In their 2019 paper introducing the volume The Precursors of Indo-European Kloekhorst and Pronk suggest 34 shared innovations common to the non-Anatolian branches of Indo-European.
They argue that these innovations confirm the “Indo-Anatolian” hypothesis, and form the basis for a common protolanguage for the non-Anatolian branches which they call “classical Proto-Indo-European” (“cl.PIE”).
The authors propose a time gap of 1000-1200 years between a proposed Proto-Indo-Anatolian language and classical Proto-Indo-European.
I have taken it upon myself to review this paper in as systematic a manner as I am able. Although it is certainly the most robust case yet to be made for Indo-Anatolian, I have several serious reservations regarding the argument as a whole and in its specifics.
I mean no harm to either Kloekhorst or Pronk by this. Both are professionals of celebrated capacity and output. By contrast, I am an amateur, and thus this exercise can be taken as a private indulgence at best. If they should come by it, the authors are free to engage with it to any degree which they feel appropriate, but I am sure they have more worthwhile activities with which to occupy themselves.
The argument as a whole
1. We can summarise the argument in favour of Indo-Anatolian as proposed by Kloekhorst and Pronk as follows:
There are shared innovations in the non-Anatolian daughter languages…
Which point to the existence of a shared non-Anatolian protolanguage, dubbed “Classic Proto-Indo-European”…
Which can be shown by the magnitude of these innovations to have been spoken at a date of significant remove (more than a millennium) from the protolanguage which gave rise to both Proto-Anatolian and Classic Proto-Indo-European.
1a. In order for the particular case for Indo-Anatolian advanced in the paper to be proven, all three of these individual assertions must be demonstrated convincingly.
2. My view, which I will set out below, is that the authors demonstrate only the first assertion convincingly, and the second assertion partially.
2a. The third assertion remains undemonstrated, and is indeed somewhat undermined, by the specifics of the paper.
2b. In sum, the evidence provided in no way supports the authors’ assertion that Indo-Anatolian must have been spoken up to 1200 years prior to the emergence of “classical” Proto-Indo-European.
3. Instead, the evidence presented supports the idea that Anatolian was the first family of Indo-European languages to diverge from the the protolanguage as a whole.
3a. However, the timescale on which this divergence happened cannot, on the basis of Kloekhorst and Pronk’s own arguments, have constituted much more than a few generations prior to the emergence of “classic” PIE.
4. On the basis of the above, we can conclude that although “Indo-Anatolian” might be a technically correct appellation for the structure of the IE family, it is not a meaningful category.
Principal weaknesses
5. With the exception of a handful of conditioned sound changes and allophones, the authors do not suggest any radical phonetic differences among their post-Anatolian innovations. Many of their examples in fact presume an identical (or near identical) phonology in PIE and PIA.
5a. The only exception – the 19th proposed shared innovation – rests on a flawed assumption which I will discuss below in 8.
5a(i). To presume any language’s phonemic inventory would remain unchanged for this long seems unlikely in the extreme. On this basis alone, we can cast reasonable doubt the proposed time depth of 1000-1200 years.
6. Beyond this, no other evidence is presented to support a time depth of this magnitude. We are simply assured that the non-Anatolian innovation of the feminine gender is “something that cannot have happened overnight”.
6a. We are also not given any reason why Oettinger’s hypothesis of an 800 year gap – or indeed, any hypothetical gap shorter than this – should be disregarded, although the authors dismiss Oettinger’s estimate as “conservative”.
7. The argument of the paper appears to run afoul of what we might call the “linguist’s time depth problem”. It isn’t clear to me that the authors – or many others – actually grasp the magnitude of the time depths they are proposing.
7a. 1000-1200 years is roughly the depth of time between the latest stages of Old English and current Modern Standard English.
7b. It is also, very roughly, the approximate time depth between the earliest recorded Mycenaean Greek and the Classical Attic of Xenophon.
7b(i). In both examples we can point to dozens of phonetic innovations which definitively separate these stages from each other, and from their ancestors (reconstructed or attested).
7b(ii). Furthermore, if we presume an initial divergence around 4000BC, Anatolian, by the time of its historical attestation, would have experienced well over 2,000 years of evolution separate from the rest of the family. With this in mind, we would expect to see a level of divergence equivalent to that between Egyptian and Semitic, which simply isn’t evinced in the Anatolian material.
7c. As we will discuss below, many of the innovations identified by Kloekhorst and Pronk simply do not require a great deal of time to be realised. All but the most significant could conceivably have been arrived at after only a single generation, and perhaps three at most.
Those innovations in full
Thematic considerations demand that I not consider the author’s proposed innovations in the order that they have proposed them. I apologise for any irritation this may cause.
8. Sound Changes: the authors postulate only one significant phonemic change between PIA and Anatolian + cl. PIE - number 19:
Anat. *h₂ = *[qː] and *h₃ = *[qːw] vs. cl.PIE *h₂ = *[ħ] or *[ʕ] and *h₃ = *[ħw] or *[ʕw]: fricativization of uvular stops.
8a. There are immediate grounds for dispute of the argument provided in Kloekhorst (2018) to the effect that laryngeals h₂ and h₃ originally represented uvular stops in the proposed PIA, or that they represented uvular stops at any stage of the development of Anatolian.
8b. In the first place is the question of vowel colouring. Laryngeal theory was initially developed to account for differences in vowel qualities across PIE reflexes in the various branches. It is unnecessary to recapitulate the theory in its entirety here. We can content ourselves with a description of the basics, which remain fairly uncontroversial:
8b(i). h₂ in proximity to PIE e effects the lowering and backing of the vowel to a. All (or at the very least most) instances of a in PIE must be construed as some combination of h₂ and e.
