I am not competent to discuss how far the stems may be actually restored in the Finno-Ugrian languages; but since they are put at the basis of comparison, we must give ourselves a clear idea of their nature. First of all, a stem restored is a certain hypothesis which ought to be based upon strict phonetic variations characteristic of a group of languages supposed to have been issued from one and the same language. The introduction of such a hypothesis as a pra-language makes the whole reconstruction of stems little reliable. Yet it cannot be practically applied for further research and comparison of the languages where the historic variations of sounds are not established. Again, the elementary requirement is that the hypothesis may have no reverse force [137]. In spite of this, A. Sauvageot does it all the time and «stems» are put as a basis of comparison. Let us illustrate our proposition. The variations of sounds may go into different directions; e.g., s→s'→s»→h→x→k and k→k→c→ts→s, both phenomena observed in the «Altaic» languages, so that the change of consonants s and k cannot be presumed as one going in one and the same definite direction. If this is so, then the restoration of stems extending over a long period of time (in ages, centuries, perhaps millenniums), where the historic variations are not established, cannot be considered as reliable. In fact, we have already seen that the restoration of Tungus p is a mere misunderstanding. There are no facts to show that such an initial consonant ever existed in pra-Tungus. Let us now suppose that in the Finno-Ugrian pra-language k is restored as the initial consonant of some stem. We may be allowed to compare this stem with the Tungus words with the initial k only on the condition that this k is preserved from the hypothetic Tungus pra-language. However, we have seen that k may appear as an alteration of the original x, which in its turn may appear from the alteration of h and the latter may exist as a simple aspiration of the vowel. So that the initial k in Tungus may happen to be of a secondary origin and the original consonant might be zero. As shown, it might also be s, and t with their variations. Since the finding of a stem is the finding of a complex transmitted, for scientific safety one must have synchronous data (stems) for all groups compared.
Again, I am not competent to discuss the question of how reliable are Finno-Ugrian stems restored, but from the point of view of their comparison with the Tungus material it should be pointed out that the stems compared include only four groups of consonants; namely, labial, dental, glottal, and nasal [138], which are shown to be alternate. As a matter of fact, such a generalization is not yet complete, for labials alternate with the glottals and glottals alternate with the dentals which might take place in the hypothetic pra-languages. In the instance of the lateral spirant tenuis L in the Lolo languages, I have shown that I may alternate with a series of consonants including r, dental, guttural, etc [139]. So if one wants to have an ideal generalization and a really scientific one, it must be restored merely as a consonant and a vowel. Indeed, such a generalization brings to an absurdity the whole construction and for this reason one has to stop at a certain moment of generalization. However, whether one stops earlier or later, the scientific value of the hypothesis does not change. Such a grouping of stems according to the consonants, like labials, dentals, and glottals, is certainly artificial, for these groups of consonants alternate. Yet, to distinguish voiced and non-voiced, occlusive and spirant, etc., is also impossible for the practical achievement of proving the existence of common words. Therefore the generalization is arrested at a certain point where it may still have a scientific appearance and not keep comparatists very restricted by the rigidity of the system.
In this position it is not noticed that another theoretical objection comes from a different side; namely, since the stems are actually reduced to five types and two syllables, and since the number of groups of languages compared is not numerous (Finno-Ugrian, Turk, Mongol, and Tungus), the chance of finding common words is greatly increased because the number of combinations in these conditions is very limited. In order to demonstrate the common words, one must bring a very large number of cases exceeding at least two or even three times the number of probable coincidences mathematically calculated and predicted. Indeed, this is an elementary scientific requirement which must be observed by any one who is looking for reliable conclusions. This side of the problem is not discussed nor probably foreseen by A. Sauvageot. It is merely rejected without discussion.
137. Such was, for example, the case with J. Nemeth, who postulated that the alteration s→x in Tungus may have a reverse force in the sense of k→s which has permitted the restoration of kam~sam to prove that saman has originated from the Turk ham. To be more exact, the unfortunate saman~kam had to prove the phonetic law k~s. Such an elementary lapsus on the part of J. Nemeth has led B. Laufer to further abberration with the problem of the word saman. This fact shows how dangerous it is to operate with the restorations and fnrther using of hypotheses. (Cf. «Sramaua-Shaman,» pp. 111-113.) On the other hand, this fact also shows that A. Sauvageot's methodology is not accidental, for J. Nemeth is stated to be also responsible for A. Sauvageot's work (vide his «Avant-Propos»).
138. A. Sauvageot has promised to bring forth another series of parallels with the initials of other groups.
139. «Phonetic Notes on a Lolo Dialect and Consonant L,» 1930.