The relationship between linguistics in general and the historical-comparative tradition within it is quite extraordinary. The story has often been told how historical and ‘comparative’ linguistics—at one time the dominant, almost exclusive, activity in the discipline—was overtaken by the new synchronistic wave a few decades ago. Yet it is strange that historical linguistics, even after its alleged dethronement, hardly suffered damage worse than novel competition for attention and support, and occasional outbursts of ill-informed temper. On the whole the diachronic study of language, now presumably visible in its true context rather than in the earlier, absolute fashion, was judged to have come through the fire of relativization unscathed. Though its generalizations had often been questioned from within—a matter to which we shall return below—it continued to be admired for its achievements, and especially those who had found other interests would profess admiration from afar and would spread a feeling that here was a tradition that should not be touched because it could not easily be improved upon and that, in fact, deserved a pedestal.
In one sense the eulogists were right; in another, their attitude spelled mischief. Pragmatically, comparative and historical study had been a success; its procedures and results were coherent and, being testable in principle, had stood the test wherever it could be applied. They were even discovered to be more general than had been thought, and not at all limited to particular languages in particular historical conditions. Yet the same practitioners who had brought about this state of affairs and whose erudition and sagacity was so plain when they were seen dealing with their data could also be found discussing their concepts and their methods in the abstract—not to mention even more ambitious forays into philosophy. Experience tells us how brittle such activities can be when contemporary, and what additional scrutiny they demand and how easily we misunderstand them when we confront them in retrospect, with altered presuppositions and different expectations, though, alas, frequently still armed with a deceptively unchanged vocabulary. We must in any case remember that tacit assumptions have a way of being at least as powerful as assumptions openly proclaimed, and that the price for missing the relationship between the two can be heavy. In this vein we shall examine certain tenets of historical and comparative linguistics; and we shall do so in a spirit of deep respect for our distinguished predecessors’ record, but also with the wariness that is called for when we discuss their own descriptions and evaluations, no matter how often repeated and how well established, of that very record.
In introducing a field of knowledge it is customary, at least as a matter of style, to begin by proclaiming some self-evident axiom. In the case of historical linguistics the traditional opener is to the effect that ‘all languages change’. It is quite clear what the original target of the statement was: it was rightly directed against the idea that languages existing in a fixed literary form and deliberately cultivated in this form (like Latin or classical Arabic) should be representative of language as a general human activity. In actual fact, Latin and Arabic have of course changed from the form on which the literary variety was once based, and live on in their ‘descendants’. But aside from this, let us have a look at the axiom and its claim to self-evidence. To begin with, let us consider some of the obvious analogies. To say that a language changes seems not unlike saying that an object or a substance (say, a rock) changes properties—color, temperature, consistency—or that an individual changes (a caterpillar into a chrysalis, and a chrysalis into a butterfly; or a sick child into a healthy adult), or that a city or a political institution changes in history; and even if this kind of thing should not be intended, the reader is still bound to understand it so. What is more, it may occur to him that the same ubiquitousness of change that is being claimed for languages may exist for some of the analogs: he may believe that all living beings change or that all cities change. In no event, however, is he in any doubt about the perseverance of the object or substance to which changeability or change is attributed. He is convinced—naively, perhaps—that a changing rock, caterpillar, child, or city is the same rock, insect, human individual, or city both before and after the change; that the substance remains identifiable as such and that only some of its accidental traits have become different. His natural inference must then be that a language, even as it is affected by change, also remains identifiable.
At this point, to be sure, certain terminological matters are explained to him, and he is told that our customary labeling may be misleading and that, along one line of descent, a language may change the name by which it happens to be known: ‘Anglo-Saxon’ (if the term is used instead of ‘Old English’) ‘becomes’ Middle English; Latin (or something like it) ‘becomes’ French or Rumanian; and so on. Still, the purpose of this warning is only to insist that while the label does not matter, there is reality to what is being labeled, namely, the line of descent. It is implied that the business of recognizing Old English (Anglo-Saxon) and Latin as ‘ancestors’ or ‘earlier stages’ of Middle English and of French, respectively, is either not a problem at all, or one that can be separated from another problem, namely, that of describing the changes along the line. On the preliminary plane on which we are at present moving there has been very little discussion of just what the criteria for establishing an ancestry-and-descent relation might be other than, perhaps, ‘inspection’. At best there is agreement that certain particular conceivable factors are no more dependable than labeling: habitat (American English is not the descendant of any pre-Columbian language), physical descent of the speakers (it does not matter what languages a person’s ancestors spoke at specified times in the past), explicit folklore rife among speakers or observers, and the like.
The fact is that the expert has at this point nothing to say to his reader. The pronouncement that all languages change, however meaningful he feels it is and whatever meaning we may be able to give to it, is not in the nature of a coherent report of some indubitable experience so long as we cannot say what it is that change is ascribed to. The quasi-truism we need, at least rhetorically, at the outset, must be less specific. It can only be to the modest effect that no two forms of speech, found in different places and at different times, are the same. In this formulation, the expression ‘different places’ should not be pressed,1 and ‘form of speech’ means what is usually called a language (or dialect) at a given time—that is, the kind of object that historians find or construct. Let us also assume that ‘sameness’ can somehow be defined, and let us agree, as perhaps we may, that the pronouncement just made is acceptable in an empirical sense.
Our subject matter, then, is forms of speech or language states, each one unique, with their coordinates of time and space, but so far without lines of a meaningful sort connecting them. Nor do we ask at this juncture how we have come by our data: whether from direct observation, from written records, or through inference.
The next step is to identify those forms of speech among which historical relations of some sort exist (‘descent’, relatedness’, but also ‘contact’). In order to do this, we must for a moment digress and anticipate, and we hope that we can do so once again by staying close to empirical matter on which we may agree without much hesitation. We may begin by observing that while we do not yet know how to determine a line of descent, we do know something about certain effects of the lapse of time on speech communities and on the speaking individuals making up speech communities. We may for instance notice that while words like sick and miss contrast with thick and myth in normal English, thick may become homophonous with sick, and myth with miss2 both (1) when ‘borrowed’ into the speech of certain German-speaking immigrants, and (2) when occurring in their English. We have little difficulty in understanding this bit of phonemic substitution; in fact we expect it, given the sound systems of the two languages in contact,3 and we expect it to occur with ‘regularity’, that is, in all the words that are either (1) borrowed or (2) learned by those speakers at that time. If we now choose to match the original shapes with the (1) borrowed or (2) imperfectly learned shapes that only came into existence with the contact situation, we conclude that a contact situation may result in a phonemic merger. What is more, if we knew nothing about the chronology of the two language states, the very fact that there is one kind of English (ordinary English) in which the two pairs of words are different, and another (the English spoken by our special group) in which they are identical would inform us that the latter had innovated.