8b(ii). h₃ in proximity to PIE e effects the backing and rounding of the vowel to o. Because of its rounding effect, it is generally presumed to reflect a labialised consonant of one kind or another.
8b(iii). Laryngeals are not directly attested (aside from some tentative and marginal instances) in any Indo-European branch other than Anatolian, where h₂ is preserved in most positions and h₃ is preserved in some cases word-initially, where it has merged with h₂. In Anatolian, h₂’s reflex can appear in both fortis and lenis variants (e.g. ḫuḫḫa- in Hittite.)
8b(iv). The precise phonological character of PIE laryngeals cannot be reconstructed with confidence. In Anatolian, they have usually been taken as uvular fricatives, which exist alongside an innovated set of labialised uvular fricatives.
8c. Whatever their precise nature, the vowel colouring properties of h₂ and h₃ must already have been productive in the protolanguage before the divergence of Anatolian, as Anatolian vowels preserve clear evidence of their effects.
8c(i). Uvular stops cannot have produced these vowel colouring effects, nor can uvular fricatives.
8c(ii). The proposed reconstructions of h₂ and h₃ as ([ħ] or *[ʕ]) and ([ħw] or *[ʕw]) – voiced and unvoiced pharyngeal fricatives with labialised counterparts – are typologically unlikely, as is their reconstruction as velar or uvular fricatives.
8c(ii)a. In Akkadian, for instance, proximity to Proto-Semitic *[ħ] and *[ʕ] (and to the voiced velar/uvular fricative *[ɣ~ʁ]) causes the vowel a to colour to e. This is precisely the opposite change to that observed in all Indo-European daughter branches.
8c(iii). We therefore, in my opinion, have sufficient reason to discount this change as evidence of a common non-Anatolian innovation. This eliminates the only major phonemic innovation given by the authors for their proposed “classic Proto-Indo-European” as opposed to “Proto-Indo-Anatolian.”
9. A more minor – but nevertheless potentially more fruitful – shared sound change in cl. PIE comes in the form of Item 21:
Hitt. amm- < *h₁mm- (< pre-PIA *h₁mn-) vs. cl.PIE *h₁m- ‘me’: degemination of *mm to *m (Kloekhorst 2008: 111234).
9a. The cited Kloekhorst 2008 makes the case for an original PIE (sic) oblique 1st person pronominal stem *h₁men-/*h₁mn- on the basis of the Hittite oblique stem amm- and the ablaut forms observed in the second person reconstructions *tue (acc./obl.) and *tewe (gen.), arriving at an acc./obl. *h₁mn-e and genitive *h₁men-e.
9a(i). This “phantom” -n- is recruited here to explain the peculiar shape of the Tocharian 1st person personal pronouns (e.g. Toch. A. nom./obl. masc. näṣ fem. ñuk gen. ni, and Toch. B. ñaś, all ultimately from Proto-Tocharian *mñäś and PIE *mne-ge.
9a(ii). Analogies of the form *mne-ge are reflected in Greek ἔγωγε, ἐμέγε (emphatic forms of ἐγώ, ἐμέ), potentially in Proto-Germanic *mek (< *me-ge, although the final -k may reflect analogy to the nominative *ek), and, potentially in Hittite ammuk (although this may also be in analogy to nominative úk).
9a(iii). Kloekhorst (2008) explains the absence of -n- in the oblique stem in other branches through means of a proposed assimilation of *-mn- to *-n-, which he ascribes to “late PIE”, and claims this change is “visible in the Ved. instr.sg. of -man-stems: e.g. raśmán- has instr.sg. rašmā and drāghmán- has instr.sg. drāghmā, both from *-mn-oh₁. When the preceding root contained a labial consonant, the cluster -mn- was assimilated to -n-…”
He also cites (after Kroonen 2006) the example of Skt. budhná < *bʰudʰmḗn, bʰudʰmnés, Avestan raogna < rewgʰmn and Germanic *bragna- < *mrogʰmn.
9b. Unfortunately, this proposed change does not stand up to scrutiny as a “late PIE” or “classic PIE” innovation as opposed to Anatolian.
9b(i). In the first place, Greek preserves reflexes of the cluster -mn- which must be inherited from the Proto-Indo-European stage. Ablaut variants of *-men- in the zero grade also continue in Greek, as we shall see below.
9b(i)a. One prominent example is ὕμνος, < *sh₂ómn̥. We also have direct evidence of this form persisting in the ablauting suffix *-men- in the word μνῆμα gen. μνήματος, which must necessarily reflect a P.H. *mnemə- and P.I.E. *mnēmn̥-. Likewise, χεῖμα, χείματος must reflect P.H.*kʰeiymə- < *gʰeiymn̥. (Both survive alongside o-grade reflexes μνήμων, χειμών.)
9b(ii). *-mn- in many cases in Sanskrit is also not resolved by syncope, but by epenthesis. In some cases, it persists (e.g. sumná-).
9b(ii). Kloekhorst’s examples from Kroonen are also not without their complications in this regard.
9b(iii)a. *bʰudʰmḗn, Kroonen argues, syncopated its -m- in oblique stems in PIE, creating the simplified stem *bʰudʰnós from whence budhná.
However, this must have happened late enough for Proto-Hellenic to escape with its *bʰudʰmḗn intact, leaving us with Ancient Greek πῠθμήν < *putʰmḗn with oblique stem *putʰmén- in which *bʰudʰm̥n- appears to have been replaced by *bʰudʰmen- by analogy with the accusative case *putʰménə (a regular outcome of the PIE *bʰudʰménm̥ in Proto-Hellenic).
