Language: Its Nature, Development and Origin

XIX.—§ 1. The Old Theory.

In the beginning was the root. This is “the result of strict and careful induction from the facts recorded in the dialects of the different members of the family” (Whitney L 260). “The firm foundation of the theory of roots lies in its logical necessity as an inference from the doctrine of the historical growth of grammatical apparatus” (Whitney G 200). “An instrumentality cannot but have had rude and simple beginnings, such as, in language, the so-called roots ... such imperfect hints of expression as we call roots” (Whitney, Views of L. 338). These are really three different statements: induction from the facts, a logical inference from the doctrine about grammatical apparatus (i.e. the usually accepted doctrine, but on what is that built up except on the root theory?), and the a priori argument that an ‘instrumentality’ must have simple beginnings. Even granted that these three arguments given at different times, each of them in turn as the sole argument, must be taken as supplementing each other, the three-legged stool on which the root theory is thus made to sit is a very shaky one, for none of the three legs is very solid, as we shall soon have occasion to see.

XIX.—§ 2. Roots.

In the beginning was the root—but what was it like? Bopp took over the conception of root from the Indian grammarians, and like them was convinced that roots were all monosyllabic, and that view was accepted by his followers. These latter at times attributed other phonetic qualities to these roots, e.g. that they always had a short vowel (Curtius C 22). I quote from a very recent treatise (Wood, “Indo-European Root-formation,” Journal of Germ. Philol. 1. 291): “I range myself with those who believe that IE. roots were monosyllabic ... these roots began, for the most part, with a vowel. The vowels certainly were the first utterances,[89] and though we cannot make the beginning of IE. speech coeval with that of human speech, we may at least assume that language, at that time, was in a very primitive state.”

The number of these roots was not very great (Curtius, l.c.; Wood 294). This seems a natural enough conclusion when we picture the earliest speech as the most meagre thing possible.

These few short monosyllabic roots were real words—this is a necessary assumption if we are to imagine a root stage as a real language, and it is often expressly stated; Curtius, for instance, insists that roots are real and independent words (C 22, K 132); cf. also Whitney, who says that the root VAK “had also once an independent status, that it was a word” (L 255). We shall see afterwards that there is another possible conception of what a ‘root’ is; but let us here grant that it is a real word. The question whether a language is possible which contains nothing but such root words was always answered affirmatively by a reference to Chinese—and it will therefore be well here to give a short sketch of the chief structural features of that language.

XIX.—§ 3. Structure of Chinese.

Each word consists of one syllable, neither more nor less. Each of these monosyllables has one of four or five distinct musical tones (not indicated here). The parts of speech are not distinguished: ta means, according to circumstances, great, much, magnitude, enlarge. Grammatical relations, such as number, person, tense, case, etc., are not expressed by endings and similar expedients; the word in itself is invariable. If a substantive is to be taken as plural, this as a rule must be gathered from the context; and it is only when there is any danger of misunderstanding, or when the notion of plurality is to be emphasized, that separate words are added, e.g. ki ‘some,’ šu ‘number.’ The most important part of Chinese grammar is that dealing with word order: ta kuok means ‘great state(s),’ but kuok ta ‘the state is great,’ or, if placed before some other word which can serve as a verb, ‘the greatness (size) of the state’; tsï niu ‘boys and girls,’ but niu tsï ‘girl (female child),’ etc. Besides words properly so called, or as Chinese grammarians call them ‘full words,’ there are several ‘empty words’ serving for grammatical purposes, often in a wonderfully clever and ingenious way. Thus či has besides other functions that of indicating a genitive relation more distinctly than would be indicated by the mere position of the words; min (people) lik (power) is of itself sufficient to signify ‘the power of the people,’ but the same notion is expressed more explicitly by min či lik. The same expedient is used to indicate different sorts of connexion: if či is placed after the subject of a sentence it makes it a genitive, thereby changing the sentence into a kind of subordinate clause: wang pao min = ‘the king protects the people’; but if you say wang či pao min yeu (is like) fu (father) či pao tsï, the whole may be rendered, by means of the English verbal noun, ‘the king’s protecting the people is like the father’s protecting his child.’ Further, it is possible to change a whole sentence into a genitive; for instance, wang pao min či tao (manner) k’o (can) kien (see, be seen), ‘the manner in which the king protects (the manner of the king’s protecting) his people is to be seen’; and in yet other positions či can be used to join a word-group consisting of a subject and verb, or of verb and object, as an adjunct (attribute) to a noun; we have participles to express the same modification of the idea: wang pao či min ‘the people protected by the king’; pao min či wang ‘a king protecting the people.’ Observe here the ingenious method of distinguishing the active and passive voices by strictly adhering to the natural order and placing the subject before and the object after the verb. If we put i before, and ku after, a single word, it means ‘on account of, because of’ (cf. E. for ...’s sake); if we place a whole sentence between these ‘brackets,’ as we might term them, they are a sort of conjunction, and must be translated ‘because.’[90]

XIX.—§ 4. History of Chinese.

In 1894 I ventured to offer my mite to these investigations by suggesting an explanation of one phenomenon of pronunciation in present-day Chinese. I refer to the change sometimes wrought in the meaning of a word by the adoption of a different tone. Thus wang with one tone is ‘king,’ with another ‘to become king’; lao with one is ‘work,’ with another ‘pay the work’; tsung with one tone means ‘follow,’ with another ‘follower,’ and with a third ‘footsteps’; tshi with one tone is ‘wife,’ with another ‘marry’; haò is ‘good,’ and haó is ‘love.’ Nay, meanings so different as ‘acquire’ and ‘give’ (sheu) or ‘buy’ and ‘sell’ (mai) are only distinguished by the tones. Edkins and V. Henry (Le Muséon, Louvain, 1882, i. 435) have attempted to explain this from gestures; but this is palpably wrong. In the Danish dialect spoken in Sundeved, in southernmost Jutland, two tones are distinguished, one high and one low (see articles by N. Andersen and myself in Dania, vol. iv.). Now, these tones often serve to keep words or forms of words apart that but for the tone, exactly as in Chinese, would be perfect homophones. Thus na with the low tone is ‘fool,’ but with the high tone it is either the plural ‘fools’ or else a verb ‘to cheat, hoax’; ri ‘ride’ is imperative or infinitive according to the tone in which it is uttered; jem in the low tone is ‘home’ and in the high ‘at home’; and so on in a great many words. There is no need, however, in this language to resort to gestures to explain these tonic differences: the low tone is found in words originally monosyllabic (compare standard Danish nar, rid, hjem), and the high tone in words originally dissyllabic (compare Danish narre, ride, hjemme). The tones belonging formerly to two syllables are now condensed on one syllable. Although, of course, Chinese tones cannot in every respect be paralleled with Scandinavian ones, we may provisionally conjecture that the above-mentioned pairs of Chinese words were formerly distinguished by derivative syllables or flexional endings (see below, p. 373) which have now disappeared without leaving any traces behind them except in the tones. This hypothesis is perhaps rendered more probable by what seems to be an established fact—that one of the tones has arisen through the dropping of final stopped consonants (p, t, k).

