Mad dog, the first palatalization, and the k- prefix

There are in Middle Chinese lexica three obviously related words (and four characters) for ‘mad dog’ (狂犬). In the Baxter and Sagart (2014) reconstruction (referred to as ‘OCNR’ below):

0279g and 0287-, OCNR *ke[t]-s > MC tsyejH,

0279g OCNR *ke[t]-s > MC kjiejH.

0335d OCNR *k<r>e[t]-s > MC kjejH.

The reconstruction of a front vowel rhyme *-et-s in all three words, as opposed to *-at-s, was forced by the chongniu IV word kjiejH, which can only reflect a front vowel. In this vocalic context, the alternation between velars and palatals was treated as a case of the first palatalization, an early Han shift according to Schuessler, applying to an OC velar initial: MC tsye- was seen as reflecting OC ke-, while in kjejH, failure to palatalize was thought to be due to medial -r-, in this case an infix (since the root is r-less): *k<r>e-. The chongniu IV reading kjieiH was anomalous: its velar initial failed to palatalize despite the front vowel and absence of an intervening -r-. Bill Baxter and I thought that kjiejH reflected OC *ket-s in a non-palatalizing late OC dialect, a few forms of which are preserved in MC.

A passage in the Jing Dian Shi Wen points in a different direction. In the database Bill Baxter and I are maintaining, under , Bill noted this passage from JDSW 1353: “之瘈。吉世反 [kjiejH]。狂也{國狗之 無不}”. This supports a word-family connection between 噬 ‘to bite’ (OCNR *[d]e[t]-s > dzyejH) and 瘈 ‘mad (dog)’. The root would be *tet-s ‘to bite’: we would have this word-family, presented below with emended reconstructions between lenticular brackets and the original in blue for comparison:

root tet ‘bite’

噬 【*m-tet-s> dzyejH > shì ‘bite (v.)’ 0336c (OCNR *[d]e[t]-s > dzyejH)

狾 【*kə-tet-s> tsyejH > zhì ‘mad (dog)’ 0287- (OCNR *ke[t]-s > tsyejH)

瘈 【*kə-tet-s> tsyejH > zhì ‘mad (dog)’ 0279g (OCNR *ke[t]-s > tsyejH)

瘈 【*k-tet-s> kjiejH > jì ‘mad (dog)’ 0279g (OCNR *ke[t]-s > kjiejH)

猘 【*k-t<r>et-s> kjejH > jì ‘mad (dog)’ 0335d (OCNR *k<r>e[t]-s > kjejH)

In ‘to bite’, *m- is the ‘volitional’ prefix *m1a– (OCNR, p. 59). This is a highly common prefix in verbs whose subject is a volitional agent. It is responsible for the voiced initial in MC. The other words all have prefixed *k- or *kə-: these are short/tightly attached and long/loosely attached variants of the same prefix. OCNR (p. 54) states that

short (*C) and long (*Cə) variants that apparently had the same morphological functions,

but notes that

The conditions of occurrence of short and long variants are not understood; for the moment we treat them as free variants.

For a few years now since the publication of OCNR, I have regarded the loosely attached forms as basic, and the tightly attached ones as the result of unstressed vowel syncope in a second prefix, with late loss of the first prefix, thus:

stage 1: *Cə-Root > stage 2: *Cə-Root > stage 3: *Cə-Root


stage 1: *Cə-Cə-Root > stage 2: *Cə-C-Root > Stage 3: *C-Root

As far as preinitials go, the OCNR reconstructions represent the stage-3 situation. There are parallel developments in the evolution from Austronesian to Kra-Dai: Austronesian trisyllables lose their second vowel through unstressed vowel syncope (in preparation). The morphological function of the k- prefix in our examples will be discussed below.

On the phonological level, the emendations in our words for ‘mad dog’ explain the failure of kjiejH to palatalize despite the front vowel without supposing a non-palatalizing OC dialect: we only need to assume that the OC initial cluster *k-t- we reconstructed for OC was still present in western Han times in one shape or other (k-t- ? k-l- ?) , so that initial k- was shielded from the front vowel when the first palatalization of velars took place. Infixal <r> still accounts for the vowel in the kjejH doublet. The palatal initial in MC tsyejH, previously thought to go back to a velar stop, can now be recognized as the regular outcome of an OC alveolar stop. In principle an OC *tet-s without a velar prefix would be enough to give MC tsyejH, but the ‘mad dog’ semantics require the velar prefix, as we will see. Alternation between the two readings, tsyejH and kjiejH, of 瘈 ‘mad (dog)’ is now treated as loosely kə- vs. tightly attached k-: loosely attached kə- falls (as often), exposing t- > MC tsy-; while tightly attached k- forms a cluster with t-, simplifying to k- at some point between early Han and MC. This gives us a useful chronological indication on the simplification of clusters.

OCNR says the k- prefix added to verbal roots “derives nonfinite forms of the verb that can be used as nouns (…)” (p. 57). This seems to apply well to ‘mad dog’, except that //are not nouns: they are non-finite modifiers of the noun ‘dog’ in the compounds 【猘狗】(《吕氏春秋·首时》, 《汉书·五行志中之上》), and 【瘈狗】(《左传·襄公十七年》, JDSW 1156). This confirms the function of k- as non-finite verbs prefixes. This prefix is probably related to velar nominalization prefixes found

in Gyalrong languages (Jacques 2021:714, 811). Japhug examples from Jacques (2021:714-5):



The dead one’ (many attestations)



The one who eats it’ (many attestations)

By our analysis, then, the Chinese words for ‘mad dog’ are interpretable as ‘biting dog’.

