Observations on the "Neogrammarian Manifesto"

by Suzanne Kemmer

Hermann Osthoff and Karl Brugmann, "Preface to Morphological Investigations in the Sphere of the IndoEuropean Languages, vol. I," 1878.

Note: Page references are to the translation in Lehmann (1967) (see Selected Bibliography).

This preface to the newly established journal Morphologische Untersuchungen had a strong influence on a generation of scholars who were evolving from 'philologists', investigators of the arcane facts of older languages, into linguists, or 'language scientists' (Sprachwissenschaftler, the term later favored in Germany).

I call this the "Neogrammarian Manifesto" because it used polemic, rhetorical language befitting what its authors wanted to introduce and to declare as a revolutionary change in ideas. Some specific claims or phrases illustrating the article's consciously groundbreaking character include the following:

The authors O&B describe what the new approach should be. They call for an investigation of

They maintain that investigators need to: (bracketed material is my view of the implicit contrast they make) The following, from late in the article, is a famous quote which was taken like a call to arms (p 202):
Therefore: only that comparative linguist who for once emerges from the hypothesis-beclouded atmosphere of the workshop in which the original Indo-European forms are forged, and steps into the clear air of tangible reality and of the present in order to get information about those things which gray theory can never reveal to him, and only he who renounces forever that formerly widespread but still used method of investigation according to which people observe language only on paper and resolve everything into terminology, systems of rules, and grammatical formalism and believe they have then fathomed the essence of the phenomena when they have devised a name for the thing--only he can arrive at a correct idea of the way in which linguistic forms live and change, and only he can acquire those methodological principles without which no credible results can be obtained at all in investigations in historical linguistics and without which any penetration into the periods of the past which lie behind the historical tradition of a language is like a sea voyage without a compass.

The elevation of 'tangible reality' over 'gray theory' is a call for empiricism, i.e. more firmly-grounded data, which they believe requires a new observational methodology. The complaint about scholars who 'observe language only on paper' is a jab at those who get all their data from books instead of listening to language to see what kinds of things happen in natural speech. (Not that these writers ever came to focus on modern spoken languages; they continued to specialize in the ancient languages, although many of their followers took them at their word and began studying the modern dialects of European languages.) They apparently dismiss the technical aspects of comparative linguistics (terminology, rule systems, etc.) as the field then had developed, or at least any of this which is not grounded in direct observation of living languages.

O&B stress that language is not just about articulation and its mechanisms; it is 'psychophysical', which, in modern terms, means it has a crucial cognitive dimension. Studying the operations of this psychophysical mechanism is necessary to acquiring some idea of the general principles of language--'what is possible in language in general'. Generalization was certainly sought by the earlier comparatists; in fact the generalizations they called 'laws' were considered the hallmark of progress from the days of Frans Bopp. But Osthoff and Brugmann mean a different sort of generalization, one going beyond the particular historical facts of a language and which embraces how languages in general can and do change. In this way they foreshadow the later idea of a theory of diachronic typology first proposed by Greenberg almost a century later.

The call to observe language use by speakers also of course prefigures the field of sociolinguistics: studying 'the way in which linguistic innovations, proceeding from individuals, gain currency in the speech community' is an important focus of attention in modern sociolinguists.

Many of O&H's contemporaries took them at their word and began to study the modern dialects, which had been neglected in favor of ancient texts. It took a long time, however, for the new field of dialectology, and its newly enriched empirical base, to start to develop into a true field of sociolinguistics with a focus on how language functions and spreads in the speech community.

O&B's aim is to do two things. First, they wish to emphasize the strongly rule-governed nature of one type of change, namely what might be termed 'ordinary sound change'. This property makes the study of language change amenable to the discovery of law-like principles, almost like those of the natural physical world. In effect, such law-like behavior makes it possible to see Linguistics as a science, which can develop its own empirical methodology for the discovery of general principles.