However this may be, the death-blow was given to the dogma of the primitiveness of Chinese speech by Ernst Kuhn’s lecture Ueber Herkunft und Sprache der Transgangetischen Völker (Munich, 1883). He compares Chinese with the surrounding languages of Tibet, Burmah and Siam, which are certainly related to Chinese and have essentially the same structure; they are isolating, have no flexion, and word order is their chief grammatical instrument. But the laws of word order prove to be different in these several languages, and Kuhn draws the incontrovertible conclusion that it is impossible that any one of these laws of word position should have been the original one; for that would imply that the other nations have changed it without the least reason and at a risk of terrible confusion. The only likely explanation is that these differences are the outcome of a former state of greater freedom. But if the ancestral speech had a free word order, to be at all intelligible it must have been possessed of other grammatical appliances than are now found in the derived tongues; in other words, it must have indicated the relations of words to each other by something like our derivatives or flexions.

To the result thus established by Kuhn, that Chinese cannot have had a fixed word order from the beginning, we seem also to be led if we ask the question, Is primitive man likely to have arranged his words in this way? A Chinese sentence, according to Gabelentz (Spr 426), is arranged with the same logical precision as the direction on an English envelope, where the most specific word is placed first, and each subsequent word is like a box comprising all that precedes—only that a Chinaman would reverse the order, beginning with the most general word and then in due order specializing. Now, is it probable that primitive man, that unkempt, savage being, who did not yet deserve the proud generic name of homo sapiens, but would be better termed, if not homo insipiens, at best homo incipiens—is it probable that this urmensch, who was little better than an unmensch, should have been able at once to arrange his words, or, what amounts to the same thing, his thoughts, in such a perfect order? I incline to believe rather that logical, orderly thinking and speaking have only been attained by mankind after a long and troublesome struggle, and that the grammatical expedient of a fixed word order has come to Chinese as to European languages through a gradual development in which other, less logical and more material grammatical appliances have in course of time been given up.

We have thus arrived at a conception of Chinese which is toto cælo removed from the view formerly current. The Chinese language can no longer be adduced in support of the hypothesis that our Aryan languages, or all human languages, started at first as a grammarless speech consisting of monosyllabic root-words.

XIX.—§ 5. Recent Investigations.

I have reprinted the above sketch of Chinese, with a few very insignificant verbal changes, as I wrote it about thirty years ago, because I think that the main reasoning is just as valid now as then, and because everything I have since then read about this interesting language has only confirmed the opinion I ventured to express after what was certainly a very insufficient study. Chinese pronunciation, including its tones, may now be studied in two excellent books, dealing with two different dialects—Daniel Jones and Kwing Tong Woo, A Cantonese Phonetic Reader, London, 1912, and Bernhard Karlgren, A Mandarin Phonetic Reader in the Pekinese Dialect, Upsala, Leipzig and Paris, 1917 (Archives d’Études Orientales, vol. 13). Karlgren is also the author of Études sur la Phonologie Chinoise (ib. vol. 15, 1915-19), in which he deals with the history of Chinese sounds and the reconstruction of the old pronunciation in a thoroughly scholarly manner on the basis of an intimate knowledge of spoken and written Chinese, and in Ordet och pennan i mittens rike (Stockholm, 1918), he has given a masterly popular sketch of the structure of the Chinese language and its system of writing.

Of the greatest importance for our purposes is the same scholar’s recent brilliant discovery of a real case distinction in the oldest Chinese. In classical Chinese there are four pronouns of the first person (I, we) which have always been considered as absolutely synonymous. But Karlgren shows that the two of them which occur as the usual forms in Confucius’s conversations are so far from being used indiscriminately that one is nearly always a nominative and the other an objective case; the exceptions are not numerous and are easily explained. The present Mandarin pronunciation of the first is [u], of the second either [uo] or [ŋo]. But if we go back to the sixth century of our era we are able with certainty to say that the pronunciation of the former was [ŋuo], and of the latter [ŋa]. This, then, constitutes a real declension. Now, in the second person Karlgren is also able to point out a distinction of two pronouns, though not quite so clearly marked as in the first person, the objective showing here a greater tendency to encroach on the nominative (Karlgren here ingeniously adduces the parallel from our languages that the first person has retained the suppletive system ego: me, while the second uses the same stem tu: te). The oldest Chinese thus has the following case flexion:

(See “Le Proto-chinois, langue flexionnelle,” Journal Asiatique, 1920, 205 ff.).[91]

XIX.—§ 6. Roots Again.

To return to roots. The influence of Indian grammar on European linguists with regard to the theory of roots extended also to the meanings assigned to roots, which were all of them of verbal character, and nearly always highly general or abstract, such as ‘breathe, move, be sharp or quick, blow, go,’ etc. The impossibility of imagining anybody expressing himself by means of a language consisting exclusively of such abstracts embarrassed people much less than one would expect: Chinese, of course, has plenty of words for concrete objects.

The usual assumption was that there was one definite root period in which all the roots were created, and after which this form of activity ceased. But Whitney demurred to this (M 36), saying that E. preach and cost may be considered new roots, though ultimately coming from Lat. præ-dicare and con-stare: these old compounds are felt as units, “reducing to the semblance of roots elements that are really derivative or compound.” As Whitney goes no further than to establish the semblance of new roots, he might be taken as an adherent rather than as an opponent of the theory he objects to. But, as a matter of fact, new words are created in modern languages, and if they form the basis of derived words, we may really speak of new roots (pun—punning, punster; fun—funny; etc.). Why not say that we have a French root roul in rouler, roulement, roulage, roulier, rouleau, roulette, roulis? This only becomes unjustifiable if we think that the establishment of this root gives us the ultimate explanation of these words; for then the linguistic historian steps in with the objection that the words have been formed, not from a root, but from a real word, which is not even in itself a primary word, but a derivative, Lat. rotula, a diminutive of rota ‘wheel.’ (I take this example from Bréal M 407). To the popular instinct sorrow and sorry are undoubtedly related to one another, and we may say that they contain a root sorr-; but a thousand years ago they had nothing to do with one another, and belonged to different roots: OE. sorg ‘care’ and sārig ‘wounded, afflicted.’ If all traces of Latin and Greek were lost, a linguist would have no more scruples about connecting scene with see than most illiterate Englishmen have now. Who will vouch that many Aryan roots may not have originated at various times through similar processes as these new roots preach, cost, roul, sorr, see?

