Only Connect

Le Ton Beau de Marot: In Praise of the Music of Language
by Douglas R. Hofstadter.
Basic Books, 632 pages, $30.

Connection, linkage: this is the great task of modernist aesthetics. “Only connect!” thinks a character in E. M. Forster’s Howard’s End. “The ordinary man,” reports T. S. Eliot, “falls in love, or reads Spinoza, and these two experiences have nothing to do with each other, or with the noise of the typewriter or the smell of cooking; in the mind of the poet these experiences are always forming new wholes.” Douglas Hofstadter––best known as the author of the Pulitzer Prize-winning intellectual fantasia Gödel, Escher, Bach: An Eternal Golden Braid––is an enthusiastic adherent of this tradition. Perhaps too enthusiastic.

In Gödel, Escher, Bach, Hofstadter used three figures from different intellectual disciplines––mathematics, graphic art, and musical composition––to anchor a series of reflections on such recursive structures as Möbius strips and iterated algorithms, such self-referential statements as the statement “This statement is false,” and the possibility of building intelligent computers. (By profession a computer scientist, Hofstadter is a leading figure in the quest for artificial intelligence.) One of the minor themes of Gödel, Escher, Bach is translation: What does it mean to translate something from one language into another? Can Lewis Carroll’s “Jabberwocky” be translated? If so, what would be the French for “‘Twas brillig, and the slithy toves / Did gyre and gimble in the wabe”?

This new book, Le Ton Beau de Marot, takes up the question of translation in more detail. Several years ago, Hofstadter became interested in a slight, short poem, “Ma Mignonne,” by the sixteenth-century French poet Clément Marot, and started wondering how it might best be translated into English. What was essential to preserve in translation, and what could be dispensed with? Hofstadter sent a copy of the poem, plus his own literal translation and an enumeration of what he felt to be the key features of the poem’s structure, to “fifty, perhaps even a hundred, friends scattered around the globe.” Dispersed throughout this enormous book are the responses of those friends, along with almost fifty different translations of the poem.

Clearly Hofstadter saw the text of “Ma Mignonne” as anchoring his reflections about translation in the same way that the works of Kurt Gödel, M. C. Escher, and Johann Sebastian Bach anchored his thoughts about self-reference in his first book. But the poem is in fact too slight to moor anything so active as Hofstadter’s mind, and he continually goes off on tangents that take him far, far away from the question of translatability. To be sure, polymathic digressions produce much of the charm of his writing: his excursuses on ambiguous newspaper headlines (“British Left Waffles on Falklands”), palindromes, dirty jokes, and various other linguistic games are funny and provocative. But when he starts telling us of his fondness for Wendy’s Spicy Chicken sandwiches, or explaining that, on a visit to Clément Marot’s home town, he bought some postcards and got his hair cut before he joined some friends for dinner––well, one begins to wonder what editors at Basic Books do to earn their money.

If readers can wade through the ephemera, however, they will discover that the key to his project, not just in this book but also in his research on cognition and artificial intelligence, may be found in a statement he cites from the mathematician Stanislaw Ulam: “What is it you see when you see? You see an object as a key, a man in a car as a passenger, some sheets of paper as a book. It is the word ‘as’ that must be mathematically formalized.... Until you do that, you will not get very far with your AI problem.” Virtually all of Hofstadter’s work revolves around this quest for “as”––which is why translation is so important.

He directs a research program on “Concepts and Cognition” that focuses on analogies and creativity, for the obvious reason that the creative thinker is the one who can find the connection, the analogy, the “as,” that links the seen and known with the previously unseen and unknown. If we explore what we do when we translate a poem from one language into another, perhaps we can come to understand how analogical thinking works well enough that we can teach computers to do it. And if we can do that, thinks Hofstadter, computers will be intelligent and the quest for artificial intelligence will have succeeded.

But can we do that? Hofstadter easily punctures the hubris of computer scientists who think they have already achieved that goal, but he is even more scornful of philosophers who, like John Searle, think it impossible. For Hofstadter, there is no principle that makes artificial intelligence unrealizable––but that has something to do with what he thinks human intelligence is.

In Gödel, Escher, Bach, Hofstadter posits what might be called a deontological theory of consciousness, one which holds that humans think by first learning and then employing certain rules. It may seem to us that our minds aren’t purely “rule-governed,” but that’s just an “illusion.” These rules that govern the mind, Hofstadter claims, express themselves through certain patterns. We recognize patterns and thereby appropriate (even if we don’t fully understand) the rules that produced those patterns. Hofstadter thinks that the ability to recognize patterns “lies very close to the core of ‘pure’ intelligence, if there is such a thing”––and thus, since computers follow rules and rules produce patterns and the recognition of patterns is the definition of intelligence, computers can exhibit intelligence.

It is not clear, of course, that mere recognition of patterns constitutes intelligence. Searle thinks it does not, and Hofstadter’s many pages attempting a refutation of Searle attack the illustration Searle uses rather than his point as such. “Intelligence” itself may not be a particularly meaningful concept; Stephen Jay Gould, in The Mismeasure of Man, has done serious damage to the notion that there is a single capacity that can be called intelligence. Hofstadter makes his work a lot easier by selecting one thing that computers could plausibly be thought to possess and calling that “intelligence.”

