Assibilation of k and g is an effect which strikingly distinguishes Frisian from its Dutch and Low German surroundings. Frisian shares this assibilation with English in many cases, but the process seems to have been different.
In this study I try to explain the effect in Frisian by hypercorrect influence from a possible substrate language. The 'late' assibilation in West Frisian since about 1450 is left out of consideration here.
Assibilation of palatal k and g in Frisian was already profoundly investigated by Theodor Siebs (1886, 1901: 1290-1302). He stated (1886: 8) that the compound k + î is impossible without an ç-like transition sound (ich-Laut). In pronouncing k, the tongue is already moving towards the situation at î, where the tip of the tongue lays against the palate. Consequently the resulting sound is a palatal stop, hardly discernable from a dental stop t, followed by a voiceless uvular fricative ç. This sound is still approached in West Frisian words like tsjerke 'church' (Dutch kerk) and tsiis 'cheese' (Dutch kaas), where ç has evolved to sj or s respectively. Apart from k in initial position also medial k may be assibilated, like in West Frisian brutsen 'broken'.
For palatal g, assibilation should yield the voiced counterpart, viz. a dental stop d, followed by a voiced uvular fricative £, which might evolve to zj or z. Where assibilation can only be expected in case of a stop, it did not occur to initial palatal g, which had become a velar fricative (, often evolving to a glide j, like in Old Frisian jeva 'give'. The same usually occurred in medial and final position (West Frisian neil, dei 'nail, day' compared to Dutch nagel, dag). A stop, however, must have been preserved in the combinations ng and gg, because here assibilation appears: West Frisian finzen 'captured', sizze 'say', widze 'cradle', compared to Dutch gevangen, zeggen, wieg.
Siebs deals with all example words he found in Old Frisian as well as in modern dialects, where he is able to attribute the great variation in spelling (like sth, sx, sz, ts, tsz, tz, z for assibilated k and ds, dsz, dz, s, schz, sz, z for assibilated g) to different development of the composing elements of the assibilation result, as also partly found in modern dialects (for k initially West Frisian ts, tsj, tj, East and North Frisian s, I, medially only West and East Frisian ts; for g medially West and East Frisian dz, z and perhaps North Frisian d). On the basis of written forms of names, he concludes (1886:47) that assibilation of k and g in front of palatal vowels e and i did at the earliest occur in the 13th century, especially because assibilation proceeded in some younger dialects. Later on, however, he puts the beginning of assibilation back to a very early period, distinguishing a younger assibilation (meitsje < makia) in West Frisian (1901: 1290, 1294f.).
For a long time assibilation played a role in the discussion about the question whether there has existed an Anglo-Frisian language community. Recently Stiles (1995: 195f.) listed two arguments against assibilation as a common development, namely
(1) assibilation must have taken place after the diversion of old au, resulting in Old English ceapian 'to buy' with assibilation of initial k, against Old Frisian kapia without it.
(2) Frisian shows assibilation of k in front of palatal vowels only, whereas Old English also shows it behind them, resulting in cirice 'church' beside Old Frisian tser(e)ke.
In a reaction to (1), Fulk (1998: 146ff.) recalled that palatalization and assibilation are separate sound changes, because when the palatal factor disappears a palatal stop reverts to velarity as long as it is not affricated.
As to the difference in assibilation in medial position (2), he supposes that the attested differences arose merely from different methods by which the resulting paradigm irregularities were resolved rather than by a difference in the affrication process proper.
This discussion not yet seeming to be decided, assibilation is primarily treated in the following as an effect in Frisian alone. Efforts will be directed to find a mechanism which explains how assibilation of palatal k and g occurred in Frisian, where it did not in the rest of continental West Germanic. Connecting it to a Celtic substrate, as Schrijver (1999) did recently for the North Sea Germanic vocalism, seems impossible here, because Celtic does not show assibilation (cf. Lewis et al. 40-43, 27-34). A solution is found in considering that a substrate may influence a new language also in a hypercorrect manner.
Hypercorrection generally indicates exaggerated pronunciation for social reasons. In fact the speaker is so anxious to avoid incorrect utterances that he applies correct rules in a faulty way. About forty years ago Labov (1966) discovered that this may lead to language change in that the faulty application becomes a new standard. For the speaker eager to gain prestige, this is usually a counterproductive behaviour which could be solved with some education.
