Google's AI system inspired by romance novels, writes poetryMay 18(AZINS) The art of conversation is one that not even all humans fully master. However, following feedback from users that Google's artificial intelligence system, dubbed "Parsey McParseface," came across as stilted and unnatural, despite (or possibly due in part to) its perfect grammar, engineers decided to teach the bot a few literary tricks. Lessons came in the form of 2,685 romance novels, fed into the neural network over several months.

Why romance novels, and how did this lead to Parsey McParseface's first poetry collection?

In an interview with Buzzfeed News, leading Google software engineer Andrew Dai explained that the simple plot structure meant that the romance novel genre made a particularly effective learning tool: "Girl falls in love with boy, boy falls in love with a different girl. Romance tragedy." If the plot remains essentially the same, the words do not. This is how the system can use the structure as a support to then learn more about the intricate nuances within the language itself.

The team, led by Dai and Oriol Vinyals, then gave the system a challenge. They provided two sentences, a beginning and an end, and asked McParseface to bridge the two concepts using a maximum of 13 additional lines. The results look a lot like contemporary poetry, albeit rather melancholy:

"there is no one else in the world.

there is no one else in sight.

they were the only ones who mattered.

they were the only ones left.

he had to be with me.

she had to be with him.

i had to do this.

i wanted to kill him.

i started to cry.

i turned to him."

Why is Google putting time and research into its AI system's conversational capabilities? In an unpublished paper, engineers explain the importance of intelligent parsing (understanding and describing the syntax of a sentence) for the automatic extraction of information, translation and other applications of Natural Language Understanding. Parsey McParseface has been hailed by developers as the most "accurate such [algorithm learning] model in the world." Poetry-wise, it still has some way to go.