The Heroic Age, Issue 1, Spring/Summer 1999

In Gildas's De excidio Britanniae, he says the spiritual life of Britain had suffered because the partition (divortium) of the country was preventing the citizens (cives) from worshipping at the shrines of the martyrs.¹ He names three such martyrs, all of whom he thought had been put to death during the Diocletian persecution (in effect in Britain from A.D. 303 to 305), although he is hesitant about the dating. Gildas can be very inaccurate about events occurring more than a century before his own time, and in this case his hesitation seems well founded: It is likely that all three died earlier and not necessarily at the same time (Frere 1987:321). The martyrs were St Alban, whose shrine was at Verulamium, and St Aaron and St Julius, whose shrine was at the city of the legions (urbs legionum). Strictly speaking, Gildas says only that Alban came from Verulamium and that Aaron and Julius were citizens of the city of the legions, but if the shrines had been in different places we can be sure that he would have said so. He was after all concerned not with the geographical origins of his three saints but with access to their relics.²

Bede adds nothing to what Gildas says about Aaron and Julius. After a very full account of St Alban’s martyrdom, his Historia gives the other two only a skimpy little tailpiece, based on the passage in Gildas and echoing its wording (Bede 1.7; Colgrave and Mynors 1991). Bede used the same passage for the entry about Aaron and Julius in his De temporum ratione. This work, in turn, was used by various continental chroniclers and martyrologists and by the compiler of the entry for the two saints in the expanded continental version of Bede’s Martyrology, put together by Florus of Lyon in the ninth century, in which their names immediately follow that of St Alban.³ Taken in isolation, the fact that Florus gave the two saints a feast day (22 June) might suggest that he used an independent tradition that gave a date for their martyrdom, but, since St Alban was already on Florus’s list under that date, it is more likely that he was simply lumping two obscure British saints together with a better-known compatriot. The two saints also seem to have been known in Ireland from an early date. An early commentator, in an Irish martyrology of a generation before Florus, notes against the feast day of a much better-known Aaron (July 1), the brother of the Old Testament patriarch Moses, that there was a St. Julius "i nAlbain".⁴

The identity of Gildas's city of the legions is not immediately obvious. The phrase urbs legionum is not known to have been used by any other writer in Roman Britain either as a place-name in the strict sense or as a recognised description of a place. It is recorded in the Middle Ages, but almost all the known instances seem to derive from Gildas.⁵ The place Gildas intended must be somewhere in the part of Britain that he knew, which certainly stretched from the Straits of Dover in south-east Britain round to Cornwall in the south-west and Gwynedd in north-west Wales but which may have extended beyond that. Even the meaning of the phrase is not quite certain. Depending on the relationship between its two terms, it could mean a legionary camp large enough to be a kind of city or a city with legionary associations. Modern scholarship has taken Gildas to have intended the former but the latter has never been excluded, and we will have to return to it. The former sense would present fewer problems. The part of Britain that Gildas knew contained many legionary encampments, but he knew it well enough to have known the difference between lesser camps and the major permanent fortresses, of which Britain had three. If his urbs legionum was a legionary camp, it must have been one of the three fortresses, despite the lack of evidence that any phrase like his was ever used of any of them (Rivet and Smith 1979:s.v. Deva, Eburacum, Isca, and passim).

Two factors suggest that Gildas meant the fortress at Caerleon-on-Usk. It was near the centre of his area of interest, and there is solid evidence of a shrine dedicated to Sts Aaron and Julius there from the ninth century onwards.⁶ That shrine may even explain why these two obscure and very British saints are unexpectedly invoked among the thousands of glosses in Irish and Latin in the marginalia of the well-known ninth-century manuscript of Priscian's Institutiones Grammaticae at the monastery of St Gallen in Switzerland. One explanation would be that Irish monks taking the manuscript to continental Europe travelled by way of Caerleon and that some event there that we can no longer guess at caused them to write an invocation to the local saints into their book.⁷