9b(iii)b. Proto-Germaninc *bragma does not necessarily reflect PIE *mrogʰ-mn-o/*mregʰ-mn-o, its shape pointing to a preform *mrogʰnom. Its only certain cognate, Greek βρέχμα, reflects a regular development of PIE mregʰ-mn̥ by way of Proto-Hellenic *brekʰmə, further supporting a retention here.
9c. Further, as mentioned above, it is clear that this change could not have encompassed Tocharian (cf. P.Toch. *mñäś). This means it cannot be employed as an example of a shared innovation of Non-Anatolian PIE, but can perhaps be used in defence of “Indo-Tocharian.”
9d. The unusual forms found in Tocharian may be explained without reference to an oblique stem in *h₁mn- via analogy to Proto-Samoyedic *minä. Uralic influence would explain the presence of -n- in Tocharian without reference to other branches.
9d(i). However, this is just as much of an argument for its retention by analogy and would not fully explain its absence, particularly in Greek, where it leaves no trace.
10. Item 20, the proposed voice assimilation reflected in Non-Anatolian reflexes of the ultimate ancestor of PIE *nókʷts (*negʷt-), also reflected in Non-Alatonian reflexes of *sogʷ(ʰ)tH to me seems sound.
10a. Hit. nekutt- can only reasonably reflect an original *négʷt- rather than *nókʷt- and the same can be said for Hit. šakuttai-.
10b. That said, this represents a fairly trivial conditioned sound change which could easily have taken place within a single generation.
11. Semantic innovations: these account for numbers 1 through 8 of the authors’ 34 proposed innovations. I won’t be treating these in order, as some are easier to account for than others.
11a. Items 2, 3, 6, and 7 concern the semantic extension of four transitive verb stems:
Hitt. ḫarra- “grind, crush” vs. non-Anatolian *h₂erh₃- “to plough”.
Hitt. laḫu- “to pour” appear vs. non-Anatolian *leuh₃- “to wash”.
Hitt. šāḫ- “to fill, stuff, plug” vs. non-Anatolian *seh₂- “to be sated”.
Hitt. šai- “to impress, prick, stake” vs. non-Anatolian *seh₁- “to sow”.
Anatolian verbal pair ēš- and eš- (respectively “to sit” < *h₁eš and “to sit down” < *h₁e-h₁s-) with non-Anatolian *h₁e-h₁s- “to sit” an s *sed- “to sit down”.
11a. Beginning with the last: *sed- is taken to be a post-Anatolian innovation, replacing the earlier root aorist/stative form h₁es- while the originally dynamic *h₁e-h₁s- takes on a stative meaning.
11a(i). As noted by Kocharov in his 2020 review of this article (and the rest of the volume it introduces), this shift in aspect of *h₁e-h₁s- accompanied by an introduction of a new root is hard to explain given Indo-European’s tendency towards regular opposition in tense and aspect.
11a(ii). It’s hard to disagree with his assessment that it would be “more economical to assume that (pre-)PIE had complementary verbs *h₁es- and *sed- showing a contrast of lexical aspect parallel to PIE *ḱei- ‘to lie’ and *legh- ‘to lie (down)’.”
11a(iii). We cannot therefore call this a “settled” example of a shared innovation of non-Anatolian, although we can class it as “pending”.
11b(i). As for the rest: it seems reasonable to presume that the first, ḫarra-, represents root with the original meaning “to crush or grind” which has developed the technical term “to plough” as an extended sense of “to crush or grind (the soil)”.
11b(ii). Kloekhorst (2008) suggests that this indicates an Anatolian divergence before the IE family had come into contact with the plough. This strikes me as plausible, even if Anatolian broke away from the rest of the PIE language continuum as late as the end of the 4th millennium BC.
11b(iii). The case of lāḫu- (which must phonologically reflect a stem /lāhʷ-/). seems similarly plausible on the face of it, extending the basic meaning of “pour water” to the more specific “wash”.
11b(iii)a. However, if this verb is cognate with *leuh₃, the laryngeal metathesis is not easily explained: /lāhʷ-/ in Hittite should theoretically descend from a Proto-Anatolian *loh₂-u. *leuh₃- would, under normal circumstances, yield *lū- in both P. Anat. and Hittite.
11b(iii)b. The authors actually cite this putative laryngeal metathesis as Item 26 on their list, after Oettinger (2013). This is one of 11 items which they describe as “promising but perhaps less forceful” than their other 23 more confidently asserted arguments. They note here that this metathesis is not clearly explicable, and we should likely suspend judgement on this word in the meantime.
11b(iv). The cases of šāḫ- and šai- seem even less clear-cut. In particular, šai-, if it is cognate to *seh₁-, seems likely to have originally carried the sense of “press into the ground”, in which case its divergent meaning indicates an innovation by extension in Anatolian (where it has also come to mean “to seal”, or of headgear “to put on”.)
11b(v). As for šāḫ- < *seh₂, it seems as likely to me that instead of a modification of an original meaning from “fill” to “satiate”, we are actually seeing complimentary reinterpretations of an originally dynamic root, “to become full”.
11c. This phenomenon of fientive/dynamic roots being reinterpreted – in manners varying from developing into purely nominal/adjectival roots, into statives, or even into transitive verbs – may also be represented in Items 4 and 5.
