Jouna Pyysalo has written a very bold and thought-provoking dissertation that revisits some fundamentals of Indo-European historical phonology. He attempts to bring Indo-European linguistics, which has been plagued in the 20th century by fanciful multilaryngeal reconstructions, back to its frugal Neogrammarian virtues. Following Zgusta, Szemerenyi, Tischler and Burrows (but strangely missing Russian monolaryngealists such as V. Dybo, S. Starostin, A.Kassian), he rejects the complex combinatorics of H1+e, H2+e, H3+e, H4+e frequently postulated by Indo-Europeanists in favor of a single laryngeal /H/ firmly attested in the so-called “Old Anatolian languages” (Hittite, Luwian, Hieroglyphic Luwian, Cuneiform Palaic and Cappadocian) but lost elsewhere. The intriguing part of the dissertation is the hypothesis that Old Anatolian /H/ covered a voiceless and a voiced variants inherited from the Proto-Indo-European (PIE) stage and that they had no coloring effect on neighboring vowels (in the languages outside of the Old Anatolian cluster) and no syllabic (vocal) allophones in PIE. To account for the statistical regularity of association between Old Anatolian h and Narrow Indo-European a, Pyysalo postulates for PIE diphonemic sets *Ha and *aH: while laryngeals as obstruents get lost in Narrow Indo-European and New Anatolian languages (Lycian, Lydian, etc.), the associated vowel survives (or changes into other vowels, according to the phonetic rules of individual IE branches). Along the way, Pyysalo also simplified a portion of the Neogrammarian legacy by deconstructing a) labiovelars – hypothetical phonemes poorly unattested in actual IE languages – as a combination of velars and /w/; b) palatovelars as combinations of velars and /y/; c) voiced aspirates as combinations of voiceless stops and the voiced laryngeal, by analogy with the more widely accepted interpretation of Indic voiceless aspirates as combinations of voiceless stops and the (voiceless) laryngeal. Pyysalo’s c) has been my conviction since I’ve realized that *bhr– in the IE cognate set BROTHER (PIE *bhreH2ter > Lat frater, Gk phrater, etc.) derives from earlier mr– (also found in Lat maritus ‘husband’, Gk meiraks ‘boy, girl’, Germ *brudi– ‘bride’, Latv marsa ‘brother’s wife’, etc.) and that the aspiration of *bh– is the result of a feature throwback from the medial laryngeal in *mreH2ter) (see Dziebel G. V.2006. “Reconstructing ‘our’ kinship terminology: Comments on the Indo-European material in A. V. Dybo’s and S. V. Kullanda’s The Nostratic terminology of kinship and affinity,” in Algebra rodstva 11, 67-68). This etymology supports Szemerenyi’s (“Introduction to Indo-European Linguistics,” 1996, 144) contention (quoted by Pyysalo, p. 398):
“Since according to our conclusions the ‘laryngeal’ was a glottal spirant h, it is also clear that the unvoiced and voiced aspirates originally represented the combinations unvoiced stop+h and voiced stop+h, which in Indo-European counted as monophonematic.”
Pyysalo ends up with the following PIE phonemic inventory:
Pyysalo (p. 58) puts Indo-European linguistics on a hot seat by observing that
While it’s rewarding to see an Indo-Europeanist openly talking about the problems with Indo-European reconstructions (instead of just praising the remarkable achievements and the scientific rigor of comparative Indo-European studies), the root of these problems, as I have extensively wrote on the pages of the Algebra rodstva almanac from 2000 on, is not in the inability of Indo-Europeanists to come to terms with the Old Anatolian data. It’s the methodological weakness of the traditional comparativist method that is holding the Indo-European historical phonology back. This weakness stems from the logical contradiction inherent in the definition of a cognate set. Sound laws are supposed to be derived from the material assembled in the least law-like manner – by simple visual inspection in search of similarities in form and meaning. This approach can detect obvious cognates but it’s not well suited to detect cognates subject to more dramatic transformations of form and meaning due to accelerated language change, great time depth or other factors. Under the current approach to linguistic comparativism, formal similarity dominates over semantic relationships. Only once a protoform is reconstructed, it’s subjected to semantic interpretation. As a result, sounds laws are constrained by a linguist’s naive perception of “similarity” in the lexical material, while proposed etymologies are often dubious on semantic and anthropological grounds.
