ. belongs to elevated language and epic, while ἀ. belongs to the language of prose. For example, Aristophanes uses κ. once only, and that in a passage parodying high tragedy (Thesm. 900); verse inscriptions use a mix of both terms, while prose inscriptions (except for the Cypriot syllabary) invariably use ἀ. This simple distinction neatly explains the second and third problems, why non-Homeric hexameter uses κ. exclusively, and why iambic and lyric poetry use both terms. However, it leaves unanswered the more central questions of why the Iliad uses ἀ. as much as it does, and why it is used only for mB. (The solution proposed in Part 4 solves these problems by turning this distinction on its head and concluding that ἀ. is the more archaic term; but, since that is a counter-intuitive hypothesis, I do not wish to draw on it yet, for fear of begging the question.)  17 Although early epics seem to be largely traditional, the study of R. Janko, Homer, Hesiod and the Hymns (Cambridge, 1982) indicates that there is at least some sense in which the Homeric epics are linguistically earlier than most other hexameter. Linguistic dating does not necessarily have a bearing on the dating of mythical content, but the latter does not directly concern us here. Table 3 presents fuller details of the attestations of ἀ. in Archaic verse. The discussion in Part 2, below, will refer back to Table 3 several times. One point that may be noted straight away is that in the Odyssey ἀδελφεός appears only in Book 4. A similar table for κ. is omitted as it would take too much space (though I have drawn up such a table).
But before the possible solutions are considered in detail, we have one final curiosity -if not an actual problem -to note. Pairs of brothers in Homer tend to be either ἀδελφεοί or κασίγνητοι, but not generally both, as Table 4 shows. In

HYPOTHETICAL SOLUTIONS
When faced with a problem like this a linguist might perhaps think first of etymologies as a place to look for an explanation; an anthropologist might think of social structures; a Homerist, of metrical formulae. These are good models for the kinds of areas we should look in for hypothetical solutions. Various considerations suggest the following as the best candidates: 1. Coincidence. The distinction is not statistically significant.
2. Metrical formulae. The words are used differently because of formulae characteristic of the Homeric hexameter. 3. Transitional change in vocabulary. The appearance of ἀ. in Homer represents a transition from a supposedly older term, κ., to a newer term. 4. Parentage. One of the two terms has a tendency to refer to half-brothers; alternatively, one of the two terms should be interpreted as a classificatory term, applying to cousins as well as siblings. 5. Inheritance. ἀ. indicates a mechanism of inheritance from ego to the children of the ἀ., or alternatively a mechanism of inheritance from the previous generation. Other hypotheses that are based on the idea of one term having a specialized meaning are, in general, very weak ones. Here are a few, exemplorum gratia: 6. Relative age (or status). One of the two terms refers to an older (or younger) brother, or a brother with greater (or lesser) status. 7. Closeness or sentiment. ἀ. refers to brothers who have a particularly close relationship, or at least a particularly close empathy at the moment of utterance. 8. Upbringing. ἀ. refers to brothers that have been brought up separately (or together), and stresses that they are none the less related. 9. Wooing. One of the two terms refers to a brother who woos a wife on behalf of the other brother.
These, too, are assessed below; but these hypotheses are suggested only by a priori considerations, and there is very little evidence to support them. Of these hypotheses, numbers 4 to 9 explore various ways of explaining the anomaly in terms of a distinction in meaning: they suppose that one term is general, while the other is more specialized. The following discussions, in rejecting these hypotheses, act as a kind of reductio ad absurdum argument: they show that the anomaly cannot be explained solely in terms of semantics.
Even if we could successfully demonstrate a hypothesis based on meaning, we would still be faced with the secondary questions of why ἀ. is avoided in other Archaic epic, and why both terms are used seemingly indiscriminately in lyric and iambic. A distinction based on register (κ. is epic, and ἀ. is prose) easily explains these secondary problems; but then that leaves the central question unexplained. It turns out that solving both sets of problems at once is challenging.

