Illegal clusters in Latin and proto-Romance and what they have to do with an Old High German problem

In my last blogpost I talked about sound laws which cross language boundaries and more specifically the early medieval language boundary between the Germanic and the Romance speech communities. In that respect I want to adduce another case which in my opinion may constitute another example of bilingual communities perpetuating a regular sound change from the one language as a sporadic sound change in the other.  In this specific instance I want to talk about a sound change which has proven to be remarkably persistent, namely the Latin soundlaw which turns /-tl-/ > /-kl-/. This soundlaw is best illustrated by the Latin reflex of the PIE instrumental suffix*-tlo/-tleh2 which has the unmarked Latin form /-culu-/ and /-cula-/.


PIE *peh3tlom > pre-Latin *pōklom >Latin pōculum

PIE *eǵhtlom > pre-Latin *veheklom >Latin vehiculum


Remarkably this soundlaw was also active in proto-Romance and turned the new */-tl-/ clusters, caused by early Romance syncope, into */-kl-./


Vulgar Latin VETULA /wɛtula/ >  *vɛtla > *vɛkla

 > *vɛkja > Italian /vɛk:ja/ <vecchia>.

                > Old French vjɛλə > French /vjɛjə/ <vieille>

 The pre-stage *vetla probably never was a phonetic reality because /-tl-/ constituted an illegal cluster in proto-Romance which was automatically substituted by /-kl-/. It is interesting to note that /tl/ is still an illegal cluster in present French when it appears initially, as evidenced by a 1998 study of Hallé. In a phonetic experiment speakers of French were asked to transcribe four nonwords containing the initial cluster /tl/ and four nonwords containing the medial cluster /dl/. More than 80 % of the participants transcribed the cluster /tl/ as /kl/, for example in the case of the first nonword.


/tlabdo/ : <klabdo>

Word medially the cluster /tl/ in modern French gained a secondary schwa as is the case with the word <atlas> /atəlas/. In that case it uses a different strategy to render the illegal cluster /tl/ pronouncable than its proto-Romance predecessor.


The tlo-suffix was also productive in the Germanic languages, where it participated in some complicated internal sandhi developments (which I will try to elucidate in my MA-thesis). One of the outcomes of PIE *-tlo is proto-Germanic *-þla. An interesting problem concerning this þla-suffix is the alternation /þl/ to / χl/ which seems to be responsable for OHG mahal in constrast to Gothic maþls. Also OHG bihal, which is thought to derive from *bīþl < *bhiHtlo, shows the same alternation. What happened here? Most Oldgermanicists regard it as an isolated soundlaw in west-Germanic, which has no clear parallel inside the continuum of Germanic languages. This, however, is not a very satisfactory explanation, because only Old High German and Langobardic, which may be regarded as an archaic dialect inside the Old High German continuum, appear to show the variant with /χ/ followed by an anaptyxic vowel /a/ consistently.


If one considers the proximity of the Old High German dialect continuum to the Romance continuum, especially pre-Italian Romance, an interesting possibility comes to mind. A Romance intermediate stage, with the substitution of /tl/ to /kl/, in the development of Old High German mahal from Germanic *maþlaz may yield a solution which accounts for the phonological intracies. Let’s assume, for my hypothesis’ sake, that the word reached early medieval Italy as Germanic *maþl(s). Seventh-century Langobardic /þ/ is consistently substited for /d/ by pre-Italian Romance speakers, as is seen in Italian:


Italian guadagnare < Langobardic *waiðanjan


Therefore I do not think that Germanic *maþl(s) entered pre-Italian Romance via Langobardic. Loans from Gothic into pre-Italian Romance, on the other hand,  substitute /þ/ word medially for Romance /t/.


North-Italian grinta < Gothic *grimmiþa


I’d like to hypothesize that Ostrogothic maþls entered pre-Italian Romance in the very early sixth century and because Gothic /þ/ was word medially perceived as /t/ the Romance speakers would have approximated the Gothic pronunciation of the word as *matl(s). This however was an illegal cluster in  Romance and was perceived as *makl. The word, audibly Germanic in origin, was picked up by the Langobardic invaders of Italy in the late sixth century. Here the anaptyxis kicks in and gives *makal. The thing that happens next is of course the Old High German sound shift which turns medial /k/ into /χ:/, spelled as single <h>or double <hh>, and that would yield the desired outcome /maχal/ <mahal>, which is found in the Langobardic laws.



Gothic /maþls/ > (Italian Romance */matl/ ) > Italian Romance*/makl/ > pre-Langobardic */makal/ > Langobardic /maχal/ <mahal>


This scenario is to my mind more plausible than to postulate an isolated Germanic soundlaw with no phonetic parallels within the Germanic dialect continuum. The form /maχal/ subsequently spread across the Old High German continuum and reached Saxony in the ninth century. Old Saxon mahal is to my mind a loan<!–[if !supportFootnotes]–>[1]<!–[endif]–> from Old High German which isn’t very surprising if one considers the fact that the Franks institutionalized the mahal as a tool of government.