Now let us suppose that history partly repeats itself and that another contact situation arises, this time involving speakers of a language similar to Japanese, (1) borrowing right and light from ordinary English into their language, or (2) learning to say those two words in speaking English; and let us suppose that the contrast disappears in the process. By virtue of the logic employed above, a matching of ordinary English with the special English of this group will lead to the judgment that the latter has innovated in merging the original r with the original l, while ordinary English has retained the contrast. And, to take the next step: if our quasi-Japanese English is directly matched with our German English, both may be seen to contain one innovation each. The matching procedure has of course become well known as the ‘comparative’ method (in the narrow sense of the word) with which to reconstruct a common ‘ancestor’ even in cases where it is not available in uninnovated shape. If we imagine for a moment that ordinary English has died out and only our two special varieties have survived (and if we assume, for the sake of the argument, that ‘quasi-Japanese’ English somehow4 preserves the distinction between words with s and words with th [θ]), we have the situation shown in Fig.1. From Fig.1 we reconstruct ordinary English, as shown in Fig.2.
Before we stop to consider the meaning of the connecting lines, a disclaimer is in order. The preceding remarks are offered only to illustrate the fact that there are known sociolinguistic situations (for instance, language contact) that result in merger, and to lay the ground work for the thesis that when historians speak of lines of descent they have processes in mind that share fundamental properties with those particular situations. No one should think for a moment that these are exactly and literally the very ones that we have here constructed for our oversimplified skeleton cases. In particular, the languages involved in this substratum5 model need not be widely different and mutually unintelligible; it has, in fact, long been pointed out that gross effects of this sort frequently fail to survive, because gradual assimilation of imperfectly learned forms of speech to a persistent standard tends to eliminate long-range effects even if such imperfect versions arise and exist during a limited period. The real breeding ground for the development in which we are interested must be a much subtler and much more pervasive condition, in which the contact is one among dialect varieties, styles, and other entities of the sort that add up to produce ranges of variation within a speech community and with regard to which the speaking individuals possess richly diversified degrees of control. Only to the extent that the crude models in Figs. 1 and 2 have their counterpart in this infinitely more finely graded context, are those models applicable. Yet, as the record of comparative linguistics shows, it is reasonable to believe that indeed there are such counterparts and that it makes sense to look upon phonemic mergers as the prototypes of linguistic innovation.
We may now return from our digression. Historical relations among forms of speech are, in part, revealed by a degree of regularity’ or absence of phonemic randomness as we translate from one language into another. Tentatively, and subject to further interpretation, we suggest that two forms of speech that are linked by such ties are so linked in one of two ways: either in such a way that one form of speech has all the innovations, or in such a way that each has at least one innovation. In the former case we have a line of descent from the retaining ancestor (or older stage) to the innovating descendant (or later stage). In the latter case we have a collateral relationship with two lines of descent emanating from a third form of speech: the common ancestor, which may be either known or reconstructed, but which in any event can be reconstructed as possessing none of the innovations.
The lines drawn in Figs. 1 and 2 represent, of course, such lines of descent in our primitive model, where the descendant languages were not typical forms of speech such as historians and comparatists usually encounter, but special languages resulting from known or essentially familiar contact situations, equated ex hypothesi with ordinary descendant languages. Yet there is no doubt that the principle carries over, since actual controversies such as the question of whether or not the Romance languages really come from (literally) Latin are actually argued in this fashion. Those who maintain that Latin is not the source but a collateral must show that there is at least one innovation in Latin that is not in the Romance languages. To illustrate the formal factors that enter into such decisions, let us imagine three forms of speech, A, B, and C, occurring, in this order, at three successive times and found ‘linked’ to one another in the sense alluded to above (Fig.3).
In particular let there be three kinds of contrasting pairs in A: (1) such that the contrast is answered, as the words are translated, by a homophone in both B and C, (2) such that it is answered by a contrast in B and by a homophone in C, and (3) such that it is answered by a contrast in both B and C, such as that shown in Fig.4. Clearly, A contains no innovations with respect to either B or C, nor does B carry any innovations with respect to C, so that A is an older stage of B and of C, and B an older stage of C.6
If, on the other hand, we let C be so constituted (C’) that the first class of contrasting pairs in A is answered by (1) a homophone in B and by a contrast in C’ (with the other two classes, (2) and (3) as before), we obtain the diagram in Fig.5, where A is separately ancestral to B and to C’. It should be noted that the relative location of B on the time scale with regard to C is predictable from the properties of B and C in Fig.4, but that the same is not true (except with recourse to glotto-chronological assumptions that do not enter here) with regard to B and C’ in Fig.5. A, of course, must antedate B, C, and C’. What is perhaps most remarkable about the line of descent is that although it is a natural and, to us, entirely familiar construct, it is not a primitive. The notion of descent is built on the notion of relationship, not vice versa.
It is perfectly true that all innovations (that is all phenomena to which we would reasonably give that name) are not phonemic mergers. If, however, we choose our lines of descent on a phonemic basis, we are frequently rewarded by the discovery that a good many innovations of different kinds—semantic, syntactic, and, generally, morphemic—occur along the same lines of descent. There are two principal reasons for relying initially on phonemic merger: first, as pointed out earlier, that we can relate it to phonemic substitution in fairly well understood sociolinguistic settings, and, secondly and most prominently, insofar as the history of the discipline is concerned, phonemic merger is the phenomenon that may be recognized as an innovation in the ‘comparative’ matching process itself. In the other areas the criteria are less clear-cut and depend more on universals (e.g., such universals of semantic change as may be generally accepted), on internal reconstruction, and on extraneous information. It is no accident that comparative linguistics, even when it continues to carry the ancient label of comparative ‘grammar’, is to such an overwhelming extent in fact comparative phonology.