Coincidence
Pro. This first hypothesis is that there is, in fact, no problem at all. Gates already noticed that ἀ. only ever refers to brothers in Homer, and he was not troubled by the discrepancy; the gender of ego should not trouble us either.
Contra. If our 20 occurrences of ἀ. were a statistically representative sample, we should ideally expect them to include c. 13 occurrences of ἀδελφεός with male ego, about three with female ego, about three of ἀδελφε(ι)ή with male ego, and perhaps one occurrence of ἀδελφε(ι)ή with female ego. That one or perhaps even two of these permutations might not be represented could perhaps be explained as coincidence; but three out of four is asking too much. A possible explanation for the absence of two of the four permutations is that ἀδελφεή 'sister' is not easily accommodated in dactylic hexameter. However, three forms of ἀδελφεή could be accommodated by correption, the common metrical phenomenon of shortening a long vowel at word end when followed by a vowel in the next word, as we in fact find with ἀδελφεοί in Iliad 15.187 and, presumably, in the new Archilochus. 19 Moreover, any form of ἀδελφεή could be accommodated in hexameter by lengthening, viz. ἀδελφειή, as we actually find in the masculine ἀδελφειοῦ (Il. 5.21,7.120,13.788). There the lengthening of ε is a side effect of the contraction *-όο > -οῦ: the older form *ἀδελφεόο was perfectly metrical, but a rigidly regular hexameter could not tolerate *ἀδελφεοῦ. 20 There is nothing to stop the exact same process happening with ἀδελφεή; indeed we do find late poets using ἀδελφειή by analogy with ἀδελφειοῦ. 21 It would be arbitrary to accept correption of ἀδελφεοί and lengthening in ἀδελφειοῦ while denying these on principle for ἀδελφεή.

Metrical formulae
Pro. Parry himself considered that formulaic epithets carry no meaning of their own, but are purely rhythmic devices. The relative frequency of κ. and ἀ., and the near-complete absence of the latter term in non-Homeric hexameter, suggests that κ. is the most usual term, and that ἀ. is more closely connected to metrically specific situations. As Table 3 shows, 19 of the 20 occurrences of ἀ. are in the same metrical position, after a third-foot trochaic caesura. Some lines add an extra hemipes (ἀδελφειοῦ), but this is not significant as it is metrically necessary.

Contra.
Words often gravitate to one metrical position in Homer, but that is certainly not the same thing as a metrical formula. In fact all of the occurrences of ἀ. in Homer are independent of one another except for four or perhaps five. Three (or four) are formulaic: Iliad 6.61 ≈ 7.120 = 13.788 (≈ 10.72, though this is more dissimilar). The formulaic lines are of the form ὣς εἰπὼν παρέπεισεν (or ἔτρεψεν) ἀδελφειοῦ φρένας ἥρως. 22 Two others are part of a repeated passage referring to the same individuals (Iliad 13.695 = 15.334). But even if we exclude repetitions, we are still left with sixteen independent uses of the term.

Transitional change in vocabulary
Pro. ἀ. is the standard term for all siblings in Classical Greek. Given the greater frequency of κ. in earlier Greek, it looks as though κ. had been the standard term for all siblings, and that ἀ. is a late arrival. Its appearance in Homer could represent an intermediate stage in a transition from κ. to ἀ.
Contra. ἀδελφειοῦ is extremely early, predating the contraction *-όο > οῦ. 23 Furthermore, on this hypothesis we should expect an increase in the use of ἀ. over time, displacing κ.; instead we find the opposite. The avoidance of ἀ. in later hexameter could be waved away as false archaism, but there is no actual 20 On this process see Janko (n. 17), 87-8 and 248-9 n. 41; id.,

