The lava lamp of language

At a recent dinner party, the conversation turned to the evolution of language. Not in any profound philosophical sense, but simply observing that languages change slowly over time, sometimes quite radically, and expressing wonder and puzzlement that this should be so.

Most people are aware, even if only vaguely, that the English language has evolved over centuries from “Ye Olde Englishe” of the Anglo-Saxons, as found, for example, in the tenth-century Exeter Book (illustrated here).

Ye, of course, is an amusing misspelling of “Þe”, where the Old English letter Þ should be pronounced as a hard “th” (and note that the word “olde” has never been spelled thus, except in jest, while “English” only very occasionally attracted a final “e” and more often omitted the final “h”).

We can all agree that Old English is, to all intents and purposes, unintelligible to the non-cognoscenti. Consider the final four lines on the lefthand page, above (Exeter Book, folio 112v), relating the riddle of the bookworm, which begins “Moððe poρd fρæt”, meaning “A moth ate words”. Could you have guessed that?

Anglo-Norman-Scandinavian?

And even when the language evolves into what lexicographers call Middle English, around 1150 or thereabouts, it is still pretty difficult to decipher. We have seen this with the fourteenth-century Wycliffe Bible, even though it has at least developed a more modern-looking word order.

Try reading Chaucer’s Canterbury Tales, written around 1390 (the opening lines of the prologue are pictured here): “Whan that Aprílle with his shouris sote and the droughte of Marche hath percèd þe rote ...”. Peter Ackroyd, in his Retelling, renders this as “When the soft sweet showers of April reach the roots of all things ...” (he has decided to omit Chaucer’s “drought of March”).

Other modern languages experienced their own trajectories. Pictured here is the ninth-century Canticle of Saint Eulalia (“Cantilène de sainte Eulalie”), written in Old French, the Frankish equivalent of our Old English.

It is clear, from the opening line, that this is neither Latin nor French. It reads: “Buona pulcella fut eulalia” (“a good beautiful-girl was Eulalia”), “Bel auret corps bellezour anima” (“a beautiful body had she, and a more beautiful soul”). No modern French speaker could comfortably decipher this.

Like English, French passed through a Middle phase (the language of Rabelais and Montaigne), before becoming the Modern (or “classical”) French language that Cardinal Richelieu’s Académie française began to purify and preserve from the seventeenth century onwards.

Latin spoken by barbarians

We are probably all familiar, to some extent, with how this evolution occurred in the West, at least in its broad principles. Waves of migrating peoples — Germanic Saxons, Franks, and Goths, Scandinavian Normans — put their own stamp on an existing Latin-based language.

Italian had a similar transformation, via the late vernacular “Vulgar Latin”, into what is best described as Medieval Latin, as found (for example) in the Placiti Cassinesi (“Monte Cassino Decrees”), written in the 960s (one is pictured here).

“I know that these lands [kelle terre], which are contained by these boundaries [kelli fini], for thirty years have been possessed on the part of Saint Benedict”, it reads, in a language that cannot be described as either Latin or Italian.

As each city-state developed its own version of Italian, it was the Florentine of Dante Alighieri (a younger contemporary of Chaucer’s) that prevailed, after the foundation of the philological Accademia della Crusca in Florence in 1583.

Here is the opening page of the Divina Commedia from an edition published in 1472, thus closely comparable to the edition of The Canterbury Tales pictured above. This is the Italian equivalent of Middle English.

“Nel mezo del camin di nostra vita” (“Midway upon the journey of our life”, in Longfellow’s translation of 1867), “mi ritrovai per una selva oscura” (“I found myself within a forest dark”), “che la diricta via era smarrita” (“For the straightforward pathway had been lost”). A curious mix of Latin and proto-Italian.

What about Greek?

I still find it charming when — despite all of this evidence for the tortuous development of western languages — friends and colleagues learn, with surprise, that modern Greek-speakers are just as likely to be challenged by the Odyssey of Homer as we are. Surely they can simply read it?

The Greeks, of course, suffered no early influx of Franks, Saxons, or Normans to alter their language. And though the Byzantine empire of the seventh and eighth centuries was beset by Slavs and Bulgars in the west and Islamic Arabs in the east, their language was unaffected, continuing to hark back across the millennia to Classical Greek. The writings of Anna Komnena, for example, from around 1120, are perfectly intelligible to a classicist.

Ironically, having survived the medieval rise of Sunni Muslim power in the Levant, the death knell of the Byzantine Empire was sounded in the west, with the depredations of the unruly crusader armies in the thirteenth century. The Peloponnese fell into Frankish hands (becoming known as Morea) and the rising power of the Ottoman Turks finally consigned Byzantium to oblivion in 1453.