Even so, we are not yet rid of all of our unspoken assumptions. In particular, the recognition of merger is not a simple matter, because what observation typically yields are conflicting regularities. Assuming, for the sake of the argument, some direct knowledge of a form of speech known as proto Indo-Iranian (it is actually only obtained through reconstruction), we find that in a certain chronologically later form of speech, Old Persian (the language of the royal Achaemenian inscriptions), certain proto Indo-Iranian contrasts are matched by homophones: both d and dh by d, for example. But the case of words with pI-I ģ and ģh is not so clear: in some words g and gh are both matched by OP z, whereas in others, OP has the same d that also corresponds to pI-I d dh. One suspects that the two classes of words reflect two historical situations, or two diverse channels of transmission. It is customary in such cases to assign primacy to one of these channels and declare it the true line of descent, and let the other channel (or channels) represent a body or bodies of loanwords. In this specific instance the merger of g gh with d dh is considered genuine and a sound change, while the z-words are classified as borrowings from the language or dialect of the Medes.
There are material reasons for this assignment, but they are a matter of external and accidental knowledge. It is doubtful whether the true line could be distinguished from the effects of borrowing in a formal way. Sheer numbers are no guide, since a language may be ‘swamped with loanwords’, and there is increasing skepticism concerning any a priori argument of the kind that says that syntactic or semantic structure, presumably immune to replacement by borrowed matter, will reveal the true ancestry. The study of loan translations and our growing acquaintance with the mechanisms that impart common traits to languages forming an areal complex have cast a fundamental doubt over all such notions. One may wish to rely on the criterion of the basic vocabulary to decide which is borrowing and which change along a primary line, on the plea that individuals and populations continuing to speak their language will borrow basic vocabulary last, whereas to acquire a new language would seem to mean to acquire the basic vocabulary first of all. In any event, as we have seen, since the same phonemic substitution plays its role both in assimilating loanwords and in shifting to another language, there will be phonemic regularity on both sides. Apart from the matter of the basic vocabulary, there is, in other words, no easy formal criterion to decide whether the z-words are loanwords in a d- dialect or vice versa.
In one sense, this does not matter much. It turns out that we are free to treat the two strains in the vocabulary, even though we find them combined in the same form of speech and in the same texts, quite as if they represented two different descendant languages appearing in separate bodies of text. This recourse to the assumption of dialect borrowing is not, as is sometimes suggested, a shamefaced and arbitrary way out of embarrassing contradictions that cannot be handled by supposedly ordinary means, and thus an admission of methodological bankruptcy. On the contrary, it is a normal and constant necessity, and in no way destructive to the classical assumptions. It does not, for example, invalidate the comparative method for a moment. The two components of the Old Persian vocabulary are simply made to function as two separate witnesses in the task of reconstructing proto Iranian and proto Indo-Iranian. There is, to be sure, the technical problem of deciding between (rightly) recognizing the same proto entity coming through along two channels of transmission and (wrongly) setting up only one channel of transmission but contrasting entities in the reconstructed proto languages, one to account for the words in which z in most other Iranian languages is matched by an Old Persian z and another for those in which it is matched by a d. There are technical remedies, based on doublets (that is, one and the same morph passing through both channels [-zana/-dana (‘race’) comes close]7), or based on the occasional association of more than one competing regularity in different segments of one and the same morph.8
Likewise, we need not choose—although, on the basis of partly extraneous yet decisive considerations, we in fact do choose—between classifying Middle English as a descendant of Old English and classifying it as a descendant of Old French, since nothing is lost by using the ‘Anglo-Saxon’ vocabulary of Middle English as a Germanic language and the French-based vocabulary of Middle English as a Romance language, along with the other Germanic and Romance languages, in the reconstruction of Germanic, and of (Vulgar) Latin, respectively. Things like this are done: the otherwise imperfectly known Latin of Illyria has left loanwords in Albanian, and these loanwords, complete with the phonemic substitutions attendant upon their transfer into Albanian, simply take their places alongside the rest of the evidence, presumably direct-line, for Vulgar Latin.
Looking back, the following points seem worth stressing. Rather than claim to be able to recognize, by inspection, two forms of speech as successive stages of one language and then be caught in difficulties and contradictions, we have found it more consistent and more congenial to the observable practices of historical linguistics (though not always to its professed tenets) to define descent as a relationship between two ‘related’ forms of speech such that only one of them shows innovations. This presupposes an ability to recognize innovation: encouraged by the observation of certain sociolinguistic situations (contact, variation, code switching, dialect learning and unlearning, etc.), we feel that we can interpret correspondences between a phonemic contrast in one form of speech and a homophone in another as the effect of merger in the latter. Thus we use merger as the innovation par excellence—not because there are no others (there are, and they are, once recognized, at least as important) but because phonemic merger is identifiable ipso facto. As the ancestry-and-descent relationship is sharpened by relativizing it in terms that are slightly more primitive, we also realize that it does not immediately link entire forms of speech but only, as it were, their constituent strands. Limiting ourselves, once again and perforce, to words and their phonemic shapes, we are more directly sure of the descent of each of the two components of the English vocabulary than we are of the descent of English as a whole or, for that matter, of what it means to assign a unique and exclusive ancestry to a form of speech as such. Even without the customary appeal to pidgins and creoles, the concept of the mixed language cannot simply be shrugged off; if it has real weaknesses they must surely be more hidden than that.
The preceding remarks are a plea for asking how two languages must be constituted for us to declare one of them, with a reasonable choice of metaphor, to be ‘descended’ from the other. This seems preferable to asking, given that one language is descended from another, what properties two such languages turn out to possess. The differences between the two styles is neither great nor unfamiliar, and our preference is simply a tribute to the definitional nature of the whole question, designed to keep us from getting bogged down in a pseudo-question. This advantage is even clearer as we take a closer look at the process of sound change, to which we have already attached such special importance, and which, in turn, appears as a facet of language descent: as Italian is descended from Vulgar Latin so are Italian sounds specifically descended from Vulgar Latin sounds. The concepts that come into play are, again, best handled as truly definitional (as indeed they have long been handled in the course of substantive work) and not in the pseudo-factual form in which they have encumbered much of the general debate. Not unlike the pronouncement that ‘all languages change’, the pronouncement that ‘sound change is regular’ has been presented as a finding, though, in this case, not as a self-evident one. The implication is that sound change might just as well have been expected to be ‘irregular’, but was discovered to be regular instead. Those who deny the finding still agree on the factual nature of the expectation that they merely claim has not been borne out. This framework leaves room for three processes: (1) regular sound change, (2) irregular sound change, and (3) change other than sound change, that is, absence of sound change.