Second, O&B want to create a larger framework of understanding in which regular sound change is just one process of change. They presented analogy as a kind of opposing force, which, in effect, can 'mess up' the results of regular sound change so that forms appear that are unexpected given the predictions made by regular sound laws. They feel that simply invoking 'exceptions' to the sound laws is, although common among comparatists, an unacceptable practice. So they urge a search for motivated exceptions, and a study of the psychological motivations for such exceptions.

Probably the idea that the article is best remembered for is its strong and explicit claim about the exceptionless nature of sound change. The key paragraph reads:

First, every sound change, inasmuch as it occurs mechanically, takes place according to laws that admit no exception. That is, the direction of the sound shift is always the same for all the members of a linguistic community except where a split into dialects occurs; and all words in which the sound subjected to the change appears in the same relationship are affected by the change without exception.

The claim is meant to be very strong, although there are some hedges in this passage. For example, there is the phrase "inasmuch as it occurs mechanically". 'Mechanically' in their usage appears to mean taking place 'automatically, like a machine', given the right input conditions. That is, if the phonetic conditions are met, the change naturally and inevitably occurs. But by 'inasmuch as', do they mean to the extent that it occurs 'mechanically'? If so, it would seem to suggest that there could be some degree to which a given sound change happens non-mechanically. If so, then they would allow some 'wiggle room' for factors that could make a particular sound change not completely regular. Or is 'inasmuch as' just an underspecific way of indicating 'seeing as; because'?

I think that with this caveat they were trying to allow for sporadic sound changes like dissimilation and metathesis--sound changes, to be sure, but obviously not regular. These seem to involve some psychological aspect or as we would term it now, processing factor, at least when they first occur and begin to spread. These types of irregular change were well known and of course O&B would want them to be excluded from their generalization, as they make explicit in Footnote 5. It is not that clear to me that one can distinguish the two types, regular and irregular, merely by identifying them respectively as 'non-psychological' vs. 'psychological', because presumably there is some cognitive processing involved in any speech production. Still, it is reasonable to hypothesize that there is nevertheless some qualitative difference, and to look for evidence of different kinds of processing, now that we have better empirical methods of studying speech.

A side-note: My own theoretical orientation makes me suspect differences between regular and non-regular processing are never very sharp when you look at them very closely: there is typically a gray area between them where some phenomena fall. This is an issue that is central in the big debate between dichotomists like Pinker (following traditional generative dichotomism), and his opponents the connectionist modelers as well as usage-based linguists and psychologists like Bybee, MacWhinney, and Bates. Still, I think the evidence adduced by both sides must be taken into account. Although the brain does not work by sharp distinctions, leaving the extreme dichotomist position unlikely, there could be a noticeable difference in the extremes of the continuum that could account for differences in regular and irregular processing--in this case, the way that typically regular sound changes like assimilation, vs. typically irregular changes like dissimilation, arise and spread.

In the quote above there is also an exception phrase allowing for lack of regularity in the case of dialect splits. So if a sound change is spreading through a group, but the group is at that time splitting into somewhat separate speech communities, then it need not affect all the members of the one-time or larger speech community. One of the subgroups could be bypassed by the sound change.

This view seems reasonable enough theoretically, although in practice it allowed those arguing for the absolute regularity of sound change to account for exceptions by appealing to 'dialect admixture'--an easy (perhaps too easy) claim to make when older texts were copied and recopied by scribes from different areas. It is also difficult in real life to determine when a dialect split has actually occurred, so presumably in a border area where one is in progress, different speakers will behave differently: some will have the change, some won't. Presumably in this case, if O&B's thesis is correct it would make the prediction that a particular speaker exhibiting the innovative sound change should have not only that feature, but numerous other features that link his/her speech to that of other people affected by the sound change, and which separate it as a dialect from the next 'dialect' without the change.

