What do you think?
Rate this book
Understanding Natural Language
Understanding Natural Language
191 pages, Hardcover
Published January 1, 1972
2 people are currently reading
94 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 3 of 3 reviews
June 10, 2009
I read this book when I was 19, and it completely knocked me over. Computers could understand English sentences and respond to them in an apparently intelligent way. I hadn't realized it was possible.
"I want to work on that too!" I decided. And I still am.
____________________________________________
As Ursula Le Guin says, true voyage is return. Since posting this review, I've had several conversations with people who've asked what system would be today's equivalent of Winograd's SHRDLU, which could take part in a conversation about moving toy blocks around on a simulated table-top. I found it disturbing that I couldn't come up with anything. Having discussed the question with many friends and colleagues in the research world, I feel more and more strongly that something has gone wrong. We are no longer trying to build these ambitious, inspiring AI systems, which make an honest attempt to emulate deep human understanding. Instead, we're focusing on narrow technical issues, and implementing them inside much simpler applications, which do things like airline flight reservations, restaurant bookings, and calendars. If this were just a stepping stone to the ambitious systems, there would be no problem. But that doesn't appear to be true. For the last 20 years, pretty much all the work has been of this technical kind. We never get to the cool stuff, and people are starting to forget it was ever there.
Well... no use complaining, do something about it! Together with a few like-minded colleagues, I've started work on an Open Source initiative that we're calling "Back to the Future". The first stage will be SHRD2, a reconstruction of Winograd's classic SHRDLU, but done using modern tools, and with spoken input. (We also hope to have a real robot). We already have some of it working. When we're far enough along, we will start on a new application, where the goal is to be able to play Bridge and talk about it. Bridge is a difficult game, which involves thinking about knowledge, belief, communications, deception and planning. It's challenging, but we want to do something that will stretch us.
We're building the SHRD2 and Bridge applications using the Regulus platform that we developed at NASA and elsewhere. I'm posting about progress on my Regulus blog.
"I want to work on that too!" I decided. And I still am.
____________________________________________
As Ursula Le Guin says, true voyage is return. Since posting this review, I've had several conversations with people who've asked what system would be today's equivalent of Winograd's SHRDLU, which could take part in a conversation about moving toy blocks around on a simulated table-top. I found it disturbing that I couldn't come up with anything. Having discussed the question with many friends and colleagues in the research world, I feel more and more strongly that something has gone wrong. We are no longer trying to build these ambitious, inspiring AI systems, which make an honest attempt to emulate deep human understanding. Instead, we're focusing on narrow technical issues, and implementing them inside much simpler applications, which do things like airline flight reservations, restaurant bookings, and calendars. If this were just a stepping stone to the ambitious systems, there would be no problem. But that doesn't appear to be true. For the last 20 years, pretty much all the work has been of this technical kind. We never get to the cool stuff, and people are starting to forget it was ever there.
Well... no use complaining, do something about it! Together with a few like-minded colleagues, I've started work on an Open Source initiative that we're calling "Back to the Future". The first stage will be SHRD2, a reconstruction of Winograd's classic SHRDLU, but done using modern tools, and with spoken input. (We also hope to have a real robot). We already have some of it working. When we're far enough along, we will start on a new application, where the goal is to be able to play Bridge and talk about it. Bridge is a difficult game, which involves thinking about knowledge, belief, communications, deception and planning. It's challenging, but we want to do something that will stretch us.
We're building the SHRD2 and Bridge applications using the Regulus platform that we developed at NASA and elsewhere. I'm posting about progress on my Regulus blog.
Read
March 20, 2017This text a very precious artifact. All the secular problems, but seen from a different perspective.
August 27, 2014
SHRDLU still is the most important Natural Language Interface to date. This book is personal, thought-provoking and exciting. It is not possible to read it and not pick several new ideas.
Displaying 1 - 3 of 3 reviews