Technology started talking back when Siri was introduced. But "conversational agents" like Siri can't yet carry a conversation. While machine-learning systems can follow pre-scripted repertoires and adapt to user patterns, other elements of language—such as sarcasm—don't yet compute.
"To be effective, computers might need to acquire some social inferences," said Marilyn Walker, UC Santa Cruz computer science professor and leader of the Natural Language and Dialogue Systems Lab in the Baskin School of Engineering.
To create those capabilities, Walker mines online social sites to develop statistical algorithms that capture the flow of natural conversations. A team of her graduate students, the SlugBots, were picked to compete in Amazon's $2.5 million Alexa Prize Challenge to develop a "socialbot" that could turn the company's voice-controlled devices into chatterboxes.
Improving conversational algorithms also advances fundamental science, noted Pranav Anand, associate professor of linguistics. Working with Walker and UC Santa Cruz cognitive psychologists Jean Fox Tree and Steve Whittaker, he brings a "rich intuition about language" to develop these new algorithms. Ultimately Anand hopes their work will improve studies of language itself.