This automated poetry-writing computer system is so good that most readers ‘strongly prefer’ its verses to those of Shakespeare. Or at any rate, that’s what its creator claims. Jonathan Law investigates…
On a bone-cold day in March the Wikiworm brought us some much needed cheer by digging out “The Bookseller/Diagram Prize for Oddest Title of the Year … a humorous literary award that is given annually to the book deemed to have the oddest title.” The winds cut like a skinning knife, the skies lowered with snow: but who would not be warmed by the thought of the 2002 winner Living with Crazy Buttocks or 2003’s The Big Book of Lesbian Horse Stories or indeed Proceedings of the Second International Workshop on Nude Mice (1978)?
I’m in the book business myself, so this stuff, great as it is, was not really new (just occasionally I’ve run into the prize’s founder, Bruce Robertson, at book fairs – the undeniable longueurs of which inspired him to begin his collection of strange titles back in the 1970s). However, I’d quite forgotten one of the more piquant details in Worm’s account – the moves to disqualify the 2008 winner, The 2009–2014 World Outlook for 60-milligram Containers of Fromage Frais, on the grounds that it had been created by a computer, rather than by its listed author, Philip M. Parker. And what I didn’t realize at all was the sheer range and volume of work produced by this Parker, a professor of management science with a background in marketing and economics; indeed with some 200,000 titles to his name, he could probably claim to be the most productive ‘writer’ in history.
A little Googling shows that Professor Parker has also concocted the following page-turners, any one of which could surely have been a contender for the Diagram:
The 2007-2012 World Outlook for Rotary Pumps with Designed Pressure of 100 psi or Less and Designed Capacity of 10 gpm or Less.
Avocados: A Medical Dictionary, Bibliography, and Annotated Research Guide
The 2007-2012 Outlook for Golf Bags in India
Webster’s Albanian to English Crossword Puzzles: Level 1.
Oculocutaneous Albinism – A Bibliography and Dictionary for Physicians, Patients and Genome Researchers.
The 2007-2012 Outlook for Premoistened Towelettes and Baby Wipes in Greater China
These, and many, many like them, have been created by Parker’s patented book-writing system – a modus operandi designed to eliminate what he refers to with some scorn as the “costs associated with human labour, such as authors, editors, graphic artists, data analysts, translators, distributors, and marketing personnel”. Essentially, Parker creates the template for a particular type of book – a handbook on a rare medical condition, or a survey of the sales outlook for the nichest of niche products – and then uses his algorithms to trawl the Internet and his own vast databases for content. The computer decants this into the prepared mould, takes care of grammar and format, and – hey prestissimo – you’ve something that looks like a book.
The beauty of the Parker system is that it is not so much print on demand as write on demand: only the title need exist until somebody, somewhere admits to the hole in their life that can only be filled by, say, The 2007 Report on Wood Toilet Seats: World Market Segmentation by City. An order is placed – and the computer creates a unique literary product that is then dispatched to the lucky punter. Parker has estimated that the total cost of producing a book in this way, which might sell for upwards of £200, is something like 12p.
Until now, Parker’s published works have all been in fields that might be considered friendly to the algorithmic approach – medical and marketing texts, solvers for crosswords and other word games. But – and this is the real subject of this post – recent years have seen him move into something altogether more intriguing. Using a notion of ‘semantic webs’ based on graph theory and a new suite of programs that he has nicknamed ‘Eve’, Professor Parker is now making a bold assault on the literary genre that might seem most inimical to his methods – poetry.
To demonstrate his belief that perfectly good poetry can be written by program, Parker has devised a set of heuristics to “mimic what I think my brain does when it is asked to write a poem on a particular topic using a particular poetic form (as assigned to me in grade school or college)”. In very simple terms, a set of algorithms is used to search over the semantic web associated with the chosen subject – ‘dogs’ perhaps, or ‘Charles Darwin’, or ‘Intimations of Immortality from Recollections of Early Childhood’. The results are then filtered through a further set of constraints imposed by a given poetic form – chiefly metre, scansion, and line-count. In this way, Parker can create a ‘poem’ based on (virtually) any word in the English language, in any one of a series of demanding technical forms. The professor of management science claims, on this basis, to be the author of over 1.3 million poems.
To see how this might work in practice, you can click through to the website of Parker’s Toto Poetry Project and have a go yourself. Put a word or subject into the box, click search, and lo, you have a set of digital poems – in forms ranging from the traditional sonnet, limerick, or haiku to some choice new ones of Parker’s own devising (for more on which, read on).
So, to start with the altogether obvious, what do Parker’s patent poetry-writing programs make of the subject ‘Dabbler’? On the simplest level we get this seven-word diagonal acrostic:
Well, I suspect most of us Dabblers are five or six or seven of those, so full marks for accuracy, at least. Moving up a notch or two on the technical side, we get this enigmatic rondelet – a French form consisting of a seven-line stanza alternating lines of four and eight syllables:
skilful in administration.
field engineering department.
Traffic conditioning function,
composition of transmission.