11c(i). *mer- means “to die” in all non-Anatolian branches, but in Anatolian (Hittite) has the sense “to disappear, vanish.” Hittite nekutt- “twilight, dusk, dawn” is clearly cognate to the root which becomes “night” in all other branches, *nékʷt-/*nokʷt.
11c(ii). In the case of nekutt- we are actually lucky enough to have direct evidence of an origin in a fientive verb: Hittite has also a verb neku- meaning “to become evening”.
11c(iii). These two words both appear to derive, ultimately, from the root *négʷt- “naked”. Hittite nekumantae “to remove one’s clothes” continues this meaning and appears to be a re-verbalisation of a verbal adjective *nekumant- (< *negʷ-went-, “having nakedness”).
11c(iv). If we reconstruct *négʷ- as an originally fientive verbal root (rather than purely nominal/adjectival) “to become naked”, or “to become bare, stripped”, we can see the later nominal forms arising from the sense of “to become stripped of light”.
11d. In this context, we might also construct a similar path of evolution for *mer-: from fientive “to become hidden” (or similar) to an active intransitive “to depart, vanish”.
11d(i). In terms of innovation, the semantic extension to “die” as a euphemism is extremely likely, but this is again something that could have happened in relatively short order.
11d(ii). Both meanings could also have been innovated separately from an originally fientive root.
11e. Supporting this possibility, we have the Hittite adjectival root, maršant- “deceitful, dishonest, profane”, and its derivative maršaḫḫ- “desecrate, profane, falsify”.
11e(i). Although there is no stem marsa- evident in Hittite, Kloekhorst derives maršant- from a putative P. Anat. stem *mr̥sa-, reflected marginally in Luwian and Lycian, and points to a PIE origin in *mr̥s-o, cognate to Sanskrit mr̥ṣā “in vain, falsely” and Sanskrit marṣ- “to forget” (along with Lithuanian už-mirštri and Tocharian märs-, both “to forget”).
11e(ii). Kloekhorst also identifies a root maršē- “to become corrupt” which appears to reflect a PIE “stative” verb in -eh₁-, < *mr̥s-eh₁.
11e(iii). If we assume a fientive meaning of the underlying root *mer- (“become concealed, become hidden”) we can easily interpret the extended stem in -s as an aorist with the simple meaning “hidden”, without reference to time or state.
11e(iv). The stative extension *-eh₁- as seen in maršē likely reflects a modification of the PIE instrumental, originally reflecting “with hidden-ness”, which has been re-interpreted as a causative. The semantic extension of such a meaning to encompass falsehood and deceit is documented elsewhere (compare Arabic kfr originally meaning “to cover”).
11e(v). Less baroquely, we can also make a case for *mer- retaining its meaning “to depart, disappear” in at least one marginal case in Indo-Iranian: *maryas (reflected in Skr. marya OP. marika presumably < *mer-y-ko-), which has the meaning of “young warrior” and “stallion”, and could easily be taken to reflect the meaning “one who leaves (the tribe)”.
12. At this point we are equipped to discuss the significance of item 1: participles:
Hitt. participle suffix -ant-, which forms both active and passive participles, vs. cl.PIE *-e/ont-, which is only active.
In fact, productive uses of this suffix in Hittite seem to be passive in almost all cases. Unusually, they are also almost always perfective in sense, reflecting an attained state, whereas reflexes of *-e/ont are aorist or imperfective in other branches.
12a. Note “Hittite” and not “Anatolian” because Luwic languages preserve the *-ent suffix only sporadically, as in the name of the storm god Tarhunt- (/trχʷnt-/), which is a lexicalised participle of the verb *terh₂-u- “prevailing, conquering, able” – notably an active participle.
12a(i). The handful active participles in -ant in Hittite appear to be either lexicalised as nouns (as in *tr̥h₂-w-ónt-s) retained from P. Anat.
12a(ii). The few verbs in Hittite which form an active participle: “drinking” (akuwant- < aku-), “eating” (aduant- < ed-) and “being” (ašant- < eš-) may also appear with passive meaning (e.g. “eaten”). This can be explained as either a retention of obsolete meanings in common everyday verbs, or as a result of the ambiguous transitivity of these verbs.
12b. The productive Luwian participle suffix -(i)mma- (which has a generally passive meaning) appears to have cognates in Balto-Slavic, Albanian, Tocharian, Greek, and possibly in Indo-Iranian. All may well share an origin in a putative mediopassive participle *-m(e)no , reflected in I.Ir. -āma, P.B.S. *-mo, and Greek -μέν-, Tocharian A -mām.
12b(ii). A similar formation appears to be preserved marginally in Italic, given the shape of Latin words alumnus (linked to alō “to nourish”, perhaps indicating originally “one who is nourished or receives nourishment”) vertumnus (from vertō, “to turn”) and femina.
12c. Although the specifics of the evolution of this element are not clear, it seems prudent to conclude that Proto-Anatolian inherited both of these morphemes from PIE, as did other branches, and generalised them in its own way.
12d. The presence of the fossilised active participles in Luwian - along with some examples of the generally attributive -ant in Hittite also having active meaning, supports this, and suggest that regardless its original function (whether active, passive, or both) the -ant participle was generalised to a perfective passive form in Hittite (its active functions performed by other elements).
12d. We must instead assume that both *-ent- and *-men- began their lives in PIE as elements used primarily with verbal bases to create verbal adjectives. This function can be called upon to explain the presence of both suffixes as adjectival formations in the “Caland system”.