There is, however, another kind of hypercorrection, which is compulsory for many people who learn a new language. This is because they are unable to discern some phonemes of the new language. A known example is Dutch uu [y], which is never mastered by the majority of foreigners, who then mostly escape to [u] or [i]. On the other hand, English [ð] and  are unknown in Dutch, resulting in frequent learners' utterings like dis, zis 'this', ting, sing 'thing'.
The same occurs with loanwords, like English goal, which penetrated Dutch as a sport-term perhaps already at the end of the 19th century. Dutch does not have the voiced stop [g], which is therefore usually replaced by its 'normal' i.e. its etymological representative, the voiceless fricative [x], resulting in a pronunciation [xo.ul]. There appear, however, also variants with initial [k], as documented by kool (de Coster 1992). Some examples of this effect also appear in Dutch dialects, according to Weijnen 1996:
kaskon < French Gascon 87
karwei < German (Ein)geweide 87
keukele < German gaukeln 92
It should be clear that neither of the solutions ([x] or [k]) is correct, but without mastering [g] the Dutch speaker was forced to choose between them. In the following I will call the case of the more exaggerated forms like [k] 'supercorrection', in order to distinguish it from the non-compulsory hypercorrection. Accordingly the 'normal' forms ([x] etc.) can be labelled 'subcorrection'. It must be noted that the choice between these two forms might be influenced by sociolinguistic factors: the higher the status of the foreign language, the earlier one would expect supercorrection instead of subcorrection.
Now which is the Germanic sound that "proto-Frisians" were unable to repeat? A candidate may be aspiration of stops, because the position of the tongue during the pronunciation of a palatal stop, as described above, is precisely the position arising at strong aspiration. We therefore will treat aspiration in West Germanic in the following paragraph.
A case of supercorrection remarkable in this framework is reported by W. de Vries (1942: 79), who told that in coming back to the city of Groningen in 1896, he found some younger people had started to aspirate stops: kat 'cat' became khath and even khats. In the last form s is clearly a supercorrect representative of aspiration!
Today Germanic languages like English and German and also the eastern dialects of the Netherlands have strongly aspirated stops. This can even make a t from non-aspirated areas misunderstood as d. So indeed a reason for supercorrection!
About the history of aspiration, not much seems to be known. As usually not having phonemic status, it seems even today being noticed only by researchers who know the non-aspirated forms from other languages. So Löfstedt (1931: 221-25) reports North Frisian p, t, k to be aspirated in the initial position, but not medially. In a local Low German dialect in East Frisia, Remmers (1997:83ff.) finds p, t, k strongly aspirated as a single consonant in the initial position of an accented syllable. In my own experience, West Frisian is never aspirated, and in Sater Frisian only occasionally some speakers aspirate. From the province of Groningen I heard persons from high and low social levels applying strong aspiration.
In contrast to Romance lenition of consonants, Bichakjian (1980: 213) sees in West Germanic a trend of what he calls 'fortition' (hardening), starting with the gemination of stops succeeding from glides and ending in the High German sound shift. From this, one is tempted to date todays relatively strong aspiration in parts of West Germanic back to the days of the Germanic conquests.
A possible test could be in the way the relevant consonants are adopted in early loanwords, because when stops in West Germanic were strongly aspirated at that time, one could expect West Germanic voiced stops to be borrowed as voiceless stops in Romance and Celtic. Indeed Gothalania became Romance Catalunya [Catalonia], but a list of loanwords failed to produce further examples.
Also Celtic really does not show initial k (often written c) from Old English g. On the other hand one would expect voiceless stops from Celtic and Romance to be borrowed as voiced stops in Germanic. The few examples of Celtic loanwords in English, however, show voiced stops in Celtic too. Also loanwords from Latin generally yield corresponding consonants in Germanic (Kluge 1913: 26). Sometimes, however, words with initial t, p were borrowed in Germanic with initial d, b, such as Dutch degel, deppen, boegseren, bolder and some placenames like Doornik, Demer. So it seems that pre-Germanic p and t were not always identified as Germanic p and t. This suggests variable pronunciation in the substrate, which leads to the following paragraph.