But how is sound change defined? Surely not as ‘change in sound’, since every linguistic change must be that, and since the ability to distinguish sound changes from other linguistic changes is nevertheless rightly praised as a fruitful achievement. The replacement of gluff by glove, of cweðan (cp. quoth) by say, of holpen by helped, of ey by egg, all constitute alterations in sound, but they are precisely not classified as sound changes within the meaning of the pronouncement on regularity. On the other hand, the replacement of hring (and, much later wring) by ring is said to be a sound change in the technical sense. Why?
Before answering this question, it may be necessary to justify the use of the term ‘replacement’ as an equivalent of, or perhaps even an improvement on, ‘change’. We have already encountered the difficulty inherent in the term ‘change’ when applied to forms of speech in toto: it is the difficulty of identifying the quasi-genetic substance to which the accident of change is attributed. For this reason we prefer the notion of correspondence: as we translate—and surely this is a question of translation, presupposing the existence of a theory of translation of some sort—from, say, English into Dutch, we observe that thumb and dumb ‘go to’ words with d- in Dutch; and as we follow, specifically, what we called a line of descent, we may note that hring and wring ‘go to’ or ‘become’—these words now taking on the familiar chronological sense in the special context of translation from stage to stage—ring. So did gluff go to’ glove, cweðan to say, holpen to helped, and ey to egg. Only, this way of putting things is not equally familiar in all cases. It is quite customary to say that hring becomes ring, and even that gluff and holpen become glove and helped (though it is probably more natural to say that ‘the singular of gloves’ goes from gluff to glove, and ‘the past participle of help’ from holpen to helped). What we traditionally do in these cases is (I) keep the meaning constant and record how the morphs, one from each stage, are paired in replacement.9 The same may of course be done for cweðan: say and for ey: egg, and it is done when we say that a certain meaning used to be expressed by cweðan but is now expressed by say, or that an ovum was once denoted by ey and later by egg. It is, however, more common in these instances to switch to a different principle of classification (II) and keep the morphs constant,10 following them through their denotations and their altering pattern of co-occurrence. We will say, for instance, that cweðan became obsolete (except in be-queath and, in a sense, in quoth) ; that secgan-say changed its meaning, in part by losing certain nuances (‘relate’, ‘speak [truth, lies, etc.]’) and in part by taking on new ones (like the simple quotative ‘he said’); or, finally, that ey became obsolete, and that egg is a loanword from Scandinavian.
If, however, we adopt style I for all these examples, we may proceed to classify them according to certain special relationships that may or may not exist among the paired (that is, the replaced and the replacing) morphs. Clearly, a special relationship of a simple sort (1) exists for glove and helped: in each of these histories one allomorph has replaced another, inside the morpheme as it were, so that glove- now occurs outside the plural glove-s, to which it was once limited, and so that - ed, instead of -en, is found after one more verb than was the case before.11 A special relationship of a different kind (2) exists for both ring and, again, glove, as we should have discovered if all the material had been in: the difference in the morphemic shape of the paired morphs recurs. Hring-ring is matched by hrēow-rue, hrefn-raven, etc.; the (incidentally, more recent) replacement gluff-glove is matched by staff-stave.12 No special relationship is present in cweðan-say and in the context as constructed in ey-egg, and certainly in a case like ēam (the old word for ‘uncle’) -uncle.
It is well known how this classification is interpreted. Ring is called a sound change. Glove and holpen are classified as analogic changes; say is a semantic change; and egg and uncle are loanwords. The reason why glove is not also called a sound change is that the replacement of f by v occurs in instances where f alternates with v, but not in all such cases (wife), let alone in cases where there is no alternation.13 In short, the reason why hr > r is considered a sound change is that it is a residual case:14 it is not built on an alternation nor is there a conflicting regularity that would call for the setting up of a separate channel of transmission (in other words, hr- always goes to r-). It is considered a sound change because it is ‘regular. We already know how to interpret competing regularities. We are free to consider the replacement of g gh d dh by d, as well as the replacement of g gh and of d dh by z and d, respectively, either as examples of ‘sound change’ along two lines of descent (‘true Persian’ and ‘Median’) or as an example of dialect borrowing interfering with the true line of descent for Old Persian. If this is how we reason, there is no occasion for surprise in Leskien’s formula that sound change occurs without exception, since it is a tautology; and of the three processes envisaged above only two would be properly defined, namely, (1) regular sound change, and (3) change other than ‘sound’ change. How, then, has it been possible again and again to raise the question of the regularity of sound change with the thought that not only is it a matter of observable fact (rather than of definition) but also that sound change can be irregular and that Leskien was wrong?
This is not the place to study in detail the arguments with which Chen and Wang wish to prove that the neogrammarians (and, presumably, scholars like Meillet and others after them) “underestimated the extent of exceptions to the so-called ‘exceptionless’ sound laws.”15 But ‘extent’ is the last thing that matters; the literature is full of cases in which the truly ‘regular’ development is reported buried deeply under the surface. In any event, Chen and Wang themselves are champions of ‘lexical diffusion’. Their principle of explanation is that processes that in retrospect are judged to be regular sound changes, or that may be predicted to end up as such processes, do not affect all morphs at “one and the same time” or, shall we say, at the same pace, but rather spread from morph to morph ‘until’ they become regular. Some of their material may be open to criticism, but much of it is greatly to the point. Only, it is a mistake to believe that, however misleading Leskien’s wording may have proved to his readers, lexical diffusion is incompatible with neogrammarian principles as practiced.
In Glarus German, ‘ew’ and ‘iw’ have almost merged by sound law (the formulation is that of Leonard Bloomfield, a defender of the regularity principle) into y:, but for the words for ‘deep’ and for the name of the local knoll known as the ‘Kneeridge’, both of which have œj (the word for ‘knee’ itself has y:).16 This shows “that the y: for old ‘ew’ is really an importation”—the ‘real’ Glarus form is the ‘exceptional’ œj. (The surrounding areas have y: throughout.) Of course, once these two items, too, are borrowed with y:, the dialect borrowing will have become a sound change. The traditional argument can only be which is which: have the Glarus speakers shifted to the surrounding dialect so that the two exceptional words are loanwords from old Glarus German (as, on a gross level, French—namely the Latin of Gaul—contains Gaulish loan words), or do we decide, as Bloomfield would like to do but for the precarious material, that all the other words are borrowings so long as at least one item with œj is left in the speech of the locality? It should be obvious that the quarrel is not about the events in dialect history, which remain the same, in all their presumable complexity, in both descriptions. Rather, the issue is the wording. To be sure, there remains, if one wishes, the criterion of the basic vocabulary: the survival of the local place-name, Kneeridge, in its ancient form, may be taken as justification to view the process as borrowing (the most familiar terms stay to the last) rather than as language shift or learning (the most familiar words are changed first). But this merely opens up the depths of possible disagreement on the notion of basicality itself.