The proper definition of a root seems to be: what is common to a certain number of words felt by the popular instinct of the speakers as etymologically belonging together. In this sense we may of course speak of roots at any stage of any language, and not only at a hypothetical initial stage. In some cases these roots may be used as separate words (E. preach, fun, etc., Fr. roul = what is spelt roule, roules, roulent); in other cases this is impossible (Lat. am in amo, amor, amicus; E. sorr); in many cases because the common element cannot, for phonetic reasons, be easily pronounced, as when E. drink, drank, drunk or sit, sat, seat, set are naturally felt to belong together, though it is impossible to state the root except in some formula like dr.nk, s.t, where the dot stands for some vowel. Similar considerations may be adduced with regard to the consonants if we want to establish what is felt to be common in give and gift (gi + labiodental spirant) or in speak and speech, etc.; but this need not detain us here.

In my view, then, the root is something real and important, though not always tangible. And as its form is not always easy to state or pronounce, so must its meaning, as a rule, be somewhat vague and indeterminate, for what is common to several ideas must of course be more general and abstract than either of the more special ideas thus connected; it is also natural that it will often be necessary to state the signification of a root in terms of verbal ideas, for these are more general and abstract than nominal ideas. But roots thus conceived belong to any and all periods, and we must cease to speak of the earliest period of human speech as ‘the root period.’

XIX.—§ 7. The Agglutination Theory.

According to the received theory (see above, § 1) some of the roots became gradually attached to other roots and lost their independence, so as to become finally formatives fused with the root. This theory, generally called the agglutination theory, contains a good deal of truth; but we can only accept it with three important provisos, namely, first, that there has never been one definite period in which those languages which are now flexional were wholly agglutinative, the process of fusion being liable to occur at any time; second, that the component parts which become formatives are not at first roots, but real words; and third, that this process is not the only one by which formatives may develop: it may be called the rectilinear process, but by the side of that we have also more circuitous courses, which are no less important in the life of languages for being less obvious.

In the process of coalescence or integration there are many possible stages, which may be denominated figuratively by such expressions as that two words are placed together (that is—in non-figurative language—pronounced after one another), tied together, knit together, glued together (‘agglutinated’), soldered together, welded together, fused together or amalgamated. What is really the most important part of the process is the degree in which one of the components loses its independence, phonetically and semantically.

As ‘agglutination’ is thus only one intermediate stage in a continuous process, it would be better to have another name for the whole theory of the origin of formatives than ‘the agglutination theory,’ and I propose therefore to use the term ‘coalescence theory.’ The usual name also fixes the attention too exclusively on the so-called agglutinative languages, and if we take the formatives of such a language as Turkish, as in sev-mek ‘to love,’ sev-il-mek ‘to be loved,’ sev-dir-mek ‘to cause to love,’ sev-dir-il-mek ‘to be made to love,’ sev-ish-mek ‘to love one another,’ sev-ish-dir-il-mek ‘to be made to love one another’—who will vouch that these formatives were all of them originally independent words? Those who are most competent to have an opinion on the matter seem nowadays inclined to doubt it and to reject much of what was current in the description of these languages given by the earlier scholars; see, especially, the interesting final chapter of V. Grønbech, Forstudier til tyrkisk lydhistorie (København, 1902).

XIX.—§ 8. Coalescence.

The various degrees of coalescence, and the coexistence at the same linguistic period of these various degrees, may be illustrated by the old example, English un-tru-th-ful-ly, and by German un-be-stimm-bar-keit. Let us look a little at each of these formatives. The only one that can still be used as an independent word is ful(l). From the collocation in ‘I have my hand full of peas’ the transition is easy to ‘a handful of peas,’ where the accentual subordination of full to hand paves the way for the combination becoming one word instead of two: this is not accomplished till it becomes possible to put the plural sign at the end (handfuls, thus also basketfuls and others), while in less familiar combinations the s is still placed in the middle (bucketsful, two donkeysful of children, see MEG ii. 2. 42). In these substantives -ful keeps its full vowel [u]. But in adjectival compounds, such as peaceful, awful, there is a colloquial pronunciation with obscured or omitted vowel [-fəl, -fl], in which the phonetic connexion with the full word is thus weakened; the semantic connexion, too, is loosened when it becomes possible to form such words as dreadful, bashful, in which it is not possible to use the definition ‘full of ...’ Here, then, the transition from a word to a derivative suffix is complete.

English -hood, -head in childhood, maidenhead also is originally an independent word, found in OE. and ME. in the form had, meaning ‘state, condition,’ Gothic haidus. In German it has two forms, -heit, as in freiheit, and -keit, whose k was at first the final sound of the adjective in ewigkeit, MHG. ewecheit, but was later felt as part of the suffix and then transferred to cases in which the stem had no k, as in tapferkeit, ehrbarkeit.

The suffix -ly is from lik, which was a substantive meaning ‘form, appearance, body’ (‘a dead body’ in Dan. lig, E. lich in lichgate); manlik thus is ‘having the form or appearance of a man’; the adjective like originally was ge-lic ‘having the same appearance with’ (as in Lat. con-form-is). In compounds -lik was shortened into -ly: in some cases we still have competing forms like gentlemanlike and gentlemanly. The ending was, and is still, used extensively in adjectives; if it is now also used to turn adjectives into adverbs, as in truthful-ly, luxurious-ly, this is a consequence of the two OE. forms, adj. -lic and adv. -lice, having phonetically fallen together.

It may perhaps be doubtful whether the G. suffix -bar (OHG. -bari, OE. bære) was ever really an independent word, but its connexion with the verb beran, E. bear, cannot be doubted: fruchtbar is what bears fruit (cf. OE. æppelbære ‘bearing apples’), but the connexion was later loosened, and such adjectives as ehrbar, kostbar, offenbar have little or nothing left of the original meaning of the suffix. The two prefixes in our examples, un- and be-, are differentiated forms of the old negative ne and the preposition by, and the only affix in our two long words which is thus left unexplained is -th, which makes true into truth and is found also in length, health, etc.

XIX.—§ 9. Flexional Endings.

Thus the Scandinavian passive originates in a coalescence of the active verb and the pronoun sik: Old Norse (þeir) finna sik (‘they find themselves’ or ‘each other’), gradually becomes one word (þeir) finnask, later finnast, finnaz, Swedish (de) finnas, Dan. (de) findes ‘they are found.’ In Old Icelandic the pronoun is still to some extent felt as such, though formally an indistinguishable part of the verb; thus combinations like the following are very frequent: Bolli kvaz þessu ráða vilja = kvað sik vilja; “Bolli dixit se velle: B. said that he would have his own way” (Laxd. 55). In Danish a distinction can sometimes be made between a reflexive and a purely passive employment: de slås with a short vowel is ‘they fight (one another),’ but with a long vowel ‘they are beaten.’ A similar coalescence is taking place in Russian, where sja ‘himself’ (myself, etc.) dwindles down to a suffixed s: kazalos ‘it showed itself, turned out.’