A gignetic approach to the reconstruction of kinship terminological systems and to linguistic comparativism broadly focuses on this initial stage of cognate set composition and uses semantic variation as an important counterpart to formal variation. Semantics is not an afterthought but an essential criterion actively contributing to a more in-depth definition of a cognate set. A more complexly defined cognate set – in reality a combination of two or more one-dimensional cognate sets – should yield a more sophisticated, complete and sound laws of linguistic change in a particular family of languages.
Pyysalo (p. 80) feels himself on a firm ground when he declares:
It’s precisely when the doubts seem to recede completely that the traditional comparativist method shows its limitations, while the gignetic method creates a breakthrough. The first cognate set listed by Pyysalo contains forms with the initial velar (Slav *kosti ‘bone’, Lat costa ‘rib’) (conveniently omitted by Pyysalo), while the second one cannot be divorced from PIE *pek’– (Skrt pášu, Avest pasu ‘cattle’, Lat pecu ‘cattle’, pec?nia ‘money’, Goth faíhu ‘money, movable goods’, OHG fihu, OEng feoh, ONorse f? ‘livestock, property, money’, Lith pekus ‘cattle’). The perfect semantic fit between the two sets can be clearly observed in such forms from the first group as Gk ??? ‘flock of sheep’, ???µ?? ‘guardian, herder’.
Having the alternation such as *peH-/*pek’– in mind, we can observe the same correspondence between, for example, Hitt haluga ‘message’ (a word with obscure etymology under the known sound laws) and PIE *k’lewo– (Gk kleFos ‘glory’, klu? ‘I hear’, Skrt srávas, Old Arm lu, lsem, Slav *slovo ‘word’, etc.). Similarly, IE *H3osdo– ‘branch’ (Hitt hasduir, Arm ost ‘Ast, Zweig’, Gk o?zdos ‘Ast, Zweig’, Goth asts ‘branch, palm branch, leafy branch’, etc.) shows parallel forms with an initial velar (possibly, labiovelar, which would be consistent with the labial component of /H3/ inferred from such forms as Gk o?zdos): OHG questa ‘bunch’, ONorse kuistr ‘branch’, Slav *gvozd ‘forest’ (> Russ gvozd’ ‘nail’). Also, Hitt huidar, HLuw huidar ‘wild animals, fauna’, Palaic huidumar ‘Lebe, Lebenwesen’ can be linked to PIE *gwiwo– ‘life’ in addition to forms such as ONorse vitni ‘creature’ (referenced in Pyysalo, p. 79-80) resulting in the identification of /H3/ with labiovelar.
Pyysalo (pp. 91-92) uses another well-known IE cognate set to demonstrate the lack of connection between vowel quantity and vowel coloring in PIE. Hitt mehhur/n ‘time, noon’ corresponds to Skrt m?ti, mím?te ‘to measure’, m?tr? ‘measure’, Lat m?tior ‘I measure’, m?ni ‘in the morning’,Goth mitan ‘to measure’, m?la ‘measure of grain’, OHG mezzan ‘measure’, Lith metai ‘year’, Alb mat, mas ‘I measure’, mot ‘year, weather’, etc. Again, this set finds its counterpart in PIE *meg’h– usually glossed as ‘great’ (Hitt megi– ‘big’, Gk megas ‘great, large, mighty’, Lat magnus, Gothic mikils, Old High German mihhil, Skrt mahan, etc. Pyysalo treats this set separately from the meH-set on p. 408. But both sets neatly derive from PIE *meg’h– or *meH– ‘to measure’; some forms chose the semantic path of TIME’, others the semantic path of SIZE and STRENGTH. The Gothic forms show a perfect morphological match between the me-l and the miki-l extensions reinforcing the phonetic similarity. Whether we postulate a law of velar weakening in some to-be-determined phonetic environments (potentially in non-accented syllables), a combination of a plain velar and a laryngeal (voiced in Pyysalo’s interpretation or voiceless in Szemerenyi’s interpretation) yielding the Indo-European palatovelars and labiovelars, or an entirely different ancient phoneme, Hittite (mehhur/n) doesn’t appear any more archaic in its phonetic development from the PIE root than Sanskrit (mahan) or Old High German (mihhil). Old Anatolian uniqueness is reduced to providing supporting evidence that, the TIME subset, too, contains the trace of an ancient consonant.
From a formal perspective one can conclude that the Indo-European material shows alternation between –ek’-/-eg’-/-ek– and –a-/-o– in correspondence sets that include Old Anatolian /H/. The “loss of the laryngeal” is only one path taken by non-Anatolian languages, the other one being a velar reflex. Both Puhvel and Kurylowicz are correct.