In the twelfth century, Geoffrey of Monmouth’s mendacious but hugely influential Historia regum Britanniae made the Caerleon identification part of the accepted history of Britain. Geoffrey mentions the City of the Legions more than once.⁸ (The phrase can be given initial capitals straight away in this case because it is so obviously a place-name.) He says that the City of Legions was on the River Usk, implying that it must have been Caerleon, and in a passage that echoes what Gildas says about the martyrs, he adds that in King Arthur’s day the city had two principal churches, one dedicated to St Aaron and the other to St Julius. Geoffrey knew Caerleon personally, but this looks like a characteristic stroke of his imagination, creating another bit of the millennia of pseudo-history he needed out of the passage in Gildas, the obvious etymology of the place-name Caerleon, and the local cult of the two martyrs. The two churches in particular look like inventions for the sake of his story: The evidence of the cult, such as it is, suggests a single church dedicated to both saints, but Geoffrey wanted a church of canons as the metropolitan church of Wales.⁹ and a church of nuns to which Guenivere could flee from Arthur’s wrath. What Geoffrey said about the two saints found its way into the work of many other mediaeval historians (Tatlock 1950:235), but nothing testifies more strongly to its seductive power than the effect of this part of his book later in his own century on Gerald of Wales. Gerald (Dimock 1861-1891: Vol. 6; Thorpe 1976), too, had seen Caerleon for himself, but, into the lively details of what he saw, he interpolated a description of the two lost churches made up of a mosaic of snippets from Geoffrey’s Historia.¹⁰ He did this despite reporting in the same chapter the well-known tale of how, when a local man afflicted with unclean spirits had St John’s Gospel placed on his lap, the evil spirits all flew away "like so many birds", but, when Geoffrey’s book was put there instead, the demons would return and "alight all over his body, and on the book too."

The post-mediaeval period has seen almost universal agreement that Gildas's urbs legionum must have been Caerleon, although cautious scholars have sometimes noted for the record that he might have meant Chester, whose legionary fortress was at or just beyond the north-east edge of his apparent area of interest. There are one or two passages in the works of later authors that could be taken as supporting Chester, notably one in Bede’s Historia where he describes the place as civitas legionum, a phrase very similar to Gildas's in form and meaning (Colgrave and Mynors 1991, 1:2). We will return to this.

David Dumville has pointed out, however, that taking the city of the legions as Caerleon-on-Usk is at odds with what seems to be implied by the rest of the literary and archaeological evidence for the period (Lapidge and Dumville 1984:78, 82), and much the same is true of Chester. It is difficult to think why "the citizens" should have been unable to make their way to either place when Gildas was writing, unless much more of western England and Wales was then permanently in Anglo-Saxon hands than any historian has suggested. This is a serious matter. Whatever mistakes Gildas may make about the history of Britain in earlier generations, his De excidio, with all its limitations, is much the best historical source for the history of Britain in his own lifetime, and its testimony is in principle capable of outweighing all the rest of the evidence put together. Dumville was right to say, therefore, that this is a problem that urgently demands a solution: If Gildas means what he has been taken to mean, a great deal of fifth and sixth-century British history needs rewriting.

The most recent attempt at the problem ingeniously suggests the citizens' difficulties were with a Saxon settlement in the upper Thames valley (Higham 1994:103-6), but whatever difficulties a settlement in that area might have created for British pilgrims coming to Caerleon from the east, it would not have caused the kind of comprehensive exclusion that Gildas implies from Caerleon, and still less of course from Chester. There is, however, a simple solution: that the urbs legionum is the third major permanent legionary fortress in Britain, York.

That identification, unlike that of Caerleon or Chester, is entirely consistent with the rest of what is believed about the period, although no exact date can be given either for the completion of Gildas’s book or for the time when York fell permanently into Anglo-Saxon hands.¹¹ The time when Gildas wrote has been the subject of a great deal of dispute, but the dates most commonly supported are close to A.D. 540. York fell to the Anglo-Saxons in the course of the westward expansion of a settlement in the east of what was later to become the Kingdom of Deira and, later still, Northumbria. The conquest of York must have been one of the decisive moments in that expansion and might even have been what prompted the expanding community to declare itself a kingdom. If the urbs legionum is York, then the loss of the shrine of Sts Aaron and Julius might have been very recent when Gildas wrote and all the more bitter for that.

