AI and literary translation

Tim Parks (Afterthoughts, February 14) claims that large language model AI can translate literary prose as well as human translators. Having fed GPT-4 a passage from an Italian novel, he declares that “The software nailed it perfectly” and notes that AI seeks “to overcome the computer’s lack of human experience with a predictive mathematical model.” That lack cannot be overcome by a device that has no body and cannot detect an author’s intentions from context.

The demonstrable shortcomings of GPT-4’s translations suggest that some of these cannot be remedied by more advanced versions of AI. Here is a sentence by Proust that David Lodge used in Language of Fiction to show why prose could be as hard as poetry to translate well:

Mais ce bonsoir durait si peu de temps, elle redescendait si vite, que le moment où je l’entendais monter, puis où passait dans le couloir à double porte le bruit léger de sa robe de jardin en mousseline bleue, à laquelle pendaient de petits cordons de paille tressée, était pour moi un moment douloureux.

Here is GPT-4’s translation:

But this goodnight lasted so little time, she came back down so quickly, that the moment when I heard her coming up, then when the light rustling of her blue muslin garden dress, with its small braided straw cords, passed through the double doors of the hallway, was a painful moment for me.

Improved AI may eliminate unidiomatic phrases such as “lasted so little time.” But “came back down” is wrong, because Marcel’s mother came up and went back down afterwards. A bot cannot understand the narrator’s point of view and physical location. “Then when” for “puis où” separates the rustling from the “moment” rather than making it the next part of it. Here is C. K. Scott Moncrieff’s translation:

But this goodnight lasted for so short a time: she went down again so soon that the moment in which I heard her climb the stairs, and then caught the sound of her garden dress of blue muslin, from which hung little tassels of plaited straw, rustling along the double-doored corridor, was for me a moment of the keenest sorrow.

Transferring the verb from sound to narrator, who first “heard”, then “caught”, the sound, continues the description of the moment with a parallel structure. And “little tassels of plaited straw” is of course far better than “small braided straw cords”. The bot retrieves words; it knows nothing about dresses.

Consider, finally, the closing clause: étAIT pour MOI un moMENT doulourEUX. Note the rhythm, see which word comes last and hear the vowels of “douloureux”. The flat “was a painful moment for me” will not do! The sentence postpones its predicate to enact the narrator’s wish that the moment could have lasted longer. Anyone trying to improve Scott Moncrieff’s translation slightly would need to appreciate why the sentence works as it does, something AI cannot do.

Parks overestimates the abilities of GPT-4 and seems to undervalue his own: “you have to admire the creativity of the scientists who have produced a machine that demonstrates how little of what you do is creative”. I would admire the ingenuity of the engineers and would insist that what a good literary translator does, whether one calls it creation or re-creation, is certainly creative.

Roger Greenwald
University of Toronto

The magic of science

Jennan Ismael’s review of The Impossible Man: Roger Penrose and the cost of genius by Patchen Barss (February 21) notes that Einstein’s struggles to find symmetrical equations for general relativity and their confirmation by the eclipse of 1919 are “a vivid example of the magic of science”, which they are.

An even more striking example is that of gravitational waves. In 1916, in his paper on general relativity, Einstein used mathematics to predict the existence of gravitational waves, the disruption of space-time by massive accelerating objects. After a century of advances in physics, engineering, materials science and other disciplines, his prediction was confirmed when gravitational waves, which had been inferred by the measurement of radio emissions in the 1970s, were actually measured. In September 2015, the US National Science Foundation Laser Interferometer Gravitational-wave Observatory physically sensed the undulations in space-time caused by gravitational waves generated by two colliding black holes 1.3 billion light years away.

James Smith
Columbus OH

Violets

Wading into the “violets/violer” plagiarism controversy, Derek Meyer (Letters, March 14) misreads the opening of the second paragraph of Finnegans Wake.

In “Sir Tristram, violer d’amores, fr’ over the short sea, had passencore rearrived from North Armorica on this side the scraggy isthmus of Europe Minor to wielderfight his penisolate war”, the noun “violer” (“now archaic … a player of the viol”: Shorter OED) is in apposition to “Tristram”; Joyce puns on “viola d’amore”, but “amores” alludes to Ovid, whose loves are of course adulterous. “Violer” echoes “vicus of recirculation” in the first paragraph and “amores” morphs into “Armorica” (ie Brittany), itself punning on “America”; “rearrived” because Tristan, healed as “Tantris” by Isolde, returns to claim her as a bride for his aged lord, King Mark, thereby violating his undeclared love for her (and hers for him), only to later violate Mark’s trust by starting an utterly consensual affair with her; “penisolate war” echoes “Peninsular War” (“penisola” is Italian for peninsula”), but puns on “penis”: in the SOED, “penis” immediately follows “peninsulate”!