Outside of rGyalrong, Konnerth (2009:101-107) finds that the derivational functions most often associated with the ‘gV-‘ prefix are participant nominalization, derivation of adjectival modifiers, action/event nominalization, and derivation of adverbs. These functions are consistent with non-finite derivations out of verbs.

Similarly, Matisoff (2003:138) writes that prefixed k(h)ə-, the reflex of his *g- in Tangkhul Naga, is highly productive, serving “for nominalizations (…) and relativizations, but does not appear with ordinary finite main verbs”. This combination of evidence supports a non-finite k- verb prefix at proto-Sino-Tibetan level.

Further, this ST prefix may be compared to the stative, non-finite ka- prefix of PAN (Zeitoun).

In closing, the exceptions to the Chinese first palatalization have MC kji-: they do not palatalize despite the front vowel. There are several sources, which either were not velars, or had an intervening segment other than -r- between the velar and a front vowel in western Han:

  1. C.qi- (type B, no -r-)

  2. k.m-ri (already kli- by western Han time)

  3. k.ti- cf. root to ‘bite’ below.

Removal of the ‘mad dog’ example does not disprove the ‘non-palatalizing dialect’ hypothesis, but it weakens it.

Cite this post as: Laurent Sagart, "Mad dog, the first palatalization, and the k- prefix," in Sino-Tibetan-Austronesian, 09/11/2021,


Baxter, William H. and Laurent Sagart. 2014. Old Chinese: a new reconstruction. New York: Oxford University Press

Jacques, Guillaume. 2021. A grammar of Japhug. Berlin: Language Science Press.

Konnerth, Linda Anna. 2009. The nominalizing prefix gV- in Tibeto-Burman. PhD dissertation, University of Oregon.

Matisoff, J. (2003). Handbook of Proto-Tibeto-Burman. Berkeley, Los Angeles, London: University of California Press.

‘Bicycle’ to ‘bike’, and the limits of regular sound change.

The development of bike [bajk] out of  bicycle [‘bajsɪkl] has occasioned some etymological speculation in print and online.  The two principal proposals seem to be (a) that bicycle was reduced from a trisyllable to a monosyllable through cumulative loss of its medial syllable [sɪ] and of final syllabic [l], thus bicycle > bicycle  > bic [bajk] (here); and (b) that bicycle was pruned to bicycle , the result being [bajk] rather than [bajs] because the phoneme underlying [s] was a velar /k/ (Hausman 1976).  This last, generatively-inspired proposal is not hugely plausible. If the underlying phoneme was /k/, why was the velar brought to phonetic light after pruning, when the etymological connections of –cycle were least apparent ? why does /k/ never surface in cycle and its derivatives ?

Here is  a third account, which I believe makes better historical sense. It agrees with the first as regards pruning  of final syllabic [l], but differs from it in supposing syncope of the second, unstressed vowel [ɪ], rather than loss of the entire second syllable [sɪ] in bicycle. The two changes, l-pruning and ɪ-syncope,  applying in whatever temporal order, resulted in [bajsk], with [k] resyllabified as part of the monosyllable’s coda. The modern form bike in turn results from the simplification to [jk] of the  cluster [jsk], unattested in word-final position, bringing it in line with like, hike, pike, Mike etc.

There are interesting lessons in this. When polysyllabic words see a marked and sudden increase in their frequency, as for instance following fast societal acceptance of a technological innovation, as here, or for whatever other reason, there will be pressure on them to  become shortened. Shortened forms can take hold very fast. Thus in French, the five-syllable masculine noun coronavirus [kɔʁɔnaviʁus], in use at very low frequencies since c. 1965, suddenly underwent a sharp frequency increase at the beginning of the Covid-19 pandemic in December 2019; I first heard it shortened to trisyllabic [kɔʁɔna] on March 15, 2020 (the new trisyllable was itself subjected to competition from disyllabic [kɔvid] shortly afterwards, in April or May, 2020, and at the time of writing, [kɔvid]  appears to have gained the upper hand). Meanwhile in the UK, a parallel reduction of coronavirus to disyllabic rona appears to have been underway, as shown by this  journal article dated July 28, 2020.

The main tools language has at its disposal to shorten words are pruning and unstressed vowel syncope. One works from a word’s edge—or both edges, as in rona—, the other from the inside. Both affect words that are too long for their frequency,  without any stateable phonological conditioning.  Syncope will result in unattested or unacceptable consonant clusters, and prunings may affect syllabification, in ways that require repairs in order for the shortened word to fit the language’s canon and phonotactics. Strings will be resyllabified, clusters simplified. As the clusters will be rare or unattested, they will tend to simplify in unique ways. Although they cannot be regarded as ‘regular’, these are common historical changes affecting the sound of words.

  • Hausmann, R. B. (1976). An Etymological Brainteaser: The Shortening of Bicycle to Bike. American Speech, 51(3/4), 272. doi:10.2307/454976 
Cite this post as: Laurent Sagart, "‘Bicycle’ to ‘bike’, and the limits of regular sound change.," in Sino-Tibetan-Austronesian, 19/06/2020,