Advocates of ‘sporadic’ sound change (that is, of the separate validity of class 2, above) will have their claim neither proved nor faulted until they can specify the circumstances under which competing regularities are to be interpreted neither as (a) ‘analogic’ play among allomorphs the morphophonemic nature of which creates phonemic recurrences, nor as (b) representing different channels of transmission, including the one dubbed the main channel, but instead as (c) sporadic ‘sound’ change. If this is only a case of labeling, in all sobriety, regularities of nonanalogical origin for which we cannot name or care to construct a reasonable historical background (the Medes or the Scandinavians), ‘sporadic’ means no more than ‘somehow competing’ and is noncommittal. It is, however, sometimes said that such processes tend to have special phonetic properties; reportedly, they are distant dissimilations and assimilations, metatheses, haplologies, and the like. In actual fact, however, these so-called minor sound changes seem to offer no more difficulty to interpretation than the major ones such as, say, ordinary and ‘regular’ contact assimilations. The often-cited Greek and Latin haplologies of the type hēmimédimnon > hēmédimnon (‘half-medimnos [a measure]’), nūtrītrīcem > nūtrīcem (‘nurse [acc.]’) are apparently like other sound changes in being typologically constrained (they may be said, if one so wishes, to help maintain a favored syllable structure for words), as well as in having transparent analogical re-formations for exceptions (hēmi-médimnos [‘remade’] also occurs). As a matter of fact, the ‘exceptions’ to the more ordinary sound laws are often far richer than what we have here. This, however, is not even the issue. It seems that sporadic sound change is either a contradiction in terms or merely a traditional and not particularly well chosen collective designation for other than main-channel material.
It is odd that Leonard Bloomfield, who did so much to reinterpret nineteenth-century principles in clearer and more consistent language, should have thought the minor sound changes so “very different from those covered by the assumption of [regular] sound change”; odd, also, that he may have continued to regard this ‘assumption as material rather than definitional.17 Part of the explanation is that the tradition on this point was both exceptionally powerful and at the same time encumbered with another, extraneous problem: that of graduality. In classical nineteenth-century linguistics the sounds of language were, after all, frequently thought of as forming a substance (liable to alteration in time) to which speakers only give utterance, almost as if they were performers reciting some preexistent text. Consequently, the question usually asked was not what it is that speakers do, in the exercise of ‘normal speech activity’,18 that looks in retrospect like sound change, but rather how speakers cope with alteration when they discover it. Graduality, that is, continuous change in imperceptibly small steps or something of the sort, was postulated to explain why speakers will accept the merging of precisely those phonemic contrasts the observance of which seems to be the essence of ‘normal speech activity’ as long as that observance lasts (it goes against the grain to imagine that speakers will begin to make click and lick alike, and yet this is what happened to hl-/l-, and later to kn-/n-): if the speakers had been aware of the aberration they would have corrected it, and it was graduality that prevented awareness. This speculative notion of necessary physical graduality has had remarkable staying power. To what extent sound changes move in fact through small or infinitesimal stages—in the sense that individual speakers go through the paces as they live through the change—is an empirical question so far as measurements of variation in ‘apparent time’ are concerned; but there is nothing empirical about proclaiming that all sound changes must be physically gradual as some phenomena disappear and are replaced along what the historian selects as his line of descent.19 Moreover, it is entirely characteristic of classical historical linguistics that for many of its contexts the issue of graduality does not matter. The formulation of sound changes and the validity of the resulting alteration in question could or could not have been gradual.
One reason for this lies in the familiar truth that a distinction may be made between phonemic and (merely) phonetic changes and that the distinction is fruitful. Subphonemic change, we often say, may for certain purposes be disregarded. It is perhaps not relevant to the phenomenon studied here. Subphonemic change is, in a sense, without shape so that its incidence or absence cannot be clearly ascertained in a given case; and although it may have direction (and is often reported as having direction), it is, for instance, capable, in theory, of having its direction reversed. All this leaves us without a clear-cut criterion to decide whether or not innovation has occurred, except in some statistical sense—a sense that is clearly uncongenial to the fundamental historical-comparative practices, whatever standing and potential it might otherwise have. The traditional reliance on phonemic changes is surely not just an accident of scholarly history, though we must also remember that many of the virtues of our familiar methods were born of a necessity: recorded material from the past yields little subphonemic detail, and our procedures for reconstruction have naturally been judged first and foremost by their ability to give us the kind of information for unrecorded speech forms that written records give us for known ones.
A phonemic merger, as we have seen, is a different matter. Not only do we claim that we know whether or not the absence of a contrast is a fact for a given form of speech, as compared with another (cognate) form of speech, but also we think of the merging of a contrast along a line of descent as irreversible, because its reversal, further down the line of descent, would be a split, and, so we say, we neither know nor can construct a mechanism for phonemic split per se. This is, to be sure, a circularity. Lines of ancestry and descent were defined on the basis of merger in the first place, and it is only in deference to that definition that wherever we find a (possibly subsequent) form of speech with a contrast where another (possibly earlier) form of speech has no contrast, we posit a collateral rather than a directly ancestral relationship for the two. We had of course arrived at the definition because of our interest in the workings of phoneme substitution and because of our desire to find the irreversible process we need for purposes of reconstruction. Hence we must assign what would otherwise qualify as a case of spontaneous split to a relationship other than descent. We say, for example, that a population A may acquire the contrasts of a form of speech B20—that is, that it may learn a new language or dialect perfectly’ (rather than ‘imperfectly’). True to our framework, we record neither ‘change’ (i.e., a conflicting regularity) affecting B-words borrowed into A, nor ‘sound change’ occurring in a descendant of B,21 nor even sound change occurring in a descendant language of A such that this language has come about by ‘total borrowing’ from B. Instead, we only say that B has not undergone sound change even as it has spread to new speakers. If the acquisition of B by A-speakers is less than perfect with regard to some other contrast, and there is a substitutive merger, this will then constitute a sound change affecting B (and not A).