Parentage
Pro. The notion that the two terms indicate some distinction between siblings on the basis of which parents they share, or how many parents they share, is the solution that has been suggested most frequently. There are four main alternatives here.
(a) One of the two terms means 'half-brother' (that is, only one shared parent), while the other means 'full brother'. 24 This is the most general form of this hypothesis.
(b) More specifically, ἀ. means 'uterine brother', as suggested by the word's etymology (see Part 3, below); earlier still it would have meant just 'uterine'. By implication, κ. would be more general, which is why the exact relationship is sometimes clarified by an adjective (κ. ὄπατρος 'agnatic brother', κ. ὁμογάστριος 'uterine brother', and other constructions beyond Homer). 25 It is also sometimes clarified by the compound αὐτοκασίγνητος 'full brother'. 26 (c) κ. means either 'agnatic sibling' or 'uterine sibling'. If the former, the occasions where κ. is qualified with ὄπατρος would be emphatic, while ὁμογάστριος would represent a reversal of the more fundamental meaning; if the latter, then vice versa. 27 (d) We should interpret κ. as a classificatory term, referring to collateral as well as co-lineal relatives, including cousins as well as siblings. 28 At Iliad 15.545 Hector gives commands to his κασίγνητοι, including one Melanippus who is in fact a cousin. One scholion reports that this classificatory usage is Ionian, and a use of κ. in Herodotus 4.104.3 seems to confirm this. Moreover, in Homer κ. is regularly juxtaposed with ἔται, a term for collateral kin, and with ἕταρος, perhaps also to be construed as a kin term. 29 The interpretation of κ. as classificatory stems ultimately from Fustel de Coulanges's argument that early Greece had a clan system, and his interpretation of γένος as 'patrilinear clan' to match his analysis of the Roman 24 This is the presumption of Miller (n. aristocratic gens. 30 In support of this hypothesis, the only reflexes of PIE *b h réh 2 tēr 'brother' in Greek -φρήτρη, φράτηρ, and related words -undoubtedly refer to an institution much larger than the nuclear family, though its exact nature remains very uncertain. 31 Contra. Gates has already assessed these suggestions, and firmly rejected all of them. 32 (a) Both terms are used indiscriminately for full brothers and half-brothers. Out of 22 occurrences of κ. where both parents are certainly known (or where αὐτοκ. is used, which implies that both parents are shared), eight refer to half-brothers; 33 and as Table 3 shows, out of 19 occurrences of ἀ. where both parents are known, four refer to half-brothers. This does not indicate any preference one way or the other. In particular, note the cases of Hector, Paris and Helenus: all three are full brothers, but Paris is an ἀ. to Hector, while Helenus is a κ. (Iliad 6.102, 7.48). (b) A glance at the last column of Table 3 will show that the etymology of ἀ. cannot possibly have any bearing on siblings' parentage. On the four occasions where ἀ. refers to half-brothers they are agnatic, not uterine. Homer and Archilochus preserve not the slightest trace of the word's etymology. 34 (c) The suggestion here is that κ. stresses either shared paternity or shared maternity. The fact that κ. can be qualified by both ὄπατρος and ὁμογάστριος makes this a very doubtful proposition in the first place. In Homer κ. refers to a sibling who shares both parents with ego 10× (including αὐτοκ. 3×); shared father but uncertain maternity 11× (including αὐτοκ. 2×); shared father but certainly different mothers 10× (no uses of αὐτοκ.); and shared mother but uncertain or problematic paternity 3×. The last category, where shared maternity is stressed, is the most pertinent here. In Iliad 24.47, κ. appears with ὁμογάστριος; Iliad 19.293, Briseis refers to her 'three brothers [κ.], whom one mother bore along with me'; Iliad 3.238, Helen refers to her brothers Castor and Pollux as αὐτοκασιγνήτω. The last reference raises problems over paternity because of the legend that the paternity of the pairs of twins Helen/Clytemnestra and Castor/Pollux is divided between Zeus and Tyndareos. αὐτοκ. would normally stress the fact that both parents are shared, and Helen's use of the term is the more striking for that. These figures firmly disprove any link between κ. and shared maternity. Shared paternity is undoubtedly the most common way of thinking of sibling relationships in Greek; but the fact that a subtle joke about it can be put in Helen's mouth implies that it is a generalization based on social norms, rather than a linguistic necessity. 30 N.D. Fustel de Coulanges, The Ancient City, trr. A. Momigliano and S.C. Humphreys (Baltimore, 1980(Baltimore, [1864). 31 φρήτρη is paired with φῦλον 'race, tribe' at Il. 2.362-3 ('so that φρήτρη may aid φρήτρη, and φῦλον aid φῦλον'), and ἀφρήτωρ appears at 9.63. On the meaning of φρήτρη and its obscure relationship with the Attic phratry, see Andrewes (n. 29); Donlan (n. 1); S.D. Lambert, The Phratries of Attica (Ann Arbor, 1993), 269-71. On the question of whether *b h réh 2 tēr was classificatory or not, see n. 73, below. 32 Gates ( In any case early Greek had a separate term for 'agnatic brother', ἀμφιμήτωρ. The term is not used in epic but appears in Aeschylus, probably as an archaism; later poets follow his usage. 35 (d) This is the hypothesis that κ. was originally a classificatory term. However, there is now an overwhelmingly strong consensus against the idea that Greece ever had a clan system. In particular, Donlan and Patterson have shown that the tradition of interpreting γένος as 'clan' is without foundation, and that Dark Age kin groups revolved around the οἶκος rather than either clans or nuclear families. 36 The precise referents of φρήτρη and ἀφρήτωρ in Homer are obscure, but since they refer to structures larger than the οἶκος, they have no implication for the meaning of κ. 37 In pairings of κ. with ἔται and ἕταρος there can be no presumption of hendiadys: compare κ. ἄλοχόν τε, used 2× of Zeus-Hera, where it is plain that the intent of the formula is to distinguish the two terms sharply, and also the Hesiodic opposition of κ. to ἕταρος. 38 Moreover, some passages that stress co-lineal kin, at the expense of or even to the exclusion of collateral kin (Il. , choose to use κ. rather than ἀ. 39 The result is that out of 64 appearances of κ. in Homer, the evidence for a classificatory meaning depends entirely on Iliad 15.545. The interpretation of this line is bound up with the question of whether κ. was classificatory in spoken Ionic, and so a dedicated discussion of the point is deferred to Part 3, below.