But more important for someone who wants to understand Le Ton Beau de Marot is Hofstadter’s linking of patterns with rules, because that is the source of Hofstadter’s peculiar notion of what poetry is and what the translators of poetry have to do. He hates free verse, for example, because it follows no rules, and therefore has no patterns. (And therefore, by his own definition, must not exhibit intelligence.) This is hardly a helpful description of what William Carlos Williams in Paterson, T. S. Eliot in The Waste Land, and Ezra Pound in the Cantos were up to, writing poetry whose rich patterns exist even if they do not recur with regularity and predictability: their poems have patterns without discernible rules. Writing in traditional forms is no less creative; but the point to be made here is that human creativity sometimes involves the use of patterns that are not rule-governed, and the task of creating artificial intelligence is much harder than Hofstadter thinks.

Two illustrations will help to show how difficult it is for the author to grasp the notion of patterns that deviate from rules. He has a great fondness for Vikram Seth’s The Golden Gate (1991), a novel in verse written entirely in a stanza derived from Aleksandr Pushkin’s Eugene Onegin (1831). But when Seth writes, “How can I (careless of time) use / The dusty bread molds of Onegin / In the brave bakery of Reagan?” Hofstadter thinks the first and third lines quoted need an extra syllable each, so he pauses to improve them: “How can I (careless of fashion) use / The dusty bread molds of Onegin / In the brave new bakery of Reagan?” Leaving aside the question of whether being careless of fashion is the same thing as being careless of time, we should still note that Hofstadter’s mechanical sense of rhythm demands that a line not have two stressed syllables in a row (time-use, brave-bake). Of course, good poets vary stresses all the time to avoid mind-numbing repetition over the length of a poem––especially a book-length poem. But if pattern-recognition is the heart of intelligence, and patterns must follow rules, then the disruption of pattern is a sin against intellect.

My second illustration comes from Hofstadter’s impossibly long discussion of how Dante should be translated. Notwithstanding the fact, which he cheerfully admits, that he has read none of the Purgatorio or Paradiso and only some of the Inferno, he howls in outrage when he notes how few translators stick with Dante’s linked terza rima in their English versions. Hofstadter can’t imagine why none of the widely used translations of Dante keep to this form, and he thinks their neglect is a kind of betrayal.

But the answer is simple: Italian is a very rhyme-rich language (in large part because so many of its words end in vowels), while English is rhyme-poor. The translator who reproduces in English Dante’s rhyme scheme is continually forced into unnatural word order and peculiar word choice. (Compare Dorothy Sayers’ persistently clunky terza rima with the more flexible forms preferred by John Ciardi or Mark Musa and you’ll see what I mean.) For the same reason, translators of French poetry into English routinely replace the six-beat alexandrines common in French with the pentameter that suits the rhythms of English. (W. H. Auden was right to say that alexandrines always sound comic in English.) Elsewhere in the book, Hofstadter acknowledges the difficulty deciding what it means to produce a “faithful” translation, but he is so insistent about how Dante should be rendered into English because the rule-governed pattern of the poet’s terza rima is more important than euphony, clarity, or narrative coherence. But since the pattern is perfectly accessible to people who don’t know a word of Italian, why bother to translate Dante at all?

Of course, Hofstadter isn’t really interested in the product of the act of translation, but the act itself. He hopes, by investigating his own translations and those of his friends, to understand better how the human mind makes analogies, and thereby to forward the quest for artificial intelligence.

And on the question of why we should want computers to be intelligent, he says something rather bold and surprising: “Why should a race of benevolent if fleshless beings be any less worthy of being considered our ‘children’s children’ than the potential gang of music-polluters, lyrical rapists, priest-assassins, and mob-frenzied jackbooted genocide perpetrators that might spring forth from the celestial union of my eager sperms with your equally eager ova?” (Hofstadter’s prose isn’t always this out-of-control, but often it is.) Though he mentions his children often and dedicates the book to them, he does not seem to relish turning over the world to them, and he seems more willing to trust children he made than those he begot.

Here again is Hofstadter’s debt to modernist aesthetics: the poet (or the computer scientist) as rival creator to God. He never asks whether a group of potential “genocide perpetrators” is likely to create beings characterized primarily by “benevolence,” and his hope seems to be that we can make computers intelligent by teaching them patterns and make them good by avoiding teaching them greed. And if virtue were merely the absence of vice, rather than the fruit of spiritual and moral discipline, that hope might be a believable one. Alas.

One thing more. A substantial portion of this book is devoted to a kind of tribute––seemingly out of place in this context, but moving nonetheless––to Hofstadter’s late wife Carol, who died rather suddenly from a brain tumor in 1993, leaving him to raise their two young children alone. One is left wondering whether Hofstadter is perhaps envisioning computers as electronic companions who will not die and leave us behind. Hofstadter is at heart a Gnostic: he thinks of our bodies as merely the containers of our immaterial essence, and wants us to consider computers as potential children (and spouses?) because our sense of human identity should have nothing to do with our genetics and everything to do with “something more abstract, something based on a certain way of thinking and feeling and caring, irrespective of the physical medium in which it [our identity] is embedded.” A person’s identity dies, Hofstadter thinks, because it was “embedded” in an all-too-fragile “physical medium.” Here again God screwed up, and we will not repeat His mistakes: If the AI people get it right, our cyber-children and cyber-friends and cyber-spouses may have to grieve our passing, but we will never have to grieve theirs.

Alan Jacobs is Associate Professor of English at Wheaton College.