Roughly and schematically, these are some of the rules of what is by no means an empty game even if the rules are by way of definitions. There is a conviction that this way of determining ancestry is not inappropriate to social, political, and demographic history and that the trees that result are not too far removed from the pedigrees and tables of descent with which students of history have traditionally animated their chronicles.
The elaborate phonologies that are the glory of historical and comparative linguistics do of course contain many instances of phonemic split or conditioned sound change. But these follow Polivanov’s Law; they are, in other words, not independent, spontaneous split processes per se, but consequences, or, better, aspects of mergers. The specific nature of the relation between a split and the merger with which it belongs must in part be viewed as a matter of notation. Accepting the more usual notational conventions, one distinguishes two types. The first (1) is characterized by the circumstance that the splitting-off allophone is the very one that does the merging (with an outside phone), as when (1a) d before -er goes to ð in early modern English, with the result that weather (with an old d) and (bell) wether (with an old þ) become homonyms; din other environments (= other allophones of d) appears in modern English as d, occurring in environments in which ð (from old þ) also exists: load, loathe. It is clear that the split of d (load, but weather) depends on the effective merger of weather with wether, etc., to attain phonemic standing, since until this merger is a fact, the (presumably) special allophone of d before -er is still just a positional variant of /d/. Just so, (lb) Indo-European splits, in early Germanic, into */þ and *t—the latter (for instance) after *s, where *t merges with *d. But this old *d goes to *t generally (that is, not only after *s; =[z]), so that we have, in one of the Germanic descendants, Gothic, ast- (‘branch’; from and ist (‘is’; from *-t), and, on the other hand, þah- (‘be silent’; from *t-), but tah- (‘tear apart’; fromSee Fig.6. The difference between (la) and (lb) is only that in (la) the splitting phoneme of the earlier stage and the merged phoneme of the later stage occupy merely adjoining places in the kind of phonological framework that one might be tempted to set up as descriptively common to both stages. In (lb), on the other hand, they are the ‘same’.
Polivanov’s Law covers a second kind of split (2), which is on the surface different from the first: here it is a merger in the conditioning environment that makes former allophones contrast with one another (Phonologisierung), without necessarily or essentially involving any physical alteration in the segments that are thus changing their status. In Indo-Iranian the older (proto Indo-European) sequences *pe *po *kwe *kwo have turned into pa pa ča ka, respectively; *k w has split into č and k. It is customary and, indeed, natural to stress the compensatory function of this variety of phonemic split, on which we tend to look as separate from, if connected with, the merger. We say, in fact, that the vowels *e and have merged (as it happens, into a) and have, in the process, produced one homonymous pa, while the earlier *kwe (> ča) has remained different from *kwo (>ka). That the difference is regarded as having shifted from the vowel to the consonant is a tribute both to the circumstance that the ‘same’ *e’s and *o’s have resulted in homonymy elsewhere (e.g., after *p), and to the phonetic facts. But these are not guaranteed to be clear-cut. If, as is only plausible, the ‘allophone’ of a after č was different (in ‘onset’, degree of fronting. . .) from that occurring after fc, the matter takes on an indeterminate aspect such that only a further appeal to the č’s and k’s in still other environments could furnish some vague guidance. Apart from this, it is notationally possible to retain the vowel phone as the seat of distinctiveness while making the preceding consonant dependent on the vowel; in other words, it is possible to decree that it is the lowering and centralization, to the extent that it took place, of both *e and *o that was sub-phonemic. True, such a notational choice is not pleasing even in this instance and would be quite absurd in others; but the fact that it can be entertained at all gives us some understanding of how the two kinds of split process are related. For if the first (1) may be represented by Fig.7, the second (2) might be represented by Fig.8, where, at the later stage, +a is a vowel with distinctive fronted on-glide, and k is = [č] before +a and = [k] before a.
We may say in conclusion that as some of the generalizations from historical and comparative work were stated they were bound to offend common sense and provoke attacks on common-sense grounds. We asked ourselves earlier how it was that these attacks never really succeeded in shaking the edifice that had been erected in laborious trial- and-error fashion. The answer lies partly in the fact that the generalizations were problematic as such: the language in which they were couched necessarily had a history of its own, replete with traditional motifs (descent, speakers’ awareness, language as a physical object or substance, the naturalness of alphabetlike segmentation, etc.) not to mention traditional reticences; and while it must have been intuitively clear, at the time of formulation, just how they fitted the data, they became vulnerable as they were misunderstood later. The edifice itself has, by and large, survived and developed further on its own immense merits, which are chiefly those of internal consistency and historical concreteness. It does not stand and fall with the generalizations of the past since it was never put up in the first place by simply carrying out prescriptions. But this is not to say that generalizations are useless and that existing ones should not be reexamined. We have pointed to a few concerns in such a reexamination. The most pressing ones seem to be of two kinds: there are definitional tautologies—of a wholesome sort—to be recognized both in the business of identifying a line of descent and in separating so-called sound changes from other changes; and there are subtle relationships to be interpreted between the phenomena of variation and of replacement at the crucial point where the synchronic study of speech communities and the domain of the historian abut.
Although long known as ‘comparative grammar’, historical and comparative linguistics became more and more preoccupied with phonology as time went on. We have tried to show that this was not an accident and that it was instead connected with the fact that some sound changes are mergers and that mergers, under certain implicit or explicit assumptions, may be seen as ‘irreversible’. We might also say that sound changes, unlike some other change processes, produce effects that are ipso facto recognizable as innovations, and that it is on this powerful property that the so-called comparative method22 of reconstruction—that is, of reconstructing the phonemic shape of morphs and of certain morph sequences—rests. The de facto adoption of this principle (however formulated) is the best illustration of the process of formalization that linguistics underwent, so early in its history and “against the predisposition and expectation of the discoverers.”23 Precisely because it was not expected, it caused intellectual discomfort, and there was a good deal of discussion of the chances for comparative reconstruction of entities other than morph shapes. There is much to be said, and even more yet to be found out on this subject.