12d(i). Indeed, traces of the second element are visible in some instances in Hittite, for example in gimm “winter”, attested only in the dat.loc. singular gimmi and elsewhere supplanted by gimmant-, must reflect PIE *gʰ(e)y-men “cold” whence also Greek χεῖμα, Latin hiems, containing the element *-men- in assimilated form -mmi.
12d(ii). The fact that Hittite gimmant- had apparently overtaken gimm- early on in the language’s history provides an interesting illustration of the likely meanings of both suffixes.
12d(iii). Melchert (1983) discusses the development of this suffix in Hittite, noting that it has a primarily deverbal form which creates verbal abstracts. In other languages (e.g. Sanskrit, Greek), it can be seen in fossilised form in both verbal abstracts and rare agent nouns.
12d(iv). The function of *-ent- appears to have been used to create attributive verbal adjectives equivalent to English present participle/gerund formations in -ing, “the x-ing y, the y is x-ing.”
12d(v). In gimmant- we appear to see a form which might have originally meant something like “coldness” or “the cold one” (from a stem *gʰey-, “cold, become cold”) which has been re-interpreted as a verb meaning “be(come) winter” and then extended as a verbal adjective “winter-ing”.
13. “Where,” you might well be asking, “is he going with this?” To put it as succinctly as possible, I believe we can explain Hittite’s unusual situation by reference to the transformation of a system of roots and derivations which was largely verbal in semantic terms to one which was largely nominal and which has left a trail of irregularity and destruction in its wake - most notably in the form of various different Caland system formations.
13a. Bozzone (2016) builds a compelling case for how this process could have occurred. Her evidence is mostly confined to Indo-Iranian (particularly Sanskrit), but shows how various elements within this system could become “stranded” through the evolution of various semantic roles.
13b. For our purposes we need only note that evidence of such a process is already present in Hittite: the various derivatives of the hypothetical stem marša- attest to a process by which a single root is extended and re-analysed multiple times, leaving some derivations “stranded” in the process.
13b(i). The case of gimmant- (and its verbal derivative) provides another, more limited example of the process, which is also paralleled in the adjective ḫarki-/ḫargai, “white, bright”, a clear cognate of of Greek ἀργός and, Sanskrit rjrá, and Tocharian ārki/ārkwi.
13b(ii). The Hittite and Tocharian reflexes clearly exhibit a Caland variant in *-i- in contrast to the Greek and Sanskrit variants in *-ro- (although i-variants in Greek and Sanskrit do occur marginally). A derived stem ḫargnu- “to make white” exhibits a form without the -i-, and the fientive ḫarkiješš- “become white” exhibits *-eh₁-sh₁-, the “stative” (< instrumental) suffix combined with an imperfective. We also see clear traces of the other typical “Caland” morphemes.
13c. The derivational system of which the elements which became participles in the various IE branches is clearly already in operation in Anatolian, with parallel results to other IE families.
13c(i). The apparent discrepancy in the development of the -ant participle can only be addressed within this context. All evidence points to participle elements originally developing from forms acting as deverbiatives, of no fixed aspect or tense, which were generalised into different participle forms in various branches.
13c(ii). If *-ent- did indeed originally carry an attributive meaning, its evolution into an aorist/present participle is fairly intuitive. Its development of a passive sense in Hittite is less expected, but must be a later development or specialisation that must replace a more naturally active original sense.
14. At this point we must take a break to discuss one of the two syntactic innovations mentioned in the paper, because a relation or variant of this suffix must also be the root of the Hittite “ergative” suffix -anza (which also has parallels in Luwian, Lycian, and perhaps also in Palaic). This peculiar manner of marking neuter agents is given as Item 22.
14a. Lopuhaä-Zwakenberg, in the same volume in which the article we are discussing appears, traces the “ergative” (which she calls the agentive) to the Proto-Anatolian semantic suffix *-ont, which she refers as a “personalising suffix”.
14a(i). She concludes that the disparity between Anatolian and other Indo-European languages indicates that:
Like Anatolian, PIE did not allow neuter nouns to stand as agents.
Anatolian “filled” this paradigmatic gap by grammaticalising the *-ont- suffix to mark neuter agents.
All other branches extended the neuter accusative/nominative case *-om to cover the gap.
14a(ii). This points to a common innovation in other branches and to an early split for Anatolian.
14b. To arrive at this conclusion, Lopuhaä-Zwakenberg proposes two a-priori explanations for the difference between Anatolian and other branches: either PIE did allow neuter agents, and Anatolian innovated by using the *-ont suffix to mark them, or PIE also disallowed neuter agents and marking them with the accusative *-om represents an innovation in other branches.
14b(ii). She discounts the first possibility because “one would expect the loss of agents to happen in semantically inanimate nouns, rather than in morphologically neuter nouns”, and cites that even in PIE there is an imperfect alignment between semantic inanimacy and the morphological neuter gender, with reference to the lemmas *dʰéǵ-m- “earth” (feminine in most branches) and *peḱu “cattle” (morphologically neuter but semantically animate).
14b(iii). She concludes “Thus one would expect the agent of all inanimate nouns to be lost, and one would expect the agentive construction to be determined by animacy rather than by morphological gender.” She cites one example from Cuneiform Luwian which refutes this: morphologically animate tiia̯mm(i)- appears in agentive stance ti-ia̯-am-ma-an-ti-iš as part of the formula “the sky and the earth must wash their mouths”.
14c. I have to say that I agree with Lopuhaä-Zwakenberg’s conclusion that PIE must have originally disallowed neuter agents, but disagree with her argument here.
