We consider again the two diagrams seen before:

There is a clear analogy between them. This makes us expect that grammaticalization pushes linguistic acts and operations down the teleonomic hierarchy, thus converting them into automatic processes. Grammaticalization would then be another instance of the many processes of automation that characterize complex human activities.

Spelling out the analogy, we note that the diachronic relation between less and more grammaticalized constructions corresponds with the diachronic relation of more controlled and more automatic execution in learning a skill. As was seen at the end of the relevant section, automation is also routinization by frequent practice. Now grammatical formatives are, on the whole, more frequent than lexemes. They are practiced so often that, given invariant conditioning, their automation in the course of language acquisition is almost inevitable.

Before we strive for more precision here, let us note the role that such a direct link between grammaticalization and automation would play in linguistic theory: It may provide the sought explanation of the irreversibility of grammaticalization. Recall from the table of controlled and automatic processing that one of the differences between controlled and automatic behavior lies in the fact that the former can produce variant output, is flexible and may be adapted to environmental conditions, while the latter generates invariant output, is rigid and decreasingly amenable to willful change. Regaining control over something one does automatically is hard or impossible (Schneider & Chein 2003, §3). To take two linguistic examples: Controlling the tongue position for vowels of a given height takes a phonetician; all others will either just be able to imitate two vowels that only differ in frontness, or they will never learn this tongue movement. And it takes a linguist to willfully supply the wrong plural allomorph to a noun, producing e.g. anten instead of ants as the plural of ant; all the others will simply always produce the correct form. The linguist’s caprice would, in fact, be an example of degrammaticalization: What is actually totally conditioned by the morphological context would develop a new kind of variation. The explanation for the all-but-inexistence of degrammaticalization in everyday language activity is therefore simply that grammaticalization is a case of automation, and automation is irreversible for reasons having to do with human neurology.¹

Although the basic idea behind this account appears to be correct, some weaknesses and possible misunderstandings must be dispelled. First of all, grammaticalization is something happening at the level of the language system used by a speech community; to the extent that it changes the conventions of the society it is a case of ‘sociogenesis’ (Feilke et al. 2001:2). Automation is a process happening at the level of the individual mind and physis; to the extent that it changes an individual during his lifetime, it is an ingredient of ontogenesis. The two concepts are, consequently, on clearly different levels of analysis. The similarity discerned between them is, first of all, an analogy. Explaining properties of grammaticalization by properties of automation appears to presuppose that the former concept may legitimately be subsumed under the latter. The question is therefore what the bridge is between the individual mind and the conventions of the society. Putting it bluntly: If the language system in use in a speech community changes over time, does the linguistic competence of its members likewise change over time? And if so, does it happen at the same pace? The received doctrine is that linguistic change generally proceeds very slowly and that a given diachronic process may take generations or centuries to change a language system. If so, it would be hard for a linguistic change to correspond to some process going on in the individual mind.

Diachronic change has often been compared with first language acquisition. However, the evolution of grammar in primary language acquisition is not simply a kind of grammaticalization.² As far as generalizations over grammar acquisition are warranted, the following seems to hold: In the first stages of language acquisition, the child takes the holistic approach to chunks he is confronted with. The analytic approach is gradually introduced and stepwise complements the holistic approach. As a consequence, complex forms which first were learnt as unanalyzed wholes later become amenable to analysis. Once this is achieved, other complex constructions may be formed on analogy with the former, and thus a rule of grammar is acquired (cf. Tomasello 2003, ch. 8). This is an important aspect of the acquisition of grammar; and it has nothing to do with grammaticalization. Quite in general, children are not the motor of linguistic change (s. Bybee 2010, ch. 6.6 among many others). At least in occidental societies, the initial phases of a grammaticalization process, the recruitment of a lexical construction and its metaphorical use in new contexts, are a privilege of educated adult speakers (s. Lehmann 1991 for data from contemporary German).

