[299]
If the only method of reconstruction is the comparative method, by the same token comparison has no other aim than reconstruction. Unless they are to be sterile, correspondences observed between several forms must be placed in a chronological perspective and result in the reconstitution of a unique form, as we have already insisted several times (cf. p. [16] ff., p. [272]). Thus in order to explain Latin medius (‘middle’) beside Greek mésos (‘middle’), it was necessary to postulate an earlier (though not necessarily Proto-Indo-European) form *methyos, from which both medius and mésos could be derived. If the comparison is between two forms belonging to the same language rather than different languages, the same applies: thus Latin gerō (‘I bear’) and gestus (‘borne’) point back to an earlier stem *ges- once common to both.
It should be noted in passing that comparison based on phonetic changes should always be supported by morphological considerations. In examining Latin patior (‘I suffer’) and its past participle passus, one takes into consideration the past participles factus (‘made’), dictus (‘said’), etc. because passus is a formation of the same kind. On the basis of the morphological connexion between faciō (‘I make’) and factus (‘made’), dı-cō (‘I say’) and dictus (‘said’), etc., it becomes possible to set up a similar connexion at an earlier period for patior and *pat-tus. Likewise, if the comparison is morphological, it should be supported by appeal to phonetics. The Latin comparative meliōrem (‘better’) can be compared to Greek hēdíō (‘sweeter’) because phonetically one goes back to *meliosem, *meliosm and the other to *hādioa, *hādiosa, *hādiosm.
[300]
Linguistic comparison is not, then, a mechanical operation. It implies bringing together all the information relevant to constructing an explanation. But it should always result in a hypothesis couched in a formula which purports to reconstruct an antecedent. Comparison should always return to the reconstruction of linguistic forms.
But does this investigation of the past aim at the reconstruction of the actual, concrete linguistic forms of an earlier period? Or is it limited merely to abstract and fragmentary claims concerning parts of words; for example, that Latin f in fu-mus corresponds to a Common Italic þ, or that the first sound of Greek állo and Latin aliud was already an a in Proto-Indo-European? Its task may well be limited to the latter; we might even say that its analytic method has no other aim than establishing these partial connexions. However, from a number of isolated facts of this kind it is possible to draw more general conclusions. For example, a series of cases like Latin fu-mus allows us to say with certainty that þ occurred in the sound system of Common Italic. Similarly, we can say that Proto-Indo-European had in what is called pronominal flexion a neuter singular ending -d, which was different from the adjectival ending -m; and that again is a general morphological fact inferred from a collection of isolated cases (cf. Latin istud, aliud, vs. bonum; Greek tó ← *tod, állo ← *allod vs. kalón; English that, etc.). But we can go further. Having reconstructed these various correspondences, it is possible to synthesise all those which bear upon a form in its entirety. Thus we can reconstruct whole words (e.g. Proto-Indo-European *alyod ), flexional paradigms, etc. In order to do this, one must take various independent facts and group them together. If, for example, one compares the various parts of a reconstructed form like *alyod, one notices a great difference between the -d, which raises a question of grammar, and the a-, which raises no such question. A reconstructed form is not given as a whole. It is a sum total of historical phonetic arguments which can always be broken down into its constituents. Each part of the reconstruction can be called in question and remains open to examination. Hence reconstructed forms have always provided a faithful reflection of the general conclusions which are applicable to their particular case. The Proto-Indo-European word for ‘horse’ has at different times been held to be *akvas, *ak1vas, *ek1vos, and finally *ek1wos: only the s has remained uncontroversial, and the total number of sounds in the word.
The aim of reconstruction is thus not to restore a form for its own sake, which would be in any case rather ridiculous, but to crystallise and condense a series of conclusions which are held to be correct in the light of the evidence currently available. In a word, it records the progress made to date in our science. Linguists have no need to be defended against the bizarre accusation of attempting to reconstruct Proto-Indo-European in its entirety, as if they intended to use it as a language. Since that is not what they have in mind when studying historically attested languages (for they do not study Latin linguistically in order to be able to speak it well), it is even less plausible in the case of isolated words in prehistoric languages.
Moreover, although reconstructions remain open to correction, they cannot be dispensed with if one hopes to gain some idea of the language as a whole, and to what linguistic type it belongs. It is an indispensable way of representing in a relatively simple form a host of general facts, both synchronic and diachronic. The general outlines of Proto-Indo-European immediately become apparent from the reconstructions as a whole. For example, it is clear that the suffixes were formed from certain elements (t, s, r, etc.) and not others, and that the complications of vowels in German verbs (cf. werden, wirst, ward, wurde, worden) conceal in the main the same original alternation: e — 0 — zero. Furthermore the study of the history of later periods is much easier as a result. Without reconstructions to begin with, it would be much more difficult to explain changes that have taken place since the prehistoric period.
[302]
§2 Degree of certainty of reconstructions
There are some reconstructed forms which are quite certain, and others which are disputable or clearly problematic. As we have just seen, the degree of certainty to be attached to a reconstructed form as a whole depends on the relative certainty attributable to the partial reconstructions involved in the synthesis. In this respect, two words are hardly ever comparable. There is a difference between two such informative Proto-Indo-European reconstructions as *esti (‘he is’) and *didōti (‘he gives’); for in the latter the vowel of the reduplication is open to doubt (cf. Sanskrit dadāti, Greek dídōsi).
In general, more doubt is cast upon reconstructions than is warranted. There are three facts which should boost confidence in them.
The first, which is of capital importance, has already been indicated earlier (p. [65] ff.). Given any word, one can distinguish clearly its constituent sounds, how many of them there are, and where they begin and end. The objections raised by linguists given to peering into their phonetic microscopes have been dealt with previously (p. [83]). Doubtless in a group like -sn- there are fleeting transitional sounds present; but it is antilinguistic to bother about them. The ordinary ear does not catch them. Above all, speakers are invariably in agreement about how many sounds there are. We can say, then, that in the Proto-Indo-European *ek1wos there were just five distinct differential elements which speakers recognised.
[303]
The second fact has to do with the sound system of each language. Every language has an inventory of sounds fixed in number (cf. p. [58]). In Proto-Indo-European, each element in the system appears in at least a dozen reconstructed forms, and sometimes in thousands. So we can be sure we have identified the whole inventory.
Finally, in order to identify the sound units of a language it is not necessary to be able to characterise them positively. They are to be considered differential entities having the property of not being confused with one another (cf. p. [164]). That is what is essential: so much so that one could designate the phonetic elements of a reconstructed language simply by figures or symbols of any kind. In Proto-Indo-European *ĕk1wŏs, it is useless to inquire about the exact quality of the ĕ, whether it was open or close, how fronted it was, etc. As long as there are not various kinds of ĕ, that is unimportant, provided ĕ cannot be confused with any other distinct phonetic element in the language ( ă, ŏ, é, etc.). This amounts to saying that the first sound of *ĕk1wŏs did not differ from the second sound of *mĕdhyŏs, or the third sound of *ăgĕ, and so on. This sound could, without specifying its phonetic character, be classified and designated by a number in a table of Proto-Indo-European sounds. So the reconstructed form *ĕk1wŏs means that the Proto-Indo-European form corresponding to Latin equos, Sanskrit açva-s, etc. comprised five specific sounds drawn from the inventory of sounds of the Proto-Indo-European language.
Within these limits, then, our reconstructions are fully valid.