Once, in my halcyon days of doctoral studies, I gave an informal presentation that ended not with the typical slide “any questions?”, but rather “any answers?”.
The talk’s topic fitted into my overall research into the word order of ancient Indo-European languages. My task was to acquire, present, evaluate and compare the available evidence, upon which I could then conclude with new findings or insights. So, my audience of Edinburgh experts might have been expecting a presentation concerned with what their junior colleague had come to understand. What they got instead was well-researched frustration about something that I didn’t understand – and still don’t.
So, forgive the reminiscences and selfish return to gone glory days; I have unfinished business with a curious piece of ancient syntax. It saw words dissected and spread around their sentences, its motives unclear. In what follows, I’ll set out what it looks like in our historical texts, before turning to how it theoretically got the better of me.
Dis-secting
Did our linguistic forebears feel flexible in how they could construct sentences? Abso-bloody-lutely.
In our historical sources, older members of the Indo-European family (to which Modern English also belongs) display great variation in the relative order of their words. The permitted permutations make it necessary to speak only of patterns and preferences, rather than hard syntactic rules.
The same flexibility could potentially be applied to an individual word, which might find itself cut and split up. Such lexical dissection is known as tmesis (‘t-me-sis’), from the Ancient Greek for ‘cutting’. It thereby shares a root with anatomy, dichotomy and atom. Occasional instances of tmesis occur in English for humorous emphasis: the aforeused abso-bloody-lutely is one victim of the verbal knife, as are ri-goddamn-diculus and boo-f*cking-hoo.
Higher up the family free, tmesis is sometimes documented in ancient languages like Latin. One famous example is attributed to Ennius, an early writer of the Roman Republic. Among the supposed fragments of his writings is the half-line:
While also slotting his words into a metrical template, Ennius playfully employs tmesis to suit the meaning; he splits the word for ‘skull, brain’ (cerebrum) in two.
Another oft-quoted example of tmesis comes from Sextus Pompeius Festus, a Roman linguo-geek living in the 2nd century AD. Festus documents an artifact of Latin that, to his ears at least, by then sounded old-fashioned:
ob vōs sacrō, in quibusdam precātiōnibus est, prō vōs obsecrō, ut sub vōs plācō, prō supplicō”
‘Ob vōs sacrō, in certain prayers, stands for vōs obsecrō, just as sub vōs plācō stands for supplicō’ (Festus’ epitome of De verborum significatione, O)
What Festus is getting at here is that the words obsecrō (meaning ‘I beseech’) and supplicō (also meaning ‘I beseech’) could be interrupted by the object pronoun vōs ‘you’. The resultant phrases were confined, though, to the archaic language of prayers. An English equivalent would be like inserting us into forgive and deliver into the Lord’s Prayer, and praying ‘for-us-give our sins’ and ‘de-us-liver from evil’.
Festus, a darling of historical linguists today, gives two further examples. This time, they come from the foundational laws laid down in Rome’s early years:
ut in lēgibus: trānsque datō, et endoque plōrātō
‘As in the laws: trānsque datō (‘and he shall surrender’) and endoque plōrātō (‘and he shall implore’)’ (Festus’ epitome of De verborum significatione, S)
The verbs for ‘surrender’ and ‘implore’ are cut in twain by the presence of -que, meaning ‘and’. As with religion, legal language is quite used to preserving linguistic archaisms.
There is a notable difference between Ennius’ assault on cerebrum and Festus’ remembered formulations like ob vōs sacrō. In the second, tmesis is carving at the joints; it makes a clean break within obsecrō, namely between its initial prefix (ob) and the root verb (sacrō). The break is at a morphological boundary within the word.
That break is present within hundreds (thousands?) of English words, as part of its Germanic and Latinate inheritances, although prefixes in English differ as to how lively and malleable they are. We can wield the prefix re– (redo, rewatch, reorder) far more dextrously than the prefix ob- (obvious, oblige, obfuscate).
The same break cannot be seen in cerebrum. Tmesis of such an intact word would’ve felt as strange as dividing English skull or brain and positioning the parts across your line of poetry.
Re-cognising
Now, here’s the thing: the ob vōs sacrō pattern, in which a prefix and its verb bookend their clause, pops up across the Indo-European family, far beyond Latin.
