of the words in question; and from the result attained by the action of analogy. I will discuss these three points briefly in the above order.

:

First, in regard to the classification according to the psychic processes: much that MISTELI has brought forward on this subject may serve to introduce its discussion, which has not yet begun in earnest. I will only emphasize one point here it is important to distinguish whether a transfer of form has taken place of itself, so to speak (as is the case in the greater majority of instances), or whether the speaker, finding the form which is demanded by the phonetic laws for some reason inconvenient, seeks for some other formation, and as the result of this search a transfer of form takes place. An example of the latter sort is the Latin dative and ablative plural in -abus, which frequently occurs in deabus, filiabus and libertabus, and in isolated instances in other words. As is most clearly shown by the passages in NEUE's Formenlehre der lateinischen Sprache, 2nd edition, 1, page 22, these dative-ablative forms arose where a distinction from the corresponding forms of the masculine was needed. There was no objection to saying di deaeque, deorum dearumque, deos deasque; but what should be said in the dative and ablative, -dis disque? There was a similar drawback to the use of filia in wills or other provisions quae pertinent ad necessitatem juris. Suppose, for example, provision must be made in case a son or sons, a daughter or daughters are living. Should it be worded: "filio seu filiis, filia seu filiis exstantibus"? It was evidently in such and similar predicaments that the forms in -abus originated, and the process can hardly have been other than the following: another form is sought instead of dis and filiis, which in special cases could not be employed; and this form is suggested in consequence of the connection subsisting in the mind of the speaker between the series filiae, filiarum, filiis, filias, and duae, duarum, duabus, duas. In ordinary speech the ablative duabus can exercise no attractive force on mensis and the rest, because their is is protected by connection with the is of the second declension, which has the same significance. Not until this connection is for some special reason dissolved, does duabus exert its attractive force. The old grammarians are

therefore quite correct in saying that the forms deabus etc. were created differentiae causa; but the impulse toward differentiation was not able to evolve new and original formations, only imitative ones, after existing models. This impulse to differentiate can accordingly be classed among the motives which are active in the construction of forms by analogy. (Cf. MISTEĻI as quoted above, page 472.)

We find a second ground of classification in the constitution of the words in question, that is, in the conditions which must be present in the words before any action of analogy can take place. Under this head we must ask first of all whether words connected only by sound, and also whether words connected only by sense, can influence each other through the working of analogy. I should be inclined to answer the first question in the negative, the second in the affirmative. To illustrate the first, Misteli gives a good example (page 434), which I will repeat here:

“Although καθίζω, ἐκάθισα forms in the future καθιῶ, -θιsis, -diet, as if xad were the root and to the ending, as in Badilw, Badiouμai, so that scarcely a shadow (in the ) of the root sed remains; yet despite the identity of ending, xádɩe, καθίζον, -θίζω, -θίζων, -θίζοις have not the remotest connec tion with, for example, a πρόρριζε, πρόρριζον, τρίζω, -ρίζων, -pilots; the gulf between noun and verb cannot be bridged over by any amount of phonetic identity, and it is only because we regard this as self-evident that we can speak of purely phonetic analogy."

As to the second point, it is at least clear that endings whose function is identical enter into association, even without phonetic similarity; thus, áɣávos arises from analogy with the dative plural in -ots, while there is no seductive similarity of form between ot (in aywot) and ots. Whether the same can be observed in word-stems (e. g. whether the form of the adjective "good" can influence the form of the adjective “bad”, or the like) must be more accurately investigated. CAROLINA MICHAELIS (Studien zur romanischen Wortschöpfung, page 35) assumés such an attractive force in the case of the Italian greve, which would accordingly owe its e to the influence of the e of leve. In the second place, we must remark that in

inflected words the associative action can start either from the word-stem or from the endings, and in this connection a distinction must be drawn between material and formal analogical construction. 1) An example of material formation by analogy is the Greek oést, which came from the previously existing form ἡδύσι through the influence of ἡδέος, ἡδές, ἡδέων. In the singular the forms ήδύς, ηδύ, ἡδύν were able to resist the attractive force (although oéa does occasionally appear); but in the plural, where, after the assimilation of the accusative to the nominative, house was the only case with u, that constituent part of the forms (all belonging to one series) which was felt to contain the chief significance was made uniform. The innumerable formal constructions by analogy are illustrated by forms like ayavots, Herzens etc.

