The impossibility of Artorius

Diachronic phonology doesn't care about your feelings, part n

A comment chain on my last Arthur-related post draws my attention to a philological argument for Arthur < Artorius. It’s quite simple: long ō before r gives u in Welsh borrowings from Latin.

I’d be willing to accept this if it wasn’t for the fact that:

1. It isn’t actually true.

2. Artorius doesn’t have a long ō.

We’ll come back to this later.

Trying to get to the bottom of this (it’s quoted uncritically in many places with the reckless abandon only afforded to etymologies the author wants to be true) took some time. But I’ve now ironed it out, along with some kinks in parts 1 and 2. We’ll explore those later on.

The whole issue is another example of my favourite pet hate: bad historical chronology. I’ve griped before that linguists often have a very poor sense of relative time. Historians are even worse, and don’t appear to be able to appreciate that events must by necessity happen in a certain order.

With this in mind, I can now explain very exactly and precisely why the name Arthur can’t derive from the Roman nomen gentili Artorius. This post is going to be extremely boring, so I hope you’re prepared.

Nō, ū

Long story short, ō > ū isn’t a conditional change in Welsh borrowings from Latin words. It’s a regular sound change in early Common Brittonic which occurred in all positions as part of the early restructuring of the Proto-Celtic vowel system.

Long story long: there are about half a dozen words where we can see a hypothetical change ō > u in Welsh borrowings from Latin. Here’s a selection:

dolur < dolōr- (borrowed from the oblique stem, nominative is dolor)

llafur < labōr- (similar deal)

addurn < adōrn-

ffurf < fōrma

llurig < lōrica

However, this isn’t just confined to ō after r as is often cited. Dolur and llafur both have close relatives dolurus and llafurus which are presumably borrowings from their adjectival forms dolōrōsus and labōriōsus. Labōriōsus at least would seem to prove that proximity to r isn’t necessary to trigger the change. So let’s look at what’s actually going on.

(N.B. I’ve ommitted the oft cited awdur < au(c)tōr as its form suggests it’s a later “learned” borrowing that’s been transformed by analogy to older forms. A genuine Latin borrowing of this word would have become **ɔdur in Old Welsh and **odur later on).

Proto-Celtic to Proto-Brittonic

Some of the most significant developments from Proto-Celtic to early Brittonic are made in its vowel system. In particular, long vowels undergo the following shifts:

1. *ū merges with *ī in all positions

2. *au, ā (and possibly *eu) merge as ɔː (possibly *au > *ā first)

3. *ai > *ɛ:

4. *oi and *ou > *ō > *ū

These shifts are significant to us because they temporarily restore *ō to early Brittonic. In Proto-Celtic, PIE *ō had already been absorbed into *ā (*ū in final syllables).

This means we have a distinct time window within which words borrowed into Brittonic with a long *ō would naturally see this long *ō raised to *ū.

We can determine exactly when this window would have been with reference to Latin glosses of Celtic words. *ou makes the change first, with references to both Loundinion and Londinion already occurring in the first to second century. All instances of *ō would become *ū by around 300 AD – the change is already visible in names recorded in the Antonine Itinerary made around this time, along with the one which produced ɔː.

This is interesting to me because it reveals that another common Celticist assertion that Latin borrowings in Welsh were all sufficiently early to be carried along by normal Brythonic sound changes isn’t 100% right. There are a few other Latin borrowings where we would expect to see *ō > *ū if this were the case, but we don’t. Here’s a few:

cystrawen < cōnstruenda

cysergu < cōnsecrāre

cyffes < cōnfessiō

cyffaith < cōnfectiō

All of these words suggest an OW and Late Common Brythonic precursor stage with a short *u. In this regard they’re treated no differently from similar words with a short Latin o (e.g. cynnig < condīcere, cyngredawr < congregātiō).

I can only conclude they were borrowed late enough for phonemic vowel length to have collapsed in Vulgar Latin (with allophonic nasalisation before a sequence -s+F- also contributing to the raising of *o > *u). One example which appears to confirm this is cusyll < cōnsilium, which is at least orthographically treated as preserving an OW/CB long *ū, suggesting it was borrowed much earlier.

This little diversion is necessary because it helps us establish 2 things:

1. Many instances of Latin long ō were borrowed into Brythonic at an early enough point to be captured by the general shift *ō > *ū.

2. But not all Latin borrowings occurred in this period.

This is important to us insofar as Artorius is concerned because of the specific chronology of Latin’s development (alongside that of Brittonic) and how the pronunciation of Latin vowels changed over time.