An indirect treatment, certainly, but I think it’s a grower. Rather less oblique is this little quinzaine (a poem consisting of three lines, comprising a total of 15 syllables):
Dabblers are several sprays.
Are they concordant?
Do you care?
Two very pertinent questions, I’m sure you’ll agree (the answer, I take it, is ‘Not really’). Next up, two of the more ingenious verse forms devised by Prof Parker himself:
Dabbler: Master, can I become adroit?
A dilettante, without exception, you were!
An authority, you have not been!
An instructor, you are not!
A babbler, you will continuously be!
That is a Yoda – a four-line poem based on the distinctive inverted speech patterns of the Jedi Grand Master.
Can I pass a pulse, published so pupils might see
Fully cheerful allusions, perhaps educating, and oh but revealed free?
Dipper or picker look all too absolute. But is fumbler?
Macerator, lover? – or dampener, twiddler? Darn! I excluded mumbler!
I wonder irrigator and otherwise disturber, but frankly ought I
I – state concepts so – abstracts crafted from computers spry?
And that – which reads a bit like some of the more recent work of Geoffrey Hill – is a Pi: a verse form in which the letter counts of the words follow the sequence in the celebrated ratio (3.1415926 and so on).
Trusting that this has whetted your appetite, let’s go on to see what Parker’s algorithms make of our esteemed editors. I’m afraid the program seems to have something of a crush on young Brit:
That’s another Parker-devised form – a Zed – and a piece to prove the old critical nostrum that strict formal constraints need do nothing to inhibit the expression of passionate feeling.
I’m afraid that Gaw fares a good deal worse in this ‘phrasal cinquaine’:
to watch intently
a fathead or flathead
and this simple acrostic:
He is also the putative subject of this weird but strangely effective rondelet:
take into consideration
consent to publish agreement.
to perceive by mental vision.
On the promising subject of Worm, who after all inspired this post, I can offer you a nonet – a somewhat vermiform piece in which the first line has nine syllables and each subsequent line has one syllable less:
slides which children ride at blazing speeds
a wisp or fibre
On the surely inexhaustible subject of Malty, ‘Eve’ takes a predictable line of attack, producing this ‘mirror cinquain’:
and this tight octosyllablc couplet:
Really pickled and boozy.
But also raddled and woozy.
I could go on and on, but as a final test let’s see how the Parker machine fares with that most unforgiving of strict literary forms, the limerick:
There was a young gal from Croaker,
She needed a word for soaker.
So smart that woman from Croaker.
There was a young gal from Khotan,
She needed a word for rotan.
No, it’s britisher!
That literate gal from Khotan.
There was a young girl from Balty,
She wanted a word for malty.
Could it be spunk trunk?
But settled on drunk!
That literate gal from Balty.
There was a young lad from Solvang,
And he wanted a word for slang.
Such as bajingo?
But then thought lingo!
That literate lad from Solvang.
Of course, it’s no real surprise that a computer can work within certain formal and semantic constraints to produce something that reads a bit like poetry. If you can teach a machine the rules of chess, there’s no reason why it shouldn’t also learn the requirements of a quinzaine or limerick. What this doesn’t address, unfortunately, is the whole question of literary value – are Parker’s ‘graph theoretic poems’ any good, as poems?
Parker himself seems to be in two minds here. At times, he is careful to minimize any claims on this score, arguing that the poems were created “for educational or didactic purposes” only – most obviously, to illustrate a variety of complex poetic forms but also to help students of English acquire and retain vocabulary. Elsewhere, however, he takes a bolder line, arguing that his programs have been designed with an inbuilt tendency to create poems that will satisfy the author (programmer) and match the tastes of his readers. Can literary merit be defined in terms that will translate into programming language? Parker is quite sure that it can. Aesthetic value, he argues
… is a mathematical problem of constrained optimization, where an economic utility function is being maximized. [Also] there is a portfolio problem given that there are exogenous social constraints, especially the perceived preferences of the reader … In economics, this is similar to a matching function between “buyer” and “seller” where there needs to be an equilibrium of sorts between consumers … Solving this problem becomes important when there are many poems that each maximize the utility function, or when social constraints are more “important” than the traditional utility maximization problem.
More provocatively still, Parker maintains that his poetry-writing computer passes the Turing test in terms of being indistinguishable from a poetry-writing human. Blind reviews suggest that his own “graph theoretic poems are basically indistinguishable from traditional poems … and in some cases judged by many to be of better ‘quality’ than traditional poems.” Indeed, Parker goes so far as to state that reviewers have shown a “strong preference” for his own work over that of Shakespeare, even when the authorship is revealed – a finding that he attributes “to the fact that most people generally do not like Elizabethan sonnets.” His final thoughts on the matter are expressed with an infuriating mildness: “my best guess is that graph theoretic poems, matching genre and topic, are no better than any others, but they are really not much worse either” (my italics).
Having conquered the world of poetry, Professor Parker is now planning to extend his method to the writing of romantic fiction.