On the homonymy of ‘put’ and ‘suck’ in Proto-Indo-European


 PIE *dheh1- ‘to put, make’ and *dheh1-(i-) ‘to suck (mother’s milk)’ look like two separate verbal roots with a very different meaning, which happen to be homonymous. Yet recent investigations have shown that the morphological behaviour of both verbs is more similar than was previously thought. This article offers a re-evaluation of the evidence and explores the options to bridge the semantic gap between ‘put’ and ‘suck’.

Of course, the existence of two homonymous roots is quite conceivable. Since Pre-PIE sound changes may have merged sounds which were distinct at an earlier stage, it would only be natural if we found some degree of homonymy in PIE. Other PIE roots which have an identical form but which are regarded as two different etymological entities, e.g., by LIV2, are *med-'to measure ' and 'to get satiated' , *peuH-'to cleanse' and 'to rot' , and *teh2-'to steal' and 'to melt' . Yet did 'put' and 'suck' really have a completely different morphological profile in PIE? Our handbooks admit that the roots coincide in two important formal aspects: they share a radical verb stem (root aorist *dheh1-'put' versus root present or aorist *dheh1-'suck') and reduplicated formations, viz. a reduplicated present for 'put' in all languages and a reduplicated present for 'suck' in Anatolian (Mallory/Adams 1997: 382). Recent investigations have furthermore proposed the reconstruction of an original i-present meaning 'to put' on the basis of evidence from several branches of Indo-European. It seems worthwile, therefore, to reassess the evidence for both roots and to reconsider the semantic gap between them.
Beforehand we must settle a terminological matter, viz. the exact meaning of the terms used for 'suck' . Depending on the different agents and experiencers of the action, a three-way semantic distinction can be made between (1) 'to suck milk from a teat' , with a new-born mammal as the agent (German saugen, Dutch zuigen, French sucer), (2) 'to nurse, suckle, make a young suck milk from a teat' , with the mother or wet-nurse as the agent and the new-born as experiencer (German säugen, Dutch zogen, French allaiter), and (3) 'to lactate, have milk that can be sucked by a young' , with the mother as experiencer (cf. Gamkrelidze/Ivanov 1995: 487, Martirosyan 2010. Since the English glosses 'suck' and 'suckle' are used promiscuously for both meanings depending on the author, and since 'suckle' is ambiguous as to its valency (transitive or intransitive), I will define the terms in the following way, which has no other goal but to provide clarity within the context of this paper: 1=suck, 2=make suck, 3=lactate. Of course, the distinction is not always expressed in the lexicon of individual de vaan Indo-European Linguistics 7 (2019) [176][177][178][179][180][181][182][183][184][185][186][187][188][189][190][191][192][193] languages. A fourth term that plays a linguistic role is (4) the milk-giving organ, the 'udder' , 'teat' or 'breast' , which can also be regarded as agentive. Finally, there is (5) the result of milking, basically 'milk' .
Another preliminary clarification is in place regarding my position in the debate on the form and function of verbal i-stem derivatives and on the reconstruction of the PIE verbal system in general. I assume the existence in PIE of athematic i-presents with ablaut of the type *CC-éi-in the singular and *CC-i-́in the plural. Some of the best known instances of roots with and without an i-suffix are *dheh1-vs. *dheh1-i-'to suck' (as discussed in this paper), *kreh1-vs. *kreh1-i-'to sift' , *peh2-vs. *peh2-i-'to protect' , *peh3-vs. *peh3-i-'to drink' , and *seh1-vs. *seh1-i-'to let go ' (cf. Rasmussen 1989: 19-63). Following the work by Kloekhorst 2006and Kloekhorst 2008, Lubotsky 2011, de Vaan 2011, and Kloekhorst & Lubotsky 2014, I furthermore acknowledge the existence of a verbal category with o/zero ablaut of the suffix *-i-, sg. *CC-ói-, pl. *CC-i-. It represents an Early PIE intransitive verbal stem with o/zero-ablaut (which one may call a "perfect" or a "prestage of the canonical PIE perfect") derived from an earlier athematic i-present (Kortlandt 2010: 43, 373-382). The stems in *CCói-, *CC-i-́surface either as a present, as in the Hittite present type dāi/tii̯ anzi- (Kloekhorst 2006(Kloekhorst , 2008, or as a perfect, as in Indo-Iranian (Lubotsky 2011). There is no agreement on which meanings the i-suffix may have conveyed in Pre-PIE or Early PIE, though chances are that it added an imperfective or progressive aspectual meaning to the meaning of the root. Kortlandt (2010: 378) argues that *-(o)i-"apparently contributed a sense of directionality", and surmises that it may in origin be identical with the verbal root for 'to go' .
