Lit-bots rewrite the age-old story of authorship

Around the time Tiffany Chan started training a bot to write like a little-known Victorian writer for her English master’s thesis, Harry Potter came up in the news.

A company had trained a bot, also known as a web robot, to write in the style of JK Rowling after feeding it all seven Harry Potter tales. The resulting chapter, which included memorable scenes such as Ron eating Hermione’s family, caught the media’s attention.

Chan approved of the imaginative project. Bot writing experiments have been around for a long time, she says. And Chan, an English student with no formal computer science training, was in the midst of her own attempt to deconstruct the very text other students would be studying.

Messing with literature

“In some circles of literary studies, the text is like a sacred thing and you’re not supposed to mess with it,” she says. “I like messing with things so it’s appropriate.”

Chan was training her bot to write like Grant Allen, a nineteenth-century writer who produced more than 30 novels, including his own technology focused work, The Type-Writer Girl, which, among other things, explored the influence of typewriters on authorship and the production of literature.

“What is the role of the machine in relation to a human author?” Chan asks. “I see a connection between how the type-writer impacted writing in the nineteenth-century and what bots are doing in the twenty-first.”

How to create a lit-bot

Chan used an artificial neural network, modelled loosely on the form and behaviour of human brains, to generate text in Allen’s style. To do that, she had to “train” the software by feeding the bot 31 of Allen’s books, which Chan found on the free Project Gutenberg site.

The bot, in turn, analyzed patterns in the text and constructed a model of that data. It was then able to generate new text imitating those patterns. The results are often absurd—and sometimes accidentally profound. Chan says when she was starting out, she would “feed” the bot phrases. When asked about the meaning of life, for example, the bot answered as follows:

“The meaning of life is so understood in silence. She was a living of the first moment in the sea before her soft complete instinct to her own dear old American window."

Chan was amused. “The first sentence seems almost vaguely deep,” she says. “And then you get random affection for a window.”

A nestful catch

In the end, Chan’s bot wrote 10 paragraphs in Allen’s style although she says it was mostly gobbledygook. Take the following for instance.

“I was 22 and without employment,” bot-Allen wrote. “I would not say by this that I was without occupation. But the idea of his father was over, the wind was a nestful catch.”

Chan may not have succeeded in producing text worthy of Allen’s name, but she did impress her thesis supervisor, English professor Jentery Sayers, who says Chan had reimagined the form and content of graduate work.

“I could be mistaken, but I believe Tiffany is the first student in Canada to build and submit a literature bot for their graduating master’s project,” Sayers says. “You can not only read her argument but also ‘execute’ or ‘run’ her research. It’s incredible.”

Collaborative, not antagonistic relationship

Chan, meanwhile, doesn’t think bots will be replacing human writers of literature any time soon.

“What stands out to me, both in my own work and in the Harry Potter bot, is that we find bot writing delightful precisely becauseit's so absurd and different from what we would expect a human to say,” she says. “I think of it more like a collaboration between human artist and machine rather than an antagonistic relationship.”

Chan has since created a Twitter bot called @TaySEliot, which combines Taylor Swift lyrics with T.S. Eliot poetry—her funniest bot project yet. All of the experimentation has paid off. UVic Libraries recently hired Chan to work in its Digital Scholarship Commons, where she teaches workshops and tinkers with new technologies.