Reviews for Generating Natural Language Descriptions with Integrated Text and Examples

The average rating for Generating Natural Language Descriptions with Integrated Text and Examples based on 2 reviews is 4 stars.

Review # 1 was written on 2016-01-04 00:00:00 Jello Paul I doubt anyone else is going to review this book, so I thought I would. The problem with reviewing one's own book is that it's impossible to get any distance to it. So some of the time I'll gloat, and some of the time I'll whine, but I'm afraid that's inevitable. You were warned. I'm not going to talk about the technical aspects: there are a couple of reviews about those in academic journals. Instead, I'll first give a quick non-technical overview, and then present the back-story. So, what we are trying to do here is create a toolkit that lets you build speech-enabled software. By that, I mean that you talk to the computer, and it does something more than just transcribe the words. For example, it might translate what you said into another language, answer a question, or carry out a command. To do something like this, the machine has both to recognize the words, and also understand what they mean. I'm not going to get into philosophical issues about what it means to "understand". If the machine responds in an appropriate fashion most of the time, then as far as I'm concerned it's understanding you. Philosophers have spent a lot of time thinking about this, and our work has direct roots in the philosophy of language. Switching to a practical point of view, if you want to carry out a conversation with a computer there are two extreme positions you can adopt. At one end, most phone-based commercial systems (e.g. the ones that some airlines use to let you make reservations) have the machine ask questions which the human responds to. So it asks you where you want to go, where you are leaving from, what day you want to travel, etc. This works fairly well, but you are hardly having a conversation at all; you're not doing much more than filling in a form. It doesn't work too badly for airline reservations, but imagine trying to control a robot this way! A typical conversation might be something like this: ROBOT: What should I do? You can say "move", "pick up object", "check status"... HUMAN: Move. ROBOT: Should I move right, left, straight on, or at an angle? HUMAN: Left. ROBOT: How far should I go? Say a number of metres. HUMAN: Four metres. ROBOT: Moving. What should I do? You can say... HUMAN: Check status. ROBOT: What status would you like to check? You can say "tires", "battery"... HUMAN: Battery. ROBOT: My battery is at 58% charge. What should I do... At the other end, you'd really prefer it if the human was allowed to say anything they wanted to the machine, and have it respond appropriately; our hypothetical human/robot conversation might now look like this: HUMAN: Well, so now you need to move right --- I mean left --- about, ah, about three or four metres. ROBOT: I'll move a bit more than three metres left, correct? HUMAN: Good. Now, how much charge do you have left on your battery? ROBOT: It's at 58%. Unfortunately, we are still a long way from being able to build HAL 9000; attempts to do this kind of thing produce systems which fall flat on their faces most of the time. So what we've done is to aim for a compromise, which we call controlled language user-initiative. We let the human take the initiative in the conversation, but only express themselves within a restricted range of language. Continuing the example, we'd end up with a dialogue like this one: HUMAN: Robot, move three point five metres to the left. ROBOT: Moving three point five meters left. HUMAN: Check battery level. ROBOT: Battery level is 58%. Summing up, what we're doing is building a toolkit for constructing this kind of spoken dialogue system; in more familiar terms, if you think of the dialogue system as a website, then we want to build the analog of Dreamweaver. The way we do it is based on formal theories of grammar. I won't say anything more about that, though, but go on to describe how the project happened. When we started off in early 2001, the two main people involved were me and my colleague Beth Ann; she was working at NASA Ames Research Center, and I was at a start-up in Cambridge, England. NASA already had a platform for building grammar-based speech-enabled systems, but they weren't happy with it; there were technical problems, and messy issues to do with intellectual property rights. The IP issues were such that we couldn't use it at all at the start-up, so we decided we would go back to the drawing board, and build a new one. The plan was to attack on two fronts. I would start building the new system at the start-up; at the same time, Beth Ann would use the old platform to check out the feasibility of the solutions we thought we could see to some of the technical problems. We decided that the new platform, which we christened Regulus, would be Open Source, in order to avoid all the IP hassle. Things started to go wrong at once. The start-up had agreed in principle to the plan of making things Open Source, but as soon as we had some concrete software they started to wonder whether it might be better to keep it proprietary. There was an endless series of conversations between the start-up's lawyer and his opposite number at NASA. It didn't help that neither one had any idea of what the software did, and we also had a feeling that the start-up's lawyer was stalling on purpose. In the end, Beth Ann, who was visiting Europe for a conference, decided to visit Cambridge so that she could have a face-to-face