On the syntactic level, I assume that Pre-PIE had an ergative case system, in which the agent of a transitive clause was marked differently from the agent of an intransitive clause, the latter sharing the marking of a patient (Vaillant 1936;Kortlandt 2010, passim;Willi 2018: 504-540). This assumption explains, among other things, why the PIE nominative equals the accusative in the neuter gender: these were inanimate nouns which did not normally occur as an agent. This theory entails that the PIE nominative continues the Pre-PIE endingless absolutive (the patient marker). The Pre-PIE genitive ending *-s was the ergative marker for animates, which became the animate nominative marker *-s in canonical PIE (Beekes 1985). The PIE instrumental ending *-t, which is preserved in Anatolian and became "ablative" *-d but also, by lenition before obstruents, "instrumental" *-h1 (probably [ʔ]) elsewhere, served as the ergative of inanimate nouns (Kortlandt 2010: 40). The accusative marker *-m probably goes back to an earlier allative affix of Indo-Uralic.
As to the verbal conjugation, I adopt the view that (what became) the athematic present and aorist featured in transitive verbs (Kortlandt 2010: 391), corresponding to what became the Hittite mi-inflexion, whereas the thematic present and the perfect were used for intransitive verbs, corresponding to the subsequent Hittite ḫi-inflexion. At this early stage, the subject of athematic verbs indicated the transitive agent. In the thematic present and the perfect, the subject referred to the experiencer or the patient of the action, whereas the agent was expressed by means of the dative (if animate) or the instrumental (if inanimate), cf. Kortlandt 2010: 102, 399, Barðdal et alii 2014 In the following sections I will summarise the evidence for the meanings 'put' resp. 'suck' in reflexes of *dheh1-in the main branches of Indo-European, before reaching a conclusion in the final paragraph. As far as this survey concerns undisputed reconstructions for *dheh1-'to put' , I will remain brief and the evidence will not be treated exhaustively.

Armenian
a. 'put' . The PIE root aorist is continued in Armenian, 3sg. ed 'put' , middle edaw 'was put' . b. 'suck' . diem 'to suck, drink mother's milk' allows for several different reconstructions, among which are *dheh1-, *dheh1-i-, and *dhih1- (Martirosyan 2010: 239). For the two Classical Armenian variants dal and dayl 'colostrum, beestings' , Martirosyan 2010: 229-231 argues that only dal represents an inherited formation, with vocalisation of the laryngeal in a zero grade *dhh1-l-. Latin fēcundus 'fertile' shows the gerundival suffix -undus after a stem *fē-k-. None of the hitherto proposed etymologies for gerundival -cundus which regard its /k/ as a reflex of a PIE lexical root are convincing (Weiss 2009: 444, fn. 75, Jasanoff 2010, Brachet 2016; the other gerundives in -cundus are fācundus, iōcundus, īrācundus, rubicundus, and verēcundus). In view of the probability that the k-affix arose in PIE roots in a final laryngeal (Kortlandt 2018), and since the root *fē-is the only one that also shows suffixal -k-in Latin in other morphological categories (pf. fēc-, pres. fak-), it seems likely that *fēkondos provided the starting point for -cundus, possibly together with fācundus from the root *bheh2-. From these two forms, the suffix would then have spread to other gerundives meaning 'having the property of' .