14c(i). The two examples she gives to illustrate the PIE semantic/morphological concordance with animacy marking are not the strongest: Hittite evidence points to an original neuter gender for *dʰéǵ-m- (as we would expect given that the feminine is a post-Anatolian innovation, see below), and *peḱu most likely originally referred to “movable wealth, that which can be taken in hand” as evinced by *peḱ also carrying the meaning “to pluck” of wool or hair, and must therefore have been semantically inanimate before its secondary meaning “livestock, cattle” developed.
14c(i)a. We can actually use these examples to suggest a much greater concordance between semantic and morphological animacy in an earlier stage of PIE: *dʰéǵ-m- becomes morphologically (and partly semantically) feminine in other branches perhaps by association with the “Earth Mother”. *peḱu remains morphologically neuter, but comes to have semantically animate referent.
14c(ii). However, this to my mind actually reinforces the conclusion that PIE did not permit neuter agents, rather than undermines it.
14c(ii)a. Rather than this implying a gap in the language’s paradigm, I think we should suppose that PIE, in earlier stages, used the “Anatolian solution” of the *-ont suffix.
14c(ii)b. Kloekhorst (2008) notes that it’s hardly possible to escape the conclusion that this suffix must be related to the participle, but that the nature of the relationship is uncertain. I believe I can now offer an elegant solution to this in light of (11-13) above.
14c(ii)c. If we presume the *-ont element to have originally had a primarily attributive function, its application in the creation of neuter agent statements suggests it originally carried a meaning something like “x, by means of y, verbed z”, where x is an assumed agent, y is an inanimate noun marked with the *-ont suffix and z is the object or patient of the verbal argument.
14c(ii)d. In other words, the “problem” of a neuter agent is resolved by suborning the proposed neuter agent to the verb as a kind of verbal adjective (I am told a similar construction occurs in modern Japanese).
14d. This seems to me to be a promising assumption on many counts: it might allow us both to explain, with further investigation, the precise circumstances of the emergence of the PIE noun classes/system of gender. It may also prove to be fertile ground in untangling the specifics of “original” PIE morphological alignment.
14e. It is also, perhaps, one of the most compelling arguments both for an early split of Anatolian from the rest of the family, and for a particularly early date of this split.
14e(i). Once again, however, I am not convinced at all that it is evidence that requires a 1000 year time depth to explain – a few centuries is probably enough. The processes which govern the breakdown of the “old” system in PIE are already well underway by the time Anatolian would have broken off, and would have continued in relative isolation on a divergent path for another 1000 years or so before our evidence of the family becomes available.
15. Morphological innovations account for items 9 through 18. Of these, some are easier to address than others.
16. Item 9, the non-Anatolian innovation of the feminine gender, is at this point confirmed, and represents perhaps the single most obvious and substantial argument in favour of the Indo-Anatolian hypothesis.
16a. However, as mentioned above, it is simply not necessary to assume a 1200 year separation to account for such an innovation.
16b. Noun classes can emerge or collapse fairly rapidly under the right conditions. In the case of PIE, it seems reasonably clear that the feminine gender is an extension of classes of collective and abstract nouns in *-h₂.
16c. For this class to develop a distinctively feminine sense requires only a simple 2 step process:
1. Collective nouns become a popular or predominating source for female names.
2. Animate nouns are reanalysed as “masculine” by comparison.
16d. This process can reasonably be assumed to have taken place within the course of only 100 years or so, perhaps slightly longer, perhaps slightly shorter. In light of what we’ve discussed in (14) above we might wish to extend this period out to 200-300 years, perhaps a little more, perhaps a little less. But we certainly do not require a time depth of 1000 years to fully account for the process.
17. Item 10 (Anatolian retention of *tiH- in the nominal stem of the second person singular personal pronoun as opposed to *tuH- in other languages) is to my mind firmly established by Kloekhorst (2008).
17a. However, this represents a trivial analogical change, and if anything is an argument for a recent split of Anatolian from other branches; otherwise we are to presume that this retained vocalic archaism has been lovingly preserved for well over 2000 years by the time we receive news of it in Hittite, Palaic, and Luwian.
18. Item 16 (conjunctions šu, ta as a parallel to Cl.PIE demonstrative pronouns *so/*to-) can be immediately dismissed - ta appears to carry on the PIE instrumental *toh₁ (Rieken 1999) while ša reflects an assertive particle *h₁su (Zimmer 1994).
18a. The existence of a reflex of *toh₂ suggests Anatolian had inherited the original system of PIE demonstratives but lost them by the period of recorded Hittite (as per Melchertt).
18a(i). The authors’ inclusion of this item feels like a particularly egregious reach as they cite Watkins (1963) whose suggestion has since been obsoleted. They are surely aware of the works which accomplish this because they cite them frequently.
19. Item 18 (Hittite’s retention of an allative in -a, which is reflected only as fossilised *-o in *pro “before” *upo “down (to)” and *h₂do “to(ward)” in other branches) can likewise be easily dismissed: this cannot even correctly be called an innovation.
19a. As with many other items on the authors’ list, this can be used as a piece of evidence for Anatolian’s early divergence. However, on its own, it is not a change that requires a great deal of time separation; an independent allative case in *-o is very likely to have become blended and confused in the other branches with other lative cases at a fairly early stage.
19b. I should also note that no affirmative case has ever been made that allative *-o, was “innovated out” of non-Anatolian PIE in a single instance. It can therefore only be securely identified as another retained archaism of Anatolian, rather than a shared innovation of the rest of the family. Further work would be required to demonstrate this.
19. Items 11, 12, 13, refer to specific athematic reflexes in Anatolian which have been thematicised in all other branches. While interesting in and of themselves they do not necessitate a particularly vast time depth to have come about.