Tò Χρoνικòν τoû Moρéως (“The Chronicle of the Morea”) is a poem written in the 1300s in mainland Greece. It begins, “I am going to tell you a great tale; and if you are willing to listen to me, I hope it will please you” (in lines 1–2 of the Bern Burgerbibliothek manuscript pictured here).

It is difficult for the layman to appreciate, but scholars characterize the Chronicle’s Greek as plain and simple, based upon the oral vernacular rather than the Greek classics. But more than that, the writer (apparently a Frank who had settled in the Peloponnese) employs novel forms of words and erratic grammatical tense-switching — the Greek equivalent, perhaps, of the Canticle of Saint Eulalia.

Sadly, while Greece languished under Ottoman rule for centuries, her language deteriorated under Turkish influence and fragmented across highlands and islands. Local dialects diverged in disparate areas: Pontic, along the southern Black Sea coast, and Mariupolitan, along the northern; Cappadocian, in central Turkey; Cycladian, amongst the eastern islands; and Cretan, not to mention the mainland of Attica. Meanwhile, the western isles had fallen under the rule of the Venetians, who renamed Kerkyra as Corfu, for example, and it is interesting that the first standard grammar of Modern Greek to be attempted was written by the Corfu-born Rome-resident Nikolaus Sofianos in the 1540s, though it wasn’t published until 1874.

The eventual liberation of Greece from Turkish rule is an extraordinary story, but from the linguistics point of view, it is sufficient to observe that it was not until the 1850s that a Modern Greek language based on spoken Greek was proposed, and the debate over the katharévousa (“purefying”) was only brought to a conclusion in 1976.

It is unsurprising, then, that ancient Greek, though perfectly intelligible to Anna Komnena in 1120, is now as alien to a modern Athenian as Chaucer is to us.

The emperor’s new clothes

Everywhere people are falling over themselves to justify new and exciting uses for AI technology. Many of these are obviously trivial and contrived. Fake video abounds, students employ “content generators” to write their essays for them, and unscrupulous publishers create false scientific publications. Meanwhile, acres of creative writing are silently mined as raw material to train robots to emulate humans. All of this places a strain on existing data centres and boosts carbon emissions at a time when we are supposed to be conserving energy.

Of course, there are undoubtedly good uses of AI technology in the analysis of large scientific data sets, in the detection of patterns and potential correlations, and in the prediction of possible future data trends.

None of this, it seems to me, applies to the Arts and Humanities. Here, research is necessarily driven by human ingenuity.

There are no large scientific data sets that defy human comprehension. No raw statistics whose “crunching” requires machine assistance. Just, it seems, a fair number of researchers anxious to jump on a band wagon and draw upon whatever funding might briefly be available.

Earlier in the year, I became aware of a massively multi-authored paper (published in the journal Nature Vol. 645) entitled “Contextualizing ancient texts with generative neural networks”. I was immediately repelled by its jargon-heavy content, the kind of writing that I have usually found to be designed to exclude all but those with a vested interest. (I commented on this phenomenon back in June.)

Nevertheless, I was intrigued by the charm-offensive mounted on this paper’s behalf by Professor Dame Mary Beard, who pronounced that “it promises to be transformative!” And so, not one to admit defeat easily, I decided to have another go.

Robotic interpretations

The purpose of the paper is to announce a new computerized tool named Aeneas, a “multimodal generative neural network for contextualizing Latin inscriptions”. According to the authors, it is designed to solve the following epigraphic problems:
“letters, words or entire sections of an inscription may be lost over time, with the full extent of the missing text often being unknown. High levels of human mobility, absence of explicit dates and the frequent use of abbreviations, a hallmark of the Latin epigraphic habit, further complicate our interpretation of these inscribed artefacts”.

The writers are not particularly forthcoming about Aeneas’ method, but they cite an altar from Mainz (CIL XIII, 6667) as a successful example of its use. This I find confusing, since the altar (as can be seen here) is entirely undamaged and does not seem to suffer from any of the aforementioned “problems”.

It reads:
Deab(us) Aufan(is) / et Tutelae loci / pro salute et in/col(u)mitate sua / suorumq(ue) om/nium L(ucius) Maiori/us Cogitatus b(ene)f(iciarius) / co(n)s(ularis) vot(um) sol(vit) l(ibens) l(aetus) m(erito) / Idibus Iuli(i)s / Gentiano et / Basso co(n)s(ulibus)

“To the Aufanian goddesses and the guardian of this place, for the health and safety of himself and all his family, Lucius Majorius Cogitatus, beneficiarius of the legate, gladly, willingly, and deservedly paid his vow on the Ides of July when Gentianus and Bassus were consuls” (i.e. 15 July AD 211).