A similar case is the Romanic future: It. finiro, Sp. finire, Fr. finirai, from finire habeo (finir ho, etc.), originally ‘I have to finish.’ Before the coalescence was complete, it was possible to insert a pronoun, Old Sp. cantar-te-hé ‘I shall sing to you.’

A third case in point is the suffixed definite article, if we are allowed to consider that as a kind of flexion: Old Norse mannenn (manninn) accusative ‘the man,’ landet (landit) ‘the land’; Dan. manden, landet, from mann, land + the demonstrative pronoun enn, neuter et. Rumanian domnul ‘the lord,’ from Lat. dominu(m) illu(m), is another example.

XIX.—§ 10. Validity of the Theory.

Now, does this kind of explanation admit of universal application—in other words, were all our derivative affixes and flexional endings originally independent words before they were ‘glued’ to or fused with the main word? This has been the prevalent, one might almost say the orthodox, view of all the leading linguists, who may be mustered in formidable array in defence of the agglutination theory.[92]

Against the universality of this origin for formatives I adduced in my former work (1894, p. 66 f., cf. Kasus, 1891, p. 36) four reasons, which I shall here restate in a different order and in a fuller form.

(1) Nothing can be proved with regard to the ultimate genesis of flexion in general from the adduced examples, for in all of them the elements were already fully flexional before the coalescence (cf. ON. finnask, fannsk; It. finirò, finirai, finira; ON. maðrenn, mannenn, mansens, etc.). What they show, then, is really nothing but the growth of new flexional formations on an old flexional soil, and it might be imagined that the fusion would not have taken place, or not so completely, if the minds of the speakers had not been already prepared to accept formations of this character. I do not, however, attach much importance to this argument, and turn to those that are more cogent.

(2) The number of actual forms proved beyond a doubt to have originated through coalescence is comparatively small. It is true that not a few derivative syllables were originally independent; still, if we compare them with the number of those for which no such origin has been proved or even proposed, we find that the proportion is very small indeed. In the list of English suffixes enumerated in Sweet’s Grammar, only eleven can be traced back to independent words, while 74 are not thus explicable. Anyone going through the countless suffixes enumerated in the second volume of Brugmann’s Vergleichende Grammatik will, I think, be struck with the impossibility of any great number of them being traced back to words in the same way as hood, etc., above: their forms and, still more, their vague spheres of meaning, and on the whole their manner of application, distinctly speak against such an origin.

As to real flexional endings traceable to words, their number is even comparatively smaller than that of derivative suffixes; the three or four instances named above are everywhere appealed to, but are there so many more than these? And are they numerous enough to justify so general an assertion? My impression is that the basis for the induction is very far from sufficient.

(3) This argument is strengthened when we are able to point out instances in which, as a matter of fact, flexional endings have arisen in a way that is totally opposed to the agglutinative, which then must renounce all claims to be the only possible way for a language to arrive at flexional formatives. See below (§ 13) on Secretion.

(4) Assuming the theory to be true, we should expect much greater regularity, both in formal (morphological) and in semantic (syntactic) respect than we actually find in the old Aryan languages; for if one definite element was added to signify one definite modification of the idea, we see no reason why it should not have been added to all words in the same way. As a matter of fact, the Romanic future, the Scandinavian passive voice and definite article present much greater regularity than is found in the flexion of nouns and verbs in old Aryan.

XIX.—§ 11. Irregularity Original.

It will be objected that the irregularity which we find in these old languages is of later growth, and that, in fact, flexion, as Schuchardt says, is “anomal gewordene agglutination.” Whitney said that “each suffix has its distinct meaning and office, and is applied in a whole class of analogous words” (L. 254), and in reading Schleicher’s Compendium one gains the impression that the old Aryan sounds and forms were like a regiment of well-trained soldiers marching along in the best military style, while all irregularities were the result of later decay in each language separately. But the trend of the whole scientific development of the last fifty years has been in the direction of demonstrating more and more irregularity in the original forms: where formerly only one ending was assumed for the same case, etc., now several are assumed. (See, e.g., Walde in Streitberg’s Gesch., 2. 194, Thumb, ib. 2. 69.) And as with the forms, so also with the meanings and applications of the forms. Madvig as early as 1857 (p. 27. Kl 202) had seen that the signification of the grammatical forms must originally have been extremely vague and fluctuating, but most scholars went on imagining that each case, each tense, each mood had originally stood for something quite settled and definite, until gradually the progress of linguistics made away with that conception point by point. In place of the belief that the original Aryan verb had a definite system of tense forms, it is now generally assumed that different ‘aspects’ (‘aktionsarten’), somewhat like those of Slav verbs, were indicated, and that the notion of ‘time’ differences was only afterwards developed out of the notion of aspect: but if we compare the divisions and definitions of these aspects given by various scholars, we see how essentially vague this notion is; instead of being a model system of nice logical distinctions, the original condition must rather have been one in which such notions as duration, completion, result, beginning, repetition were indistinctly found as germs, from which such ideas as perfect and imperfect, past and present, were finally evolved with greater and greater clearness.

Similar remarks apply to moods. All attempts at finding out, deductively or inductively, the fundamental notion (grundbegriff) attached to such a mood as the subjunctive have failed: it is impossible to establish one original, sharply circumscribed sphere of usage, from which all the various, partly conflicting, usages in the actually existing languages can be derived. The usual theory is that there existed one true subjunctive, characterized by long thematic vowels -ē-, -ā-, -ō-, and distinct from that an optative, characterized by a formative -iē-: -ī-,[93] and that these two were fused in Latin. But, as Oertel and Morris have shown in their valuable article “An Examination of the Theories regarding the Nature and Origin of Indo-European Inflection” (Harvard Studies in Classical Philol. XVI, 1905) it is probably safer to assume for the Indo-European period substantial identity of meaning in the modal formatives iē: ī and the long thematic vowels -ē-, -ā-, -ō-, which were then continued undifferentiated in Latin, while on the one hand the Germanic branch has practically discarded the forms with long thematic vowel and confined itself to the i suffix, and on the other hand two branches, Greek and Indo-Iranic, have availed themselves of the formal difference and separated a ‘subjunctive’ and an ‘optative’ mood.

XIX.—§ 12. Coalescence Theory dropped.

