Can we teach computers to understand what we really mean?
UC Santa Cruz researchers are working to make that a reality, with help from linguistic cues culled from the online world, say Pranav Anand, associate professor of linguistics, and Marilyn Walker, professor and graduate program director of computer science and director of UC Santa Cruz’s natural language and dialogue systems lab.
Virtual assistants such as Siri aren’t trained to “get” irony or make common-sense connections. Telling them, “It’s dark” doesn’t lead them to turn the lights on.
Making computers more adept at understanding the contextual richness of conversations and commands will require close cooperation between computer scientists and linguists, Walker said. Only by joining forces can they sprinkle computers with “conversational fairy dust.”
Humans have strong social behaviors and instincts that don’t stop when they talk to Siri and other virtual assistants. By tapping a rich vein of online data from blogs and tweets, programmers can learn how people talk to each other, and use that to design better agents.
For instance, you could “inch up the sarcasm meter” of a sentence such as, “You already know that” if you added the word “But” to the beginning and a patronizing term such as “dearie” to the end.
“This is very exciting,” Anand said. “It means your tweets about Monday’s debates and your latest camping vacation are contributing to the scientific study of language. That research will lead us to the next generation of virtual assistants.”