When computers speak, hear, and understand

While HAL quickly turned into a malignant representation of technology , we have also been treated to futuristic glimpses of other more helpful and reliable computers. For example, in the Star Trek TV series and movies about space adventure in the 23rd century (and beyond), the computer is omnipresent on the Enterprise, answering requests for data in any domain and always knowing whether a request is directed to it or to another entity in the room. The common trait of these advanced systems—besides having sufficient " intelligence " to navigate the galaxy, negotiate cosmic-scale peace treaties, and analyze medical data well beyond the current abilities of our best medical experts—is they all communicate through speech. Imagine Captain Kirk striding onto the bridge and asking Mr. Spock to pass him the keyboard so he can query the computer. Whether Star Trek, Star Wars, or 2001, the assumption is future keyboards and displays will no longer be required or even desirable for human-computer interaction. One could imagine the ultimate goal for conversational interfaces would be to build a computer system that understands speech and communicates as well as humans do [2]. After all, when two humans engage in a conversation, we employ communication techniques and skills unthinkable for today's digital machines. These techniques allow both the speaker and the listener to work together to understand the speaker's meaning and for both parties to establish common ground for the conversation [1]. Clearly, that common ground is built on past conversations, the immediate surroundings and context of the current conversation, as well as the cultural background of each party. While incorporating this knowledge into conversational interaction with a machine would be impossible today, we can expect future computers to achieve human levels of speech and even surpass them. Well-educated humans usually have a single area of expertise in which the extent of knowledge runs deep but can also refer to facts and historical context (more superficially) in other domains. Thus, a computer scientist can speak at length about memory parity, caching, and canonical representations and also make reference to Leonardo da Vinci's influence to his peers, along with how many minutes it takes to roast a leg of lamb. Yet this same person would most likely be unable to diagnose a medical condition , give the Latin name for every type of flower, or cite the specifics of a trade treaty between two countries. …