In the historical part I have already mentioned some instances of coalescence explanations of Aryan forms which have been abandoned by most scholars, such as the theory that the r of the Latin passive is a disguised se, which would agree very well with the Scandinavian passive, but falls to the ground when one remembers that corresponding forms are found in Keltic, where the transition from s to r is otherwise unknown: these forms are now believed to be related to some r forms found in Sanskrit, but there not possessed of any passive signification, this latter being thus a comparatively late acquisition of Keltic and Italic: these two branches turning an existing, non-meaning consonant to excellent use in their flexional system and generalizing it in the new application.[94]

The explanation of the ‘weak’ Gothonic preterit from a coalescence of did (loved = love did) was long one of the strongholds of the agglutination theory, Bopp’s original collocation of these forms with other forms which could not be thus explained (see above 51) having passed into oblivion. Now we have Collitz’s comprehensive book Das schwache Präteritum, 1912, in which the formative consonant is shown to have been Aryan t, and the close correspondence not only with the passive participle, but also with the verbal nouns in -ti is duly emphasized.

The impossibility of explaining the Latin perfect in -vi from composition with fui has been demonstrated by Merguet (see Walde in Streitberg’s Gesch., 2. 220). Instead of this rectilinear explanation, scholars now incline to assume an intricate play of various analogical influences starting from a pre-ethnic perfect in w in isolated instances.

Many have explained the case ending -s as a coalesced demonstrative pronoun sa or, as it is now given, so; the difficulty that the same s denotes now the nominative and now the genitive was got over by Curtius (C 12) by the assumption that sa was added at two distinct periods, and that each period made a different use of the addition, though Curtius does not tell us how one or the other function could be evolved from such a pronoun. The latest attempt at explanation, which reaches me as I am writing this chapter, is by Hermann Möller (KZ 49. 219): according to him the common Aryan and Semitic nominative ended in o and the genitive in e, but to this was added in the masculine, and more rarely in the feminine, the pronoun s as a definite article, so that the primitive form corresponding to Lat. lupus meant ‘the wolf’ and lupu ‘(a) wolf’; later the s-less form was given up, and lupus came to be used for both ‘the wolf’ and ‘wolf’ (similarly presumably in the genitive, if we translate the presumed original forms into Latin lupis ‘the wolf’s’ and lupi ‘(a) wolf’s,’ later lupi in both functions). In Semitic, inversely, an element m, corresponding to the Aryan accusative ending, was added as an indefinite article, the m-less form thus becoming definite, but in the oldest Babylonian-Assyrian the distinction has been given up, and the form in m is (like the Latin form in s) used both definitely and indefinitely. Ingenious as these constructions are, the whole theory seems to me highly artificial, and it is difficult to imagine that both Aryans and Semites, after having evolved such a valuable distinction as that between ‘the wolf’ and ‘a wolf,’ expressed by simple means, should have wilfully given it up—to evolve it again in a later period.[95] Fortunately one is allowed to confess one’s ignorance of the origin of the case endings s and m, but if I were on pain of death to choose between Möller’s hypothesis and the suggestion thrown out by Humboldt (Versch 129), that the light (high-pitched) s symbolized the living (personal) and active (the subject), and the dark (low-pitched) m the lifeless (neutral) and passive (the object), I should certainly prefer the latter explanation.

Hirt (GDS 37) also thinks that the s found in Aryan cases is an originally independent word, only he thinks that this se, so was not originally a demonstrative pronoun, but the particle, which with the extension i is found in Gothic sai ‘ecce,’ and as it can thus be compared with the particle c in Lat. hic, it is clear that it might be added in all cases—and as a matter of fact Hirt finds it in six different cases in the singular and in all cases in the plural except the genitive. Hirt makes no attempt at explaining how these various case-forms have come to acquire the signification (function) with which we find them in the oldest documents; “the s element had nothing to do with the denotation of any case, number or gender, and only after it had been added to some cases and not to others could it come to be distinctive of cases” (p. 39). In other words, his explanation explains just nothing at all. The same is true with regard to the ‘particles’ om or em, e, o, i, which he thinks were added in other cases, and when he ends (p. 42) by saying that “this must be sufficient to give a glimpse of the way in which Aryan flexion originated,” the only thing we have really seen is the haphazard way in which this flexion is formed, and the impossibility at present of arriving at a fully satisfactory explanation of these things. I should especially demur to the two suppositions underlying Hirt’s theory that Aryan had at one period a completely flexionless structure, and that the same sound when occurring in various cases must have had the same origin: it seems much more probable to me that the s of the nominative and the s of the genitive were not at first identical.[96]

That item of the coalescence theory which probably appealed most to the fancy of scholars and laymen alike was the explanation of the personal endings in the verbs from the personal pronouns: we have an m in the first person of the mi-verbs (esmi) and in the pronoun me, etc., and we have a t in the third person (esti) and in a third-person pronoun or demonstrative (to); it is, therefore, quite natural to think that esmi is simply the root es ‘to be’ + the pronoun mi ‘I,’ and esti es + the other pronoun, and to extend this view to the other persons. And yet not even this has been allowed to stand unchallenged by later disrespectful linguists, headed by A. H. Sayce (Techmer’s Zeitschr. f. allg. Sprwiss. i. 22) and Hirt. As a matter of fact, the theory is based exclusively on the above-mentioned correspondence in the first and third persons singular, while the dual and plural endings do not at all agree with the corresponding personal pronouns and the endings of the second person can only be compared with the pronoun through the employment of phonological tricks unworthy of a scientific linguist. Even in the first person the correspondence is not complete, for besides -mi we have other endings: -m, which cannot be very well considered a shortened -mi (and which agrees, as Sayce remarks, much more closely with the accusative ending of nouns), -o and -a, neither of which can be explained from any known pronoun. There is thus nothing for it except to say, as Brugmann does (KG § 770): “The origin of the personal endings is not clear”; cf. also Misteli 47: “The relations between personal endings and the independent personal pronouns must be much more evident to justify this view.... The Aryan language offers direct evidence against the assumption that a sentence has been thus drawn together, because it uses in the verbal forms of the first and third person sg. pronominal stems which are otherwise employed only as objects, and, moreover, would here place the subject after the predicate, though in sentences it observes the opposite order.” Meillet expresses himself very categorically (Bulletin de la Soc. de Ling. 1911, 143): “Scarcely any linguist who has studied Aryan languages would venture to affirm that *-mi of the type Gr. fēmi is an old personal pronoun.”

The impression left on us by all these cases is that many of the earlier explanations by agglutination have proved unsatisfactory, and that linguists are nowadays inclined either to leave the forms entirely unexplained or else to admit less rectilinear developments, in which we see the speakers of the old languages groping tentatively after means of expression and finding them only by devious and circuitous courses. It is, of course, difficult to classify such explanations, and the agglutination or coalescence theory has to be supplemented by various other kinds of explanation; but I think one of these, which has not received its legitimate share of attention, is important and distinctive enough to have its own name, and I propose to term it the ‘secretion’ theory.