Take its cousin, Ancient Greek. Its greatest representative, the poet Homer, attests an employment of this pattern that is much livelier than fossilised fragments. Separating a single prefix and putting it at the beginning seems to have been a valid grammatical option for Homer (whoever he/she/they was/were).
‘For Patroclus had sent fear into them all’ (Iliad 16.291)
ἐν δ᾽ ἐρέτας ἐπιτηδὲς ἀγείρομεν, ἐς δ᾽ ἑκατόμβην θείομεν en d᾽ erétas epitēdès ageíromen, es d᾽ hekatómbēn theíomen
‘Into it let us gather sufficient rowers, and onto it let us put a hecatomb’ (Iliad 1.142–3)
The poet here has cut up verbs like ἐνίημι (eníēmi) ‘to send in’ and ἐναγείρω (enageírō) ‘to gather in’, and shoved their two components to the edges of the clause. While acknowledging the many inevitable exceptions and variants, this ‘P … V’ pattern exemplified by Festus and Homer is common enough to be noteworthy.
Homer crafted his verses according to a strict metrical template, yes, but the frequency of tmesis and the general principle that his works couldn’t be unintelligible gibberish together indicate that this was a genuine part of general Greek grammar. It was at least grammatically acceptable, perhaps sometimes preferable, in everyday speech and among the wider population of speakers.
If so, then it wasn’t to last. Very few post-Homeric examples of this pattern can be found.
κατά με φόνιος Ἀΐδας ἕλοι katá me phónios Aḯdas héloi
‘May murderous Hades take me’ (Sophocles’ Oedipus at Colonus 1688–9)
Besides, Homer already cast a long shadow that we cannot remove the possibility that Sophocles here was simply imitating the epic poet. From the classical era onwards, we don’t find this pattern, and Greek prefixes remain firmly glued to Greek verbs.
The general trend of this kind of tmesis is therefore a decline, although Homer also appears to be its chronological peak. Curiously, Myceanean Greek (the language behind Linear B, spoken centuries before Homer) to date displays no sure instances of the pattern.
In later alphabetic Greek writing, this would be rendered ἐπιδέδασται (epidédastai). Such Mycenaean examples from the second millennium BC only show prefixes like ἐπι- immediately preceding their verbal colleague. They form what we recognise as ‘compound’ verbs.
Our sources are Mycenaean Greek are few in number, inviting the speculation that we simply haven’t yet dug up its examples of the ob vōs sacrō pattern. Nonetheless, what we do have is so consistent in not attesting it. On this basis, it has been suggested that Mycenaean Greek was innovative and ahead of the grammatical curve when it came to compound verbs, compared with the later-but-conservative Ionian dialect of Homer.
Over in India and Central Asia, Vedic Sanskrit, the language of the Vedas and a cousin of Latin and Greek, makes great use of prefix-like elements that are flexible in their position relative to the verb. A compound verb in Vedic Sanskrit may be formed with more than one such element (I believe two is the recorded record in Vedic, but four in Classical Sanskrit). The default position is for that item to stand right before the verb, but there is also a noticeable tendency to place it away, at the very beginning of their common clause.
‘The quick messenger who ran forth to you, O Mitra and Varuna’ (Rigveda 8.101.3)
वि (ví) ‘apart, away’, प्र (prá) ‘forward’, परि॑ (pári) ‘around’, नि (ní) ‘down’, परा (párā) ‘away, along’ and others contribute to the meaning of a Vedic verb, often simply by adding the action’s direction. The प्र and अद्र॑वत् in the second example here can together be translated as ‘ran forwards’ or more literally ‘forwards … ran’, but new senses can emerge from the compounding. If two prefix-like elements are present, only one is allowed in this remote position.
‘Around and forwards run well, for winning the prize’ (Rigveda 9.110.1)
Compared with Latin’s fragments and Homeric Greek, the Rigveda exhibits a great deal more arrangements of the ob vōs sacrō pattern.