19a. Thematicisation of previously athematic nouns was common in Anatolian and productive in Hittite in every period, indicating that Anatolian shared a regularising tendency common to all branches towards thematic stems.
19b. Since thematicisation is relatively trivial and by no means uniform across the branches, to use any particular athematic noun in Anatolian as a contrastive example of a shared innovation when it appears thematised in other branches is a tricky operation – as we’ll see below.
20. Item 11: the change *h₁ekus (whence Hitt. ekkuš Luwian āššu-/azzu, Lycian esb-) to *h₁ekwos (whence all other non-Anatolian reflexes) is well attested.
20a. The athematic u-stem is preserved in Greek ὠκύς and Indo-Iranian āśu- (Skt.), āsu- (Av.), all meaning “swift, swiftly” (which in Vedic is used in a nominal sense as a doublet for aśva).
20b. What appears to be a bare root version is also evinced by Latin ocior, “swifter”, which must originate with a Proto-Italic ōk-.
20c. All of these forms point to a reduplicated *h₁oh₁ḱús, which itself suggests the original root *h₁ek- is another verbal root (“go swiftly” or “make haste” perhaps?) which survives only with deverbal meanings.
20d. Regardless of the precise route of development, the thematicisation of *h₁ekwos outside of Anatolian must be viewed as a trivial analogical development, which the authors acknowledge elsewhere.
20e. If it can be regarded as a common innovation, it cannot be one separated from Anatolian by a particularly great gulf of time, and there is nothing to stop thematicisation being an independent innovation across branches.
21. Item 13 Hittite ḫuwant- reflecting an athematic *h₂uh₁-ent- v.s. thematic *h₂wéh₁n̥tos cannot be called upon as evidence of a non-Anatolian innovation because it is shared by Greek ἀείς, ἀέντ- (< P.G. *awēəts, *awént-).
21a. Both the Greek and Anatolian (Hittite) reflexes must be generalised from the accusative stem of an original *h₂wéh₁-n̥t-s, *h₂uh₁-ént-m, *h₂uh₁-n̥t-ós, while other branches show a basic stem *h₂wéh₁-n̥t-o.
21b. Are we to take this as evidence of a putative Greco-Anatolian branch (possibly in conjunction with the wider Balko-Caucasian group, giving us Helleno-Thracio-Albanio-Armeno-Anatolian)? Only time will tell, but the evidence seems to point us inexorably toward this conclusion.
22. The case of Item 12, P. Anat. *jéwg- yielding Hittite jūk, in obvious opposition to non-Anatolian *jug-ó, is perhaps the most interesting of the three.
22a. It’s notable that the Hittite word here is not only a root noun, but a root noun of a slightly divergent form: *jéwg- is generally reflected in nouns meaning “(yoke) pair”, a meaning shared by the Hittite reflex.
22b. In the languages where forms from both *jug- and *jéwg- are preserved, we do find that they frequently form doublets: possibly reflecting analogy between two forms which would be close to identical in many descendant languages.
22b(i). Greek, e.g., contrasts ζυγόν with ζεῦγος, Latin iugum contrasts with iūgerum, a singular back-formed from the plural iūgera, which presupposes an original, unattested singular *iūgus, *iūgesus.
22b(ii). It appears as though the Hittite jūk was later thematicised to juk-a-, with a regularised neuter nom/acc singular jūkan.
22c. Frustratingly, jūk does not appear to survive in plural forms, meaning we cannot be sure that this is another instance of a doublet in the same pattern.
22d. If this is a root noun, it must reflect a fossilised sense of an original verbal root meaning “to pair (together)”, and reflect a noun formed directly from a root aorist.
22d(i). Does this present evidence of a shared innovation outside of Anatolian? Not conclusively, as we only have evidence here of a retained archaism, as in (17) and (19) above, among others.
23. Item 17, the presence of the elements *sm/*si in pronouns, explored by de Vaan in the same volume, is intriguing but seems muddled. There is marginal evidence of these forms - or forms like them - existing in Hittite second and third plural possessive pronouns (which de Vaan explicitly mentions) although this is by no means certain and likely merits further investigation.
24. Item 14, the Hittite genitive in -an which Kloekhorst (2017) presumes to be a number indifferent continuation of an Indo-Anatolian *-om, contrasting a number-specific innovation of this element as a plural, has been conclusively refuted by Goedgebuure (2019), who demonstrates that *-om must continue an original PIE genitive plural.
25. Item 15, the lack of thematic present verbs in *-e/o-, as argued in Kloekhorst 2017, seems plausible – especially if as Kloekhorst suggests, the *-e/o- element in thematic presents is an evolution of the morpheme that act as a subjunctive in other branches.
25a. If this is the case, Anatolian must have jettisoned this proto-subjunctive element early in its development but after the split from the rest of the family.
25b. We can therefore call this an argument for the early split of Hittite - but again, this is not an innovation/loss that requires one to imagine a thousand-year time gap.
26. Returning to syntactic innovations we can now also address Item 23: the syntax of bare interrogatives. Haug and Sidelstev demonstrate that Anatolian “bare interrogatives” represent a separate innovation from the “bare interrogative” syntax found in other branches.
26a. The uniformity of non-Anatolian bare interrogatives suggests a shared innovation and is a definite argument in favour of Indo-Anatolian - though once again the question of time must be considered.
26b. As a point of interest, the similarity in form (if not in phylogenic origin) of Anatolian bare interrogatives and their non-Anatolian counterparts is so close that one wonders if they might have been influenced by contact with other Indo-European families. Early Greek and Indo-Iranian represent the most obvious candidates.