Unfortunately such neat dialect separations in which all features line up to clearly differentiate varieties from one another are the exception rather than the rule in the real world. The idealized picture of language split in which one population first moves away and falls out of contact with its parent language, in the process developing a set of its own characteristic innovations which distinguish it clearly from its sister dialects until finally turning into a separate language, MAY have occurred in the cases of migrations to new areas such as happened with the various branches of Indo-European (although some would dispute that this neat and relatively quick separation ever occurred, even in the case of these indisputable population movements). But Europe has been developing dialects in a relatively enclosed area, both with and without large corresponding population movements, for many hundreds of years. Dialects split along not only geographical lines, but also social lines, as modern sociolinguists have shown us, and changes in natural classes of sounds do not have to proceed all at the same pace.

The progression of the Second Germanic sound shift, otherwise known as the High German Consonant shift, shows the stepwise progression of related sound changes clearly. The series [p], [t], and [k], and their geminated versions, underwent lenition to corresponding fricatives and affricates, depending on position in the word. (Compare Low German pipe 'pipe', with the original p's preserved as in the other branches of Germanic, with standard High German Pfeife 'pipe', in which initial [p] was affricated to [pf] and medial [p] was fricated to [f]. (Geminate p's also turned into affricates, as in Apfel 'apple'.) However, the changes in this class of stop consonants spread not as a unit but in sequence, and show different 'reaches' of geographical spread from south, where all three changes started, to north. [p] underwent such changes first and its changes spread farthest, followed by [t], and [k] changed slowest and spread least far to the north. The result was the phenomenon called "the Rhenish fan" in which the change isoglosses create a fan-shaped distribution of dialects along the Rhine, each roughly wedge-shaped area showing a different degree of the overall change in stops: those to the south showing more of the parts of the change and those to the north showing fewer, in a pattern reflecting the degree of spread.

The last part of the highlighted passage above makes a clear claim that a sound change in a given speech community must affect all words meeting the right conditions. Dialect admixture in texts might once again be appealed to in order to account for exceptions, but here at least a clear distinction in principle is possible: the prediction would be that any given speaker should exhibit a sound change across the board for all items of vocabulary in which the right phonological conditions apply. This prediction has been in modern times falsified. The work of Wang, Labov, Bybee and others clearly point to the existence of the phenomenon of lexical diffusion, in which some sound changes move gradually through the lexicon of a group and of individual speakers. I suppose a die-hard modern-day Neogrammarian might appeal to 'speakers borrowing individual lexical items from another dialect', much as present-day generative historical linguists like Anthony Kroch appeal to speakers internalizing 'two different rule systems', one with the change (their own "real dialect") and one without (a dialect borrowed for the purpose of communicating with those for whom is is their "real dialect"). But it has been shown (e.g. by Bybee) that sound change is influenced by the usage factor of lexical frequency. This kind of systematic influence could not be readily accounted for by a random phenomenon like individual borrowings (or separate dialect grammars for that matter).

Some scholars suggested that sound change only looks regular after the fact, at the end of its progression through the lexicon and the community. The Neogrammarian account would seem to assume a rather sudden change in the speech of an individual, in which all words were affected, and a rather rapid diffusion through a speech community (until/unless it is undergoing a split defined by other linguistic differences besides the change in questionm, as noted above.) Still, sound change does show a surprising degree of regularity even when we look close-up at real speech communities. Labov, in his article "Resolving the Neogrammarian Controversy" (in Language in, I think, 1979), sought to solve this apparent paradox by proposing that some (phonetically defined) types of sound changes are very quick and very regular, "Neogrammarian"-type changes, and others, still rather regular (unlike dissimilation and metathesis) are subject to slower, diffusion-like spread. Bybee has her own typology of types of regular and less regular sound changes that as far as I recall, does not line up with Labov's. The issue is not completely settled, in either Sociolinguistics or Historical Linguistics, but at least lexical diffusion and its relation to frequency is widely acknowledged.

However the issue will ultimately be resolved, there is no doubt that Osthoff and Brugmann's claim, along with the rest of their claims and rhetorical flourishes, caused quite a stir in comparative linguistics.

