came to be understood as a part of the formative.

There is no reason for believing that this process of irradiation, which plays such a prominent part in the development of nominal suffixes and case endings, should not also have been influential in new formations based on verb stems, and yet little has been made of this in actual practice.

If, in the first place, we discard the connection of the Italic imperfect with IE *bheu- altogether, the next thought that suggests itself is that the suffix *-fā- Lat. -bā- gets its ƒ (b) from some root ending thus, from which it became detached by wrong analysis and spread to associated words. The correspondence of Lat. b and Osc. ƒ of course suggests that such a root must have ended in IE bh. Moreover, since -bam is always preceded by a long vowel (ā, ē, î), it is likely to be one in which one of these vowels preceded the bh. Of extant Latin verbs scribo 'I write' would be most serviceable as such a pattern, with which cf. Osc. scriftas Umbr. screihtor nom. pl. 'scriptae' and Gr. σkapıçάoμai 'I scratch'. As to the a of the formative, we would have to assume that Lat. er-a-m < *es-a-m and Italic *fā-m < *bhu-ā-m were not the only examples of their kind, but that there was a larger category of imperfects in-a- in existence at one time." Among these was Italic *scribh-a-m, *scrib-a-m if transliterated into Latin. This could have been misdivided scrī-ba-m and caused the formation of the similar sounding sci-ba-m from scio, and then also i-ba-m from eo. In support of this *scribh-a-m could be mentioned the abovе Gr. σкаρiçãoμa, which was based on the same IE a formation, and of which the imperfect oxapıyaóμŋv for *σkapupāμny would presuppose the same base as *scribh-ā-m. After Lat. sci-bam and ibam were then formed e.g. amā-bam and monē-bam, and after the latter in turn emēbam : emĕre and audiēbam : audīre.

However, while the patterning of sci-ba-m after *scrib-a-m is thus not at all impossible, there are in turn equally weighty objections against this hypothesis of the origin of the Italic imperfect. The assumption of a larger category of Italic imperfects in -ām, in spite of Gröber, Archiv. f. lat. Lexic. 1. 228, is anything but certain,10 and in spite of Gr. éσkapıpαóμηy it cannot be denied that an imperfect *scribh-ā-m Lat. *scrib-a-m is not by any means so well authenticated as an IE *bhu(u)ã-m, for which can be cited not only the Lithuanian preterite buvau pl. buvome, but also OBulg. ba (3 pl. conditional), Irish bg, and the old Latin injunctive fuam, fuas, etc., which is structurally identical with

On the possibility of this, cf. Stolz 289.

10 For more probable explanations, cf. Meyer-Lübke, Rom. Gram. 2, 282, and Lindsay, Lat. Lang. 491.

the Lithuanian and Irish forms and, eventually, Italic *-fa-m, although found only in volitive meaning." If we add to this consideration that *scribh-a-m 'I wrote' would hardly have been frequent enough to give rise to the whole category of Italic imperfects, we can say that the disadvantages of looking for the origin of the imperfect in this direction to some extent outweigh the advantages of thus being able to explain the stem form of the verb in amā-bam, monē-bam, and sci-bam as well as the durative meaning of the formation.

Let us now turn to the other alternative, that of so modifying the current opinion as to remove the difficulty of the stem form of the verb and the durative meaning. If this can be done, it is after all more probable that the imperfect came from *bhua-m, although it is not altogether out of the question that *scribhā-m played a subordinate part. Let us assume that Osc. fufans 'erant,' representing *bhu-bhu-ā-nt, is the oldest imperfect in -fā-. In Latin of course this would have become *fu-ba-nt, but erant < *era-nt later took its place. If this *bhubhya-nt Italic *fu-fant was a pluperfect as to origin, we at once have the solution of the difficulties. In the first place semantically. The perfect of 'grow, become' is equivalent to 'be', and its pluperfect is therefore equivalent to the imperfect of 'to be'. Thus in RV 2.12.9 yó víçvasya pratimā'nam babhū'va, sá janāsa indraḥ 'he who is a superior of everything, he, o men, is Indra' the Sanskrit perfect babhū'va is 'he has become and therefore is' and is therefore equivalent to ásti 'he is'. Similarly Gr. φύω in Il. 4. 484 ἀτάρ τέ οἱ ὄξοι ἐπ ̓ ἀκροτάτῃ πεφύασιν ‘branches have grown (i.e. are) on its top'. That the corresponding pluperfect is the equivalent of the imperfect of 'to be' is shown e.g. by Hes., Theog. 152 κεφαλαὶ δὲ ἑκάστῳ πεντήκοντα ἐπέφυκον ‘fifty heads had grown (and therefore were) for each one' i.e. 'each one had fifty heads'. This equivalence of the perfect system of *bheu- and the present of *es-, of the pluperfect of the former and the imperfect of the latter, was no doubt the starting point for the incorporation of forms of *bheu- into the paradigm of *es, and for the almost complete syncretism in some instances. In the Italic period, however, *bheu-, i.e. *fu-, was not yet found in the present system, and this makes it all the more probable that the perfect and pluperfect, in which Italic *fu- has displaced *es-, has been the connecting link between *es- and *bheu-. Reverting now to *bhu-bhua-nt Osc. fu-fa-ns its imperfect meaning from its perfect stem is fully explained by its being originally a pluperfect of *bheu-.