It may help to test this hypothesis if we consider how Gildas expected his readers to recognise his "city of the legions". He might have assumed they would know where the shrine of Sts Aaron and Julius was; if so, his words would mean "that particular legionary city (of the three possible candidates) where you know these martyrs are buried." Yet it would be simpler, more normal, and sustain a parallel (Gildas liked parallels) with what he appears to be saying about Verulamium if he were offering his "city of the legions" as a place better-known than its shrine, so that those readers who did not know where the shrine was could locate it from the city.¹² If so, urbs legionum will be the equivalent of a name, a reasonably well-known and unambiguous descriptive phrase. Such a phrase should fit only one place, and, therefore, if it means what it has always been taken to mean, one and only one of Britain's three legionary fortresses. Surprising though this may seem, it does so; and the place that it fits is York.

We may take the two words in Gildas's phrase separately. The first, urbs, is a civil, rather than a military, term, and York, in addition to being the military headquarters for northern Britain, was a Roman colonia and, therefore, in Roman administrative terms, a city. Moreover, it was no ordinary city but a major administrative centre: Since the Romans divided Britain into two provinces in the first century A.D., York had been the administrative capital of the north and so the most important city in Britain after London (Mann 1961:316-20; cf. Frere 1987:242-43). It could even be called a metropolitan city, both because it was the chief city of its province and also in something of the modern sense as well. Also, in the late empire, urbs was sometimes used to describe cities that fell into one or both of those categories, in the latter case by extension from a secondary sense of urbs as the city (i.e., Rome; Lewis and Short 1886).¹³ When a writer of the period wanted to speak of an ordinary city as opposed to a major city, he would be likely, particularly if his Latin was good (and Gildas’s was excellent), to choose a word with different connotations from urbs. The word civitas was available to describe ordinary, as opposed to major, cities, although it could also be used to describe all cities, both major and ordinary together.

There is no evidence that either word was applied in any sense to Chester or Caerleon under the Roman empire. Neither was a colonia nor a municipium nor the chief city of a province (however insignificant or short-lived) nor even of a tribe (Stevens 1937:202-3). Their modern names, which are both cognate with castra legionis ("the camp of the legion"), suggest that they were thought of primarily as military centres,¹⁴ and there is some evidence to suggest that castra and civitas (in its inclusive sense) were alternative categories in Britain during the final years of the empire.¹⁵

In Gildas's initial description of Britain, he claimed that it contained twenty-eight cities, which he called civitates.¹⁶ He will have been following an official list of some kind, which will have used civitas inclusively. York would certainly have been on any list of that kind: There is nothing to suggest that either Chester or Caerleon was.¹⁷ Elsewhere, however, Gildas seems to observe some kind of distinction between civitas as any city and urbs as a great city, although it is difficult to be sure because on many occasions his usage looks as if it may be constrained by a literary or experiential context.¹⁸ Almost three-quarters of his uses, for instance, quote or paraphrase the Bible. Two instances of urbs, the passage under consideration here and a reference to Aquileia as the place where Magnus Maximus was executed, seem likely to reflect established late Roman usage, though there is no reason to suppose that Gildas would have disagreed with it. (Aquileia, according to Ausonius (Green 1992:171),¹⁹ was the ninth most important city in the world.) It is reasonable to suppose, therefore, that Gildas's urbs legionum embodies a recognised Romano-British distinction between York as the legionary city in the full sense and Chester and Caerleon, possibly as lesser cities (civitates rather than urbes) but more probably as mere fortresses (castra).