Instead of the modern concept of lexical diffusion, or splitting their unitary idea of regular sound change into completely regular and quasi-regular types of change, the Neogrammarians appealed to a different opposing force to regular sound change, namely analogy. Analogical change in language is an accepted fact of change in modern historical linguistics, although it no longer plays the same role of 'preserving' the hypothesis of absolutely regular sound change.

The following lists sum up the contrast that O&B drew between sound change and its theoretical 'opposite number', analogy. In contrast to the posited unitariness of sound change, analogy was recognized as more complex, disparate, and psychological in nature. Sound change is:

"Predictability" above does not, of course, refer to the predictability of the change itself, but predicting, for the analyst, the way a given item of a particular sound shape should be affected (or not) by a posited regular change.

Analogy, on the other hand, is:

Sound change and analogy can be seen as functionally opposing forces; sound change operates via the natural process of speech, without regard to the meaningfulness of words or word parts, and as a result it sometimes creates morphological irregularities. In some of these cases of irregularity, speakers 'repair' the morphology via analogy to make it regular again, and the repair becomes conventional. The result looks as though the sound change never happened or happened differently from the predicted, regular process we see in other words. In effect, we get apparent exceptions, that are actually not really exceptions to a regular process but the after-the-fact result of a different kind of process.

Unfortunately, where analogy is going to 'kick in' to repair the morphological system is unpredictable, and related dialects and languages will show different repairs and at different times. The only predictable diretionalities are that sound change makes morphology become more irregular, and analogy moves it in the direction of greater regularity over time. Since both forces are operative, languages do not reliably move in a single direction of becoming either more morphologically irregular or more morphologically regular over time.

How does this 'break and repair' process work? First we can look at the widespread existence of morphological alternations that typically occur in synthetic languages like the older Indo-European languages. If a morphological category happens to have the right phonetic environment for the loss or change of a sound, for example, the sound will be lost or changed even though the result might be a fusion of morphological boundaries, or a difference between forms of a stem in different morphological categories. For example, the regular change in pre-Latin that turned g to k before voiceless consonants produced the alternation between the stems rek and reg- in the nominal paradigm for the word 'king': reks in the nominative vs. reg-is and other other oblique forms with the stem reg-. The oblique (non-nominative) forms happened to occur before case endings starting with vowels, whereas the nominative in this declension had a following voiceless consonant. Such alternations are found through the paradigms of Latin, Greek, and Sanskrit, and in the records of the older languages of other branches.

In modern terms, we can say that regular sound change, operating 'blindly' on strings of sounds, can make morphological forms less transparent. Analogy can be viewed as the restoration of transparency of the form-meaning relationship within the morphological parts of words.

Now we can look at the ways this 'restoration of transparency' mechanism can play out.

Analogy for O&B includes what is now called "paradigmatic levelling", cases in which an irregular form or forms in a paradigm is 'smoothed out' and made regular, to resemble the other forms of the paradigm. For example, in some dialects of English, the irregular forms of the verb be are regularized throughout the paradigm by making all the singular forms the same, or all the plural forms the same, or even, in some dialects, by making them all the same form throughout the paradigm. Staffordshire English as described by Trudgill, uses am throughout the entire present tense paradigm of be, and were throughout the past tense. Similarly, the Scandinavian languages levelled their present tense inflection in all verbs to the suffix -r/-er as in Jeg komm-er 'I come-PRES.SG', vi komm-er 'we come-PRES.SG'.

Analogy also includes what is now termed 'analogical extension' in which one form or set of forms 'jumps across paradigms' and applies to lexical stems it previously did not apply to. For example, some nouns that were once in a different stem class in the history of English became 'attracted' to the most productive stem class and so in effect switched declension class. This type of paradigm-shifting change is well-documented in all the branches of Indo-European, notably Slavic, in which the verbs as well as the nouns are subject to such changes.

Analogical extension changes can also be seen in terms of a '4-part analogy' in which an older form is replaced by a declensional form that fits a 4-part pattern, bringing a morphological inflectional pattern into line with another preferred pattern. For example, the plural of shoe in older English was for a while shoon. At some point the -n plural was replaced in this word by the -s plural. The analogy can be seen in these terms: 'dog is to dog-s as shoe is to X ', where the speaker presumably solves for X, choosing an X that fits a different morphological pattern from the one that was formerly conventional.