Ō tempora, ō mōrēs

Nothing lasts forever, including the Classical Latin phonemic vowel length opposition. In the dialects that would eventually become Western Italo-Romance (and thus most likely to be similar to the dialect of Vulgar Latin common in Roman Britain), the long/short vowel opposition of the Classical period eventually collapsed into a “quantity system”. For the uninitiated, this means vowels were no longer distinguished based on a one-count or two-count, but as separate sounds of equal length.

Why?

Because in Classical Latin, the monophthong vowels e/ē, i/ī, o/ō, and u/ū were already slightly different from each other. Short vowels were pronounced with a slightly more open quality (so perhaps *ɪ *ɛ *ɔ *ʊ ), long vowels with slightly more close quality (probably *ī *ē *ō *ū).

Changes to how vowels were annunciated relative to syllable accent caused this system to collapse by allophonically lengthening short vowels in open stressed syllables and shortening long vowels in closed stressed syllables, while all non-stressed long vowels became short.

Pay attention: this appears to be the reason why many assume Artorius must have had a long ō vowel. The assumption is that we’re working with a late Latin borrowing which has already experienced this process of change, and through which /ar.tó.ri.us/ became /ar.tō.ri.us/.

Unfortunately, this doesn’t really work. The emergence of the Western Romance quantity system can be fairly securely dated to the 5th century AD.

This is far too late for such a change to be registered in the change from *ō > *ū, which was most likely complete by the first half of the 3rd century. It also fails to take into account a century or more of evolution in Common Brittonic which had caused large shifts in pronunciation of its own native vowel sounds. Moreover, the only secure member of Artoria gens known to us in Britain lived in the 2nd century.

As we’ll see in a moment, the emergence of the quantity system also lead to the resyllabification of many words. This would likely have meant Artorius was taken as /ar.tór.jus/ by this time, and thus immune to allophonic lengthening.

The short conclusion to all this, I’m afraid, is “Artoriuscels btfo”. But let’s allow ourselves to get even more boring.

Let’s do some reconstructions

Artorius in CL is a fairly bog-standard word: we’d expect it to be syllabified as ar.tó.ri.us, accent on the antepenultimate -tó-. Borrowed in its classical form into the Celtic of say the late 1st/ early 2nd to 4th century (along with the lion’s share of other Latin loans to Brittonic) we’d expect it to develop one of 2 ways:

1. *Artórius

2. *Artorjos (this could either be a native Celtic resyllabification or, depending on the period, a direct latin borrowing, as an unstressed i in hiatus became -j- at an unclear point in the history of VL)

3. *Artorjəs or Artorjəh (final syllables were weakened very early, but weren’t finally lost until the middle of the 6th century, following final i-affection)

4. *Arteirjəs or *Arteirjəh (final i-affection, tail end of the 5th century)

5. *Artheir or *Arthyr (spirantalisation – compare Ambrosius yielding Middle Welsh Emreis and Emrys as a result of i-affection)

6. **Arthair or **Arthyr in Modern Welsh

Borrowed in the 5th century, Artorius could actually have been taken in one of 2 ways: *Artɔ:rius, with an allophonically long open-mid rounded back vowel, or as *Artɔrjus, in which short *ɔ from latin o remains short because of the resyllabification caused by the change of i > *j. In my mind, the second of these is more likely as the i > *j change predates the emergence of the quantity system, but it doesn’t really matter as a 5th century Common Brittonic speaker would most likely have taken them the same way, as *ɔ: existed in CB by this point (as a result of the merger of *au and *ā mentioned above) but *ɔ didn’t.

Both *Artɔrjus and *Artɔ:rius would resolve as *Artɔ:r in CB **Arthaur in Old Welsh (*ɔ: > *ɔ with the emergence of the Brittonic new quantity system around 600 AD, but becomes allophonically long in a stressed position with the loss of final syllables and thus diphthongises to *au). By Modern Welsh, it would likely have reduced to Artor again.

This doesn’t work for the Artorius supporters for reasons we’re now in a position to explore. Let’s get even more boring.

Aneirineurysm

Perhaps the earliest attestation of Arthur in literature comes from Y Gododdin, in which he’s mentioned as a superior fighter to one of the 300 warriors at Catraeth.

This specific warrior is the consonantally-profuse Gwawrddur (I presume a reflex of Proto-Celtic *Wāridūros, perhaps meaning something like “strong dawn” or “distant east”). In this mention Arthur and Gwawrddur are rhymed with each other. In all other early surviving Welsh literature, Arthur is also rhymed with words containing this vowel, which must be OW *ʉ:.