Under the late empire, when ecclesiastical organisation followed the structure of imperial administration, York's civil status made it an episcopal see. In 314 its bishop attended the Council of Arles, in whose records he was named first among the British delegates (Haddan and Stubbs 1869-78, 1:6-7).²⁰ That suggests York was the most important episcopal see in Britain at the time, although that status may have been only a temporary consequence of the favour of the Emperor Constantine, who had been proclaimed emperor there eight years before. Any bishop's see, however, must have had places of Christian worship, where the cult of martyrs would naturally have been fostered, even when, as at York, no particular dedications are known. It would be entirely natural if, as the civil structure of the empire disintegrated further and further, an awareness of York's religious status reinforced or even determined the use of the word urbs in a writer like Gildas, to whom religion was even more important than Romanitas. There is no direct evidence of bishops or shrines in Chester or Caerleon before the ninth century.²¹

Gildas's second word, legionum, also points to York, although less decisively. The word is plural, and York could claim to be a base for more than one legion in a way that Caerleon and Chester could not. York alone, after Roman military dispositions had settled down at the end of the first century, had had two legions based in it (IX and VI, successively), and it had also, as was natural for the legionary fortress nearest to the threat from beyond Hadrian's Wall, been most favoured with flying visits from emperors with accompanying favourite legions. The use of the plural may therefore, like the later competition between the Archbishops of York and Canterbury over the number of processional crosses each was entitled to have carried before him, embody a claim to superiority over rivals, reminding hearers that neither Caerleon nor Chester could properly be called more than castra legionis, "the fortress of the legion."

When all these things have been taken into account, we must surely conclude that in Gildas's time the shrine of Sts Aaron and Julius was in York, and that the cult of the two martyrs in ninth-century Caerleon has to be explained in some other way. Even if the ninth-century cult was based on authentic relics, it is easy enough to imagine plausible explanations. There must, for instance, have been later points in the divortium of Britain when a pagan Anglo-Saxon ruler who controlled York could have cemented a truce or alliance by giving the relics of the two saints as a good-will gesture to Britons who cared about them much more than he did.²²

The problems raised by Gildas's urbs legionum may be further illuminated by considering later phrases similar to Gildas's, and particularly such phrases in two other books written before the first known record of the shrine at Caerleon.

The first of these is Bede's civitas legionum, which shows that Britain's first great historian thought of Chester as a city of more than one legion. However, since modern archaeology says otherwise, Bede must have been misled, presumably by inadequate evidence. The status of his phrase is uncertain, but it seems more likely to be a description than, like Gildas's phrase, a recognised place-name equivalent. We can be certain that it is not a name: as it happens, in the very sentence that includes the phrase civitas legionum, Bede gives the British and the Saxon names for Chester, as Carlegion and Legacestir respectively (and observes that the British name is the more authentic). Nor is there any reason to suppose the phrase was a place-name equivalent. Bede, writing two centuries after Gildas, on the wrong side of the country and on the wrong side of a major cultural divide, was not well-placed to learn of such things. Moreover his Historia, whether from lack of interest or lack of information or both, shows little interest in legions or Roman cities; and much of what he does say on those subjects, including his assertion that Britain had twenty-eight cities (civitates), is simply taken over from Gildas, whom he recognised as the historian of the Britons (Bede 1.1--the 28 cities, Colgrave and Mynors 1991; 1.22--Gildas, Colgrave and Mynors 1991). It is difficult to know why he used the phrase civitas legionum of Chester, but he might have guessed that it was one of Gildas's twenty-eight cities from its size, that it was a legionary city from its British name, and that Latin descriptions of legionary cities put the word for legion in the plural by Gildas's urbs legionum. However that may be, there is no reason to take Bede's phrase as implying that the shrine of Sts Aaron and Julius was in Chester.