Vedic Sanskrit’s sister language is Avestan, the oldest representative of the Iranian branch of Indo-European. Its ancient scriptures exhibit a few instances of the ob vōs sacrō pattern, such as with 𐬁 (ā), meaning ‘at, to’. Here, it stands alone and first:
ā-mо̄i rafəδrāi zauuə̄ṇg jasatā
‘Here come to my calls for support’ (Yasna 28.3)
Other venerable great-aunts of the Indo-European family display this preference for preposed prefixes, with Hittite among them. But even centuries later, in the Late Antique/Early Medieval European era, we find tmesis.
For example, in fourth-century Gothic, the oldest thoroughly documented Germanic language, a few instances can be found:
𐍃𐌰𐌴𐌹 𐌲𐌰𐌻𐌰𐌿𐌱𐌴𐌹𐌸 𐌳𐌿 𐌼𐌹𐍃 𐌸𐌰𐌿𐌷 𐌲𐌰-𐌱𐌰-𐌳𐌰𐌿𐌸𐌽𐌹𐌸 𐌻𐌹𐌱𐌰𐌹𐌳 saei galaubeiþ du mis þauh ga-ba-dauþniþ libaid
‘He who believes in me, though he dies, will yet live’ (John 11:25)
In these Biblical lines, we see three compound verbs, 𐌲𐌰𐍃𐌰𐌹𐍈𐌰𐌽 (gasaiƕan) ‘to see’, 𐌳𐌹𐍃𐍃𐌹𐍄𐌰𐌽 (dissitan) ‘to seize’ and 𐌲𐌰𐌳𐌰𐌿𐌸𐌽𐌰𐌽 (gadauþnan) ‘to die’, broken up by intervening material. Unlike in Homeric and Vedic verse, this only involves short functional words, like the question particle 𐌿 (u) and the indefinite pronoun 𐍈𐌰 (ƕa), related to English what.
I will offer only these as relevant examples from the Germanic branch, tactfully sidestepping the question of whether the modern phenomenon of phrasal verbs in Germanic (e.g. English lift off, bring on, think up, dress down) is in some way connected to ancient tmesis. I don’t think so, but don’t feel qualified to state that definitively.
Then, finally, we turn to my beloved Old Irish. This medieval language, the ancestor of Modern Irish, Scots Gaelic and Manx, does incredible things with its verbs. Among them, we have a very minor pattern that again sees a prefix and its verb stand at a distance.
fordon itge Brigte bet
‘May Brigid’s prayers be upon us’ (Thesaurus Palaeohibernicus II.348.4)
ad cruth caín cichither
‘Fair form will be seen’ (Serglige Con Culainn 694, oft quoted in the literature, including by me, but I now worry it’s a ghost quote)
Once again, we see prefix-like particles (here ad and for) standing removed from the verb. The position of that particle is not the oddity for a change; what’s unusual for Old Irish is that the verb is in the final position. The standard pattern is for the verb to come early on, either first or immediately following an element like ad.
ad·cither a-suthine-som
‘Their eternity is seen’ (Würzburg glosses 1b15)
As with Vedic Sanskrit, these prefixish things are numerous in Old Irish (a stack of several is common), and often their presence is obligatory. The compound verb in the first and third examples, meaning ‘to see’, never appears without the added ad.
Across the branches of Indo-European, we unearth historical usages of the pattern seen in Latin ob vōs sacrō: a prefix-like element, then other material, then the verb.
I have to say “prefix-like”, because really, they’re not prefixes as we know them. To be a prefix implies being fixed – unable to be moved and to function as a standalone word. Our accumulated evidence suggests that this was not the case for prefixes across the oldest Indo-European languages. Instead, they were once all free and independent.
Although today we make a terminological distinction in English between prefixes (e.g. understand, forgo) and prepositions (e.g. under the table, for the book), the members of the two groups tend to look very similar. This is indicative of their original status; in the earliest days of Indo-European, there was no such distinction. All such directional and relational words worked like adverbs. They were accented and were not bound to one position relative to a noun or a verb. Instead, they independently contributed their own particular meaning to a noun, a verb or the sentence as a whole.
This situation endured for centuries and passed into the emerging Indo-European languages; we see their old flexibility most clearly in ancient members like Vedic Sanskrit and Ancient Greek. On this basis, ‘preverb’ or ‘local particle’ are alternative terms used, avoiding the dependent quality of prefix. Out of habit, I will now switch to preverb, although it’s hardly a perfect term. The pre- bit of preverb is based on the words’ typical position before the verb that they cooperate with – typical, but not exception-less.