OW ʉ is descended from CB ū from PC *ou and *oi, sometimes via an intermediary stage ō. This *ū becomes ʉ: at some point during the mid 5th to mid 6th century AD, before being reduced to ʉ in opposition to the reflex of CB short *u with the emergence of the Brittonic quantity system around 600. It becomes long again in stressed final syllables prior to accent retraction in the 11th century.

If that hasn’t put you to sleep you’ll probably have figured out that for Arthur (/Arthʉ:r/) to rhyme with Gwawrddur (/Gwaurðʉ:r/), it must necessarily be descended from a word which contained a long *ū in the Common Brittonic stage. Which none of our worked examples of the evolution of Artorius do.

At this point, I think we’re well and truly done. I hope I’ve demonstrated there is no way, given what we know of both Celtic and Latin historical chronology, to derive Arthur from Artorius at any period of borrowing.

Where does this leave us?

By a very roundabout route, it leaves us back where we started. Before finishing though I’d like to correct some of my own errors and also revisit the Arthwys question with slightly clearer eyes.

We’re going to need to descend into another, even more boring layer at this point and consider Old Irish.

I was slightly wrong about Artúr

Short version: Artúr, genitive Artuir, must derive from Primitive Irish *Arturos via *Arturah. It behaves as a predictable reflex of an Old Irish o-stem in this regard (with a raised u and slender r in the genitive reflecting an earlier vanished *-i).

This means it must have been borrowed directly from a Latinised form, Arturus. It can’t reflect a borrowing from Artorius.

In its classical form would have been taken as *Artorjos in Primitive Irish and continued as an io-stem probably something like **Artorj, **Arturi where rj is my attempt to convey a palatalised r. In its late Latin form, *Artɔrjus would likely end up something like **Artuar, genitive **Artuari, and would have continued as such in Middle Irish.

Artúr also can’t be considered a borrowing of some stage of Arthwys (which reflects late CB and Primitive Welsh *Artʉ:is or possibly *Artwɨ:s, see below). This would’ve made its way into Primitive Irish as either *Artwis and become OI **Artj (palatalised t) or as the earlier form *Artowissus/*Artowissos, which would have become something like **Artoi(s) or **Artui(s).

I was also slightly wrong about Arthwys

I got my change order backwards. It would have developed into *Artwɨ:s (possibly, see below) by the mid-6th century. Stem vowels of the first element of compound names syncopated in the middle of the 6th century. *rt spirantalised to *rth some time closer to the beginning of the 7th. This gives us a progression thus (I’m using *-us as the base but the same rules also apply to *-os):

1. *Arxtowidtus (earliest Proto-Celtic)

2. *Artowistus (early Proto-Celtic cluster simplification, potentially, although see below)

3. *Artowissus (Proto-Brittonic, possibly, although again see below)

4. *Artowissəs (weakening of vowels in final syllables beginning mid-late 5th century)

5. *Artowíssəh (stress retracts to penultimate syllable concurrently with weakening of final syllables)

6. *Artowīs (final syllables lost, mid-late 6th century, stressed short vowels become allophonically lengthened)

7. *Artwɨ:s (new quantity system phonemicises *i > *ɨ beginning around 600 AD, though see below for the development of *wi)

Anyway…

This leaves us in the following position:

1. We have an Irish name Artúr, derived from pseudo-Latin Arturus

2. This pseudo-Latin name – based on the genealogies of its earliest bearers – was likely adopted from the Britons.

3. It cannot reasonably be said to originate with Artorius.

4. But we do have an early king in the North called Arthwys…

5. … who just happens to have lived around the right time to be the namesake of one of the earliest recorded Artúrs.

The circumstantial evidence that Art(h)urus is simply a Latinized or “higher register” version of early Welsh **Art(h)wɨ:s is already strong. But we do have two more points in its favour.

ʉ wot m8

OW ʉ and ʉ: are usually derived from CB *ū, as discussed at torturous length above. They’re “close central rounded” vowels, distinguished allophonically by length according to stress before being phonemicised by 11th century accent retraction. ɨ and ɨ: are its unrounded equivalents. Both fall together as ɨ, ɨ: in Middle Welsh and beyond.

However, Arthwys is a phonologically unusual name: the sequence -wy- in Welsh, (/wɨ(:)/ or /wi(:)/ today depending on dialect and historical stress) generally evolves from CB *ē, which diphthongised to uɨ in the second half of the 7th century. In Arthwys however it derives from *-wi- via *-wɨ:-. In the cognate gwŷs/gwys this has clearly been understood as consonantal w (strengthened to *gw- initially) + short ɨ, lengthened in a monosyllable due to stress.