We should notice at this point that four centuries after Bede, William of Malmesbury used Gildas’s phrase in describing a rebellion against Edward the Elder. The rebellion, which took place in A.D. 924, was clearly at Chester, since Edward died a few days later at Farndon-on-Dee (Stenton 1985:339). In his account of the incident in his Gesta regum in 1120, William calls the place urbs legionum (William of Malmesbury 133.1 and passim; Mynors 1998); but since William admired Bede and intended his book as a continuation of Bede’s Historia, he may have coined the phrase either as a conscious variant on Bede’s phrase, which he uses quite often elsewhere, or by working independently from similar information (Gransden 1974:166-85, especially 169). Either way, his phrase is no guide to what Gildas had meant when he used the same words six hundred years before.

The other book that demands serious consideration is Nennius's Historia Brittonum,²³ which was put together about a century after Bede wrote his Historia. The Historia Brittonum has had a bad press recently, particularly from David Dumville, who has argued that what it says about fifth and sixth-century Britain is worthless except as a record of what people in later centuries thought happened then (Dumville 1986:1-26; 1977:173-92).²⁴ That is a sweeping claim, but we need not address the whole of it here, since Dumville acknowledges that the different sections of the Historia Brittonum need individual assessment, and we are concerned only with a single section, the much-discussed list of Arthur's battles in Chapter 56. In that list, the ninth battle is said to have been fought at the urbs legionis ("the city of the legion").²⁵ There is general agreement that the list of battles derives from a Welsh battle-listing poem, which seems to have rhymed on the names of the battlefields (Jones 1964:3-21).²⁶ Omitting some Latin inflections that would have had no equivalent in the Welsh, two consecutive names rhyme on -as (Dubglas and Bassas), and at least three on -on (Celidon, Guinnion, and legion).²⁷ Rhyme tends to give stability in transmission, so Nennius's rhyming names are likely to be the closest part of his chapter to the underlying Welsh poem.

There is a consensus that Nennius's urbs legionis translates an original Welsh cair legion. The Welsh phrase as a whole was current as a place-name (Rivet and Smith 1979), its individual words correspond to Nennius's Latin ones,²⁸ and legion, as we have seen, is also supported by the apparent rhyme. It may be added that one important group of manuscripts of the Historia glosses urbs legionis with brittannice Cair Lion dicitur ("it is called 'Cair Lion' in the British language"; Dumville 1985:104).²⁹ It might be objected that if Nennius's original read cair ("a fortified place"), he ought to have translated that by castra, a cognate Latin word similar in form and meaning; but Nennius's own book shows that the sense of cair can vary with its context. What in one place he calls Cair Guorthigirn, in another he calls arx Guorthigirni ("the citadel of Vortigern"), and his castellum Guinnion ("Guinnion fort") might represent a cair Guinnion in his original.³⁰ When he tries to identify Gildas's twenty-eight cities, he gives an introductory sentence in Latin followed by twenty-eight place-names in Welsh; the introduction calls the cities civitates and every name begins with cair, which here at least must be equivalent to civitas.³¹ The battle-listing poem may well have implied that its cair was a city too. We may ask why Nennius did not call the place civitas legionis, since he used civitas to introduce his list of cities, in which both the major British sites called Cair Legion are included. However, unlike Gildas, Nennius seems to have had no sense of a distinction between urbs and civitas (He uses both words for the same place more than once.), and he appears to have a slight preference for the former.³²