Over and over, though, we see the old adverbs fusing with verbs into single words (‘univerbating’), becoming prefixes as we today would recognise them. When the Indo-European languages each enter the historical record, the ob vōs sacrō pattern is already rare or not to last. We see it most clearly in the Greek, Sanskrit and Hittite record; in Latin, Gothic and Old Irish, prefixes are predominant and tmesis is a fossil.
From this historical perspective, the term tmesis is actually a misnomer. The opposite process is in fact true; examples like ob vōs sacrō aren’t instances of compound verbs cut up, but rather preserve an older grammar in which prefixes and verbs haven’t yet been stitched together.
Ex-plaining
I can now get to the thing that haunted my otherwise serene PhD days.
What we have with the examples given is evidence for syntactically independent and moveable preverbs. Within that independence, there is a significant tendency for one preverb to appear first in its clause. In the Latin, Ancient Greek, Vedic Sanskrit, Avestan, Gothic and Old Irish examples, the element in question is right at the beginning, preceding all else.
It does so with such frequency and across such a spread of languages that we have the necessary grounds to think that this pattern was part of the word order of their common ancestor. That is, just as we reconstruct its sounds and words, I believe and have argued that we might reconstruct such arrangements of words back to Proto-Indo-European itself.
Put another way, the claim would be that Proto-Indo-European speakers, six-thousand-ish years ago, had a syntactic ‘habit’ of shifting a single one of these directional words (equivalent to English towards, down, out, away)to the outset of their sentences. This we can call ‘preverb fronting’. It endured into several of their prehistoric language’s many offspring, but ultimately gave way to bolted-on prefixes. From its historically documented descendants, we gain indirect evidence for the original habit – note that I can’t formulate it as an obligatory rule.
Any attempt at the reconstruction of prehistoric syntax is fraught with theoretical difficulties. Every piece of historical support for it must be critically examined, to determine whether tmesis/preverb fronting really was an archaic element of their grammar.
An alternative explanation could be that it’s due to metrical reasons and to composers mucking around and making awkward words fit – a possible factor for Homeric Greek and Vedic Sanskrit. Another alternative is interference from and imitation of other languages – a possible factor for Gothic (its Bible being a translation from Greek) and Old Irish (its scholars being steeped in Latin). The business of historical syntax must consider all other potential factors for all relevant examples before proposing that some word-order oddity in our historical texts really ‘was syntactic’ in its cause. A familiar old headache has returned for me.
Nonetheless, preverb fronting avoids these pitfalls. It’s hard to dismiss all the occurrences all as poetic artifice or interference. En masse, a shared syntactic inheritance seems more plausible. It’s too common in Homeric and Vedic not to have been a viable option in the grammar of the community at large, beyond a handful of wordsmiths. Also, the two Gothic examples above do not mirror the Greek originals, and I look very sceptically on the claim that the few instances of Old Irish tmesis were modelled on Latin. Tmesis in Latin was fragmentary, hardly a feature worth copying.
As an aside, I still suspect the opposite was actually true: that preverb fronting was so common in an earlier stage of Old Irish that it was wholly or partly responsible for the standard grammar with the verb placed early on. This ‘Verb Subject Object’ order is now typical of the Celtic languages. In my preferred model, preverbs were so thoroughly fronted at first that eventually their verbs moved to join them.
So, having seen off the theoretical competition, the coast seems clear for us to claim that preverb fronting was indeed a feature of Proto-Indo-European word order. From this prehistoric popularity, it was passed on and ‘inherited’ by later speakers of Latin, Ancient Greek and the rest. This raises the question: what was it for? What was the linguistic benefit of putting these particles first?
I don’t know!
Having proposed it for Proto-Indo-European, I’ve struggled to state why it existed as a pattern within that language’s syntax. This was the question behind my pleading “any answers?” slide. I don’t lack possible explanations; there are several, but individually or collectively they don’t pass muster.
