To refer is human

Having started to read Half-Blood Prince, aloud, at bedtime, to A., I’ve accepted that I have no choice but to wade through the flabby prose and clunky exposition that forms the first two-thirds of each of the books, before the bombs she’s laid finally get to explode in the final third. But that doesn’t mean I can’t whine a bit. Example:

Then, three years ago, on a night very like tonight, the Prime Minister had been alone in his office when the portrait had once again announced the imminent arrival of Fudge, who had burst out of the fireplace, sopping wet and in a state of considerable panic. Before the Prime Minister could ask why he was dripping all over the Axminster, Fudge had started ranting about a prison the Prime Minister had never heard of, a man named “Serious” Black, something that sounded like “Hogwarts,” and a boy called Harry Potter, none of which made the remotest sense to the Prime Minister.

Any halfway-decent editor would be tearing out their hair at a paragraph like that. It’s a mess. There’s never much music in Rowling’s prose, but this is a really good example of a specific, systematic problem of hers: choice of referring expressions.

Doing (machine) natural language generation, one of the hidden but very tricky little problems is deciding on referring expressions. In a narrative that’s anything beyond childishly trivial, there are far too many ways to refer to things. Deciding what’s clear, unambiguous, helpful, but also what reads well, is a hard algorithmic issue. It’s hard enough just to decide when it’s okay to use a pronoun or not, but then there are indexicals, proper names, official titles, bits of metonymy, and so on. Too many choices. The fallback position when referring to people is to just use their name pretty much all the time; it’s clear and unambiguous. Clunky as hell, but style usually comes somewhat behind clarity when it’s a choice between the two in a machine generation context.

People, on the other hand, usually work it out quite quickly. Anyone who writes for a living, particularly so. But not Rowling, it seems. She chooses referring expressions like a not-very-good machine, dropping the same ones with no regard for variation and flow. Look at that paragraph again. There are four references to the ‘Prime Minister’; the last is particularly dissonant. The paragraph comes well into a chapter which is told with a restricted third-person point-of-view: it’s the Prime Minister’s chapter. We’re in his head, and yet Rowling is so unsure of her ability to maintain point-of-view, she seems to feel the need to keep reminding the reader. The effect, besides stumbling prose, is to feel as a reader that point-of-view is constantly shifting and unreliable. Rowling’s restricted algorithm might be allowed a little more choice if she gave the Prime Minister a name, even, but she seems not to want to.

It’s always been so. A perfect example from Goblet of Fire:

Ron and Hermione seemed to have reached an unspoken agreement not to discuss their argument. They were being quite friendly to each other, though oddly formal. Ron and Harry wasted no time in telling Hermione about the conversation they had overheard between Madame Maxime and Hagrid, but Hermione didn’t seem to find the news that Hagrid was a half-giant nearly as shocking as Ron did.

There’s no point-of-view there. There are just names, and way too many of them. It’s not prose; it’s a plot stock-take, an Excel spreadsheet.

Bah. And while I’m grumbling: ‘Department for the Regulation and Control of Magical Creatures’; ‘Department of Magical Law Enforcement’; ‘St. Mungo’s Hospital for Magical Maladies and Injuries’. Clunky, clunky, clunky, and referred to as such, in full, every single time. No-one who lived in her world and actually had to say such things would bother. Not many of us go to places like ‘St. Bartholemew’s Hospital for Human Unpleasantnesses and Unfortunate Occurrences’. We go ‘to the hospital’, or (if we’re a Brit) just ‘to hospital’.

Life’s too bloody short as it is. Growl.