However, before this will appear really plausible, it will be necessary 11 So Brugmann 539.

also to explain its form. How can a form that looks as though its structure is identical with e.g. amā-ba-nt have been a pluperfect? The answer must explain two facts, the a of the base *-bhua-, and the first syllable *bhu-, fu-.

Of the two questions thus brought up, the first need not detain us long. That present or aorist stems make their way into the perfect system is such a common occurrence, that it need not cause surprise in this one instance. To mention only a few representative Latin examples taken from Brugmann, Gr. 2. 32.470, lambī gets its nasal from the present lambo, po-posci its sc from the present posco, and sternuī its nu from the present sternuo. It is therefore only one out of many similar cases if *bhu-bhua-nt got its a from the base *bh(u)uā- found in the Lithuanian preterite buvo (third pers.), Ir. ba (third sing.), Lat. fua-nt, OBulg. bạ (third plur. condit.).

The first syllable of *bhu-bhyant must have been the reduplication.12 It is true that the IE reduplicative vowel for the perfect probably was always e,13 and that Skt. ba-bhu'va shows that this particular verb had a reduplication with e in the IE period, while an e in the perfect reduplication of *bheu- in Italic times is assured by the Umbrian fe-fure1 (third pl. fut. perf.). No doubt *bhu-bhya-nt took the place of an earlier *bhe-bhya-nt in one of two ways. Possibly the practice in Latin of substituting for the reduplicative vowel the root vowel, wherever this was identical in present and perfect systems, as e.g. in cu-curri:curro, goes back as an alternative method of reduplication to the Italic period, for not only the Faliscan fi-fiked, but also Osc. fi-fikus shows the same type of reduplication as cu-currī and *bhu-bhyānt. In this case it would be necessary to assume that, as in the Skt. perfect su-şvā'p-a: svap-, the consonantal u of *bhu-bhya- had affected the vowel of the reduplication as much as a vocalic u; or else that *bhu-bhuyānt alongside of *bhe-bhyant had affected the reduplicative vowel of the

12 That *bheu- had a reduplicated perfect is of course beyond doubt in view of Skt. ba-bhū'va, Gr. ré-pūka, and Umbr. fut. perf. fe-fure (3. pl.). Against this evidence Lat. fui and similar forms in all languages which are prone to give up reduplication are of no moment.

13 Cf. Brugmann 24.

14 See Buck 170, on the reduplication in the perfects of Oscan and Umbrian. 15 If *fu- was already considered a perfect of *es- in Italic times, there was no identity of root vowel, as is pointed out by Buck. Whether the change of the reduplication on *fe- to *fu- would have taken place in a verb without any present tense, such as *fu- would be if not associated with *es-, is hard to say. Perhaps, in view of this additional difficulty, the second suggestion of the reason for the change of *fe- to *fu- is the more probable.

latter. However, the u of *bhu-bhya-nt may as well have been due to contamination with unreduplicated forms of the same root. After *bhe-bhya-nt had become *fe-fa-nt in the Italic period,16 its structure was obscured and the influence of obviously related forms like Osc. Umbr. fust (fut. perf.) or Osc. fusíd (impf. subj.). changed *fe-fă-nt to *fu-fă-nt Osc. fu-fa-ns," and this form displaced *fe-fa-nt because of the identity of its first vowel with that of forms like *fusēt.