In Arthwys, it eventually settles on a form that looks exactly as it would if it had developed naturally from *ē. However, it appears to have been tackled in a number of different ways by contemporary speakers. It looks like it might have originally been taken as Artʉɨs or Artʉ:s if spelling variants are anything to go by – perhaps indicating an allophonic rounding of ɨ: that eventually assimilated back to -wy- by analogy to other words containing this sound. This would explain /Arthʉ:r/ in Y Gododdin at any rate.

You still need to explain that r tho

I’m just getting to that.

As I briefly touched on in my original post, the notion that CB Artʉ:s could have made its way into latin as Arturus is by no means farfetched. A sequence of VsV in Latin is highly unusual thanks to pervasive rhotacism. This change began in intervocalic sequences but was eventually lexicalised throughout most of the language via analogy. In fact, by the classical period it became so ingrained that VsV as a sequence of sounds only occurs as a result of the reduction of -ss- following a long vowel, a handful of loanwords (mostly foreign toponyms), and as the result of the assibilation of a dental in some participle forms.

In fact, the sequence -usu- in particular (As in a hypothetical Artusus) is particularly rare even in this group. It occurs almost exclusively in the perfect passive participle of a handful of verbs with stems in -t and -d (e.g. abstrūsus < abstrūdō, obtūsus < obdtundō, confūsus < confundō).

To the anxious ears of Sub-Roman British aristocrats, a name like Artusus would have sounded rustic in the extreme, and there would have been more than enough impetus for a helpful scribe to elevate it with such a small change.

Note: there’s another possibility to explore here another time – *Artowissos/us would ultimately derive from an even early Celtic stem *Artowidtos/us. The reflexes of this unusual *-dt- are by no means stable in Brittonic: while both Britonnic and Goidelic generally land on *-ss- (reducing to s most of the time), this is by no means a regular change.

Part of me wonders if *dt might not have originally passed through an intermediate stage *dz or *dr, which would have become *jr in some positions and been restored to *ss or even *st in others, explaining its occasional survivals. This would have eventually given us a (dialectical?) *Artwɨ:r/*Artʉɨr by the back door. This is an especially attractive possibility if we consider that the original Arthur likely wasn’t an early “Welsh” name, but a Cumbric one.

The apparent Pictish “version” of Arthur, Gartnait, is instructive here. While Brittonic strengthened initial *w to *gw, presumably. by first velarising w until it developed a fricative quality *ɣw, it appears that this was occasionally registered in the second elements of compounds as well, but eventually dropped. It’s possible that *ɣw originally developed in all positions before being strengthened to *gw- initially while *w or *ɣw was analysed as a lenited variant and eventually weakened back to *w.

There’s some evidence that Pictish did the same thing with initial *a (although w in Pictish names doesn’t appear to have velarised). The n in Gartnait might be explicable as an outcome of a borrowed velarised *w in non-initial position, with the final -t reflecting a divergent resolution of Proto-Celtic *dt.

But this post is boring enough as it is and I’m not currently in a position to gather up all the examples.

Where does this get us?

I don’t expect to change any minds with any of this. Theories surrounding legendary men of power tend to take on special emotional significance for their adherents. They’re part of how we relate to these kinds of stories.

And that’s fine.

What motivated me to do this was a desire to sniff-test the ideas of a dear friend, and also to satisfy an irritation caused by a particularly enduring game of academic telephone.

Historical linguistics is a science, and we should treat it like one. It doesn’t do anyone any good to simply repeat ideas that haven’t been subject to real critical assessment. Regardless of how much we do or do not like them.

With that in mind, here’s the conclusions we can draw based on the available evidence:

1. On philological grounds we have to reject the idea that Arthur derives from Artorius. I’m sorry, but it simply isn’t possible to get from one to the other.

2. This is backed up by historical attestations of gens Artoria in Britain. Its only known member had died hundreds of years before any version of the nam Arthur is known (and Arthur in any case is never latinised as Artorius).

3. While we lack a “smoking gun” proof in the form or an early Artus or Artusus, there’s ample circumstantial evidence that Arthur arises as a variant of the name that eventually becomes Arthwys – whether by Latinising tendencies, back-borrowing from Irish, or Aneirin changing the last -s to -r to get his rhyme to work.

See, I told you this would be boring.