It seems certain that Nennius would have understood cair legion to mean Chester. He names Chester and Caerleon among the twenty-eight cities of Britain, and calls them Cair Legion and Cair Legeion guar Uisc ("Caerleon-on-Usk") respectively.³³ That implies that a cair legion without guar Uisc or any other qualification meant Chester to him and the audience he was writing for. This line of reasoning has made modern scholars almost unanimous that when Nennius wrote urbs legionis, he was saying that there was an Arthurian battle at Chester; scholarly disagreement has focussed entirely on whether he was right.³⁴ Some, pointing to the mobility of fifth-century warfare, and to Gildas's assertion that the Saxons raided right across Britain as far as the western ocean (Gildas § 24; Winterbottom 1978:27), have insisted that such a battle was at least possible. On the other hand, it has often been said to be very unlikely. One distinguished authority has said forthrightly that Chester was no place to meet Saxons in the fifth century, and therefore concluded that Nennius’s phrase was an "almost certain" interpolation into his source; another suggested this supposed British victory was a confused memory of the Battle of Chester in 613 or 616, which was a British defeat (Bromwich 1963:93; Jackson 1959:8). Since the urbs legionis seemed to be the most straightforwardly identifiable place in Nennius’s whole list of battlefields, such doubts about his reliability naturally made scholars very suspicious of what he said about the other battles. It has been suggested that he invented battlefield names or supplied them from miscellaneous sources, perhaps to replace names lost in transmission, or to provide Arthur with a total number of battles thought desireable for a great warrior,³⁵ and although scholars who put these ideas forward have always made it clear that they were only suggestions, the absence of serious counter-arguments has given them a status very close to received ideas.

The whole edifice, however, is shaky. First, although we may agree that Nennius would have understood cair legion as Chester, his source was composed by someone else, and these arguments take no account of what those words might have meant to that someone else. Second, there is some reason to suppose that Nennius himself may not have been referring to Chester. The main purpose of his list of Arthur's battles was clearly to tell readers where the battles were fought. He repeatedly uses a formula emphasising that the battles were past but the names are present: "The nth battle was fought in x, which is called. . . ." He names five of the nine battlefields like that, and names the region in which one of them was situated in a similar way. All the other battlefield names may have been current too: he never implies the contrary. It fits in with this that the essential element of each name is in Welsh, since any other language would have made the names less comprehensible. Apart from urbs legionis, only one name is wholly in Latin, and it is immediately followed by a Welsh equivalent: bellum in silva Celidonis, id est Cat Coit Celidon ("the battle in the Caledonian Wood, that is the battle in the Caledonian Wood"). Perhaps Nennius was not sure whether his original's Cat Coit Celidon was a description or a name: translating and quoting provided for both possibilities.

Cair legion, however, presented no such problems. For Nennius and his circle in nearby Gwynedd, Chester must have been a very important place, and its Welsh name one of the most familiar in Britain. To provide his readers with the complete set of comprehensible Welsh names that he obviously wanted, he had only to reproduce the name his source put in front of him; but he chose instead to translate it. That demands explanation, and I suggest that the best explanation is that, knowing the Welsh phrase would make his readers think he was speaking of Chester, Nennius translated it into Latin to make them realise that he was not.

We cannot tell whether the Welsh poem or some other source prompted Nennius to this apparent refusal to endorse Chester . We can, however, be confident that he did not feel able to suggest any alternative location for the Arthurian "city of the legion". Such a location would have completed his list of place-names, and if he had had it, he would surely have given it to his readers. Instead he took pains, as he did at other points where we might guess he had doubts about his source,³⁶ to reproduce exactly what his source said, which in this case meant providing a descriptive phrase that reproduced precisely the ambiguity of his original. His phrase was much more ambiguous than Gildas's, because (as we have seen) Nennius apparently did not distinguish between urbs and civitas, and his source's uninflected legion would have applied equally to those places that had associations with several legions and to those that had associations with only one. Nennius will certainly have known that his phrase described, even if it did not name, both Caerleon-on-Usk and Chester, and presumably that it might have applied to other places that he did not know of.

It is worth noticing that a refusal to provide the specific name his list asks for tells strongly against any theory that Nennius invented battlefield names or supplied them from miscellaneous sources, even though this present essay itself might seem to provide more grist for that mill. It might be suggested that Nennius's urbs legionis is a modification of Gildas's urbs legionum, which Nennius, perhaps without knowing the place referred to, had recognised as a kind of place-name from the distant past, the very kind of thing he wanted as a battlefield name. However, if he had decided to introduce Gildas's place into his list, it is difficult to see why, having altered his main source by doing so, and having pointlessly altered Gildas's urbs legionum to urbs legionis,³⁷ he should have been unwilling to translate the altered phrase into Welsh to match his other place-names. In reality, all theories of this sort look unlikely. If Nennius was so alert to and so scrupulous about the possibility that his source might mislead his readers that he refused even to reproduce the wording that his source put in front of him, we have no grounds for thinking that he carelessly or deliberately corrupted it in more radical ways.