The writers correctly report that the altar is a well-attested type of military inscription, but more curiously they celebrate the fact that “Aeneas’ predictions successfully situate the inscription within this broader epigraphic habit, with a dating estimate (214 CE, within the expected range) and geographical attribution (correctly placing the stone in Germania Superior, with related alternatives in Germania Inferior and Pannonia) that align well with historical expectations”.

The writers divulge none of the actual “predictions” or “historical expectations” to which they obliquely allude. But later in the paper, they again emphasize that “Aeneas’ dating average for this altar is 214 CE, which is well within the 10-year range the model is trained on, and its top-3 geographical attributions are Germania superior (correct), Germania inferior and Pannonia superior”.

Remember, the text of the altar itself explicitly dates it to AD 211. There should be no confusion about this. Am I missing something here? To my mind, Aeneas has done significantly worse than any student who might simply search for this inscription in (for example) the online Heidelberg Epigraphic Database, which gives the correct date and provenance.

The writers praise Aeneas for “the historically enriched depth of the parallels it retrieves”. In the case of the Mainz altar, “even more striking”, claim the writers, “is Aeneas’ top parallel identification for this text: another votive altar, dedicated in 197 CE by beneficiarius Iulius Bellator and found near the same location in Mainz”. (They are alluding to AE 2017, 1072.)

It reads:
Deabus Aufan/is pro salute / et incolumitate / sua suorumq(ue) / omnium Iul(ius) / Bellator be(neficiarius) / co(n)s(ularis) libens posuit / Idibus Iuli(i)s / Laterano et / Rufino co(n)s(ulibus)

“To the Aufanian goddesses, for the health and safety of himself and all his family, Julius Bellator, beneficiarius of the legate, gladly set (this) up on the Ides of July when Lateranus and Rufinus were consuls” (i.e. 15 July AD 197).

Again, Aeneas’ modus operandi is not divulged, but I might guess that, like any moderately skilled epigraphy student, it searched for other dedications to the Deae Aufaniae, for example. Or perhaps it searched on the phrase pro salute et incolumitate. And maybe in conjunction with the donor’s rank of beneficiarius consularis. Certainly, the parallel texts reportedly offered by Aeneas would suggest this method (nine are listed, inconveniently by Heidelberg code, rather than the standard CIL or AE references used by generations of epigraphers). At any rate, it’s hard to know why we need AI technology for this.

Filling in the blanks

One of the features of Aeneas emphasized by the writers is that it allegedly “enables restoration of text sequences of unknown lengths”. Offhand, I can think of half-a-dozen inscriptions with significant gaps or lacunae that researchers have attempted to fill (such as this one). By contrast, neither of the altars cited by the writers has missing text, so they arbitrarily removed the words loci pro from the first one (leaving Tutelae ---/--- salute), in order to “test” Aeneas’ powers of restoration. They excitedly report that five “contextually and linguistically accurate” alternatives were proposed. This seems a rather poor result.

Our moderately skilled epigraphy student would surely predict, as Aeneas did (but only in four of its five attempts), that the word salute should be preceded by pro. (Out of 2,976 instances of the word salute in the Clauss-Slaby Database, it is preceded by pro in 2,920 cases.)

The same student would surely then realize, as the Aeneas program seems to have, that the careful layout of the inscription requires a three- or four-letter word after tutelae. There are not many options. The deity Tutela Augusta is occasionally encountered, which is presumably why Aeneas’ third choice is Aug. (I imagine it realized the four-letter maximum, as its fourth choice is sacr, which in football would be termed a “Hail, Mary” attempt, desperately relying on the stonemason’s incompetence over epigraphic precedence.) Sadly, its first choice, its best guess, is to leave a blank, which makes nonsense of the inscription.

Its second choice is the correct one. I’m not sure if that makes good odds or not. It perhaps falls into the same category as Dr Johnson’s dog walking upon its hind legs: “It is not done well, but you are surprized to find it done at all”.

New horizons?

Unfortunately, more is planned. I’m afraid I don’t agree with the writers when they claim that “Aeneas demonstrates the transformative potential of AI in augmenting historical research”. Nor with Dame Mary when she claims that “Aeneas opens up entirely new horizons”. Based on this paper, it doesn’t seem promising at all.

Incidentally, the Nature paper is not error-free. The plural of beneficiarius consularis is not beneficiarii consulares, since each man is still attached only to one legate, and it is not clear why the writers assume that such men were “usually close to retirement”. Nor is it clear why they think that tutelae pro salute is “the more commonly attested version of the formula”, since the Clauss-Slaby Database has precisely one inscription that matches this pattern (CIL XIII, 11015).