In the example given by O&B on p. 207, they suggest that the historically unexpected nominative plural forms of the Greek and Latin nouns meaning 'horse', namely hippoi and equi, respectively, were modelled on nominative pronominal forms ending in -i. This is also a case of analogical extension.

Although analogy was recognized by the earlier comparative grammarians, it received a new (in fact higher) status in O&B's article. First, it was seen as having a systematic part to play in a larger explanatory framework in which it functioned as a natural complement to sound change: it was now recognized as an important way that speakers adjusted their language functionally. Second, O&B swept away the limited and incorrect view that analogy occurs in recent language history only. The tendency was still strong among comparativists, like their antiquarian-oriented forebears of classical philology, to treat the oldest Indo-European languages as the most 'perfect' languages, languages not subject to processes of morphological change that would change or deform the 'original' system. (They viewed many morphological categories as a sign of the "developed" language, presumably because they so highly respected the Greeks and Romans for their culture and civilization, and these peoples happened to have languages with a great deal of morphology.) The earlier comparatists saw processes like levelling and analogical extension as just recent 'mistakes' that crept into the systems via imperfect thought processes of modern people. This is why they referred to analogy as "false analogy", which reflects their value judgement that analogy was somehow 'invalid'.

It seems that three strands combined to yield the firm belief in "the more ancient, the better" that was so widespread in the 19th century. There was the intense preference for the classical languages inherited from the classical philologists, the best scholars of Greek and Latin of the time, who often were the teachers of the comparative linguists. Then there was the idea that any human endeavor, institution, or phenomenon that was much earlier was 'more pristine', more original, and therefore better than the modern 'corrupted' incarnations of it. O&B in this article introduced the opposing modern idea that language change is not language degeneration. This is an idea that is still not really internalized by modern societies at large, including our own, but one which was recognized immediately as sensible by the new scientifically-oriented linguists. Finally, it is clear from writings of the time that there was a not-uncommon belief prevalent at least until the Neogrammarians that, the further back one went in the study and reconstruction of Proto-Indo-European, the closer one approached to the form of what must have been the original language of human beings. It was commonly accepted that this language was Hebrew, the language of the oldest Judeo-Christian scriptures. (This belief, too, is still widely prevalent even today, for religious reasons, and is becoming more so in the U.S. as fundamentalist creationist views gain ground, even in the body politic.)

It is made very clear in O&B's polemic that earlier times of thousands of years ago are not to be treated, as they had been, as some hermetic and alien world detached from the present. The periods in which the earlier Indo-European languages were spoken were times in which things and people were subject to the same forces as found now in the world around us. The past was full of people like us, using language as we do today. There were children who had to learn the language as any children do, and for them the language wasn't ancient, it was their modern reality (cf. p.206). So the status of such a language from the present-day point of view, namely 'ancient' as we view it, is irrelevant to its nature and functioning in earlier times.

This idea was a perfect fit with the uniformitarian tenets of two new sciences emerging in the 19th century, geology and biology, whose respective intellectual pioneers Lyell and Darwin broke away from the religiously-based view that in the distant past (actually not-so-distant; it was thought to be only about 5000 years ago) was a unique creation-period governed by supernatural forces not amenable to scientific investigation or explanation. The belief in Hebrew as the original language, before the catastrophe of Babel turned the world polyglot, was also scriptural in origin. O&B do not mention religious views on language. But it is clear that their thinking is scientific and not religious: they recognize the notion of 'degeneration' as purely a value judgment and as such irrelevant to the study of language; and they recognize the total lack of observational evidence that earlier languages, even the Indo-European ancestral language, the most ancient language yet studied at that time, were different in kind from modern ones. And if the most ancient languages resembled modern languages in the kinds of ideas and categories they expressed, the kinds of morphological structures they had, and the kinds of changes that sounds and categories underwent, then there is also no evidence that human beings in Indo-European or other ancient times were markedly different from modern ones. They see no reason to posit forms, as their contemporaries were doing, of a "primitive" or "pre-primitive" language.