Importantly, one doesn’t need Anatolian evidence to establish a connection between IE *pa– ‘protect’ and IE *pek’– ‘livestock’. Overall, PIE *meg’h– or *meH– ‘to measure’ supports Pyysalo’s contention that Old Anatolian /H/ stood for both voiceless and voiced variants, but it undermines his belief that the advancement of the Neogrammarian agenda of the scientific study of Indo-European historical phonology can be achieved by tackling Old Anatolian data. Contra Pyysalo, the key to progress in Indo-European linguistics lies not in the preferred analysis of Anatolian data (obviously, it needs to be further studied but it shouldn’t command any priority over any other IE branches) but in the improvement of comparativist methodology itself.
Pyysalo comes face-to-face with the situation of a dual reflex of PIE “laryngeals” in languages outside of Old Anatolian when he correctly observes that Lycian sometimes shows /0/ where Old Anatolian has /H/ and sometimes a velar. Examples without the loss of a laryngeal include
They are contrasted with the examples where Old Anatolian /H/ corresponds to Lyc /x/ (phonetically a voiceless velar, not a fricative [Melchert, Anatolian Historical Phonology, 40]) such as Lyc xuga next to Hitt huhhas ‘grandfather’ and Latin avus and Lyc xawa next to CLuw haui– ‘sheep’.
Pyysalo’s solution is to ignore the unmistakable similarity in form and meaning between the Old Anatolian and Lycian words and to make Lyc xuga and xawa unrelated to any other standard IE terms for ‘grandfather’ and ‘sheep’ but instead connected to obscure Hesychian forms. In the case of xuga, he relates it to
This solution is hardly convincing and contradicts the opinion of the majority of Indo-Europeanists. On the other hand, Lyc /x/ as a parallel to Anatolian /H/ remains tenuous unless we can show that the correspondence of the laryngeal to a velar is systematic in these cognate sets. And we indeed find that IE cognate sets GRANDFATHER and SHEEP are likely incomplete. PIE *Howi– (Skrt a?vi-‘sheep’, Arm hoviw (*ou?i-p?-) ‘shepherd’, Gk ???, ???, Lat ovis, OEng ewi, OHG ouwi, Lith avi?s, a?vinas, Slav *ovica ‘sheep’, etc.) can be compared with PIE *k’uwon-/*k’un– ‘dog’ (Skrt s?v??, s?(u)v??, Arm šun, Gk ????, ?????, Goth hunds, OEng hund, Lith šuo? (Gen. šun?s), Toch A ku, Toch B kunder. The morphological derivation of the DOG forms from the SHEEP forms by means of an n-extension is straightforward, while the semantic link is natural considering that proto-Indo-European pastoralists, with all certainty, used their dogs as shepherds and modern Europeans continue to call several dog breeds ‘shepherds’.
The situation with Lyc xuga ‘grandfather’ is more complicated. While the cognation of Hitt huhhas with Lat avus, Goth awo ‘grandfather’, OIrish (h)aue ‘grandchild’ is clear, a number of other forms, including Lyc xuga (< *xuxa, with voicing developing in an intervocalic position [Kloekhorst A. “Studies in Lycian and Carian Phonology and Morphology,” Kadmos 47 (2008), 125]), are similar enough to warrant serious consideration as belonging to the same group. Skrt susa ‘grandmother, grandfather, progenitor’ (from PIE *suHsiya), IE *suH– ‘son’ and Alb gjysh (from PIE *suHsos) ‘grandfather’ provide a case in point.