A third ground of classification is found in the result of the transfer of form, according as the original form is wholly supplanted by the imitative formation, as is the case with ἐλύσαμεν, which probably took the place of an older *ἔλυσμεν; or both forms exist side by side, as in the genitive senatus and senati. The question also comes up, whether an intermediate form can arise through the mutual influence of two forms, a species of formation which has received the name of "formation by contamination" [Contaminationsbildung]. An example would be the Latin jecinoris by the side of jecur. As the Sanskrit shows, where the stems yakan and yakṛt exist side by side, the Latin paradigm was once jecor, *jecinis, and jecinoris is "contaminated" from both stem-forms.

But these and similar attempts, which may possibly be made, to classify the whole mass of analogical formations, cannot possess any considerable value for the practical application of the science, since the first task must be to collect material systematically in the newly explored field. I think such a collection would have the greatest likelihood of success if the inflectional forms of a definite linguistic period could be taken up, and the inquiry instituted, in what analogical formations each individual form was either actively or passively involved. This would be the easiest way to obtain a comprehensive view

1) OSTHOFF in particular (following PAUL) called attention to this classification in the lecture we shall mention below.

of the different series of forms which exist, or once existed, in the consciousness of the speakers. We should then find that all the cases of a word taken together form a series (from which fact we can explain, among other things, the leveling processes [Ausgleichungen] which take place between strong and weak cases), and also the corresponding cases in several (although not in all) subdivisions of declension, as we can see from the transfer of the locative au of the u-stems to the i-stems in Sanskrit (kavaú formed after the analogy of bhānaú). We should also find that nouns which belong together in meaning are so firmly welded into a group that occasionally even their cases undergo a leveling process. Thus the case-ending in ur (or us) of páti “husband”, jáni “wife”, sákhi "friend", i. e. pátyur, jányur, sákhyur, has certainly followed the genitive of the nouns of relationship, like pitúr. (Cf. WACKERNAGEL, Kuhn's Zeitschrift, 25, page 289.)

In the verb we should find these interchanges [ Verschlingungen] in still greater abundance. We very soon see that not only the forms of one mode constitute a series, e. g. πεποίθαμεν formed from πέποιθα in place of * πέπιθμεν, and so with the corresponding form of different indicatives, e. g. λuoe, whose & was derived from pepe 1) (cf. MISTELI as quoted above, page 436), but that also the tense-systems of the individual verb influence each other in such a way that differences which have arisen phonetically, and which are so great that they threaten to break up the system of the whole verb, are adjusted; thus the old déλλw, eßaλoy has been supplanted by Báλλw, eßaλov; cf. JOHANNES SCHMIDT, K. Z., 25, page 153. In the same way we find that the corresponding tense-systems of different verbs are connected in the mind of the speaker; hence e. g. the lengthening of the i in aorists like άpipatat, which very probably arose after the model of ácikradat, so that now the same rhythm exists in all the forms which belong together.2)

1) BRUGMAN, Morph. Unt., 1, page 161, derives this ε of the sigmatic aorist from the original e of the perfect, and following him, GUSTAV MEYER, Gr. Grammatik, page 402. [Transl.]

2) In accordance with the above, my former statement in the Altindisches Verbum, page 110, must be modified.

If such a consideration gives us an idea of the series and network of formations under which the word-forms are ranged in the mind of the speaker, we shall at once be able to formulate an important methodic principle (which has often been formulated; v. the passage in MISTELI's article, page 408), viz., the principle that the forms which stand outside the network of series, those, that is, which do not belong to the inflectional systems, have in their favor the probability that they will exhibit the unimpeded action of phonetic laws. At the same time, it must be clear from the few examples I have adduced that in all, even the oldest periods of the existing languages (and why not also in the Indo-European parent speech?), we may expect to find formations by analogy. It is true they will appear most frequently in more modern periods, because there a coïncidence of the form-systems is favored by the more advanced mutilation of the endings. Yet we must again emphasize the fact that we are at present confined to general impressions and approximate estimates, since an exhaustive and classified material is not yet at hand.

I come to the third notion, phonetic laws. In opposition to the view of CURTIUS, LESKIEN and others have, as we saw, brought forward a doctrine which can be most simply expressed as follows: phonetic laws in themselves admit of no exceptions. This phrase, which will be tested later, requires explanation. In the first place, it is self-evident that in order to become acquainted with regular phonetic development in its purity we must subtract all such results of analogical action as were described in the preceding section, and then we must consider that the natural development of sounds can be best studied in those languages which are as nearly as possible in a state of nature. The literary languages are less adapted to this aim, because they always possess a mass of borrowed words, borrowed either from foreign languages, or from related dialects, or from former periods of the same language, which are now only represented in literary monuments, a borrowed mass, much of which has been so absorbed into the native material of the language that it is no longer felt by the speaker to be foreign. Which of us, even though he be a linguistic

DELBRÜCK, Introduction to the Study of Language.