The field of unidirectionality or otherwise of grammaticalization raises the more general question of the directionality or orientation of changes in the significative system. However, before we embark on it, we will review the analogous problem area in phonology. This need not be done systematically. It suffices to see that there are some phonological processes which are non-oriented (or reversible) and others which are oriented.

Non-oriented phonological processes

Three types of phonological processes may be adduced which are bidirectional. The first is the pair of syncope and anaptyxis, i.e. the elimination and the insertion of a vowel between consonants adjacent to a stressed syllable. The following table displays two examples of syncope, producing the consonant groups /kl/ and stop + /m/, resp., and two examples of anaptyxis inserting a vowel in a consonant group composed of these same classes of consonants.

So syncope and anaptyxis run clearly in opposite directions. Here as in other cases of variation, if one has historical data, one can know the direction of change. In the case of French miracle and Italian miracolo, we assume that both are based on Latin miraculum, so it is French which applied a change. Even this, however, is not certain. It could well be that Vulgar Latin – the actual ancestor of both languages – already had synchronic variation miraclo ~ miracolo, so that either language preserved one variant. If one is faced with synchronic data of this kind, one cannot know whether syncope or anaptyxis is at work. The case is comparable to conversion, whose direction may be ascertained historically, but which synchronically comes out as multiple categoriality.

The next pair of such phonological processes is monophthongization and diphthongization. The table displays two examples of monophthongization from Old to Classical Latin and two examples of diphthongization from two other languages where, however, the very same vowels and diphthongs are involved.

So monophthongization and diphthongization run clearly in opposite directions and may affect the same vowels and diphthongs in the same environments, viz. in a stressed syllable. Here as in the previous case, it is only documented history that can tell us about the actual direction of the variation.

The last example of non-oriented phonological variation is assimilation and dissimilation. The double example displays an original /k/ which becomes /t/ by assimilation and, in the second row, an original /t/ which becomes /k/ by dissimilation.

This example has the handicap that its second half, although based on historical data, involves reconstructed forms. Moreover, the opposite processes do not occur in exactly the same phonological environment. Here even more research is required than in the two cases seen before. At any rate, the conclusion must be that there are cases of phonological variation which are non-oriented so that changes involved can go in opposite directions.

Oriented phonological processes

We are now coming to types of phonological variation which are oriented in that they change affected phonological units only in one direction. The first is debuccalization of /s/, i.e. the change from /s/ to /h/. Two historical processes of this type are well-known. The first concerns the change of prevocalic /s/ into /h/ in Ancient Greek. For words like Lat. sex ‘six’, septem ‘seven’, an initial /s/ can be reconstructed for Proto-Indo-European beyond any doubt. In Ancient Greek, cognate words start with /h/, e.g. heks ‘six’, hepta ‘seven’ etc. So there was a phonological change of /s/ to /h/ in word-initial position in Proto-Greek.

While this example involves reconstruction, the following Spanish change occurs in historical times: syllable-final /s/ → /h/ in Andalusian. Thus Castilian dos ‘two’ is pronounced [doh] in Andalusian, and Castilian quioscos /kjoskos/ ‘kiosks’ sounds [kiohkoh] in Andalusian. Incidentally, a similar change of word-final /s/ being converted into /h/ occurred in the transition period from Proto-Indo-Aryan to Vedic; but this, again, involves reconstruction. At any rate, the type of phonological change /s/ → /h/ is well-documented, in different environments.

If this were a non-oriented variation, we should find evidence for a change /h/ → /s/. No such change is known. As long as no relevant case appears, we have to conclude that debuccalization of /s/ is a unidirectional phonological process.

Another candidate for a unidirectional phonological change is obstruent devoicing in the syllable coda. This is a phonological process which is well-established for German, Dutch, Polish, Russian and many other languages (s. Wikipedia). Here are some German examples:

This is a regular phonological process in the languages concerned, and its direction is beyond doubt. The opposite process, viz. obstruent voicing in the syllable coda, is unknown. It is true that there are processes of obstruent voicing; but they occur in a different context.

The interim balance here is the following: There are some types of phonological variation which are non-oriented in that they shift phonological units both from A to B and from B to A in the very same context. And there are other types of phonological variation which are oriented in that they shift phonological units from A to B, while there is no shift from B to A in the same context.

To tighten the conditions on directionality of a type of phonological process, we require that possible reverse processes happen in the same environment as the process whose directionality we are testing. The above irreversible changes fulfill this condition. Final obstruent devoicing is theoretically more interesting than debuccalization of /s/ because it affects an entire class of segments instead of only one.

From among the reversible changes adduced above, only syncope and anaptyxis clearly fulfill the condition of applying in the exactly same context. The examples of monophthongization and diphthongization adduced appear to fulfill it, too; but more research is required. Assimilation and dissimilation appear to happen in different environments; so this is probably not a valid example of a reversible phonological process. At any rate, the more restricted the context in which the processes in question happen, the higher their theoretical interest.