Wilhelm Bruckner, Charakteristik der germanischen Elemente im Italienischen (Basel 1898).


Benedicte Nielsen, “On Latin instrument-nouns in */-lo/”, in: Indo-European word formation; proceedings of the Conference held at the Universisty of Copenhagen, October 20th -22 nd 2000, eds. James Clackson and Birgit Anette Olsen (Copenhagen 2004) 189-213.


Pierre A. Hallé e.a., “Processing of Illegal Consonant Clusters; a Case of Perceptual Assimilation?”, Journal of Experimental Psychology; human perception and performance vol 24, no 2 (1998) 592-608.


Ti Alkire and Carol Rosen, Romance Languages; a historical introduction (Cambridge 2010).

<!–[if !supportFootnotes]–>


<!–[if !supportFootnotes]–>[1]<!–[endif]–> The native Old Saxon term is fortunately also attested as madal.


7 responses to “Illegal clusters in Latin and proto-Romance and what they have to do with an Old High German problem

  1. Ah, ik zie dat jullie niekla al besproken hebben…

  2. Beste Peter en Olivier,

    Jullie discussie is mij een waar genoegen geweest. Een doorslaggevend stukje bewijs zou mijns inziens wel eens Fins niekla ‘naald’ kunnen zijn, aangezien dat op het eerste gezicht op Proto-Noords *ne:hlo:- lijkt te wijzen. Graag jullie commentaar.

    Hartelijke groeten uit Kopenhagen

  3. Dear Peter Alexander,

    I’m a bit confused as to what you are positing. You say the long vowel in þvál, stál, héla and mál is probably the result of lengthening before retroflex /l/. But, importantly, what about the long vowel in tár then? And why then do the following ON words (lacking preconsonantal /x/ originally) all have short vowels?

    bil (*bilam) ‘moment’
    dalr (*dalaz) ‘dale’
    gil (*gilam) ‘deep narrow glen with a stream at the bottom’
    hol (*hulam) ‘cavity, hollow’
    hvalr (*hwalaz) ‘whale’
    kol (*kulam) ‘coal’
    salr (*salaz) ‘hall’
    spil (*spilam) ‘tablet’
    tal (*talam) ‘number; conversation’
    þel (*þelam) ‘the nap on woollen stuffs’
    val (*walam) ‘choice’
    valr (*walaz) ‘the slain on the battlefield’

    Yes, insofar these words survived they ended up having vowel lengthening, as we can see in the modern Scandinavian languages. But there it is of a different quality. Compare Norwegian mål and stål with Norwegian tal [ta:l] and val [va:l].

    Moreover, we still need an unambiguous example of vowel lengthening through loss of preconsonantal /þ/. There is ON ból, but it would be begging the question to derive it from PGM *buþlam. For if the discussed variation between /tl/ and /kl/ is indeed Pre-PGM, then deriviation from *buhlam might be valid as well.

    I’m not sure what you mean when you say “phonetically it would make the most sense that only preconsonantal ð lengthenes a vowel, otherwise typologically you’d expect some form of lenition throughout the phonology.” Do you mean you would expect some form of lenition in the case of preconsonantal /x/, rather than vowel lengthening? If so, why then do we have þvál, stál and héla instead of *þvagl, *stagl and *hegla?

    Concerning Old Saxon mahlian: you are probably right, in that it had a connotation of solemnity. However, that is still less of a differentiated judicial meaning than one might expect if the word were a loan from Old High German.

    The biggest difficulty I have with your otherwise interesting hypothesis is that it relies so much on (a chain of) loans to have taken place. Apart from the loan from Gothic into pre-Italian Romance, you posit that the Langobards replaced their native term with *makal, after which northern Old High German speakers replaced their native term with *makal, which shifted to mahal, after which Old Saxon speakers replaced their native term with mahal.

    And wasn’t Gothic still spoken in the area when the Langobards arrived in the early sixth century? I would think that the Gothic rulers themselves still said maþl, which the Langobards would have understood as their own native term, so why would the Langobards instead copy the term the Romance speaking subjects used?

    Also, in retrospect, I have difficulty believing that Romance speakers heard Gothic maþl (is your maþls a typo?) as *matl, and yet since /tl/ was illegal they heard *makl. To me that’s like saying a colourblind person perceives blue as green, and yet since he cannot perceive green, he sees yellow.