Since grammaticalization is a kind of variation in a language, its instantiation in individual language use is the same variation produced by the individual. As usual, one speaker innovates in using a certain construction under relaxed semantic conditions. The variant diffuses through the speech community to the extent it is taken up by other members of the same social group.³ Now the new variant of the operation, construction or formative in question may be more grammaticalized, i.e. more grammatical than its source according to the parameters of grammaticalization. Its functions then have less to do with the content of the message to be conveyed and are more related to the system underlying the construction of the message. This produces uniformity of the conditions of use of the variant, more precisely, uniformity of the relation between the cognitive-communicative problem and its solution under given contextual conditions. The more a speaker hears the new variant by others and uses it himself under like linguistic conditions, the more psychological conditions for its automation in his mind are fulfilled. The speed with which this happens depends on the frequency and uniformity of the new variant and is roughly comparable to the speed with which someone acquires a new non-linguistic habit. Depending on lots of individual and societal factors, this may take between a few weeks and many years. New linguistic variants are acquired in this way at all linguistic levels. However, it is the specific systematic uniformity of the conditions of use of the new variant which leads to the consequence that, among all the changes affecting a language in a speech community, this kind of change leads to automation of its product in speakers. However, in the initial phase of a grammaticalization process, the degree of automation is low.

The new variant may turn out to be an ephemeral fashion. For instance, in the 1980s, it became fashionable both in spoken and written German to use the discontinuous adposition von X her ‘from X’ in a limitative function, as in (Lehmann 1991, §2.4).

At the time, the construction was clearly being grammaticalized. Among the symptoms was the reduction of the discontinuous preposition to its initial component and the broadening of its function to topicalization without any specific semantic role of its complement. At the time where the fashion was observed, it was impossible to predict whether it would gain a foothold in the system. To judge from today, this has apparently not happened. Instead, from the 1990s on, the fashion has been loosing ground, and today only some remnants are occasionally heard.

As made explicit in the table of controlled and automatic processing, automatic processes are hard to vary. The only changes that may easily apply to them are increasing automation and loss. Here a distinction may be made between loss of a formative in a construction and loss of a construction. The former, as exemplified by the loss of the first component of the discontinuous French negation ne … pas ‘not’ or by the loss of several conjugation desinences in the same language, is the logical endpoint of the reduction process that grammaticalization is. Loss of a construction, as exemplified by loss of the Latin gerundive construction of the debitive type () in French, is an instance of a habit becoming obsolete.

Automated skills are forgotten just like controlled skills unless they are regularly used. In other words, automation in the use by members of the speech community does not protect a grammaticalized unit against loss.

Alternatively, a certain change may survive the period of a mere fashion and take firm hold in the speech community. Then the feature in question will be transmitted to the next generation. From the point of view of a child learning the language, it constitutes an integral part of the language system. It will then be learnt just like any established feature of the language system and will be automated in correspondence with its degree of obligatoriness (cf. Feilke et al. 2001:6-8, Tomasello 2003, ch. 8.2 and Bybee 2007, ch. 7). Given that the grammaticalized item has by now lost its original emphasis and extravagance, the next generations may proceed in its grammaticalization, which will lead to increasing automation. It is, thus, the adult member of a speech community who first automates the use of a grammaticalized variant; but it is the language-learning child who assigns it a stable place in the language system.

Once the item is strongly grammaticalized, its use is highly automatic. It may be so to an extent that it is no longer accessible to control. This entails that its automaticity can no longer be relaxed. Here we have, indeed, a causal explanation of why grammaticalization is, in general, irreversible.

¹ Levelt (1989:22) speaks of “cognitive impenetrability” of low-level processes in speech production.

² Feilke et al. 2001:11f offer a list of aspects in which primary language acquisition is not analogous to language change. Bybee (2010, ch. 6.6), too, examines similarities and, more importantly, differences between the two.

³ The locus where diffusion starts is the communicative event, where one interlocutor takes up a linguistic feature used by the other one (Schmid 2015, §5.1). Given the difference in memory imprint (Table 3.2), this works more easily for lexical than for grammatical constructions. In the case of the latter, only immediate repetition, as is typical of first-language acquisition, would forestall oblivion.