The important factor seems to be that voicing of stops alternated in the substrate language. "Alternate voicing" due to sandhi is wide-spread and even appears in Germanic languages. In the context of this article it is important to realize that the phenomenon does not always appear simultaneously and in the same way in different languages. This means that an assibilation process as outlined above could have been caused by a great variety of substrate languages. At this stage we will stick to the best documented of the possible candidates for the substrate of Frisian, which is Celtic. Both Celtic branches show lenition (soft mutation) of the consonant originally in intervocalic position, but develop it in a very different way (Lewis et al. 127-47). Because of the geographic situation, however, we only need to consider P-Celtic here.
It is generally agreed that sandhi caused Celtic mutation (Ball et al. 55), but scholars still hesitate on whether it had already occurred in continental Celtic, although Gaulish shows some indications of it (Lambert 47f). Where so little is known about mutation in Gaulish and where we have no clear indication yet as to whether the proto-Frisian substrate should have been nearest either to Gaulish or to insular Celtic, we may turn to the latter for more information.
In modern Celtic, lenition is applied according to syntactic rules (Lewis et al. 130-47). Initial lenition, however, has caused many analogical forms (Lewis et al. 129f). So in spoken Welsh sometimes borrowed words are lenited or, the opposite, the borrowed initial is taken as a lenited form. Occasionally even radical forms are interchanged when they have the same lenited form, like Welsh men for historically correct ben, Gaulish benna 'waggon'.
In Welsh soft mutation makes initial p, t, k, g to b, d, g, Ø. Thus 'a dog' is Welsh ci, but 'his dog' becomes ei gi e. In the same way ardd is the lenited form of gardd 'garden'. Although not denoted in older literary periods, this lenition is assumed to be old, for some traces already appear in Old Irish (Lewis et al. 127). When such mutation processes also existed in the proto-Frisian substrate, then its speakers will often have pronounced [g] instead of [k] or zero instead of [g] in trying to copy the language of the Germanic conquerors, but unconciously maintaining their own rules as to sandhi, etc. The resulting misunderstanding then easily could have given way to a supercorrect reaction like [k'] > [tI] or [g'] > [d£].
It is striking that assibilation in Frisian did not affect t and p like it did in High German. The most important reason in the case of t might be that its assibilation product [ts] would be quite similar to [tI], the product of k. Likewise [pf], the product of p, would be quite similar to f (although the last named objection did not prevent assibilation of p to arise in High German). Also the assibilation product of k is further away from its original form than those of t and p which therefore could more easily revert to simple stops. In addition to that, words with initial [k] are much more numerous in Celtic than those with p and t each, which will have made lenition of initial k the most frequent mutation phenomenon and therefore the most difficult habit to get out of. That would have lead to problems with the West Germanic superstrate, where initial k also was abundant.
In case of g the effect of mutation, which makes it zero, seems to be shocking enough in itself to trigger supercorrection.
As for the process of assibilation, it could be supposed that within the already Germanized field pockets of the substrate language may have existed during a long time after the Germanic conquest, as is reported for the High German area. From such pockets, successive waves of supercorrection might have been issued, which could present an alternative explanation for the varying realization of assibilation in the modern Frisian dialects. In this respect it has to be considered, that each supercorrect form will have been pushed strongly by having a higher social status than the alternative - the subcorrect form.
At this stage one has to return to the contribution of vowels. Tradionally the driving force of assibilation was sought in a pure phonetic mechanism in front of palatal vowels, as clearly expressed by Siebs (par. 2). But neighbouring languages do not show assibilation at that position. So there should have been another factor, and as discussed above, supercorrection could account for that. Apparently the process of assibilation was hampered by following back vowels, which seems obvious from an articulation point of view.
Hypercorrection contains a component not due to sociolinguistic processes but to the incapability of a person to copy the exact sounds of a foreign language. When this problem is solved by exaggerating the foreign pronunciation, the process is called 'supercorrection' here. In the other way when the pronunciation of the new language is adapted to the pronunciation of the substrate language, it is called 'subcorrection'.
This approach is applied to explain assibilation of k and g in Frisian as supercorrection of strongly aspirated stops in the Germanic superstrate by speakers of a substrate language which is supposed to show variable pronunciation of stops. This is elaborated for the case that the substrate would have been Celtic, which is known to show variable voicing of stops due to soft mutation (lenition).
Ball, Martin J., Nicole Muller. 1992. Mutation in Welsh. New York: Routledge.
Bourciez, Edouard. 19464. Éléments de linguistique Romane. Paris: Klincksieck.
de Coster, Marc. 1992. Woordenboek van jargon en slang. Amsterdam: Bert Bakker.