Also, some of the references have come unstuck. No. 71, “Elliott, T. in Epigraphy and Digital Resources (eds Bruun, C. & Edmondson, J.) 78–85 (Oxford Univ. Press, 2014)” — this is the title of Elliott’s paper, not the Bruun–Edmondson volume, which is The Oxford Handbook of Roman Epigraphy. No. 99, “Haensch, R. Steine sind. Benefiziarier, Matronae Aufaniae und die Topographie des Römischen Mainz. Z. Papyr. Epigr. 203, 293–296 (2017)” — the title is actually “Sag mir, wo die Steine sind”. Perhaps we might start with a proof-reading AI?

Caledonia, land of cold and rain (troisième et dernière partie)

It has been a busy twelve months and I find that I have managed to neglect Caledonia. Book One: The Ninth Legion, a wonderfully inventive bande dessinée by Corbeyran and Despujol — une chose impardonnable (see Caledonia, land of cold and rain (deuxième partie) from last December).

When we left the Ninth Legion, centurion Lucius was planning to track down the source of the Caledonians’ power on the Island of Death. “If we manage to unravel its mystery”, he explains, “we’ll undoubtedly gain the upper hand over our adversaries” (p. 32).

After a fierce mêlée (pp. 33–34), a scouting party manages to capture the warrior Cirig (p. 35), and seeing that he and his existing prisoner Leta share a fondness (p. 36), Lucius orders him to be hung on a cross to force Leta’s compliance in brokering a deal with her people (p. 37). Off she runs to the dark broch (p. 38) where she delivers the Roman message, but Cirig’s rival, the dark-haired Fel, shouts angrily, “Our people have no need of an invader to dictate their conduct” (p. 39).

The sacrifice

Despujol’s beautiful graphics depict the dim broch interior, with its drystone walls and intramural galleries and blazing central hearth. It is easy to miss the ominously large skull amongst the collection of antlers on the wall. What kind of creature does it belong to?

The old crone, Isla, reveals that Leta’s young brother, Mil, was sacrificed in order to summon the hell-hounds (p. 40) — the hell-hounds that we saw attacking the Roman fort by night in Part Two. Leta is furious and blames Fel, claiming that, unable to win her heart from Cirig, he has spitefully arranged for her brother’s death.

Worse than that, Mil has died in vain, since Leta was obliged to send the hell-hounds back in order to save her own life. She tells the old chieftain that she must return to the Romans with a deal to save Cirig’s life, but Galam replies, “Such a deal is impossible. You know that” (p. 42).

“But the Romans are here. They’ll never leave,” she says. “You have to negotiate. Or we’ll disappear” (p. 43). Meanwhile, centurion Lucius is apprehensive that his men have started to mistrust him (p. 44), when Leta returns with news that her father wishes to conclude an honorable treaty with the Romans (really?): “You will meet him at the new moon” (p. 45).

Legion on the march

Centurion Lucius arrives at the headquarters camp of the Ninth Legion, where the overweight, fur-clad legate Deodatus Faustus is having another cup of wine. He complains that, every time someone lifts the flap of his tent, they let the cold wind in, “And without fail, it never brings good news” (p. 46).

Lucius proposes that he can negotiate a treaty with Galam, and if Faustus entrusts him with the entire Ninth Legion, he can deter the Caledonians from any aggression (p. 47).

And so, Lucius sets off with his pessimistic optio beside him: “You’re rushing headlong into another trap”, he grumbles, “but this time, you’re dragging everyone along with you” (pp. 48–49). Lucius is undeterred. “This show of force is solely intended to ensure the meeting proceeds smoothly,” he explains (pp. 50–51).

The horror from the woods

The scene is a snowy plain bounded by dense forest. Lucius is convinced that his plan will work. The entire region seems quiet (p. 52).

Spotting Leta on a rock in the midst of the plain, Lucius rides over to her (p. 53). She seems defiant. “My father won’t appreciate you gathering so many soldiers for a simple meeting”, she scowls. “So many thousands of armoured men against a handful of naked warriors. Are you afraid, Roman?”

Meanwhile, the thin line of Caledonians along the forest edge doesn’t move. The crone Isla shakes her magical talisman and the forest starts to crack and groan (p. 54). Lucius’ eyes start from his head. His horse shies. “The shapes in the woods!” he cries. “Yes, Lucius”, laughs Leta. “Your eyes do not deceive you” (p. 55).

Too late, the Romans see the gigantic antlered figures with six-foot blades, lumbering from the forest (hinted at on the book cover — see Caledonia, land of cold and rain (première partie)). “I have betrayed you”, screams Leta. “This plain will soon be littered with corpses. Caledonia will be your grave!” (p. 56). Thus is the Ninth Legion doomed.