One is to satisfy ‘Wackernagel’s law’, the well-established observation that early Indo-European languages had groups of words that could not stand first in their clause. Specifically, these were object pronouns (meaning ‘me’, ‘you’, ‘us’, ‘something’) and conjunctions (‘and’, ‘but’, ‘for’). It’s generally agreed that they were in some way phonologically deficient, slightly or wholly unaccented, and therefore needed to tag onto the end of another word.
Something – anything – else had to precede such little words, so long as it was a ‘proper’ (i.e. accented) word, which could offer a host that for these clitics to lean on. The phenomenon was first noted for Sanskrit by Berthold Delbrück, then by Jacob Wackernagel across Indo-European more generally. Wackernagelian clauses are readily present in Vedic Sanskrit and Ancient Greek. See below how मा॒ता (mātā́) ‘mother’ and καί (kaí) ‘also’ stand as host to two clitics that follow them:
मा॒ता च॑ मे छदयथः स॒मा mātā́ ca me chadayathaḥ samā́
‘And you and my mother appear alike’ (Literally: ‘mother and of-me appear alike’ (Rigveda 8.1.6)
‘And for you too, god-like Achilles, is the fate to be destroyed under the wall of the rich Trojans’ (Iliad 23:80–81)
The rule is vaguely comparable with the extra do that n’t needs in English negative sentences like I don’t know. As for my preverbs problem, this explanation would say that the requirements of clitics triggered the fronting of a preverb to act as their host.
As a devotee of Wackernagel’s law, my regretful response is that this is not sufficient, on the basis of the historical evidence. The correlation between the two is not strong enough. Many of our examples of preverb fronting do indeed appear with one of these troublesome little words, but many do not, such as in the first two Rigveda examples herein. We might also predict fronted preverbs in all instances of their co-appearance with clitics, which we do not find. The prefix ἀπο- (apo-) of ἀπολέσθαι (apolésthai) ‘to be destroyed’ in the Greek example above has not detached itself and run to start of the sentence to be the host for its two clitics.
Moreover, it does not explain the perceived positional preferences of preverbs: either right next to the verb at the end, or one alone at the beginning. Nor does it explain itself: why shift a distant preverb to meet the Wackernagelian requirement? Why not simply front the next available word? A variety of hosts is indeed what we see.
Moving away from any obligatory syntactic rule, another explanation is stylistic: that this construction achieved some rhetorical or performative quality, something beneficial in narration but not obligatory. Such was the view of Louis Renou for the construction in Vedic Sanskrit:
“By highlighting the preverb, it has a privileged place in hymns that consist mainly of invocations … The separation is most prominent in stanzas of calling to prayer, where it was customary to mark, by the preverb, the invocation, the initial momentum whose content the rest of the sentence develops.”
(Renou 1933, translation my own)
This sounds good, but doesn’t (yet) convince me. What did a priest or storyteller really gain through a fronted preverb? Why not some other category of word? Far be it from me to move the goalposts and critique a renowned scholar, but this explanation doesn’t dig down to the root of the phenomenon. It does feel like it’s on the right lines, though, and I speculate about some ‘bookending’ rhetorical technique; by placing the two parts of the same verb at either end, it unified a clause for the audience.
A third explanation is simple emphasis. This view would consider the fronted preverb to be important and accordingly emphasised through its initial position, and perhaps heightened volume too. I’m not at all convinced by this. While I have no direct access to spoken Proto-Indo-European, my assessment is that the initial position of its clauses was not characterised by emphasis. Anything put there was not new and important – quite the opposite. It was topical information, already known to all parties in the conversation.
The abstract structure of the Proto-Indo-European clause that my thesis proposed begins with two slots:
[ topic ] [ clause ] … other stuff … [ verb ]
Into the first, speakers put their topical material, that which was familiar and linked the new clause back to things previously mentioned. Into the second went the words that signalled what type of clause it was. If it was a question, you put the question word there; if it was a relative clause, you put the relative word (i.e. ‘which/that/who’) there, for example the यो (yó) ‘who’ in the second Vedic example above.
This structure captures and explains a lot of the data; it works well as the prehistoric starting point for the word-order behaviour that we see in the descendants of Proto-Indo-European. But there remained for me an annoying fly in the ointment: fronted preverbs. There is nothing obviously topical about them (and always just one of them) that would see them suitably slotted into my abstract [ topic ] position.