One of the forms the discussion took is worth looking at. This is the search for ‘semantic laws’, to which Stephen Ullmann devotes a subchapter in his work on The Principles of Semantics.24 This search, he says, is “as old as semantics itself” in a general sense; but it had a special meaning for the neogrammarian and post-neogrammarian linguists because of the “existence of sound laws.” That language has two dimensions, ‘sound’ and ‘meaning’, must be a very general belief, existing on many levels from folklore to technical sophistication, though no doubt suggesting very different things to different individuals. In the general folklore and in what passes for ordinary common sense the idea that the two factors are somehow symmetrical is perhaps not so pronounced; rather, there is a familiar feeling that words (1) consist of such-and-such sounds (as a rock has such-and-such a composition, or such-and- such measurements), and (2) possess the property of meaning (or naming, or referring to) something (as a rock may have the property of whiteness, or beauty, or usefulness). The scientific profundity of this feeling is not the issue, but its all-pervasiveness is; and linguists, in the act of speculating, have often appealed to all-pervasive common sense as a source of self-evidence. On the other hand, historical linguists had developed the distinction between sound change and changes other than sound change, including change of meaning, as an important principle, and this did suggest a degree of symmetry. Since sound change proceeded by (neogrammarian) sound laws, the quest for semantic laws to match sound laws was natural enough. Ullmann reports that scholars were largely discouraged because they found a lack of symmetry in this respect. No semantic laws, of a kind to match sound laws, could be discovered. He reproduces Nyrop’s famous statement to the effect that
La science phonétique a. . . acquis une telle sûreté, que dans beaucoup de cas il est possible de prédire le développement que subira tel mot, vu qu’il n’y a pour un groupe de phonèmes qu’un nombre assez restreint de changements possibles. Il en est autrement de l’évolution sémantique; ici les conditions qui déterminent les changements sont tellement multiples et tellement complexes, que les résultats défient constamment toute prévision et offrent les plus grandes surprises.25
This is, as Ullmann implies, rather surprising in itself. Nyrop claims that there are universal, ‘panchronistic’ laws of a sort that allow us to predict sound changes; to predict, in other words, what are (and what were in Nyrop’s day) called ‘sound laws’—namely, the specific, presumably ‘exceptionless’ replacements observable at given times and places or, as we said above, characteristic of given channels of transmission. He then declares that no such laws are available to predict changes of meaning.
We need not pursue the matter of panchronistic laws since the claim for the existence of such laws in the realm of phonology in the form here made is fantastic, and since the compelling question was in any event quite different, aiming, as it did, at the semantic analogs for the specific time-bound and space-bound sound laws. Gustaf Stern, who was aware of the problem, offered a famous example, as follows:
English adverbs which have acquired the sense ‘rapidly’ before 1300, always develop the sense ‘immediately’. . . . Exceptions are due to the influence of special factors. But when the sense ‘rapidly’ is acquired later than 1300, no such development takes place. There is no exception to this rule.26
The formulation is meant to recall the form in which sound laws (that is, statements of sound changes) are written or should be written; in particular, it contains the requisite references to time and place. It does indeed parallel the kind of statement that is sometimes found in the more perceptive and interpretive historical phonologies, if the data will allow it. Without going into the technical detail, it is enough to remember descriptions of how sound changes will repeat themselves for a time: in pre-proto Greek, an Indo-European intervocalic s becomes first h and then zero; later, in the special Greek dialect of Sparta, s’s which have again arisen between vowels from other sound changes) are again changed to h (and, to round out the parallel to Stern’s formula, it is not known that the ‘same’ change has occurred again in the modern descendant—if indeed it is the ‘true’ descendant—of this dialect, Tsakonian). Stern’s observation was fruitful and to the point, and it illustrates an interesting case. But questions remain. The history of s-like sounds in Greece is not statable as a simple, single sound law, but rather as a typological concatenation of sound laws involving judgments on what constitutes ‘sameness’ from stage to stage (in what sense are the Indo-European s and the Spartan s the same?). The constituent individual sound laws do not have this form, but, then, they cover more: in fact, ever since August Schleicher, total accountability has been the professed or implied requirement for a valid historical phonology. Is it only the much-invoked greater number and complexity of semantic units or of morphemes, as contrasting with phonological units, that makes us despair of any hope of writing a full semantic history for a given language in terms of exhaustive Sternian laws?
One might well hesitate to go further, on the ground that the analogy between phonological and grammatical processes has led to trouble before. But some concepts are after all legitimate both in phonological and in grammatical analysis—such as the notion of contrast. Besides, there remains the fact that the question was posed, rightly or wrongly, by serious scholars of the past and present: if it leads into difficulties, it might be helpful to learn something about these difficulties.27
Let us recall the two styles of statement for morphemic change in general, including semantic change: we have a choice between saying (II) that say changed its meaning from the older reference to formal announcement, etc., to the more general meaning of today, which used to be expressed by cweðan, on the one hand, and, on the other, (I) that cweðan was replaced. It so happens that the former style is the one commonly used in description of semantic changes; semantic change is something attributed to a given morph, in this case, secgan-say. It is also a fact that it is the latter style that parallels the customary form of the sound law: as s is replaced by, or ‘goes to’, h in Greek, so cweðan is replaced by—one does not ordinarily say ‘goes to’—say. But if this is taken seriously, we end up with a semantic law of a disappointing sort, inasmuch as it tells us something that we do not usually question. Just as s > h is a sound law because we formulate it in such a way that all instances of s are replaced by h, just so it is a regular law that cweðan goes to secgan-say. It happens in all sentences and discourses in which cweðan occurs; and to the extent that it does not—in quoth, and certainly after the be- of bequeath—we may, and do, speak of conditioned replacement, and hence of conditioned semantic change. No doubt it is worthwhile to go on looking for patterns in semantic history. But the analogy with sound change holds no promise. We have had that all along, and it is empty.
1. There is no implication here that a speech community in a ‘given’ place is uniform.
2. Presumably with [s] for both.
3. There is, of course, a problem in predicting which is the ‘nearest’ phonemic entity; studies dealing with the phonology of loanwords are full of surprises. Suppose a source language has st as a word-initial cluster while the importing language does not: what general principles are available to predict that the replacement of ste. . . will be este. . . rather than te. . . sete. . . or what not (assuming that all these sequences exist in the borrowing language)? The real interest lies in turning the question around: it is the observed substitution that tells us which entity is ‘nearest.’
4. Perhaps as s and ‘t.’ Real Japanese speakers are said by Lehmann (1962:77) to replace English th with s.
5. Something like this is operative in Garde’s principle (1961) that a non-contrast dialect area will increase at the expense of a contrast area, other things being equal; but see note 20 below.
6. The alternative would have to be that A exists side by side with B and C (and B with C) in unchanged fashion, in violation of the truism above.