XIX.—§ 13. Secretion.

By secretion I understand the phenomenon that one integral portion of a word comes to acquire a grammatical signification which it had not at first, and is then felt as something added to the word itself. Secretion thus is a consequence of a ‘metanalysis’ (above, Ch. X § 2); it shows its full force when the element thus secreted comes to be added to other words not originally possessing this element.

A clear instance is offered in the history of some English possessive pronouns. In Old English min and þin the n is kept throughout as part and parcel of the words themselves, the other cases having such forms as mine, minum, minre, exactly as in German mein, meine, meinem, meiner, etc. But in Middle English the endings were gradually dropped, and min and þin for a short time became the only forms. Soon, however, n was dropped before substantives beginning with a consonant, but was retained in other positions (my father—mine uncle, it is mine); then the former form was transferred also to those cases in which the pronoun was used (as an adjunct) before words beginning with vowels (my father, my uncle—it is mine). The distinction between my and mine, thy and thine, which was originally a purely phonetic one, exactly like that between a and an (a father, an uncle), gradually acquired a functional value, and now serves to distinguish an adjunct from a principal (or, to use the terms of some grammars, a conjoint from an absolute form); my came to be looked upon as the proper form, while the n of mine was felt as an ending serving to indicate the function as a principal word. That this is really the instinctive feeling of the people is shown by the fact that in dialectal and vulgar speech the same n is added to his, her, your and their, to form the new pronouns hisn, hern, yourn, theirn: “He that prigs what isn’t hisn, when he’s cotch’d, is sent to prison. She that prigs what isn’t hern, At the treadmill takes a turn.”

Another instance of secretion is -en as a plural ending in E. oxen, G. ochsen, etc. Here originally n belonged to the word in all cases and all numbers, just as much as the preceding s; ox was an n stem in the same way as, for instance, Lat. (homo), hominem, hominis, etc., or Gr. kuōn, kuna, kunos, etc., are n stems. In Gothic n is found in most of the cases of similar n stems. In OE. the nom. is oxa, the other cases in the sg. oxan, pl. oxan (oxen), oxnum, oxena, but in ME. the n-less form is found throughout the singular (gen. analogically oxes), and the plural only kept -n. Thus also a great many other words, e.g. (I give the plural forms) apen, haren, sterren (stars), tungen, siden, eyen, which all of them belonged to the n declension in OE. When -en had thus become established as a plural sign, it was added analogically to words which were not originally n stems, e.g. ME. caren, synnen, treen (OE. cara, synna, treow), and this ending even seemed for some time destined to be the most usual plural ending in the South of England, until it was finally supplanted by -s, which had been the prevalent ending in the North; eyen, foen, shoen were for a time in competition with eyes, foes, shoes, and now -n is only found in oxen (and children). In German to-day things are very much as they were in Southern ME.: -en is kept extensively in the old n stems and is added to some words which had formerly other endings, e.g. hirten, soldaten, thaten. The result is that now plurality is indicated by an ending which had formerly no such function (which, indeed, had no function at all); for if we look upon the actual language, oxen (G. ochsen) is = ox (ochs) singular + the plural ending -en; only we must not on any account imagine that the form was originally thus welded together (agglutinated)—and if in G. soldaten we may speak of -en being glued on to soldat, this ending is not, and has never been, an independent word, but is an originally insignificative part secreted by other words.

A closely similar case is the plural ending -er. The consonant originally was s, as seen, for instance, in the Gr. and Lat. nom. genos, genus, gen. Gr. gene(s)os, genous, Lat. generis for older genesis. In Gothonic languages s, in accordance with a regular sound shift in this case, became r (through z) whenever it was retained, but in the nom. sg. it was dropped, and thus we have in OE. sg. lamb, lambe, lambes, but in the pl. lambru, lambrum, lambra. In English only few words show traces of this flexion, thus OE. cild, pl. cildru, ME. child, childer, whence, with an added -en, our modern children. But in German the class had much more vitality, and we have not only words belonging to it of old, like lamm, pl. lämmer, rind, rinder, but also gradually more and more words which originally belonged to other classes, but adopted this ending after it had become a real sign of the plural number, thus wörter, bücher.

There is one trait that should be noticed as highly characteristic of these instances of secretion, that is, that the occurrence of the endings originating in this way seems from the first regulated by the purest accident, seen from the point of view of the speakers: they are found in some words, but not in others, whereas the endings treated of under the heading Coalescence are added much more uniformly to the whole of the vocabulary. But as a similarly irregular or arbitrary distribution is met with in the case of nearly all flexional endings in the oldest stages of languages belonging to our family of speech, the probability is that most of those endings which it is impossible for us to trace back to their first beginnings have originated through secretion or similar processes, rather than through coalescence of independent words or roots.

XIX.—§ 14. Extension of Suffixes.

A special subdivision of secretion comprises those cases in which a suffix takes over some sound or sounds from words to which it was added. Clear instances are found in French, where in consequence of the mutescence of a final consonant some suffixes to the popular instinct must seem to begin with a consonant, though originally this did not belong to the suffix. Thus laitier, at first formed from lait + ier, now came to be apprehended as = lai(t) + tier, and cabaretier as cabare(t) + tier, and the new suffix was then used to form such new words as bijoutier, ferblantier, cafetier and others. In the same way we have tabatière, where we should expect tabaquière, and the predilection for the extended form of the suffix is evidently strengthened by the syllable division in frequent formations like ren-tier, por-tier, por-tière, charpen-tier. In old Gothonic we have similar extensions of suffixes, when instead of -ing we get -ling, starting from words like OHG. ediling from edili, ON. vesling from vesall, OE. lytling from lytel, etc. Consequently we have in English quite a number of words with the extended ending: duckling, gosling, hireling, underling, etc. In Gothic some words formed with -assus, such as þiudin-assus ‘kingdom,’ were apprehended as formed with -nassus, and in all the related languages the suffix is only known with the initial n; thus in E. -ness: hardness, happiness, eagerness, etc.; G. -keit with its k from adjectives in -ic has already been mentioned (376). From criticism, Scotticism, we have witti-cism, and Milton has witticaster on the analogy of criticaster, where the suffix of course is -aster, as in poetaster. Instead of -ist we also find in some cases -nist: tobacconist, lutenist (cf. botan-ist, mechan-ist).