Inheritance
Pro. This hypothesis is based on Kronenfeld's explanation for the overlap of terms for mZ and aSi in the case of the Fanti. In that case, the motivation for the specialized term was a mechanism for inheritance of property from a male to his sister's children; the brother-sister relationship was sufficiently important to offset the cost of the redundancy. Similarly in the Homeric case, we could expect to see some indication of a special relationship between the male and his brother's children, or some evidence of inheritance from a paternal uncle.
Alternatively, the specialized term could refer to a mechanism by which one or both of the brothers inherit property from the previous generation. At Iliad 15.187 ἀ. is used of Poseidon's brothers Hades and Zeus, and in a context that discusses how these three gods divided the universe up between them. This is also the only Homeric attestation of a three-way ἀδελφεός relationship.

Contra.
If the first of these alternatives held water, we should expect to find that 'brother's son' and 'father's brother' were important kin terms in Homer. In the Classical period we do find Herodotus routinely using ἀδελφιδέος 'nephew' (= Attic ἀδελφιδοῦς); and, moreover, he consistently uses it to refer -with one exception -to the son of a brother, not the son of a sister. However, the term does not appear at all in Homer, and in fact appears only once in the Archaic period. 40 The term for 'father's brother', πατροκασίγνητος, does appear three times in Homer; but that is a compound of κ., not of ἀ. Moreover, it appears only in contexts that rule out any possibility of inheritance: the uncle is always Poseidon, and ego is also divine (Apollo at Il. 21.469; Athena at Od. 6.330, 13.342). Since all parties involved are immortal, succession and inheritance are impossible.
The second alternative, a mechanism for distributing inheritance from the previous generation, would be supported if we were to find that either ἀ. or κ. has a semantic link with passages that discuss division of patrimony. But there is a serious shortage of brothers with a divided patrimony in Homer. The only pair of brothers in the main narrative who could be construed as having a divided patrimony are Agamemnon, who rules over Mycenae, and Menelaus, who rules Sparta. Their contingents are listed separately, though juxtaposed, in the Catalogue of Ships (Il. 2.569-90). But Menelaus has not inherited Sparta; rather he married into Tyndareos' family. Moreover, he and Agamemnon are making war on Troy jointly, fighting as a united family unit; if anything they would be evidence of a clan mentality, not of clearly separated nuclear families. 41 In genealogies and the like we do find a few cases where one brother migrates to a different city. This could represent either a divided patrimony, or one brother migrating because he has not received a patrimony. 42 We find separate dynasties of Assaracus-Capys-Anchises, who rule over Dardania, and Ilus-Laomedon-Priam, who rule Ilios (in Aeneas' genealogy, Il. 20.215-40); there is a division between Pelias, who remains ruler of Iolcus, while Neleus migrates to Pylos (Od. 11.253-7); and Melampus and Bias both migrate from Pylos to , though in later traditions each of the two takes a third of Proetus' kingdom. This is sparse evidence. Neither ἀ. nor κ. appears in any of these contexts. Furthermore, in the Catalogue of Ships we have several pairs of brothers who have conspicuously not migrated or divided their inheritance: Ascalaphus and Ialmenus (Il. 2.511-16); Schedius and Epistrophus (2.517-23); Phidippus and Antiphus (2.676-80); Podalirius and Machaon (2.729-33 and elsewhere); Adrestus and Amphius (2.828-34, killed at 11.329-32); Amphimachus and Nastes (2.867-75). Some of these are minor characters, arguably too minor to tell us anything definite, but not all: Podalirius and Machaon are traditional figures, and Ascalaphus' death  is an important foreshadowing of Sarpedon's.
In short, the hypothesis relating to inheritance from the previous generation rests entirely on a single reference, Iliad 15.187; and in all of the places we might hope to find support, there is none. 40 Alcm. fr. 88.1 PMG. 41 See hypothesis 4, above, on the thoroughly debunked notion that early Greece had a clan system. 42 A useful discussion of division of patrimony in historical early Greece is found in Sallares (n. 36), 196-8, 204-5. We have now exhausted all of the hypothetical explanations that are at all likely. The remaining hypotheses are given exemplorum gratia: they are unsupported by any real evidence.

Relative age or status
Pro. Relative age is one of the three typical criteria that distinguish sibling terms, as we saw in Part 1. A relative age term is therefore an obvious thing to check.
A possible, but tenuous, alternative is that κ. indicates an inequality in relative status. For example, Paris is consistently Hector's ἀ., while Helenus, arguably of lesser status, is Hector's κ.; Teucer is inferior in combat to his half-brother Ajax, and κ. is the term used of him. Perhaps κ. indicates a situation where the brothers are very unequal, while ἀ. indicates a more equal relationship.
We might allow the possibility of more flexibility in a less standard distinction like that of relative status, but there is no consistency there either. Among the gods, the sibling relationship of Apollo-Artemis is that of κ.; likewise Death-Sleep. But these are gods who routinely go together as pairs, without notable or consistent distinctions in status. (Among the gods the only ἀδελφεοί are the trio of Poseidon-Zeus-Hades.) At the mortal level, too, siblings who go together as a pair are κασίγνητοι as often as not: Acamas-Archelochus, Melampus-Bias.