meeting. We will never know what the start-up's management was originally planning to do at this meeting, since it was scheduled for late afternoon European time on September 11, 2001. Beth Ann went out after lunch to see a movie. When she came back I had the unpleasant task of telling her that the World Trade Center had been attacked, and thousands of Americans were dead. My boss offered to postpone the meeting, but Beth Ann said that she thought they might as well have it anyway; it would take her mind off things. Under the circumstances, it was impossible for my boss to adopt a tough negotiating position, and he formally agreed to make the software Open Source. It was clear, however, that my management now thought the project was less interesting, and I got switched to spending my time on other tasks. The next summer, things looked up. The NASA group had started working on Clarissa, a new speech-enabled application for the Space Station. The astronauts were keen (it was originally their idea), and Regulus seemed like it could be the right tool for the job. They arranged to invite me over for three months as a visiting scientist. We made a lot of progress with both Regulus and Clarissa. A couple of days before I was due to leave, I was called in to chat with the department's director. He said he'd been hearing good things about Regulus; could I tell him briefly what it was about, and in particular how it differed from the platform they'd previously been using? One of our key goals was software reusability, and I had a sudden inspiration. The old platform, I said, was a Saturn V; the new one was a Space Shuttle. This turned out to be the right answer, and he offered me a one-year contract to continue on the project. I resigned from my job at the start-up and went to work at NASA. It all seemed to be well on track until, on February 1 2003, we got the horrible news that the Columbia had crashed. (Amazing how many bad things happened on W's watch, isn't it?) Suddenly, NASA was in crisis mode. Even though the astronauts loved our project, and met with us regularly, it was unclear that there would be time to run an experiment on the Station. We kept plugging away, and Clarissa was close to ready; but every time we asked when they would actually be able to run it, the date was postponed. At one point, we asked if we could see the current priority list. They showed it to us, and we were both pleased and shocked to discover that we were two places behind "boost the Station into a higher orbit". We didn't ask that question again. Funding was however looking increasingly uncertain, since there was massive fallout from the Columbia disaster, and morale was terrible. I began talking with a group at Geneva University that I had previously worked with, and we put in a proposal to use Regulus to do medical speech translation. It seemed prudent to have a backup plan. We also started writing this book. Finally, on June 27 2005, we got our software tested on the Station. One of the best moments of my life, and definitely my 15 minutes of fame. We were at a conference in Ann Arbor, Michigan; there were people calling from the networks to interview us, and we got featured in newspaper stories in over 20 countries. We enjoyed it while it lasted; needless to say, two weeks later no one could remember who we were. The NASA funding position was also becoming very precarious indeed, and it was clear there was no money to continue the Clarissa project beyond the initial proof-of-concept. A large-scale test would have cost at least $50M, which was out of the question. Far more critical projects were being axed. Luckily, I had by now lined up Swiss funding. The remaining NASA money was just sufficient to let us finish the book, which was published in June 2006. Our local bookstore in Sunnyvale were absolute stars, and threw a signing party for us. Academics never have signing parties! Thank you Leigh, Jordan and other friends, that was so nice of you. I'm now in Switzerland, further developing a software platform that was largely funded by the US taxpayer. It's Open Source, so people in the US are able to use it as much as anyone else, but it still seems odd; there is, as of early Feb 2009, no US funding at all for this work. However, Obama says in The Audacity of Hope that he wants to make research a priority. We'll see what happens. If things go well, maybe we'll publish a sequel in a couple of years! Our working title is The Regulus Cookbook.

Review # 2 was written on 2018-03-17 00:00:00 Kristina Cox Despite the fact that this is in my Read shelf, I have never read this book. But I want to talk about its author, Manny Rayner. Manny is one of the greatest people I have ever known. I know for a fact that once, as he was walking in a poor part of the city, he passed a homeless man who was evidently suffering from the DTs. Manny sat with this man, on the filthy, piss-stinking scrub under a bridge, for seven (7) hours, cradling the unfortunate's head in his arms, until the fit passed. Afterwards, the man wept and tried to press into Manny's hands a few grubby coins, which needless to say Manny would not accept. So no, I haven't read Putting Linguistics into Speech Recognition: The Regulus Grammar Compiler, but how could any book about the Regulus Grammar Compiler fail to be wonderful, written by an author with Manny's heaven-drunk soul? [PLEASE FLAG THIS REVIEW, AS I AM DOING, FOR BEING ABOUT AUTHOR BEHAVIOR, AND NOT ABOUT THE BOOK.] [YOU MAY ALSO WANT TO REPORT MY BOOKSHELF AUTHORS-BEHAVING-FANTASTICALLY, AS IT CLEARLY IS AS PERNICIOUS AS ANY SHELF CALLED AUTHORS-BEHAVING-BADLY.]