  4. P.A. Kerkhof

    Dear Olivier,

    I gave the example of “needle” to indicate the loss of /þ/, not the lengthening. As to the metathesis; probably metathesized and unmetathesized variants existed alongside eachother in the history of Pgmc. The regular development of the tlo-cluster in Germanic is quite dark, so that is why I am writing my MA-thesis about it. 🙂 In case of tlo-clusters, you’d better not have any expectations to its development whatsoever. Internal sandhi in late PIE messes things up. A view on what happens in Gmc must be based on a thorough analysis of the phenomena in Slavic, Celtic, Greek and Latin, and I can tell you, that takes a lot of work! 🙂

    I know I have left out the rule /VHT/ to /V+lengthT/, My apologies. This soundlaw comes so naturally. 🙂 The other lengthenings are probably the lengthening of vowels before retroflex /l/, which happened quite late (from the 13th century onwards)? [please excuse me if I am mistaken, I am merely quoting from Noreen and de Boer’s handbooks] It is possible that also mál got its length this way if it derives from *mahla. As to your examples of ð+R that survive: that’s exactly why I spoke in terms of “tend” instead of “does”. However, phonetically it would make the most sense that only preconsonantal ð lengthenes a vowel, otherwise typologically you’d expect some form of lenition throughout the phonology.

    Old Saxon mahlian is indeed a pain in the ass. I’ll think about that. However, I think the basic meaning of the verb must originally have been a judicial one, when the noun, which yielded the verb, itself derives from the verbal root PIE *me(H)d. Otherwise the derivation wouldnt make sense. And, I believe, following Quak, the word doesnt merely means”to speak” in OS but “to speak solemnly” “to speak in a way in which you would speak in the court of law”.

    Finally, another possibility that we have to consider is that we are dealing with interference from an old loan from latin, namely Lat. macula “spot, stain”, which underwent the first Gmc consonant shift. That also would mess things up in a nasty way. 😀

  5. Dear Peter Alexander,

    Thank you for your informed reply.

    In the matter of vowel lengthening after the loss of preconsonantal medial /þ/, I don’t think PGM *nēþlō > ON nál is a particularly good example, since originally its stressed vowel was already long. Also, come to think of it, wouldn’t PGM *nēþlō rather have led to ON *náld? Compare PGM *bīþlam > ON bíldr (with gender shift). Even though the /k/ in Finnish niekla could be a secondary, Finnish development as you say, it’s striking nonetheless; I’m tempted to wager that Pre-PGM *nētlō had a variant *nēklō which was borrowed into Finnish before it evolved into PGM as *nēhlō and then into ON nál.

    You also cite Old Icelandic hvárr, but was the lost /þ/ therein really preconsonantal? As far as I know it is a contracted form of hvaðarr. I was under the impression hvárir was a secondary form. The same goes for Skáney; was the lost /þ/ therein really preconsonantal? The Old English instances seem to point to an original prevocalic position.

    Needless to say, I’m not convinced yet that ON mál could have come from PGM *maþlam instead of PGM *mahlam. I would have expected an archaic form *maðl to have shown up on one occasion at the very least. Especially with the homonym mál ‘spot, stain’ (a neuter noun as well) lurking about. And moreover because there are so many instances elsewhere in which preconsontantal /þ/ was not lost. See for example ON leðr from PGM *leþram and ON fjöðr from PGM *feþrō. But indeed, it is as you say a horrible mess.

    (And yes, there is milli, but that is a preposition, more prone to corruption, whereas mál is one of the more prominent nouns.)

    Furthermore, I’m a bit puzzled by your statement that “loss of preconsonantal medial /x/ tends not to lengthen the preceding vowel.” What about ON þvál from PGM *þwahlam, ON stál from PGM *stahlam (granted, Orel lists this one as a possible loan from WGM), ON héla from *hehlōn, ON tár from PGM *tahru(z) , ON átta from PGM *ahtau, etc.?

    Also, riskily, your hypothesis rests on Old Saxon mahal being a loan from Old High German. I have difficulty believing that. Notwithstanding the established influence of Old High German, I don´t see how mahal would be able to replace a native mathal so easily and early, especially since there is virtually no trace of mathal in Old Saxon. What we do have is the Old Saxon derivation mahlian, which looks very much native, with the OS retention of PGM *-jan and the very basic meaning ‘to speak’, instead of any differentiated judicial meaning, which would have lent support to the possibility of borrowing. By the way, it’s interesting that we find this verb in Middle Dutch as well: mellen ‘to marry, to wed’.

    As of yet I do not have an answer for you concerning OHG māllo. That is an interesting point, which I will sleep on.

    At any rate, I think your hypothesis is interesting and well thought-out, and I look forward to your reply.

  6. P.A. Kerkhof

    Dear Olivier,

    Once again thanks for your comment. The point you bring up is an interesting one; I have thought about the possibility of an inherited /tl/ : /kl/ contrast from early Proto-Germanic myself and so have other scholars, because it is the logical assumption to make when one considers the alternation :). I finally chose my Romance influence theory because it not only explains the phenomenon but gives a motivation for the alternation and the marginality of the phenomenon as well.