But why “violer”? Great warrior and famous lover, Tristan is also – according to Gottfried von Strassburg’s medieval epic – a consummate musician on many instruments. Gottfried’s Tristan und Isolde was known to Joyce (as what was not?): hence “retaled … later on life down through all Christian minstrelsy” in Joyce’s next paragraph. Joyce (a singer himself) intended his works to be read aloud – realized in performance – in order to be fully grasped.

No work, or word, however remote, is safe from Joyce’s magpie/magus logomachies: “recirculation” isn’t the half of it! As his erstwhile assistant Samuel Beckett wrote two years before Finnegans Wake was finally published, Joyce’s aim in this final work is “a matter of an apotheosis of the word” (Letters, Volume 1, 1929–1940, pp515–19, 2009).

John MacInerney
London SW19

Christopher Hill

Richard Davenport-Hines’s review of Michael Braddick’s biography of Christopher Hill (March 21) brought back to mind a memorable occasion when I was seated next to Hill at a private dinner party hosted by a Balliol graduate student. When I expressed nervousness at the prospect of supervising undergraduate student tutorials, Hill assured me that the secret of teaching was simple – all that is needed is to stay one book ahead of your student. I have never been quite able to decide whether this was an expression of cynical worldliness, kindly encouragement of a novice tutor, or a bit of both.

Sharon Footerman
London NW4

Cancel culture

Regarding Paula Keller’s review of The Cancel Culture Panic by Adrian Daub (In Brief, March 7), if the author and reviewer are going to argue that “cancel culture” is something of a myth wielded in bad faith by conservatives, they are going to have be more accurate in their case studies. Assuming that the account presented of the controversy involving the Harvard history professor Stephan Thernstrom is representative of the book’s contents, Daub has failed to offer a “good analysis” of a baseless “panic”, as claimed by the reviewer.

The review’s summary of that incident – “after one of his lectures, some of his African American students felt uncomfortable about some of his remarks, so they talked to him about it; the student newspapers reported on the incident, but the excitement quickly died down; students and professor came to some sort of agreement; Thernstrom went on lecturing” – is a highly misleading account of what actually happened.

Rather than discussing their concerns with Professor Thernstrom, a couple of students filed a formal complaint charging him with racial insensitivity. While that complaint (largely founded on his reading from plantation owner journals, self-evidently important primary sources for the study of slavery) was pending, Thernstrom found himself twisting in the wind without support from colleagues and university administrators, notwithstanding their supposed commitment to academic freedom.

Although the complaint was eventually resolved without any action being taken against Thernstrom, he elected – once bitten, twice shy – not to teach the “offending” course again. This is hardly a story that belies the reality of “cancel culture”.

Henry D. Fetter
Washington DC

The History Man

Malcolm Bradbury was far too sophisticated a novelist to have “based” any of his characters on one person in the manner posited by Richard Jenkyns (Letters, March 14) and Colin Kidd (February 28 and Letters, March 21). Indeed, Howard Kirk in Leeds is less an old fogey than an Angry Young Man; George Carmody, though conservative, is a development of the pimply wreck Louis Bates from Eating People Is Wrong (1959). Asa Briggs is an unlikely basis for Millington Harsent: a far more likely candidate as inspiration is Frank Thistlethwaite, the founding vice chancellor of the University of East Anglia from 1961 to 1980. Thistlethwaite pioneered the field of American studies (he was the founding chair of the British Association for American Studies from 1955 until 1959), and he brought to UEA a commitment to interdisciplinary departments and seminar group teaching at a time when both were seen as “radical” (ie American) developments in British university education. He also had Harsent’s “Edifice Complex”, and in 1985 he was made an honorary fellow of the Royal Institute of British Architects (a rare honour for a lay person). I don’t know if Thistlethwaite was a Labour voter, but Harsent is unlikely to be an exact copy. He is, after all, made up.

Joseph Williams
University of East Anglia

Pessimism

Oswald Spengler should be heard in any discussion of pessimism (see Kieran Setiya’s review of books by David E. Cooper and Mara van der Lugt, March 14). In Man and Technics (1932), Spengler wrote: “Optimism is cowardice”.

James Blanchard
Bar Harbor ME

The post AI and literary translation appeared first on TLS.