Cymraeg i ddysgwyr, Welsh for learners. 19882. Merthyr Tudful: Foxgate.
Fulk, Robert D. 1998. 'The Chronology of Anglo-Frisian Sound Changes' in Rolf H. Bremmer Jr. et al. (eds), Approaches to Old Frisian Philology, Amsterdamer Beiträge zur älteren Germanistik49: 139-54.
Gildemacher, Karel F. 2001. 'West Frisian Place-Names' in Horst Haider Munske (ed), Handbuch des Friesischen/ Handbook of Frisian Studies.Tübingen: Niemeyer, 155-70.
Gysseling, M. 1970. 'De vroegste geschiedenis van het Nederlands: een naamkundige benadering' in Naamkunde 2: 157-80.
Kluge, Friedrich. 1913. 'Urgermanisch, Vorgeschichte der altgermanischen Dialekte' in Hermann Paul (ed): Grundriß der germanischen Philologie3, vol. 2. Straßburg: Trübner.
Kuhn, H. 1959. 'Vor- und frühgermanische Ortsnamen in Norddeutschland und den Niederlanden, Westfälische Forschungen 12: 5-44.
Kuhn, H. 1968. 'Die ältesten Namenschichten Frieslands', Philologia Frisica 1966, 20-29.
Labov, William. 1966. 'Hypercorrection by the lower middle class as a factor in linguistic change' in William Bright (ed), Sociolinguistics. The Hague: Mouton.
Lambert, Pierre-Yves. 20032. La langue Gauloise. Paris: Errance.
Lewis, Henry, Holger Pedersen. 19894. A concise comparative Celtic grammar, Göttingen: Vandenhoeck & Ruprecht.
Löfstedt, Ernst. 1931. Nordfriesische Dialektstudien. Lunds Universitets Årsskrift, n.f. avd. 1, 26:4. Lund: Gleerup; Leipzig: Harrassowitz.
MacBain, Alexander. 19112 /1982. An Etymological Dictionary of the Gaelic Language. Glasgow: Gairm.
Remmers, A. 1997. Plattdeutsch in Ostfriesland. Die Mundart von Moormerland- Warsingsfehn. Leer: Sollermann.
Schrijver, Peter. 1999. 'The Celtic Contribution to the Development of the North Sea Germanic Vowel System, with Special Reference to Coastal Dutch'. Nowele 35:3-47.
Siebs, Theodor. 1886. Die Assibilirung der friesischen Palatalen. Tübingen: Franz Fues.
Siebs, Theodor. 1901. 'Geschichte der friesischen Sprache' in Hermann Paul (ed): Grundriß der germanischen Philologie2II:2. Straßburg: Trübner, 1152-1464.
Stiles, Patrick V. 1995. 'Remarks on the "Anglo-Frisian" Thesis' in Volkert Faltings et al. (eds), Friesische Studien II (NOWELE Supplement vol. 12), 177-220.
Tiersma, Pieter Meijes. 1999. Frisian Reference Grammar. Ljouwert: Fryske Akademy.
de Vries, Jan. 19923. Nederlands etymologisch woordenboek. Leiden: E.J. Brill.
de Vries, W. 1942. 'Iets over de verbreidheid en herkomst van het Fries en Enige opmerkingen op het gebied der Nederlandse Taalwetenschap' in G. S. Overdiep (ed), Teksten en StudiënVI. Assen: Van Gorcum.
Weijnen, A.A. 1996. Etymologisch dialectwoordenboek. Assen: Van Gorcum.
A. 1958. 'Praegermaanse elementen van de Nederlandse
Wendt, Heinz F. (ed). 1961. Das Fischer Lexikon: Sprachen, Frankfurt am Main: Fischer Bücherei.
 Bourciez 140 and 191-194 respectively.
Some examples, however, could have become invisible due to the West Romance lenition, as we see even Germanic words on initial k appearing with Romance g.
 Compared to many examples with initial g in MacBain.
 From the Celtic loanwords* to be considered, only bard and glen could be expected to have been borrowed rather early in a direct way.
*The Columbia Encyclopaedia, Sixth Edition. 2001, sub Celtic languages.
 de Vries 1992:109, 111, 69, 74. Already Weijnen (1958:8, also about both place-names) proposed a less clear pronunciation of t in the 'Belgic' substrate as possible case of transition to d. Here 'Belgic' denotes the hypothetic language of the pre-Germanic place-names found in the southern Netherlands (cf. Gysseling 1970:157).
