A comparison of the imperatives da 'take' and dai 'place' suggests that dai is here to be regarded as the bare stem. That, I imagine, is the reason why it has become customary to cite verbs of the third class in this form. When we observe, however, that verbs of the second class show such imperatives as memai 'say' and tarnai 'leave', as well as da and tarna, and that we have from consonantal stems (the first class) not only šak 'know' and the like but also pahsi 'protect', it becomes clear that i is a formative element in the imperative as well as in the indicative. In fact the imperatival i is not confined to the hi-conjugation, as witness kuenni 'strike, kill' beside kuenzi ‘he kills'.

Imperatival i, then, behaves in about the same way as the IE imperative ending dhi, of which Brugmann30 says: 'Der Ausgang dhi dürfte eine partikel sein, die mit dem als Imperativ fungierenden reinen Tempusstamm univerbiert worden war.' With this dhi we must certainly combine the ending t of Hittite imperatives such as it 'go' and wahnut 'turn'. Since there is no indication that Hittite lost final vowels, I suggest that there were originally two independent elements, dh and i, which might be used to strengthen an imperative; Hittite has preserved them both, and IE has amalgamated them to form dhi.

But what shall we say of the stem-final of dai 'he places', dai 'place', nair 'they turned', etc.? The obvious answer is that è became a before i; but of course we cannot derive dai 'he places' directly from PIE **dhe-e. That would have yielded Hittite *te or *ti. A plausible hypothesis is this: after the change of final e to i, the third personal ending of other hi-conjugation verbs was introduced in those of the third class, and then *de-i became dai. The history of the imperative would be parallel, except that the ending was i from the start. In the third pl. pret., however, the ending was er, as is shown by the frequent orthography -e-ir, and it is unlikely that *ne-er would yield nair. Here we must apparently assume analogical influence of the second class upon the stem vowel. This may be the correct explanation of the stem vowel a in the third sing. pres. and in the imperative also.

The remaining forms also of the third column of Table II were probably formed on the analogy of the second class, as follows:

The forms of the fourth column may result from a contamination of the second and third columns. Given pešti 'thou givest' and daitti

30 Grundriss 22. 3. 569.

'thou placest', it is not strange to find also paišti 'thou givest'. It may be, however, that we have here the suffix is appended to the stem in e, with change of e to a before i. The possibility that we should find here stems in ai from original ei was mentioned above (p. 223). I see no way of deciding between the three explanations.

If we are to cite Hittite verbs by their stems, it is difficult to choose between the various stems of the third class of the hi-conjugation. From the historical point of view the stem in e is most important, but for many verbs no forms from this stem are quotable. Perhaps the best plan is to choose the stem or stems that seem most characteristic of each verb; e.g. te- 'place', pe(s)- 'give', halzešš- 'call'. In doubtful cases it would be better to cite an actual form, e.g. šai 'he puts on'. At any rate we should no longer cite a secondary stem like dai- 'place'.

ON THE PHONOLOGY OF THE MESSAPIC DIALECT

J. WHATMOUGH

HARVARD UNIVERSITY

It is disputed whether Messapic is to be classed with the centumlanguages or with the satem-languages. Discussion of this problem has been unduly influenced by comparison with Albanian (a satemlanguage) and by the commonly made assumption that Albanian can only be the modern representative of an ancient Illyrian dialect. It is at least clear that Illyrian did not represent the IE palatal stops by sibilants, see Hirt, Indogermanen 2.609, and 'Stellung des Illyrischen' in Festschrift für Kiepert 181ff. (1894) where Kretschmer's view to the contrary is criticised. The fact that Messapic appears on the available evidence' to have nothing corresponding to the labiovelars or labials of the centum-languages (Lat. quinque, Gr. TEμTTÓS: Alb. pese) that represent the IE velar stops (*penke), would seem to make it a priori likely that the palatals should have been treated as in the satem-languages. But the conclusion is not inevitable. It is probable that Venetic, which was also an ancient Illyrian dialect, neither labialised the velars nor sibilised the palatals; and it has been suggested accordingly that Venetic belongs to a stratum of IE speech earlier than the cleavage into centum-dialects and satem-dialects, see Conway, Annual Brit. Sch. at Athens 8.152 (1901-2). If this view, which certainly fits the facts, could be accepted as demonstrated, it would be conceivable that Albanian might be descended from an Illyrian satem-dialect belonging to a later stratum of Indo-European speech. Other explanations, however, are not far to seek, cf. Giles in Camb. Anc. Hist. 2.26: it may be that the ancestor of Albanian has perished unrecorded; or perhaps that it was, as has been conjectured, closely related to, or

'The single plausible example penkeos, penkaheh[e (both proper names, gen. sg. masc., cf. Osc.-Lat. Pompeius, Lat. Quintus) is by itself hardly convincing, even if correctly interpreted. Messapic forms are cited from my forthcoming edition of the texts.