Another instance of Osthoff and Brugmann's scientifically-oriented stance is their call for firmer methodological grounds on which to base comparative linguistics. The too-free use of random exceptions in accounting for modern forms, without any search for an explanatory basis, is correctly recognized as antithetical to a scientific approach. O&B observe that, although appeal to analogy may well look like an 'appeal to faith', it makes far more sense to invoke analogy when we see so many instances of it in the history of language, than to try to twist and bend the sound laws to maintain the view that the early Indo-Europeans were, as superior sorts of people, untouched by the influence of the so-called 'false analogy'. They place a constraint on the use of analogy that is sound scientific method: only use analogy as an explanation of deviation from the sound laws when driven to do so; don't give up on searching for explanations. Reaching too soon for the easy analogical (or other) explanation of exceptions can easily preclude ever spotting the correct phonological conditioning factors that make what is actually a a regular sound change seem puzzling and irregular.

The potential results of an ever-deeper search for generalization was illustrated brilliantly by the case of Karl Verner's article "An Exception to the First Sound Shift" which appeared a few years before this Manifesto (and does not yet seem to be known by O&B, although Verner certainly was counted among their number as a Neogrammarian). Verner found a heretofore unnoticed generalization accounting for most of the exceptions to Grimm's law, a regularity produced by a systematic set of consonant changes in Germanic. By looking at cognates of the regular and the exceptional forms outside Germanic, he was able to discover an aspect of the original phonological environment in PIE (position of the stress in the word) which was no longer present in Germanic, but which was perfectly reflected in the set of exceptions. At a stroke, mystery was replaced by understanding, because of the insistent search for generalization and for a principled account of exceptions. This was the advantage of being 'so punctilious about the sound laws' (p. 207).

In defending their use of analogy, O&B also call for a deeper study of analogy, and hold out the hope that 'more general principles will gradually be found' and that a 'probability scale can also gradually be established' for various types of analogy. In other words, analogy is not going to be, for them, just a convenient dumping ground for cases that can't be explained otherwise, but a phenomenon in its own right that must be probed in depth. Here again we have the goal of maximal generalization firmly in sight, like any good science should have. And in laying out the "leading motives and principles" of their approach, Osthoff and Brugmann go a long way towards refurbishing the foundations of a science that had produced many great results about the structure and nature of Proto-Indo-European, but was still comparatively in its infancy.

In the subsequent development of what is now called Historical Linguistics (it has taken on a 'marked' form after the seismic shift to synchronic linguistics as the 'unmarked' kind of linguistics, post-Saussure), analogy has been probed and sub-typologized and incorporated into the general body of types of linguistic change. However, the precise psychological basis of analogy, and its role as an explanatory factior, are still under argument.

Some modern theories of linguistics (largely the usage-based ones of functional and cognitive linguistics) incorporate analogy deeply into their understanding of how speakers use language synchronically--Langacker's Cognitive Grammar, Bybee's network model of morphology, and Lamb's dynamic processing model, as well as connectionist linguistic models like those of Jeffrey Elman and Brian MacWhinney, are all essentially analogy-driven, as activations from similar forms affect the choice among alternative forms in language processing. The effects of phonological similarities in selection of a form are now recognized and incorporated centrally in the more advanced versions of Optimality Theory. Phonemes are complex but still operate in a relatively limited domain of possibilities contrained by the articulators (and, less well explored, phonological perceptual factors).

But when it comes to determining similarity on the semantic plane, and using it as an explanatory factor in processing and/or change, many linguists and psychologists still have problems with the concept, saying it is too unconstrained. Why do some analogies occur and not other conceivable ones also based on any of the myriad of similarities that humans are capable of seeing between complex concepts? These are questions that require deeper exploration into psychology before we have a full explanation of how analogy works in language, including language change.

[ jump to top ]