27. At this point, we’ve covered the author’s 23 strongest points of evidence, and one of their 11 “less forceful” contributions. The 11 secondary arguments are noted by the authors to have serious weaknesses, but we’ll recap the remaining 10 here:
27a. Item 25 (“Hitt. 1pl. -u̯en(i) vs. cl.PIE dual *-u̯e(-): development of a clusivitiy system to a plural/dual system”) can be safely dismissed: even if there is evidence that PIE’s verbal morphology included a clusivity system, there is no evidence it persisted in Proto-Anatolian (which the author notes) and until this can be found, it should be assumed that the Hittite plural is generalised from the dual.
27b. 2 (Items 29 and 30) cannot be taken to represent common innovations as of writing.
27c. 6 (Items 27 and 28 plus Items 31-34) are based on the assumption of unclear sound changes in Anatolian, or on the scenario of a consonant shift which cannot be conclusively demonstrated. (It’s interesting to me given earlier discussion that Item 27 is one of the few arguments for a genuine phonemic shift proposed between Indo-Anatolian, Anatolian, and Classic PIE, and that it is also, as with the laryngeal theory advanced in Item 19, one of the weakest).
27d. This leaves us with Item 24 - the argument of the ḫi-conjugation. Although it seems fairly conclusive that the ḫi-conjugation originates with the PIE perfect, the authors consider the reduplicated perfect class a potential common innovation in Classic PIE.
27d(i). This cannot, however, be demonstrated with confidence to be a shared innovation, and traces of “unreduplicated perfects” which must reflect survivals of the original distribution, are evinced in multiple branches.
In summary
Of the 34 shared innovations proposed, I think it possible to accept that:
Nine of these should be taken to be secure examples of shared innovation. (These are Items 2, 5, 9, 10, 11, 15, 20, 22, and 23).
Fourteen can be shown either to not be shared innovations, or to have alternative explanations which would make shared innovations unlikely in the absence of further evidence.
Eleven should be regarded either as speculative or inconclusive.
Of the 9 items which appear conclusive, only a few can be taken to constitute evidence of “major” divergence at the time of Anatolian’s split from the protolanguage.
The most fruitful of these seem to me to be Items 9 and 22, which I would hazard a guess probably represent two prongs of a related system of change. But we can also show that many of the processes which lead to divergences of Anatolian from its non-Anatolian siblings were already in operation before this split.
As I said at the outset, the particular case of “Indo-Anatolian” which the authors wish to advance consists of 3 basic premises:
There are shared innovations in the non-Anatolian daughter languages…
Which point to the existence of a shared non-Anatolian protolanguage, dubbed “Classic Proto-Indo-European”…
Which can be shown by the magnitude of these innovations to have been spoken at a date of significant remove (more than a millennium) from the protolanguage which gave rise to both Proto-Anatolian and Classic Proto-Indo-European.
In order for the particular case for Indo-Anatolian advanced in the paper to be proven, all three of these individual assertions must be demonstrated convincingly.
We can say that the first has been demonstrated with a high degree of confidence. A single shared innovation, however minor, would confirm an early split of Anatolian from its sisters. We have eight more than this, and thus can probably consider the hypothesis that Anatolian is the oldest sister of the Indo-European family to be proven.
As for the second and third prongs of the argument: we should consider that if (1) is true, (2) must necessarily also be true. After Anatolian split off from the main tree, the rest of the Indo-European family must have persisted, for however short of time, in speaking a common protolanguage.
A question of time
However, this period of time cannot have been a very long one: only 2 of these shared innovations constitute a system of changes which could not have occurred within a single generation of the separation.
Most of the demonstrable shared innovations we can point to suggest that “Classic PIE” ceased to be a cohesive dialect continuum in very short order following the departure of Anatolian. The major area of shared innovation in Classic PIE is the nominal system, but as discussed above, the non-Anatolian innovations are not an especially arduous process.
I suspect that the process of innovation which produces “Classical PIE” as opposed to “Inclusive PIE” or “Indo-Anatolian” cannot have occurred over a period any longer than around 200-300 years. Depending on other chronological considerations we might therefore place the date of the split towards the end of the 4th Millennium BC - perhaps as late even as 3000BC.
Let’s wrap this up
It occurs to me as I type this that despite setting out to refute the “Indo-Anatolian” hypothesis I have actually shown another variant of it to be true.
It may be a petty semantic argument but it’s one I’ll close out on regardless: even though it is a technically correct term, to speak of “Indo-Anatolian” is not necessarily helpful to us in trying to make more precise statements about the development of the Indo-European family: it is functionally identical to saying “middle” or “early” Proto-Indo-European, and does not allow us to think about what we are actually seeing in a way which is conducive to our better understanding of it.
It also does not allow us to make useful inferences about Anatolian, about other branches, or the nature of the protolanguage from which they developed. Sending us as it does on millennia-long wild goose chases and trying to fit data into schemas created in our own imaginations. If we’re looking for evidence of Proto-Indo-Anatolian speakers c.a. 4000BC, or even 4200BC, we are unlikely to find them, and likely to project our own superstitions onto the remains of perfectly blameless archaeological horizons in the process.
It is also, it must be said, a profoundly unaesthetic and imprecise term that fails to tell us anything about the nature of the family or its speakers. (The same can also be said of “Indo-European”, though this at least has some sense of historical continuity in scholarship). Comparative linguistics and philology are supposed to be scientific endeavours, and the terminology we use should reflect this.