Con-cluding
I remain open to the possibility that my proposal is flawed and ought to be redesigned or binned. Fronted preverbs may just be the card that collapses the tower. But that would at least depend on us identifying what exactly this word order was for.
After years of frustratingly circular thoughts and fun-but-fruitless chats with linguistic experts down a number of avenues, a convincing explanation so far defeated me. Even writing this enjoyably long post has not provoked new insights. Whether I will be the one to explain it seems less likely with each passing year. I’ve felt for so long that it’s something obvious, something exhibited by languages living today that I’ve missed or am yet to learn about.
Danesi, S. (2013). Particle-verb constructions in Vedic: The case of Ápa. Studi e saggi linguistici, 51(2). 57–100.
Hale, M. (1993). Tmesis and movement in Avestan. Indo-Iranian Journal, 36(1), 29–43.
Giannakis, G. K. (2023). At the crossroads of Linguistics and Philology: The Tmesis-to-Univerbation process in Ancient Greek. In: Giannakis, G. K., Filos, P., Güemes, E. C., & de la Villa, J. (Eds.). Classical philology and linguistics: old themes and new perspectives, Vol. 1. Walter de Gruyter GmbH & Co KG. 175–211.
Jiménez Delgado, J. M. (2023). Mycenaean Greek syntax: a diachronic perspective. In Aura Jorro, F., Del Freo, M., & Piquero, J. (Eds.)The legacy of Michael Ventris: progress and perspectives in the field of Aegean scripts and Mycenaean studies. Consiglio Nazionale delle Ricerche. 161–187.
Morpurgo Davies, A. (1981). Mycenaean and Greek Prepositions: o-pi, e-pi, etc. In Heubeck A, & Neumann, G. (Eds.) Res Mycenaeae. Akten des VII. Internationalen Mykenologischen Colloquiums in Nürnberg vom 6.–10. April 1981. Vandenhoeck & Ruprecht. 287–310.
Ram-Prasad, K. (2018). The Syntax of ‘Local Particles’ in Sanskrit: A Contribution to the Study of Proto-Indo-European Syntax. MA dissertation. University of Cambridge.
Ram-Prasad, K. (2023). Clitics and the Left Periphery in the Sanskrit of the Rigveda. Journal of Historical Syntax, 7(22), 1–53.
Renou, L. (1933), La séparation du préverbe et du verbe en védique. Bulletin de la Société de Linguistique de Paris, 34, 49–96.
Thurneysen, R. (1975). A Grammar of Old Irish. Binchy, D. A., & Bergin, O. (Trans.). Dublin Institute for Advanced Studies.
Watkins, C. (1963). Preliminaries to a historical and comparative analysis of the syntax of the Old Irish verb. Celtica 6, 1–49.
Watkins, C. (1964). Preliminaries to the Reconstruction of Indo-European Sentence Structure. In Lunt, H. G. (Ed.) Proceedings of the Ninth International Congress of Linguists, Cambridge, Mass., August 27–31, 1962. De Gruyter. 1035–1045.
Wright, J. (1910). Grammar of the Gothic Language. Clarendon Press.
Zanchi, C. (2019). Multiple Preverbs in Ancient Indo-European Languages: A Comparative Study on Vedic, Homeric Greek, Old Church Slavic, and Old Irish. Narr Francke Attempto Verlag.
Images my own or from Wikimedia. Cover image: flint blade, housed in Egypt’s Metropolitan Museum of Art, from the Naqada culture of Predynastic Egypt, 3650–3300 BC (so maybe a cutting blade contemporaneous with Proto-Indo-European).
One thought on “Ex- a Weird Old Word Order -ploring”
Wow, without your encyclopaedic knowledge of languages, that was a tough one. However, after the second read through I understood much more. I want to be at a dinner party when you are there so I can just shut up and listen to what you say. Thank you again. All best, Mark. ps, no answers, obviously!
Danny L. Bate 
Wow, without your encyclopaedic knowledge of languages, that was a tough one. However, after the second read through I understood much more. I want to be at a dinner party when you are there so I can just shut up and listen to what you say. Thank you again. All best, Mark. ps, no answers, obviously!
Sent from my iPad
>
LikeLiked by 1 person