3 As one reviewer duly notes, it has been proposed by Meiser (1998: 69) that Latin fīlius might reflect *fēlios if its -ī-was raised from *ē before i in the next syllable. Such raising is shown by subtīlis 'delicate' to tēla 'cloth' and suspīciō 'suspicion' from *subspēkiō. Yet this raising was restricted to unstressed position, which would mean that *fēlios became *fīlios in unstressed position and then replaced the stressed variant. Furthermore, this would still leave the ⟨i⟩ of Faliscan fileo and Venetic filia to be explained. Weiss (2009: 143) assumes that subtīlis may be explained by a development *-esli-> *-ezli-> -īli-in Latin.

de vaan
Indo-European Linguistics 7 (2019) 176-193 9 Celtic a. 'put' . Old Irish creitid, MW credu 'to believe' < *kred-dī-might contain the root aorist of 'to put' . The preterite 3sg. Lepontic tetu, Gaulish δεδε 'has erected' could represent the reduplicated perfect with o-grade of the root, though Matasović (2009: 86), following earlier scholars, prefers to derive the forms from *deh3-'to give' . b. 'suck' . The presents OIr. denaid 'sucks' , MW dynu, and Middle Breton denaff continue Proto-Celtic *dina-which looks like an original nasal present *dhi-n(é)h1-(thus LIV2) which would regularly have become *dini-but was reformed within Celtic to *dina-on the model of more frequently occurring nasal presents (Schumacher 2004: 273-275). It may go back to a PIE nasal formation together with Sanskrit dhinoti (see above), though it is also possible that we are looking at two independent nasal presents. The root form *di-with short vowel can be due to reanalysis of a form *dhih1-C-that arose after laryngeal metathesis in earlier *dhh1-i-C- (Matasović 2009: 99) or from vowel shortening in pretonic syllable before a resonant (Dybo's Law, cf. Matasović 2009: 86). The Old Irish preterite did 'sucked' is explained from a PIE perfect *didoi, *didī-by Schumacher. This seems to be a novel formation too, as no corresponding form exists in other Indo-European languages.

Summary
The tables below compare and summarise the evidence for *dheh1-meaning either 'to put' or 'to suck' . Each table compares the direct and indirect evidence for one particular formation. We find that the root aorist *dhéh1-and the i-perfect *dhh1-(o)i-surface both as 'to put ' and 'to suck' in Anatolian, Balto-Slavic, Italo-Celtic, Germanic, Greek, and Indo-Iranian. The k-extension is found for 'to put' in Greek and Phrygian,i-is specific to the meaning 'to suck' . Reduplicated *dhé-dhh1-is found as a verb only in 'to put' , but the nominal derivative *dhé-dhh1-presupposes 'to suck' . The lnominal is rare for the meaning 'to put' but common for 'to suck' .
The i-perfect *dhh1-oi-, *dhh1-i-exists, on the one hand, as 'to put' in Hittite and as 'to think' in Indo-Iranian, and, on the other hand, as 'to make suck, give to drink' in Slavic and Germanic. If the latter two languages indeed contain the causative suffix *-ei̯ e-(it cannot strictly be proven), then unsuffixed *dhh1-oimeant anticausative 'to suck' or 'to give milk' . The nominal derivatives in *dhh1oi-n-mean 'nourishing' , 'giving milk' .
Altogether, then, the comparison yields two main results. Firstly, the earlier view that 'to put' and 'to suck' have very different morphological profiles must be dismissed. With the exception of the l-stems, which nearly only occur for 'to suck' , all verbal formations can be found with both meanings. Secondly, within the type 'to suck' , the semantic distinction between the subtypes 'suck' , 'make suck' and 'lactate' cannot be meaningfully linked to the morphological makeup: especially in case of the root formation *dhéh1-and the l-stems derived from it, all main possibilities are represented.