No matter in which of these two ways the u of the originally reduplicative syllable of *fu-fa-nt arose (original *bhu-bhya-nt would of course also have become *fu-fă-nt), its structure became obscured, and the obvious similarity of *fu-fant to forms with radical *fu- caused the reinterpretation of its first syllable as the root, and necessarily left -fā- as suffix. Since this *fu-fa-nt was an imperfect in meaning, it was easy to attribute that in which this form differed as to meaning from other forms beginning with fu-, to that in which they differed as to form, and thus -fā- Lat. -bā- became an imperfect suffix. After this point had been reached, it became easy for this -fā- to spread to other associated verbs. Moreover, since this fu- looked like the stem or root of the verb, all other imperfects in *-fā-, which were patterned after *fu-fā-, added the imperfect ending to the root or stem. To the analysis of the naive speaker stā-ba-nt was constructed exactly like *fu-ba-nt, and similarly fle-ba-nt, i-ba-nt, sci-ba-nt. Finally the suffix could spread to dissyllabic verb stems, as in ama-ba-nt or mone-ba-nt.

With this interpretation of *-fa- the third and fourth conjugation imperfects in -ēbam or -iebam could not have been the original forms after which e.g. -abam was patterned,18 but were rather formed by analogy to those of the second. Occasions for such analogical formations were various. Some verbs had present forms both according to the second and to the third conjugations. Thus fervo fervere existed alongside of ferveo fervere,19 and the imperfect fervē-bam 'boiled, seethed', although derived 16 Cf. Brugmann 12.323.

17 Parallel with the development of the fa-imperfect went that of the f-perfect in Oscan-Umbrian. The oldest form of this I consider to be Osc. fufens 'fuerunt', which goes back to *bhe-bhy-ent with reduplication in e and secondary ending -ent instead of the original r ending. In the Italic period this became *fe-fent, which being structurally obscure was changed to *fu-fent through the influence of forms of the type Osc. fuid Lat. fuit. The next step here too was to mistake *fu- for the root instead of reduplication, and this left ƒ as a perfect suffix, after which it became possible to form similarly e.g. Osc. aíkda-fed (3. sing.).

18 So e.g. necessarily Sommer.

19 The analogical transfer of the ē is explained on the basis of the Latin, and may very well not have taken place till late in the development of the Latin language,

from ferveo, was referred instead to fervo, and caused by analogy ūrēbam: ūro 'burn'. At other times formations externally identical in second and third conjugation verbs caused the transfer of -ebam to the latter by proportional analogy. Thus the participles in -ēns -entis, e.g. monens monentis: moneo and legēns legentis: lego, were common to both conjugations,20 while in some verbs the formation of the perfect afforded a foothold. Thus vīdī vidē-bam (:video) = ēmi :emē-bam (: emo). Then in turn after the type eme-bam: emo also capie-bam: capio and audiebam : audio.

Whether Osc. fu-fa-ns, which thus represents the pattern for all Italic -fa- imperfects, was a solitary formation,21 or whether the category had become productive in Italic times, there is no way of deciding. Since Oscan and Umbrian show no other imperfect of any kind, it may be that -fā- was the general imperfect suffix of the Italic period, but as long as no other Oscan or Umbrian imperfect is forthcoming, there is no certainty. It is equally probable that Osc. fu-fans, which represents the pattern for the Latin imperfects, was still a pluperfect in Oscan, and that Oscan and Umbrian either had developed other types of imperfects of which no example is extant, or retained the IE imperfect. At any rate there was not the same motive for discarding the old imperfect as in Latin. In the latter language the formation corresponding to Gr. e-pep-o-v could not possibly have survived because all means of differentiation from the present had disappeared. Not only the augment, but the difference between primary and secondary endings22 had been obliterated from Latin, so that the forms of the present and of the old imperfect became identical. Here the imperfect in -ba-m therefore fulfilled a real want, and became the only form because it was the only clear form.23 On the other hand in Oscan and Umbrian,

even if the type amā-ba-m was pre-Italic. On the other hand, if the type legebam was also developed in the Italic period, we have only to think of the Latin forms as symbolical of the Italic-the nature of the analogy remains the same.

20 This analogy presupposes shortening of vowels before nt, but since this was already Italic (Brugmann 1.2800), this does not by any means involve us in chronological difficulties.

21 Here, as elsewhere, it would seem unnecessary to remark that the third plural, which is so often quoted because extant, represents the whole tense system. 22 The ending -m of amā-bam and the like is an exception. However, the same ending is found in the subjunctive and the future of the third conjugation, e.g. lega-m, and was not connected with past time any longer.

23 In this connection the probability suggests itself that the extreme frequency of the historical present in Latin was due to the fact that the Latin present tense perpetuated the IE imperfect as well as present in all forms except the first singular.