We've already seen how awkward
computers can be when they try to speak like humans, but researchers from North Carolina State and Georgia Tech have now developed a program that could make it easier to show them how it's done. Their approach, outlined in a recently published paper, would allow developers to create natural language generation (NLG) systems twice as fast as currently possible. NLG technology is used in a wide array of applications (including video games and customer service centers), but producing these systems has traditionally required developers to enter massive amounts of data, vocabulary and templates -- rules that computers use to develop coherent sentences. Lead author Karthik Narayan and his team, however, have created a program capable of learning how to use these templates on its own, thereby requiring developers to input only basic information about any given topic of conversation. As it learns how to speak, the software can also make automatic suggestions about which information should be added to its database, based on the conversation at hand. Narayan and his colleagues will present their study at this year's Artificial Intelligence and Interactive Digital Entertainment conference in October, but you can dig through it for yourself, at the link below.