To form a new word it is often sufficient that some existing word is felt in a vague way to be made up of something + an ending, the latter being subsequently added on to another word. In Fr. mérovingien the v of course is legitimate, as the adjective is derived from Mérovée, Merowig, but this word was made the starting-point for the word designating the succeeding dynasty: carlovingien, where v is simply taken over as part of the suffix; nowadays historians try to be more ‘correct’ and prefer the adjective carolingien, which was unknown to Littré. Oligarchy is olig + archy, but for the opposite notion the word poligarchy or polygarchy was framed from poly and the last two syllables of oli-garchy, and though now scholars have made polyarchy the usual form, the word with the intrusive g was the common form two hundred years ago in English, and corresponding forms are found in French, Spanish and other languages. Judgmatical is made on the pattern of dogmatical, though there the stem is dogmat-. In jocular German schwachmatikus ‘valetudinarian,’ we have the same suffix with a different colouring, taken from rheumatikus (thus also Dan. svagmatiker). Swift does not hesitate to speak of a sextumvirate, which suggests triumvirate better than sexvirate would have done; and Bernard Shaw once writes “his equipage (or autopage)”—evidently starting from the popular, but erroneous, belief that equipage is derived from Lat. equus and then dividing the word equi + page. Cf. Scillonian from Scilly on account of Devonian as if this were Dev + onian instead of Devon + ian.

XIX.—§ 15. Tainting of Suffixes.

It will be seen that in some of these instances the suffix has appropriated to itself not only part of the sound of the stem, but also part of its signification. This is seen very clearly in the case of chandelier, in French formed from chandelle ‘candle’ with the suffix -ier, of rather vague signification, ‘anything connected with, or having to do with’; in English the word is used for a hanging branched frame to hold a number of lights; consequently a similar apparatus for gas-burners was denominated gaselier (gasalier, gasolier), and with the introduction of electricity the formation has even been extended to electrolier. Vegetarian is from the stem veget- with added -ari-an, which ending has no special connexion with the notion of eating or food, but recently we have seen the new words fruitarian and nutarian, meaning one whose food consists (exclusively or chiefly) in fruits and nuts. Cf. solemncholy, which according to Payne is in use in Alabama, framed evidently on melancholy, analyzed in a way not approved by Greek scholars. The whole ending of septentrionalis (from the name of the constellation Septem triones, the seven oxen) is used to form the opposite: meridi-onalis.

A similar case of ‘tainting’ is found in recent English. The NED, in the article on the suffix -eer, remarks that “in many of the words so formed there is a more or less contemptuous implication,” but does not explain this, and has not remarked that it is found only in words ending in -teer (from words in -t). I think this contemptuous implication starts from garreteer and crotcheteer (perhaps also pamphleteer and privateer); after these were formed the disparaging words sonneteer, pulpiteer. During the war (1916, I think) the additional word profiteer[97] came into use, but did not find its way into the dictionaries till 1919 (Cassell’s). And only the other day I read in an American publication a new word of the same calibre: “Against patrioteering, against fraud and violence ... Mr. Mencken has always nobly and bravely contended.”

XIX.—§ 16. The Classifying Instinct.

Man is a classifying animal: in one sense it may be said that the whole process of speaking is nothing but distributing phenomena, of which no two are alike in every respect, into different classes on the strength of perceived similarities and dissimilarities. In the name-giving process we witness the same ineradicable and very useful tendency to see likenesses and to express similarity in the phenomena through similarity in name. Professor Hempl told me that one of his little daughters, when they had a black kitten which was called Nig (short for Nigger), immediately christened a gray kitten Grig and a brown one Brownig. Here we see the genesis of a suffix through a natural process, which has little in common with the gradual weakening of an originally independent word, as in -hood and the other instances mentioned above. In children’s speech similar instances are not unfrequent (cf. Ch. VII § 5); Meringer L 148 mentions a child of 1.7 who had the following forms: augn, ogn, agn, for ‘augen, ohren, haare.’ How many words formed or transformed in the same way must we require in order to speak of a suffix? Shall we recognize one in Romanic leve, greve (cf. Fr. grief), which took the place of leve, grave? Here, as Schuchardt aptly remarks, it was not only the opposite signification, but also the fact that the words were frequently uttered shortly after one another, that made one word influence the other.

The classifying instinct often manifests itself in bringing words together in form which have something in common as regards signification. In this way we have smaller classes and larger classes, and sometimes it is impossible for us to say in what way the likeness in form has come about: we can only state the fact that at a given time the words in question have a more or less close resemblance. But in other cases it is easy to see which word of the group has influenced the others or some other. In the examples I am about to give, I have been more concerned to bring together words that exhibit the classifying tendency than to try to find out the impetus which directed the formation of the several groups.

In OE. we have some names of animals in -gga: frogga, stagga, docga, wicga, now frog, stag, dog, wig. Savour and flavour go together, the latter (OFr. flaur) having its v from the former. Groin, I suppose, has its diphthong from loin; the older form was grine, grynd(e). Claw, paw (earlier powe, OFr. pol). Rim, brim. Hook, nook. Gruff, rough (tough, bluff, huff—miff, tiff, whiff). Fleer, leer, jeer. Twig, sprig. Munch, crunch (lunch). Without uttering or muttering a word. The trees were lopped and topped. In old Gothonic the word for ‘eye’ has got its vowel from the word for ‘ear,’ with which it was frequently collocated: augo(n), auso(n), but in the modern languages the two words have again been separated in their phonetic development. In French I suspect that popular instinct will class the words air, terre, mer together as names of what used to be termed the ‘elements,’ in spite of the different spelling and origin of the sounds. In Russian kogot’ ‘griffe’ (claw), nogot’ ‘ongle’ (fingernail), and lokot’ ‘coude’ (elbow), three names of parts of the body, go together in flexion and accent (Boyer et Speranski, Manuel de la l. russe 33). So do in Latin culex ‘gnat’ and pulex ‘flea.’ Atrox, ferox. A great many examples have been collected by M. Bloomfield, “On Adaptation of Suffixes in Congeneric Classes of Substantives” (Am. Journal of Philol. XII, 1891), from which I take a few. A considerable number of designations of parts of the body were formed with heteroclitic declension as r-n stems (cf. above, XVIII § 2): ‘liver,’ Gr. hēpar, hēpatos, ‘udder,’ Gr. outhar, outhatos, ‘thigh,’ Lat. femur, feminis, further Aryan names for blood, wing, viscera, excrement, etc. Other designations of parts of the body were partly assimilated to this class, having also n stems in the oblique cases, though their nominative was formed in a different way. Words for ‘right’ and ‘left’ frequently influence one another and adopt the same ending, and so do opposites generally: Bloomfield explains the t in the Gothonic word corresponding to E. white, where from Sanskr. we should expect th, çveta, as due to the word for ‘black’; Goth. hweits, swarts, ON. hvítr, svartr, etc. A great many names of birds and other animals appear with the same ending, Gr. glaux ‘owl,’ kokkux ‘cuckoo,’ korax ‘crow,’ ortux ‘quail,’ aix ‘goat,’ alopex ‘fox,’ bombux ‘silkworm,’ lunx ‘lynx’ and many others, also some plant-names. Names for winter, summer, day, evening, etc., also to a great extent form groups. In a subsequent article (in IF vi. 66 ff.) Bloomfield pursues the same line of thought and explains likenesses in various words of related signification, in direct opposition to the current explanation through added root-determinatives, as due to blendings (cf. above, Ch. XVII § 6). In Latin the inchoative value of the verbs in -esco is due to the accidentally inherent continuous character of a few verbs of the class: adolesco, senesco, cresco; but the same suffix is also found in the oldest words for ‘asking, wishing, searching,’ retained in E. ask, wish, G. forschen, which thus become a small group linked together by form and meaning alike.