7. Brandenstein and Mayrhofer (1964:157).
8. As in vispa-zana ‘containing all races,’ instead of *visa-dana, where -sp- is also Median (ibid., 12).
9. Special provision of a formally difficult kind has to be made in cases where there is both sound change and semantic change in the same etymon.
10. That is, hold on to the phonemically identical item.
11. The morphs referred to are of course - ed and -en. The vowel alternation in the stem (help/holp) should be disregarded.
12. Disregard the word staff. We are here only concerned with stave.
13. Again, this is complicated by the fact that morphophonemes may occasionally spread. The past tense dove is more recent than dived.
14. The relation of bulk to residue is of course reversible. If one believes that sound changes can be identified by inspection the non-sound change examples may be treated as residues (Bloomfield, 1933:352).
15. Chen and Wang (1975:260).
16. Bloomfield (1933:339).
17. In Bloomfield (1933:391) and Hockett (1970:247), we have Bloom-field’s memorable review of Hermann’s Lautgesetz und Analogie: “we formulate this type of change. . . by saying that phonemes change [B.’s emphasis]. This is an assumption.” But then B. goes on to allude to a good many factors in the situation (“our assumption fits the fact that. . . ”) and to conclude that “all this of course, is what Leskien meant by his laws without exception’.” This sounds as if B. had in mind something close to a definitional interpretation of this particular assumption. On the minor sound changes see Szemerényi (1970), who reaffirms their special status but does not explain how an irregular sound change can be classified as a ‘sound* change without contradiction. Latin intensives like cantitare are in no way different from hēmimédimnos and thus from innumerable analogical restorations not involving ‘minor* but ‘ordinary’ sound changes. Reduplications do indeed furnish cvcv sequences “in great numbers,” but precisely not in the word-interior positions where such sequences, or some such sequences are haplology-producing; when they do come to stand in word-interior position (*repepulī > reppulī) through prefixation, they are in large part syncopated, with the result that haplology does not apply. The point, supported by Posner (1961: 209), is that in this respect as in others the so-called minor sound changes are not essentially different from the major ones; and where they seem different at first, the exception could prove the rule. It could be, for instance, that distant conditioning (of assimilations, etc.) produces the appearance of sporadicity by leaving so much more scope for morpheme boundaries to intervene and hence for analogic change to operate (Hoenigswald, 1964:213). Compare Bloomfield (1933:391) to the effect that “it is possible that they [the minor sound changes] are akin rather to. . . analogic change and borrowing.” See also Malkiel (1968:33-45).
18. Hermann Paul’s key concept (1898:29, etc.).
19. There are other uses of the term ‘graduality’: there may be a gradual increase in the amount of vocabulary affected, in the number of speakers, localities, dialects, etc., participating, or a gradual widening of the phonological conditioning in which a change is ‘regular.’ There is room for a study of another traditional motif, namely the distinction between internally caused (‘evolutive’) and externally caused (‘adaptive’) sound changes (see, for instance, Andersen, 1972; Vachek, 1975). The prima facie reasonability of this distinction depends to some extent on the prima facie acceptance of the line of descent, and of the homogeneous speech community. But this does not begin to describe the problem fully, nor should anybody deny the existence of restrictive universals, typologies, and directionalities as ‘internal’ factors.
20. That this happens should need no emphasis, but it does of course raise questions with regard to Garde’s principle, note 5 above. In Low German, West Germanic þ and d are not distinct from each other (both are d) ; in Standard German, WGc þ is d, and WGc d is t, and yet Standard German is spreading, though perhaps by some process not intended to be covered by Garde. The example is trivial; the question is to find the limits of the principle.
21. Based on an A-substratum.
22. In the narrow, technical sense of the word.
23. Bloomfield, in Hockett (1970:425).
24. Ullmann (1957).
25. Ibid., p. 251.
26. Quoted after Ullmann (ibid., p. 254).
27. Bazell (1965:95) is quite right in saying that “the comparison with phonological change remains superficial and sterile,” and the reason, so far as I can see, is the faulty identification of homologues in the analogy. The earlier generations of scholars who may have gone astray in this respect were not the linguists who were primarily concerned with the discovery of sound laws and who hammered out Indo-European comparative phonology (and whom it is absurd to accuse of failing to see or refusing to acknowledge ‘exceptions’ in the raw material—see note 15 above), but rather those who speculated on what laws of semantic change would be like if there were any. It is entirely pertinent to analyze their efforts.
Andersen, H. 1972. Dipthongization. Language 48:11-50.
Bazell, C. E. 1965. Review of J. H. Greenberg, ed., Universals of Language. Journal of Linguistics 1:94-95.
Bloomfield, L. 1933. Language. New York: Henry Holt and Company.
Brandenstein, W., and Mayrhofer, M. 1964. Handbuch des Altpersischen. Wiesbaden: Verlag Otto Harrassowitz.
Chen, Matthew Y., and Wang, W. S-Y. 1975. Sound change: actuation and implementation. Language 51:255-81.
Garde, P. 1961. Réflexions sur les différences phonétiques entre les langues slaves. Word 17:34-62.
Hockett, C. F., ed. 1970. A Leonard Bloomfield Anthology. Bloomington: Indiana University Press.
Hoenigswald, H. M. 1964. Graduality, sporadicity, and the minor sound change processes. Phonetica 11:202-15.
Lehmann, W. P. 1962. Exercises to Accompany Historical Linguistics: An Introduction. New York: Holt, Rinehart and Winston.
Malkiel, Y. 1968. Essays on Linguistic Themes. Berkeley: University of California Press.
Paul, H. 1898. Prinzipien der Sprachgeschichte, 3e Aufl. Tübingen: Max Niemeyer Verlag.
Posner, R. 1961. Consonantal Dissimilation in the Romance Languages. Oxford: Blackwell Scientific Publications.
Stern, G. 1931. Meaning and the Change of Meaning in English. Reprint 1964, Bloomington: Indiana University Press.
Szemerényi, O. 1970. Review of Cardona, On Haplology in Indo-European. Language 46:140-46.
Ullmann, Stephen. 1957. The Principles of Semantics, 2d ed. Reprinted 1967, New York: Barnes and Noble.
Vachek, J. 1975. Zum Zusammenspiel von internen und externen Faktoren bei der Sprachentwickelung. In D. Cherubim, ed., Sprachwandel. Berlin: De Gruyter.