    This brings me to my next point; your comment on Old Norse mál. This form is not conclusive to an argument on either side. The relevant form can both be explained by the loss of preconsonantal /þ/ (Gmc. *nēðlō “needle” to OIcel. nǿl, nál) and by the loss of preconsonantal /χ/.

    Loss of preconsonantal medial /χ/ tends not to lengthen the preceding vowel (the relevant forms are quite opaque, so the phonological argument is, admittedly, not cogent), whilst auslauting /χ/ does this unambiguously (Noreen 1993) (de Boer 1920). Consider OIcel. frá from Gmc. *frah against OIcel. hlæja from Gmc. *hlahjan.

    The loss of medial /þ/ before a resonant causes vowel lengthening; Consider OE Skeð(e)nig against OIcel. Skáney, Proto-Norse hwaþr to OIcel. hvárr and Gmc. *hwaðriz to OIcel. hvárirr. But OIcel. milli from *miðli doesn’t seem te show the lengthening. It is, to be honest, a horrible mess.

    Quite interesting is the Finnish word niekla (de Boer 1920) which, if it is indeed a loan from Proto-Germanic, may point to a Pre-Proto-Germanic /tl/ : /kl/ alternation. However, there is a good possibility that /kl/ instead of /tl/ is secondary here, if one considers Finnish sieklu from Proto-Slavic *cĕdilo; so Finnish niekla is not conclusive in this regard to a Pre-Proto-Germanic alternation. This however, would place an alternation /tl/ to /kl/ quite close to the (Pre-)Proto-Germanic speech communities! An exciting thought!

    Another point I would like to make, is that an assimmilation of /-þl-/ to /-ll-/ also seems to have occurred in Old Frankish and Old High German. Preconsonantal /χ/ survives in most Frankish dialects (the Old Low Franconian loss was probably quite late), so why doesnt it show up in OHG māllo as evidenced by the Georgslied ? (the editions which give mahalo are overpresumptious because the manuscript only gives māllo). Braune’s reasoning that we are dealing with /VχV/ > /V+lenght/ is plausible but bears the difficulty that Pre-French *mallu and Old Frankish malloberg already seems to have lost the /χ/ some three hundred years earlier than most of the other h-less Old High German forms. A scenario in which /þl/ assimmilates to /ll/ in all the Frankish dialects is, to my mind, to be preferred, because it incorporates the Old Frankish evidence in the phenomenon.

    The thing that bothers me the most about my argumentation in my blogpost is that I haven’t been able to find an attestation of OHG mahal spelled with double h. However, we can posit that the Langobardic attestation mahalos contained the gemination, because geminates weren’t consistently spelled double in the relevant manuscripts. Subsequently a sort of degemination may have taken place from Romance-Langobardic to the other Old High German dialects. It is not elegant, but it is possible.

    To conclude my argument: it is very well possible to place the alternation in Proto-Germanic times and the argument might be strengthened with reference to Finnish loan phenomena. However, the linguistic aberrance within Proto-Germanic phonology and the geographically local nature of the phenomenon point to a marginal development which may have happened quite recently (Lubotsky 1998 gives these criteria for determining substrate influence for lexical items). Therefore I find it hard to assume great ancienity for the /þl/ : /hl/ alternation and the closeness to the Romance speech community, combined with the fact that the first attestation of the word with medial /χ/ is from italy (seventh century Edictus Rothari) does give some colour to my argumentation. Finally explaining the Old Norse and continental Germanic forms by assimmilation from /þl/ to /ll/ is relatively easier than invoking the complicated late! secondary loss of medial /χ/ in regard to the pre-French and Old Frankish attestations. Last but not least; it would be interesting to make a solid case for substrate influence from Proto-Finnish into Proto-Germanic. I will continue to ponder on this problem for it is quite relevant to my MA-thesis.


    Peter Alexander

  7. Yes! Very helpful post! The alternation between maþla- and mahla- had been troubling me again of late.

    However, I would rather think this alternation happened in the Pre-Proto-Germanic stage, where we woud find an original *matla- and a variant *makla-. The former would then lead to Proto-Germanic *maþla-, the latter to *mahla-, and so on in the Germanic dialects.

    If you would think this too early a date for this variation to have taken place, because you would suppose influence by Romance-speakers to be likely, then I would like to point you to Old Norse mál, which would sooner be from PGM *mahla- than from *maþla- (if it’s not a loan from a West-Germanic dialect, and I see little reason to suppose so). Scandinavia is a long way from Italy.

    Perhaps then the change from -tl- to -kl- is relatively natural.

Leave a Reply

Your email address will not be published. Required fields are marked *