XIX.—§ 17. Character of Suffixes.

There seems undoubtedly to be something accidental or haphazard in most of these transferences of sounds from one word to another through which groups of phonetically and semantically similar words are created; the process works unsystematically, or rather, it consists in spasmodic efforts at regularizing something which is from the start utterly unsystematic. But where conditions are favourable, i.e. where the notional connexion is patent and the phonetic element is such that it can easily be added to many words, the group will tend constantly to grow larger within the natural boundaries given by the common resemblance in signification.

I have no doubt that the vast majority of our formatives, such as suffixes and flexional endings, have arisen in this way through transference of some part, which at first was unmeaning in itself, from one word to another in which it had originally no business, and then to another and another, taking as it were a certain colouring from the words in which it is found, and gradually acquiring a more or less independent signification or function of its own. In long words, such as were probably frequent in primitive speech, and which were to the minds of the speakers as unanalyzable as marmalade or crocodile is to Englishmen nowadays, it would be perhaps most natural to keep the beginning unchanged and to modify the final syllable or syllables to bring about conformity with some word with which it was associated; hence the prevalence of suffixes in our languages, hence also the less systematic character of these suffixes as compared with the prefixes, most of which have originated in independent words, such as adverbs. What is from the merely phonetic point of view the ‘same’ suffix, in different languages may have the greatest variety of meaning, sometimes no discernible meaning at all, and it is in many cases utterly impossible to find out why in one particular language it can be used with one stem and not with another. Anyone going through the collections in Brugmann’s great Grammar will be struck with this purely accidental character of the use of most of the suffixes—a fact which would be simply unthinkable if each of them had originally one definite, well-determined signification, but which is easy to account for on the hypothesis here adopted. And then many of them are not added to ready-made words or ‘roots,’ but form one indivisible whole with the initial part of the word; cf., for instance, the suffix -le in English squabble, struggle, wriggle, babble, mumble, bustle, etc.

XIX.—§ 18. Brugmann’s Theory of Gender.

As I have said, man is a classifying animal, and in his language tends to express outwardly class distinctions which he feels more or less vaguely. One of the most important of these class divisions, and at the same time one of the most difficult to explain, is that of the three ‘genders’ in our Aryan languages. If we are to believe Brugmann, we have here a case of what I have in this work termed secretion. In his well-known paper, “Das Nominalgeschlecht in den indogermanischen Sprachen” (in Techmer’s Zs. f. allgem. Sprachwissensch. 4. 100 ff., cf. also his reply to Roethe’s criticism, PBB 15. 522) he puts the question: How did it come about that the old Aryans attached a definite gender (or sex, geschlecht) to words meaning foot, head, house, town, Gr. pous, for instance, being masculine, kephalē feminine, oikos masculine, and polis feminine? The generally accepted explanation, according to which the imagination of mankind looked upon lifeless things as living beings, is, Brugmann says, unsatisfactory; the masculine and feminine of grammatical gender are merely unmeaning forms and have nothing to do with the ideas of masculinity and femininity; for even where there exists a natural difference of sex, language often employs only one gender. So in German we have der hase, die maus, and der weibliche hase is not felt to be self-contradictory. Again, in the history of languages we often find words which change their gender exclusively on account of their form. Thus, in German, many words in -e, such as traube, niere, wade, which were formerly masculine, have now become feminine, because the great majority of substantives in -e are feminine (erde, ehre, farbe, etc.). Nothing accordingly hinders us from supposing that grammatical gender originally had nothing at all to do with natural sex. The question, therefore, according to Brugmann, is essentially reduced to this: How did it come to pass that the suffix -a was used to designate female beings? At first it had no connexion with femininity, witness Lat. aqua ‘water’ and hundreds of other words; but among the old words with that ending there happened to be some denoting females: mama ‘mother’ and gena ‘woman’ (compare E. quean, queen). Now, in the history of some suffixes we see that, without any regard to their original etymological signification, they may adopt something of the radical meaning of the words to which they are added, and transfer that meaning to new formations. In this way mama and gena became the starting-point for analogical formations, as if the idea of female was denoted by the ending, and new words were formed, e.g. Lat. dea ‘goddess’ from deus ‘god,’ equa ‘mare’ from equus ‘horse,’ etc. The suffix -iē- or -ī- probably came to denote feminine sex by a similar process, possibly from Skr. strī ‘woman,’ which may have given a fem. *wḷqī ‘she-wolf’ to *wḷqos ‘wolf.’ The above is a summary of Brugmann’s reasoning; it may interest the reader to know that a closely similar point of view had, several years previously, been taken by a far-seeing scholar in respect to a totally different language, namely Hottentot, where, according to Bleek, CG 2. 118-22, 292-9, a class division which had originally nothing to do with sex has been employed to distinguish natural sex. I transcribe a few of Bleek’s remarks: “The apparent sex-denoting character which the classification of the nouns now has in the Hottentot language was evidently imparted to it after a division of the nouns into classes[98] had taken place. It probably arose, in the first instance, from the possibly accidental circumstance that the nouns indicating (respectively) man and woman were formed with different derivative suffixes, and consequently belonged to different classes (or genders) of nouns, and that these suffixes thus began to indicate the distinction of sex in nouns where it could be distinguished” (p. 122). “To assume, for example, that the suffix of the m. sg. (-p) had originally the meaning of ‘man,’ or the fem. sg. (-s) that of ‘woman,’ would in no way explain the peculiar division of the nouns into classes as we find it in Hottentot, and would be opposed to all that is probable regarding the etymology of these suffixes, and also to the fact that so many nouns are included in the sex-denoting classes to which the distinction of sex can only be applied by a great effort.... If the word for ‘man’ were formed with one suffix (-p), and the word indicating ‘woman’ (be it accidentally or not) by another (-s), then other nouns would be formed with the same suffixes, in analogy with these, until the majority of the nouns of each sex were formed with certain suffixes which would thus assume a sex-denoting character” (p. 298).