What Nennius did with his source-material, however, is only a beginning. The most interesting question raised by this passage is what was meant by whoever composed the equivalent passage in Nennius’s source. That is a much more difficult matter. Nennius’s scrupulosity or otherwise with his sources does not tell us anything about how those sources treated their own source-material; and if inadequate sources could have misled a great historian like Bede, they certainly could have misled Nennius and a source whose origins lay in oral tradition. Nevertheless, could is not did, and evidence has been accumulating to suggest that Nennius and his sources may deserve to be taken more seriously than they have been in recent years (Koch 1996:246-251). It is worth considering, therefore, without presuppositions, what place the composer of an early battle-listing poem might have understood as "the city of the legion" and apparently referred to in such a way as to make a ninth-century enquirer think of that place as a legionary city that was not Chester.

The most important factor in giving an answer to this question is that the names in Nennius’s list are not, as far as we can tell, his own deductions from his source, but are directly taken from it, where they provided its rhyme-words. If Nennius’s source gave him real names, or at least, in the case of "The Battle of the Caledonian Wood", a recognised, established, and unambiguous phrase equivalent to a name, its cair legion should be a name or a name-equivalent of the same kind. But although Britain had a great many legionary encampments, their sheer number ensured that very few of them took their names from that fact alone. Their number would have made such names ambiguous, and so nearly all British place-names with legionary associations incorporate qualifying elements, usually as prefixes, as in modern Manchester and Winchester. The process can be seen at work in Nennius’s name castellum Guinnion for another of Arthur’s battles. Some of the few apparent exceptions to this process actually have quite different meanings, which would have been obvious in the early stages of their development; such as Caerleon in Morval and Carlyon, both in Cornwall, whose second element derives not from Latin legion- ("legion"), but from the plural of Cornish *legh ("flat stone", "slab"; Padel 1985:52).

The most important exceptions are of course Chester and Caerleon (-on-Usk), where in both cases the importance of long-established legionary fortresses ensured first that the place could be called simply "the" fortress of the legion, and second that that name could entirely displace an earlier native British name. The place-name record shows, as one would expect, that this took time: a common description had to turn into a recognised name-equivalent, then become a secondary name, and finally displace the original name. At York, this process was never completed, so its modern name is a descendant of Roman Eburacum by way of British (Caer) Ebrauc, Anglo-Saxon Eoforwic, and Norse Jórvík (Rivet and Smith 1979; Smith 1937:275-80); but Gildas's evidence shows that it was begun. Gildas shows that in the fifth century a name-equivalent existed that identified York as a legionary city, and that a competent author could use that name-equivalent in a book addressed to the whole of Britain, expecting his phrase to be recognised by his entire audience.

The overall pattern of warfare in Britain in the late fifth and early sixth centuries suggests that even if, as Gildas said, the invading Anglo-Saxons raided right across Britain, battles were more likely to be fought in the east than in the west, and specifically were much more likely near York than near Chester or Caerleon-on-Usk. In terms of antecedent historical probability, York (as we have seen) is a highly probable place for a late fifth-century battle between Anglo-Saxon and British forces, whereas Chester and Caerleon are very unlikely. Since the place-name evidence shows that a phrase like Nennius's could apply to all three of those places, it follows that, whatever Nennius himself thought the phrase meant, what he said about Arthur's ninth battle is much more likely to have been true of York than of Chester or Caerleon-on-Usk. We must therefore assume that if Nennius's source was reliable, his urbs legionis is Gildas's urbs legionum too; and that if, conversely, Nennius’s urbs legionis is Gildas’s urbs legionum, the principal factor that has led scholars to believe Nennius’s list of battles to be unreliable no longer exists.