Question Understanding and Siri

This week saw the introduction of Apple Computer's new iPhone 4S. Not all of the reviews have been kind. For those expecting the next hardware answer to the Android-based competition, the new phone fell short of the blogosphere's rumored expectations. However, Steve Job's final creation (announced only a day before his death) may prove to offer far more than its largely unchanged exterior reveals. This device introduces Siri-- a personal assistant that, according to Apple, "understands what you say." Okay, Apple is not above some hyperbole. Recall in 1984 that the Macintosh was introduced with the phrase "insanely great!" Well, it was. Nonetheless, Siri (whose parentage includes SRI, Inc. and DARPA) appears to be much more than some clever parlor trick. It seems to break the "understanding" barrier with some remarkable first steps.

What makes Siri an important accomplishment is not just its ability to perform speech recognition and retrieve answers (though it represents significant advancement in both areas). Siri attempts to understand the questions asked. It takes context into consideration. Here is Apple's example.

So when you ask “Any good burger joints around here?” Siri will reply “I found a number of burger restaurants near you.” Then you can say “Hmm. How about tacos?” Siri remembers that you just asked about restaurants, so it will look for Mexican restaurants in the neighborhood.

Apple Computer, Inc. (http://www.apple.com/iphone/features/siri.html)

Siri is not the first computer program to perform such feats of human-like cognitive processing. But, Siri may be the first to place such capabilities in the palm of your hand at consumer prices. It's not yet Watson-on-a-chip, but it's getting there faster than I thought. Form factor and pricing aside, this is simply remarkable--insanely great! In order to appreciate the accomplishment it might be useful to consider what an achievement question comprehension is in humans.

At a very early age we all learn the power of asking questions. In it's simplest form we learn to repeatedly (and annoyingly) ask "Why?" to almost any assertion made by grownups. We test every claim and seek deeper understandings on everything. It's fascinating to watch unless you’re the parent caught in your child's "why loop." Children also learn quite early how to understand questions (and avoid giving answers). "Who drew pictures on the wall in crayon?" Complete understanding by the guilty child followed by silence.

Of the two skills-- question asking and question understanding-- we seem to do a better job at the latter. When we ask-- "When will gas prices drop to reasonable levels?" we're asking a loaded question. What's "reasonable"? Some may think that it is unreasonable for prices to drop from a environmental viewpoint. Others might long for the days of 25 cent per gallon gasoline. The question sounds simple enough until you unpack it and discover that it is either too vague or contains an assumption (i.e. that prices will drop) or both.  On the other hand, when someone says-- "Do you know what time it is?"-- we typically understand that they are asking for the current time of day and not merely whether we happen to know the current time.

I don't want to diminish the importance of asking good questions. Good questions--carefully crafted questions-- can guide our investigations of what's true. They can define the very starting point and scope of inquiry. If you pay much attention to political discourse these days you might think that good questions are in short supply. It's a concern for a free and democratic society. More attention needs to be paid to learning to ask good questions.  Good questions are clearly very important. However, I'd like to take a moment to concentrate on question comprehension for I think it is vital to seeking and judging answers.

Once a question is asked, how is it understood? What is this simple yet amazing process? We don't even think about it. We just do it.  And, we have been doing it since we were very young. Perhaps this is, in part, why it has been so difficult for computer scientists to develop computers that can understand questions. We didn't think it was all that difficult an activity. After all, we've been understanding questions since we were asked a simple  object identification question such as --"What's this?" (as we are shown our favorite toy).

It was this problem of question understanding that Dr. Wendy G. Lehnert tackled in her 1977 doctoral thesis, "The Process of Question Answering - A Computer Simulation of Cognition." Yale University (1977). Before a human or computer can retrieve/provide an answer to a question, that question must be "understood". Lehnert developed a taxonomy of the many structural types of questions and crafted an interesting approach to conceptually analyzing the question in order to provide an appropriate answer. The notion of context is key.

Many aspects of this new application have me excited. Research and development in areas of computational linguistics, and what some are calling "computational knowledge" (see http://www.wolframalpha.com/ for one example ), are stepping out of the labs and into our practical lives. In time, I believe that we will view human thought processes differently. We may be better able to anticipate and correct for failures in our cognitive processing. This work brings into focus some of the complexities of human cognition. It may also help us see inward in a new way to better understand some of the ways we fail to understand each other.

Alright, I'm asking way too much of a device I just ordered and have yet to hold (pre-orders began today--Friday, October 7, 2011). My enthusiasm for this new application is clouding my thinking. Guilty as charged. It is, after all, just a phone. Nothing stays insanely great forever. Still, this represents a wonderful step in the evolution of our understanding of ourselves. And, that is always insanely great.