Importantly, Hitt huhhas is a reduplicative (comp. Neapolit vava ‘grandmother’ [Zimmermann 1922, 150] or USorb wowa ‘grandmother’ [Schuster-Šewc 1961; 1984, 796-797; 1987, 1682-1683; Machek 1968, 40] and hence the original form does not have to be *HewHo-. It could be *weHo-/*uHo- or even *CweHo-/*CuHo– assimilated into *HweHo-/*HuHo-. Skrt susa and Alb gjysh may, therefore, have preserved the original shape of the PIE GRANDFATHER root, while Hitt huhhas (and its direct cognates in other IE branches), so iconic in the context of the laryngeal theory, stems from earlier *suhhas. On the strength of OIr (h)aue ‘grandchild’ (Gen. aui ‘grandson’ [*awios < *awyios, with –i– developing into –e– under the influence of the following o [Maille 1910, 49-50]), úe ‘granddaughter’ (Meyer 1912, 183) and, now, IE *suH– ‘son’ (Toch we can confidently postulate a self-reciprocal meaning for PIE *suHo-/*sweHo– ‘grandfather; grandchild’. In many IE languages the GRANDFATHER-GRANDCHILD root spawned a host of derivatives including Lat avunculus ‘mother’s brother’, OIr amnair ‘uncle’ (through the assimilatory nasalization from *abn-air < *awn-air [McCone 1996, 49, 86] < PCelt *awen-tro), Welsh ewythr and OCorn eviter, Lith avýnas ‘mother’s brother,” OLith avà ‘mother’s sister, uncle’s wife’, OPrus awis (< *awio-s) “mother’s brother”, Slav *uj? ‘mother’s brother’, Alb *vella ‘brother’ (< *awnlada-< *awentlo– [Huld 1984, 128-129]). Morphologically, forms such as OIr (h)aue, OPrus awis and Slav *uj? have the same extensions as Toch A se, B soy, Gk huiús ‘(< *huios) son’ supporting the overall connection between the forms with s– and the forms with Old Anatolian h– and its accepted Narrow-IE counterparts.
The above interpretation of the PIE form for GRANDFATHER-GRANDCHILD yields to the clarification of the etymology of a widely-spread IE affinal term *swék’uro- ‘husband’s father’ (Skrt švašura ‘father-in-law’, Avest xvasura ‘father-in-law’, Prasun ?üj? [< *?u? < *?va?r < *sva?r-], Kati ??styü? [Buddruss 1976, 29-31], Gk ‘?????? ‘husband’s father’, Lat socer ‘father-in-law’, Goth swaihra ‘father-in-law’, OHG swehur ‘father-in-law’, OEng sw?or ‘father-in-law’, ONorse sv?r, OLith šešuras ‘husband’s father’, Slav *svek’?r? ‘husband’s father’, Alb vjerr, vjehër ‘father-in-law’, Arm. skesrayr ‘husband’s father’, Welsh chwegrwn, OCorn hwigeren ‘father-in-law’) and *swekr?s ‘husband’s mother’ (Skrt švašr?– ‘mother-in-law’, Avest xusr?, Waigali ??tr [dissimilated from *???r] [Buddruss 1976, 29-30], Gk ‘????? ‘husband’s mother’, Lat socrus, Goth swaihr?, OHG swigar, OEng sweger, ONorse sv?ra, Welsh chwegr, OCorn hweger ‘mother-in-law’, Slav *svekry ‘husband’s mother’, Alb vjehërrë ‘mother-in-law’, Arm skesur ‘husband’s mother’.
The morphology of Lat avunculus (< *avonculus), PIrish *awentro, Welsh ewythr and OCorn eviter, PAlb *awentlo– parallels that of *swék’uro-. The forms with –t– may in fact be analogical with the respective terms for FATHER (PIE *pHter) and MOTHER (PIE *meHter, Alb motre ‘sister’), while –c– in Latin avunculus is identical to –c– in socrus (PIE *k’). The –n– in also likely intrusive and represents a secondary “nasal infix.” We therefore can align PIE *swék’uro- ‘husband’s father’ with the Italo-Celtic-Albanian isogloss *awekro ‘mother’s brother’. Outside the Latin, Celtic and Albanian worlds, there are a few other indications that the IE terms for GRANDFATHER>MOTHER’S BROTHER and HUSBAND’S FATHER are cognates. Skrt švašura and OLith šešuras ‘husband’s father’ are assimilated from, respectively, *svašura and *sešuras just like Hitt huhhas and its counterparts in Narrow-IE languages assimilated from *suhhas (< *suHo-). Arm skesur ‘husband’s mother’ (judging by other IE forms, originally ‘husband’s father’) can be compared with k’eri ‘mother’s brother’. Until now, k’eri (< *sweriyos) was an isolate among IE terms for MOTHER’S BROTHER but in the light of the proposed reconstruction it’s naturally connected to skesur. Skesur shows the same assimilation of the earlier *k’esur into skesur as Skrt švašura and OLith šešuras.