Closeness or sentiment
Pro. Perhaps ἀ., as the less frequent term, is used to indicate a particularly close relationship between the two brothers; or, failing that, perhaps it indicates a particularly strong sentiment being expressed.

Contra.
Teucer is a κ., but the relationship between him and Ajax is clearly very close. ἀ. certainly is not used at moments of great sentiment: the one occasion on which Agamemnon refers to Menelaus as κ. (Il. 4.155) is also the moment where he expresses the greatest affection for him, when he thinks his brother is dying and vows to take vengeance.

Upbringing
Pro. The etymology of ἀ. suggests that even if it does not specifically mean two siblings from the same womb, it could be used to stress the common origin of two brothers who have been brought up separately. Paris and Hector would be a good example of this: Hector was brought up in Troy, while in later legend we hear that Paris was raised by the herdsman Agelaus. 43 Contra. The example of Agamemnon and Menelaus is sufficient to dispel this hypothesis.

Wooing
Pro. This last hypothesis is suggested by the repeated motif in many myths that one brother courts a wife on behalf of another brother. The story of Agamemnon's wooing Helen on behalf of Menelaus is particularly well developed in early epic , and is in addition the basis for the Trojan War. In Paris' case the pattern is adjusted slightly: he courts Helen in the company of a relative, Aeneas. 44

Contra.
Melampus also courts a wife on behalf of his brother, , but the term used is κ. (Odyssey 15.237; also Catalogue fr. 37.5). Though non-Homeric, κ. is also the term used in the Catalogue's account of Agamemnon wooing Helen on behalf of Menelaus (Catalogue fr. 197.5). At this point the hypotheses are starting to look rather desperate, so I shall stop here.