Conclusion
Under the traditional assumption that there were two homonymous roots of the shape *dhéh1-, their homonymy probably reflects the merger of two different pre-PIE roots. Homonymy or polysemy is a normal linguistic phenomenon, and it would be unsurprising if Proto-Indo-European possessed a certain number of homonymous or polysemous roots and words. This was the explicit position taken, for instance, by Ernout/Meillet 1959: 223, but they based it on the argument that only 'to suck' would be accompanied by i-suffixes. As we know now, i-suffixes also feature with 'to put' . In fact, both roots had a very similar morphological organisation, with a root aorist *dhéh1-, an i-perfect *dhh1-ói-, a reduplicated present, and an l-noun *dhéh1-l-. The morphological similarity between 'put' and 'suck' may, alternatively, strengthen the suspicion that one meaning was derived from the other. Since 'to put' is the more general meaning, it seems likely that 'suck, nurse' arose from semantic specialisation. A possible scenario for this development is the following. As we have seen above, we can reconstruct an imperfective meaning for the PIE verbal i-stems, and the presence of an experiencer subject in the PIE perfect. The basic meanings of the continuative, directional i-present 3sg. *dhh1-éi-t *X [nom.] and the non-volitional i-perfect 3sg. *dhh1-(ó)i-e *X [dative] may then be circumscribed as 'X is putting (toward)' and 'to X there has been putting (toward)' , Indo-European Linguistics 7 (2019) [176][177][178][179][180][181][182][183][184][185][186][187][188][189][190][191][192][193] respectively. In the context of breastfeeding, the young have to be 'put' to the breast (cf. the Hittite passage cited above), but because of the realia, the agency can be viewed in different ways. Both the mother, who 'puts' the young to her breast (especially the first time after birth), and the new-born, who 'put themselves to' or 'take to' the teats (once they have learned to do so), can adopt the role of the agent. Hence, in non-volitional 'to X there has been putting' , both the mother and the young may take the role of the experiencer, whence 'to suck' or 'lactate' (with the mother as the experiencer) or 'to make suck' (the infant being the experiencer). The same semantic polyvalence can be found in the root formations, with (1) Skt. dhatave 'to drink' ,°dha-'sucking' , Latvian dêt 'suck' , Slavic děti 'children' , (2) Germanic *dēan 'to make suck' , and (3) Latin fētus, fēcundus 'lactating' . There is no evidence that these formations adopted their semantic ambiguity from the i-stem verbs; instead, the root formation, too, could apparently be used with the opposite perspectives of 'put' ('make suck') and 'take to' ('suck'). As in the case of the PIE root *nem-'to assign, allot' , which probably got the opposite meaning 'to take, receive' in Germanic through usage in the middle voice (LIV2 453), similar variation in voice/perspective may explain the multiple valency of *dheh1-'suck, make suck' .
In active 'X is putting' , the logical subject would be the mother, but in a transitive middle such as that of Greek θήσατο 'sucked' , it can also be the infant who is 'putting himself' to the breast. The fundamental polyvalence of the verbs expressing the notion of 'suckling on the mother's breast' and their nominal derivatives would thus find a natural explanation: since the actancy in the triangle mother-breast-infant can be expressed from different perspectives, no one semantic expression has come to dominate. Gamkrelidze/Ivanov (1995: 487) cite an Ancient Egyptian parallel for this polyvalence of 'milk' and 'give milk' , viz. Egypt. mhr, which can express 'to milk (a cow)' , 'to suckle' , 'to give milk' (of a cow), 'to nurse (a baby)' , whence the terms for 'infant' and for dairy implements (Takács 2008: 464-465, TLA s.v. mhr).
Much of the preceding reasoning is of course tentative, and many details remain to be elucidated. Nonetheless, I hope to have shown that it is possible to think of pathways by which 'to put' could, in the specific usage of mammals and their new-born, and within the context of changes in PIE verbal inflexion and valency, give rise to the meanings 'suck' , 'make suck' and 'lactate' . Future etymological research may profit from these proposals.