 So West Frisian dy [di] becomes [ti] in op dy 'on that' and se [sc] becomes [zc] in wie se 'was she' (Tiersma 24f).
 Weijnen sees Celtic name elements as far north as Lüneburg (1958:20) and Drente (p. 30). Especially tempting in our case is his explanation (p. 24) of the old name Uxalia for the island of Terschelling from Celtic Uxanthos 'high' (Welsh uchel). For this and two other indications of a Celtic element in Frisia, cf. Schrijver 1999: 10f.
On the other hand Kuhn (1959:7) finds in the Dutch and Northwest German inland quite a number of what he considers to be pre-Germanic placenames with initial Indo-European p, which was lost in Celtic (cf. Lewis et al. 26). Among them even the West Frisian river-name Peasens (Kuhn 1968:22, cf. Gildemacher 167). Such a name, however, could have been copied from a pre-Celtic substrate.
 Celtic consists of two branches, called P-Celtic (Welsh, Cornish, Breton, and somewhat distant continental Gaulish) and Q-Celtic (Irish, Manx, Scottish) according to the representation of Indo-European ku (Lambert 16-18, cf. Lewis et al. 43-45).
 Based on historic evidence, however, Schrijver (1999: 9, 11) concludes that for West Flanders and Zealand the presence of British Celtic speakers is probable.
 Cymraeg i ddysgwyr 1988:8f.
The complete scheme of soft mutation in Welsh is as follows (radical forms on top):
Here Ø denotes the total disappearance of g.
 In the line of arguing of this paper the differing result in High German might be related to another substrate, viz. West Romance. In that language group initial k was assibilated in front of palatal vowels (e.g. French cerf , Rhaeto-Romance terf 'deer') since the imperial era (Bourciez 161).
So at the time of the Germanic conquest of southern Germany, the substrate language will have pronounced initial k already very sharply, causing no need to assibilate it in the (Old High German) superstrate language. Pronunciations like [kxind] for High German Kind 'child' in the extreme south-west (Wendt 102) might even be considered adaptations to the substrate (subcorrection also). Assibilation did appear, however, in medial and final k: machen 'make', Buch 'book', according to the fact that k was not assibilated in those positions in West Romance. Strikingly the product is a velar glide [x] here.
In the same era, however, West Romance shows lenition ('affaiblissement') of voiceless stops in medial position (Bourciez 166), which originally might have been a sandhi-effect too. In that way it could have induced assibilation of that other stops, yielding forms like High German Pfund 'pound', zehn [tse:n] 'ten', sitzen sit, Schiff ship.
 Counting in MacBain's etymological dictionary yields 6.2, 3.0, and 1.7 pages for words with initial C ([k]), P, and T respectively. A dictionary of modern Welsh, also counting compositions, yields a still larger difference for C and T.
 At this stage it should be mentioned that a unique situation exists in Dutch and Low German, where assibilation of stops seems to be nearly absent. In the present approach, this could mean that only subcorrection is applied, which might point to a differing sociological relation between speakers of the successive languages.
A possible reason for that could be that in heading for the rich coast, the Rhineland and the south the Germanic conquerors left unattended the poor inland, from where later on Dutch and Low German sprouted. Interestingly, this is precisely the "Northwest Block" where Kuhn (1959) claims a high density of substrate place-names.
 A Romance population still existed in the 7th century in Bregenz and Arbon, so around Lake Constance. In the 8th century, groups of Romani are mentioned as serfs in the district of Salzburg. (Bourciez 135).
A Celtic scholar had the kindness to review this paper thoroughly. The review shows that today's knowledge of Celtic cannot sufficiently explain the relevant part of my theory. Where I have still so much work to do on Saterfrisian, I don't think it wise to invest much more time on other languages.
So I leave here the material and the connections as I
see them from the Frisian side, taking the liberty to
present the conclusion paragraph of the review below.
The idea itself that a process of hypercorrection induced by language contact may be the cause for Frisian assibilation is not implausible at all, but, at this stage, and given our knowledge of P-Celtic, it does seem implausible to see the contact language as some variety of P-Celtic. Yet, it is still worth pursuing this hypercorrection hypothesis and it still may be possible to salvage it with reference to some other contact language (possibly a form of Romance, e.g. Vulgar Latin?).