LINGUISTIC EVIDENCE
Although there are anomalous distinctions in the usage of κ. and ἀ. (Part 1), these anomalies cannot be explained by chance, or in terms of metrical formulae or meaning (Part 2). We are left with the task of reviewing potentially relevant linguistic evidence.
In Ionic there is some reason to suppose κ. may have had a classificatory sense, 'cousin' as well as 'sibling'. This supposition is based on two pieces of evidence: (1) Iliad 15.545 (mentioned above, hypothesis 4), along with a scholion reporting that κ. had a classificatory sense in Ionic; 54 (2) κ. in Herodotus 4.104.3 seems to point in the same direction. 55 Of these, Iliad 15.545 lends itself to alternative interpretations. A second scholion explains the usage as an elided form of the formula κασίγνητοί [τε ἔται τε], where Melanippus would be included among the ἔται. In a similar vein Gates has suggested, on the basis of πᾶσι μάλα in the following line, the interpretation 'Hector ordered his κασίγνητοι, [cousins and] all'. By itself, then, Iliad 15.545 is not strong evidence. As for Herodotus, 1.171.6 seems to point the same way as 4.104.3; but in 1.171.6, discussing the arrangements of a temple, it is evident that the context is flavoured with ritual language. These are the only two places where Herodotus uses κ. (he uses ἀ. 125 times); furthermore, κ. does not appear in any inscriptions in Ionian Asia Minor until the end of the Hellenistic period. 56 It seems clear that κ. in Ionic was a term associated with high, archaic, ritualized and/or institutional language. When Herodotus uses κ. it is not a kin term, but a marked term for a marked kind of relationship. Gates suggests -but does not himself believe -that the writer of the first Homeric scholion may have been mistakenly generalizing from Herodotus's usage (hence the scholiast's phrase 'some say that …'). 57 This suggestion is very likely right.
In brief, there are no firm grounds for supposing that κ. was ever a kin term in Ionic. However, its use in Herodotus -and confinement to Herodotus -provide good grounds for concluding that κ. referred solely to ritual and institutional 'brotherhoods'. In its extension from the sense 'brother' to the more institutional sense, the development of κ. exactly mirrors that of φράτηρ in Classical Attic.
In the Homeric use of κ. and ἀ., there are few metrical or formulaic points to note; see LfgrE for full details. 58 ἀ. gravitates to a position after a third-foot trochaic caesura, with the exception of Iliad 2.586, as seen in Table 3 earlier. ἀ. appears in both early and late contexts: early, shown by the survival of *ἀδελφεόο as ἀδελφειοῦ; 59 late, in one line, Iliad 15.187, that has a comparatively late synizesis on Ῥέα. 60 In terms of content, ἀ. appears in a key Cyclic passage at Iliad 24.736, where Andromache foresees Astyanax's death in a manner closely echoing a fragment of the Little Iliad; 61 Janko also sees Iliad 13.694-7 = 15.333-6 as late, because of problems surrounding the character of Medon. 62 κ. is responsible for two of the five (or six) violations of Hermann's Bridge in Homer, 63  occurrences in the vocative require lengthening of -ε before a masculine caesura. 64 It is used metaphorically on at least two occasions: Iliad 6.430 (Andromache refers to Hector as κ.); Odyssey 21.216 (Odysseus promises to make the herdsmen κ. of Telemachus); perhaps also Iliad 15.545 (the supposedly classificatory usage, discussed above). 65 None of this is particularly helpful.
The distribution of both terms in the Odyssey does call for comment, however. ἀ. appears four times in the Odyssey, but only in Book 4. This is not because Book 4 is the only part of the epic to refer to siblings; κ. is spread generously throughout the epic. 66 An Analyst perspective would simply interpret the confinement of ἀ. to Book 4 as evidence that the Telemachy was composed independently of the rest of the Odyssey; but the idea that the Telemachy was composed separately has always been far too simplistic a model of the relationship between it and the rest of the epic, given the extremely strong thematic links between both parts of the poem. ἀ. cannot straightforwardly be tied to an Oresteia narrative: two of its occurrences are in Menelaus' story (4.91, 4.512) but a third is spoken by Pisistratus (4.199) and the fourth is in the narrator's voice, in the background story to Helen's drugs (4.225). In short, it is not clear what to make of ἀ.'s confinement to Book 4.
As has been stated earlier, it is probably best to look for an explanation of the specialized Homeric use of ἀ. by starting with register: by the Classical period, κ. belongs firmly to elevated language, while ἀ. is the term used in unmarked language. A look at trends in the use of both terms over time shows this clearly. Table 5 shows the use of both terms in a range of poets prior to 400. In general, the higher the ratio of κ. to ἀ., the more elevated or archaic the poetry is in tone. The ratio is high in Homer; by the fifth century, Pindar is using both terms with almost even frequency. Aeschylus, though the earliest of the Attic tragedians, and in some ways the most archaic in flavour, uses κ. the least; his ratio is lower still if Prometheus Bound is excluded from consideration. This reflects a substantial move away from epic language. The poetry of Euripides and Sophocles creates a more self-conscious kind of archaism by using κ. much more frequently: that is to say, this is clearly false archaism, since they are reversing the trend visible in Aeschylus' language. Aristophanes' sole use of κ., at Thesmophoriazusae 900, exemplifies its elevated flavour: the entire scene is a parody of Euripides' Helen, with Euripides himself as a character and playing the role of Menelaus. 67 In Herodotus, too, one of his two uses of κ. is in a context that suggests ritual language (see above). Another point of note in the tragedians is the occasional use of Ionic/Doric ἀδελφεός for Attic ἀδελφός. ἀδελφεός appears twice in Aeschylus, but both passages are incurably corrupt; 68 it does not appear at all in Euripides; Sophocles uses it twice, both in sung Doric passages (OT 160, in a chorus, and OC 535, in a kommos). The rarity of the Ionic/Doric form even in choral passages perhaps reflects a perception that it was primarily Ionic.