* Although I use here the traditional terminology I am, of course, aware of the bearing which the particular problem of Venetic and Messapic has upon the wider one of the IE gutturals.

even a dialect of, Thracian. It is at all events improbable that the ancestor of Albanian was identical with the speech of the ancient Illyrians, of which a recent valuable collection of local and ethnic names goes to confirm Hirt's view. Thus, side by side with Bapduλus (or Bápôvλis) on which almost alone Kretschmer based his opinion, seeing in ♪ [d?] the representative of an IE ĝ like Alb. ♪ as in bardi 'white', we now have also Illyrian Bargulam, Bápyaλa and other forms with g or y, in the face of which it is hard to believe that Kretschmer's account of 8 in Bapôús is correct. That Bapôvλs is necessarily cognate with Alb. bardi, Skt. bhra'jate cannot be proved; since, as a proper name, its original connotation is unknown. Ribezzo, who takes the same view of the treatment of the gutturals in Illyrian and Messapic as Kretschmer, rejects the etymology of Bapdúλλs in favour of the connexion proposed by Schulze with Illyrian Bardus. In view of the forms with g (y) there can be little doubt that & in Bapdíλs is a true plosive [d] and not a fricative [d]; but Ribezzo's own etymology of Messapic barzidihi (for Kretschmer cognate with Bapdíλλis and bardi) as derived from IE *brgh-, cf. Skt. brhánt-, breaks down when confronted with Messapic brigannas, cf. Kelt. Brigantes.

It is evident, however, that the several series of gutturals distinguished in Indo-European should, at the present stage of enquiry at least, be kept distinct so far as possible, since, so long as it is not clear whether Messapic labialised the velars or sibilised the palatals, it is impossible to say with which series the indeterminate gutturals (Brugmann's 'pure velars') should be regarded as coinciding in treatment; while, if the suggestion above mentioned be adopted, it will still be necessary, for purposes of comparison at all events, to make the same distinction.

It is also necessary to consider first forms in which it is claimed that IE gutturals or their Messapic representatives have already been identified. The familiar klaohizis, klohizis (in which ao or o stands for ou or ū as in orra: Uria, Ovpía and aoze, ozen: Uzentum, Ovčevtov) a 2nd sg. (?) optative of a sigmatic aorist from the root *kley- 'hear', must be set aside; since cognate forms with k appear both in Slavonic and Albanian, and we may have here to do either with borrowings or 'H. Krahe, Die alten balkanillyrischen geographischen Namen (1925). • Einleitung 265; Glotta 14.95n. (1925). 'Krahe 17, 83.

• Eigennamen 33n.

▾ Lingua degli antichi Messapii 23.

Cf. Whatmough, Proc. Camb. Phil. Soc. 130.2 (1925).

with an IE variation (k beside ) in the initial consonant. More convincing would be Hirt's comparison of vaikanetaos, gen. sg. masc., with Gr. oikos, Skt. veçá-ḥ, if we could be certain not only that uwas preserved in Messapic (valevas, later balevas: Valetium, veretahetis: Veretum, vasti: áσrv; oikoroihi being presumably a borrowing from Greek) but also that the diphthong or became ai. Contrast the ethnicon Poediculi and the proper names oibaliahiai[hi: Oebalia, i.e. Tarentum, Verg., G 4.125; Illyr. Oeplus, CIL 3.2891, 2900; oitinai[hi: Ital. Vtius, Vtilius, see Conway, Italic Dialects 188, 198, 257. Kretschmer apparently believes that baoçtas10 (cf. bosat, and perhaps blaozzezihi) contains an original guttural represented as a sibilant, cf. perhaps Illyr. Beucas CIL 3.7830. But, beside Illyr. Beuzas (ib. 9156), Beusas -antis (ib. x test., xi, xiv, xvii), Beuzetius -ia (ib. 9929), Ital. Buccia and Busius, Busidius, Bussenius (Conway, ID 155, 36, 34, 307) that view is untenable. The forms with c are evidently distinct from those with s, from which there is no reason to separate the Messapic forms. The same argument applies equally to daçtas (cf. dazet, dastas) which, if I understand Kretschmer aright, he regards with baoçtas as 'wertvolle Zeugnisse für den Guttural-Charakter des Messapischen'. Beside forms with -c-[k] Dacio, Daciscus, etc. (CIL 3), Duceus, Docetius (ID 34, 276) and the frequent Decius" we have forms with -s (s)- which there is no more reason to suppose connected with those with -c- than there is to suppose that -k- or even -ki- became -s- in Messapic, e.g. Dasius, Dassius, Dasiatus, Dasimius, Dasumius, all common both in Italy and in Illyria, where we find also Dasa, Dasas, Dazas, Dases, Dasianus, Dazanus, Dasmenus, Dasto, see Indices to ID and to CIL 3. Accordingly I connect the numerous Messapic names12 with s(f, s, ss, z, sz) with the Illyrian ones in s (ss, z) regarding the (Illyrian) names with c as quite distinct from both. There remains Kretschmer's view of barzidihi. Here again there are Illyrian forms, as we have seen, with g; and further, quite distinct from them, other forms with s which there is no reason to suppose represents an older g or k: Barsemis, Barsimia, Barsimsus (CIL 3.10307, D lxvi), and possibly (with an

Cf. perhaps Gr. ȧloow, root*yazik-?

10 On C and 3, see Whatmough Class. Quart. 19.68.

"I do not, of course, suggest that Dac-, Doc- (Duc-), Dec- are necessarily cognate; though Doc- may stand to Dec- in the same relation as docere to decère. 12 dassinar, daçta, daçtas, daoras, daszes, dazes, dazet, daze[, dašel, daševi, dazeziai, dazeh[i]as, dazes, daszes, dazihi, dazetis, dazetves, dazetoihi, dazihonas, dazimaihi, dazimas, dazomas, Aáσiμos, dazohonnihi, dazonnes, daços, dazohi (?) δάξου, δαζυ.