Now that the various IE forms for GRANDFATHER, MOTHER’S BROTHER and HUSBAND’S FATHER have been shown to be aligned morphologically and subject to the same assimilatory processes, it becomes clear that PIE *k’ in *swék’uro corresponds to hh in PAnatolian *suhho-. The difficult Albanian –h– in vjehër ‘father-in-law’ (sometimes interpreted as a reflex of *H4 by multilaryngealists) matches the Old Anatolian hh nicely. The present analysis furnishes a better interpretation to another puzzling form. OHG ?heim, OEng eám ‘mother’s brother’ (< *eaham, with a syncope [Schhoof 1900, 232]) (Grimm & Grimm 1889, VII, 1198; De Vreese & Boekenoogen 1910, XI, 16; Lendinara 1990, 298) undoubtably contain the formant *au(n)– ‘grandfather’, but the final segment –heim causes problems. Gothic does not show this form but, according to the Grimms, it should be *áuhaims. The latest Etymologisches Wörterbuch des Deutschen (1993, II, 830, 915, 945) seems to take the Old English form as the basis and to explain the OHG voiceless fricative as phonologically conditioned in the intervocalic position. The majority of scholars, however, tend to consider ?heim an original form and to treat it as a compound going back to *au(n)-haimaz. The latter, however, is an ad hoc form without any parallels in other IE branches. The present analysis suggests that –h– in ?heim is part of the root and corresponds to –hh- in Hitt huhhas, -g- (< *-x-) in Lyc xuga, -h- in Alb vjeher, c in Lat avunculus and the reflexes of IE *k’ in the various HUSBAND’S FATHER terms in the satem and centum languages. Finally, an aberrant isogloss with Arm ustr ‘son’ corresponding to OEng suhterga ‘brother’s son’ becomes more understandable: semantically it belongs with Toch A se, B soy, Gk huiús ‘(< *huios) ‘son’ but morphologically with the IE terms for HUSBAND’S FATHER (*swek’ro-/*suk’ro-).
To conclude, Pyysalo is correct in stressing the obstruent nature of the PIE “laryngeal,” the existence of voiced and voiceless laryngeal variants in Hittite and the polyphonemic nature of IE labiovelars and palatovelars. He is also justified in postulating diphonemic sets *Ha and *aH to explain the wide-spread correlation between Old Anatolian h and Narrow-Indo-European a. The evidence presented above, however, refutes a) Pyysalo’s (and others) conviction that “laryngeals” were lost outside of (Old) Anatolian and b) monolaryngealists’ reduction of all “laryngeals” to just one phoneme. In fact, we can provisionally postulate the following diphonemic sets for PIE:
1. *H2a/*aH2~ *k’e/*ek’ *H2a/*aH2 ~ *g’e/*eg’
2. *H3o/*oH3 ~ *kwe/*ekw *H3o/*oH3 ~ *gwe/*egw
3. *H1e/*eH1 ~ *ke/*ek *H1e/*eH1 ~ *ge/*eg
While the latter set is given here only for completeness (I can’t adduce any etymological material to support it), the former two are intimated in the correspondences laid out above. What they seem to be telling us is that PIE palatovelars and labiovelars were identical to, respectively, PIE H2 and H3. Both pairs were obstruents, both sets covary with their respective vowels (a and o) and both are “extinct” phonemes.
An example from IE numerals can illustrate the alternation between palatovelars and labiovelars in one root form:
PIE *ok’to– /*H3ek’to ‘eight’ and PIE *kwetwor-/ *kwetur– ‘four’ can be related to each other via PIE *H3eH2t-. The otherwise obscure medial a-vocalism of Lat quattuor ‘four’ fits well with –k’- in *H3ek’to, while the initial o-vocalism of *ok’to– corresponds to the labiovelar in *kwetwor-. The known Hittite corpus does not have reflexes of either roots. A connection between the numerals ‘2’ and ‘4’ in IE was noticed before (see Kassian A. “Anatolian *meyu– ‘4, four’ and Its Cognates,” Journal of Language Relationship 2 (2009): 65-78). Kassian argued that PIE *ok’to– /*H3ek’to- used to mean ‘four’ (comp. Iran *ašti– ‘(breadth of) four fingers’ (measure of length)) and he called the *kwetwor-/ *kwetur– root “enigmatic.” The enigma disappears if we accept the formal evidence presented above that IE palatovelars and labiovelars were initially found in complementary distribution with IE vowels a/aa and o/oo. Intriguingly, PIE *H3eH2t- ‘four; eight’ shows a good formal and semantic fit with Uralic *kekta ‘two’. Whether the connection between the two represents common descent between IE and Uralic or a borrowing from Uralic to IE or from IE to Uralic, it documents the progressive complexification of a numeral system with numeral ‘2’ serving as a generator of high-order numbers ‘4’ and ‘8’.