A PROPOSED SOLUTION
As we have seen, an explanation must be found in terms of something other than pure semantics. We might have hoped to find a solution in the linguistic and other data presented in Part 3; but these data only reflect the very strong associations of both κ. and ἀ. with particular registers, something that was always known. Later writers of hexameter use both terms, but only in an imitative way. They eagerly build on Homer, but in their efforts at false archaism they end up using it in ways alien to Homeric usage: ἀ. with female ego, ἀδελφειή by analogy with Homeric ἀδελφειοῦ, and at an extreme, Quintus of Smyrna's thoroughly overdone form ἀδελφειοῖο. For reasons which will now become clear, these would not have been admissible in Homer.
As we saw in Part 1, the epic connotations of κ. and unmarked status of ἀ. neatly explain why Archaic hexameter beyond Homer avoids ἀ., and why lyric is equally comfortable with both terms. However, this leaves completely unexplained the frequent Iliadic use of ἀ., and the confinement of ἀ. to male ego. Conversely, explanations of the Homeric use of ἀ., even if they were convincing, would leave unexplained the question of why non-Homeric epic avoids the word so strongly. In short, any explanation needs both to account for the anomalies in Homeric usage of ἀ., and to allow for the strong association of κ. with epic and ἀ. with prose.
Our evidence shows that in hexameter poetry, ἀ. is concentrated in the earliest texts: most common in the Iliad, less so in the Odyssey; attested in Archilochus, but not in later elegiac poets. The use of ἀ. in hexameter decreases over time, while the use of κ. increases. This is, in fact, a reverse of hypothesis 3: our solution lies in the insight that ἀ. is not a newcomer after all, but a very archaic term. Its 67 αὐτοκ. also appears in Cratinus, in a choral passage (fr. 73 decreasing frequency shows that Homer is not introducing it into epic; rather, epic language is in the process of moving away from it. This would explain why we see ἀδελφειοῦ, remnant of such an early form as *ἀδελφεόο. For this word Janko's linguistic data suggest an extremely early date: on his relative chronological scale, where the Iliad and Theogony are defined as points 0 and 3 respectively, his data place ἀδελφειοῦ at - 6.7. 69 By itself this insight is not enough, however. It does not solve the problem with which we started: why is ἀ. confined to male ego? And why are some brothers regularly ἀδελφεοί while others are κασίγνητοι, with little overlap?
The solution lies in Perpillou's etymology of ἀ. Since ἀ. has its origins in the collocation *φράτηρ ἀδελφεός 'uterine brother', as a modifier to the gender-specific term φράτηρ < *b h réh 2 tēr 'brother', both parties in the relationship must be male; therefore, ego must always be male. In legend, then, we might well see ἀ. associated with old, traditional pairings of brothers, though not necessarily exclusively.
Therefore, the final hypothesis -hypothesis 10, if you like -is as follows. In Homer ἀ. is an archaism nearing the end of its lifespan, not a new term replacing an old term; in the Iliad we see a phase where κ. is in the process of displacing ἀ. from epic, rather than the other way round. Conversely, in unmarked (prose) language, ἀ. was already in general usage in spoken language; so while ἀ. became confined to spoken language, κ. became associated with marked speech, namely epic. This explains the affinity of κ. for hexameter verse, the reduction over time that we see in the use of ἀ., and the mixed evidence in lyric/iambic.
The sequence of events must have been something like the following. Originally the term for a brother in a pair of brothers was *φράτηρ ἀδελφεός 'uterine brother' (with the implication of 'full brother': see Sibling terms prior to the generalization of ἀδελφεός, below). Both parties, therefore, had to be male. As epic verse became regularized into the Homeric hexameter, formulae involving this collocation became intolerably unmetrical, and so the noun vanished in epic, leaving just the epithet behind to stand in as a substantive. 70 At some point (probably in the Dark Age, but perhaps earlier), as a result of the loss of φράτηρ, ἀ. lost the sense of 'uterine' so completely that it could be used of agnatic brothers without jarring. κ. must have been an import from another dialect; when it became available to epic poets, its exotic character already gave it a status as a marked term. It was therefore ideal for epic usage, and in time, came to displace ἀ. in epic. At the same time, ἀ. continued to be used in spoken Ionic.
In the time of the Iliad ἀ. is simultaneously dominant in spoken language, while as a specimen of epic language it is an archaism. This is why it is on the one hand starting to be avoided in hexameter, while on the other hand it retains an association with specific pairs of brothers. By the time of the Odyssey κ. is almost completely dominant, and ἀ. is used only in one context, Telemachus' visit to Menelaus (perhaps because it is a scene that resonates with a distant past; perhaps because an old Oresteia narrative, in which ἀ. was common, has somehow flavoured the episode as a whole; or perhaps for some other, less clear, reason). In later hexameter of the Archaic period, κ. has so strongly elevated a flavour, while ἀ. is so strongly associated with spoken language, that ἀ. is entirely avoided. 69 See n. 20. 70 For the loss of φράτηρ, cf. the parallels cited in n. 45, above.
Like hypotheses 1 to 9, this explanation has weaknesses and questions that need addressing. These issues are not nearly as intractable as those that faced the earlier hypotheses, however.
(a) Mythical content. The notion that ἀ. refers to relatively traditional pairs of brothers leaves a question over Medon, whom Janko regards as a late improvisation; 71 and some Neoanalysts might be unhappy at the notion that Hector and Paris are a traditional pairing, in view of the school of thought that sees Hector as an imitation of Memnon in the Aethiopis. 72 Hector does not of course need to be a late invention, but the issue is a real one, since *φράτηρ ἀδελφεός must have been very early indeed. In these cases the problem is not one of linguistic possibilities but of a poetic device. It is conspicuous that Hector and Medon also provide the occasions for the strangely inappropriate -late? -applications of ἀ. to agnatic brothers; it could well be that ἀ. is used to bolster these characters' status and make them seem more traditional. that ἀ. belonged to a class of noun-epithet formulae where an epithet of the rhythm ()  -  sits in the third-fourth feet and a substantive is deferred to line end. 78 This would be possible in the nominative singular and the dative, but not in the accusative or the nominative plural, and the genitive looks unlikely with a single hemipes separating the words. 79 A third possibility is *φρήτηρ at line beginning, in any form. That configuration for noun-epithet formulae is uncommon, but paralleled; for example, although μεγαλήτωρ after a third-foot caesura usually either immediately follows its substantive (27×) or precedes it (19×), it also appears with a line-initial substantive (5×).
The midline formula in the first possibility is relatively inflexible, and the hexameter prefers to avoid the separation seen in the second and third possibilities. If *φρήτηρ ἀδελφεός ever existed in epic verse, *φρήτηρ probably dropped out as it became more difficult to use it in a strict hexameter. 80

CLOSING REMARKS
There are several problems with the Homeric usage of κ. and ἀ., which revolve around the central issue that ἀδελφεός is only ever used for the male sibling of a male. A lack of economy in the metrical formulaic system could be blamed on poetic artifice; but that will not do for a lack of economy in the sibling terminology system. Even if a lack of economy were admitted, the strange specificity of ἀ. is not paralleled in later Greek and would remain problematic in its own right. Trawling through likely hypotheses yields no satisfactory explanation. In particular, we may be sure that the distinction is not purely one of meaning.
It is of course reasonable to object that the solution suggested here is just one possible explanation. A competing solution may yet present itself. Sherlock Holmes, quoted at the start of this paper, should perhaps have said that 'whatever remains, however improbable, is possible.' However, any competing solution has a formidable checklist of facts that it must explain: • that ἀ. is used only for mB in Homer and Archilochus; • that ἀ. tends to be associated with specific pairs of brothers; • that ἀ. is fairly common in the Iliad but is rigidly avoided in post-Homeric Archaic hexameter (other than Archilochus, who himself is very early); • that ἀ. is used freely and in a general sense in lyric poetry and in later prose, contrary to practice in hexameter poetry.
These are the observed data; finding an elegant explanation for all of them is no trivial matter. In my suggestion, Perpillou's reconstructed collocation *φράτηρ ἀδελφεός explains the first two points. The third and fourth are explained by the 78 On this class of separated formulae see J.B. Hainsworth, The Flexibility of the Homeric Formula (Oxford, 1968), 94-8. 79 Other possibilities: syncopated forms (*φρῆτρα, *φρῆτρες) could go at line end; so could the genitive if it were lengthened by analogy with ἀδελφειοῦ, viz. *ἀδελφειοῦ φρήτηρος. It so happens that the latter would be very assonant with the only formulaic use of ἀ. attested, also in the genitive (ἀδελφειοῦ φρένας ἥρως). 80 M.L. West, 'Greek poetry 2000-700 Bc', CQ 23 (1973), 179-92, at 188, suggests a 'loose' form of the hexameter c. 1100; the demise of *φρήτηρ would be shortly after that. realization that κ. was in the process of displacing ἀ. from epic at the time of the Iliad. Now, it is true that there is a fifth datum that my hypothesis fails to deal with elegantly: • the fact that the Iliad and Archilochus use ἀ. of agnatic half-brothers, contrary to the word's etymology.
As we have seen, this issue can be accommodated by positing either (1) that the Iliad uses ἀ. to bolster the status of brothers who are relatively recent inventions (especially Medon), or (2) that ἀ. had completely lost the sense of 'uterine' by the time of the Iliad, and its application to agnatic siblings is coincidence. It must be admitted that neither of these explanations is elegant. Even so, the other hypotheses discussed here suffer from exactly the same difficulty; and as difficulties go, this one is minor by comparison with the impossibilities involved in the other hypotheses. Much uncertainty remains. We cannot know for sure the exact sequence of developments; and there remain problems surrounding φράτηρ. We cannot know whether φράτηρ (or *φρήτηρ) ever appeared in epic, and it is very unclear whether φράτηρ and *b h réh 2 tēr were originally classificatory or descriptive terms. We have seen no evidence for a link with the meaning of φράτηρ in Classical Athens, and if such a link exists it can probably only be illuminated by new Athenian evidence. It can at least be stated that if prehistoric φράτηρ was a classificatory sibling term, then ἀδελφεός, in its original sense of 'brother sharing both parents', must represent a significant move from a classificatory towards a descriptive kin system; and this move must be very early, certainly no later than the use of *ἀδελφεόο in epic. Conversely, if prehistoric φράτηρ and *b h réh 2 tēr were descriptive terms, then the later change in the meaning of φράτηρ is one of a number of parallel transformations, where terms for 'co-lineal brother' develop towards an institutional 'brotherhood' sense: this change in φράτηρ in Attic is paralleled by *b h réh 2 tēr reflexes in other languages; it is also parallel to the changes in κασίγνητος in Ionic. However, it is unclear whether such a systematic shift would be due to linguistic or social reasons.