SEATTLE, May 3, 2000 — Microsoft let its technology and some of its best minds do the talking this week at the Language Technology Joint Conference in Seattle. Along with offering demonstrations of the company’s machine learning technology, computer scientists from Microsoft Research (MSR) spoke about their work in natural language processing (NLP) and related areas. On Wednesday, Rick Rashid, MSR senior vice president, gave a keynote speech titled
“The Future – It isn’t what it used to be”
. In his speech, Rashid discussed the dramatic advances in current technology and their potential impact on human-computer interaction.
Established in 1991, MSR conducts basic and applied research in computer science and software engineering. During the past year, MSR has made progress in many research areas and contributed to numerous Microsoft products. Meanwhile, Chief Technology Officer Nathan Myhrvold, who began a sabbatical last July, recently announced that he won’t return to Microsoft full-time. Instead, Myhrvold will serve as a special advisor to Microsoft Chairman and Chief Software Architect Bill Gates, while continuing to pursue his various interests in biotechnology, paleontology and philanthropy. Rashid, who helped start MSR with Myhrvold, spoke recently with PressPass about these changes, natural language processing, the unique role MSR plays at Microsoft, and his vision of technology in the future.
PressPass: Why is Microsoft taking an interest in natural language processing?
Rashid : When you think about a computer interacting with the user, the user wants to tell the system what he wants to happen. The natural way to do that is through spoken language. Today, we mostly can’t do that because systems aren’t smart enough to understand what you’ve said. Part of what we’ve been doing is building an infrastructure that lets us provide better communication between the computer and the user.
Microsoft Research uses natural language processing in computer search technology to analyze the syntax of what a person types. We can match what provides a better connection between what they are asking and what is in the document. We have begun using these technologies to build new kinds of translation systems that translate the overall meaning of a sentence, rather than just the individual words or superficial aspects of what is said.
PressPass: What is the most exciting possibility under investigation by NLP researchers?
Rashid : One of the really exciting capabilities would be to have a computer scan a book and then answer questions about what it read. That’s something we think is feasible in the next few years. It involves analyzing the grammatical structure, or parsing, of the underlying text and creating a semantic network of what’s there, both of which we know how to do. Then we must match the semantic representation of a query against the semantic network that was built from the text.
We’ve used Microsoft’s NLP technology with Microsoft’s Encarta Encyclopedia to ask simple questions such as
“Who killed Abraham Lincoln?”
Even though Encarta might not have the phrase
“killed Abraham Lincoln,”
there may be a statement that John Wilkes Booth murdered Abraham Lincoln or that Abraham Lincoln was shot by someone. You could then match up the knowledge. I think we are beginning to see some light at the end of the tunnel in this area.
PressPass: What points did you want to stress in your keynote at this week’s Language Technology Joint Conference?
Rashid : My theme was that we are at an inflection point in terms of capabilities. We’re standing on the cusp of change that is occurring much faster than most people actually realize. In the next two to three years, the opportunities for really dramatic new applications and new capabilities will be there. The NLP field has an opportunity to participate in these changes, perhaps more so than many scientists would think.
There are three technologies where Moore’s Law — the expectation that technology doubles in power every 18 months — no longer appears to apply. With new PC-graphics cards, you are seeing a 100-fold increase in performance over three years. The transition to terabyte disc drives, which will happen over the next two or three years, will allow people to store all of the text and speech around them and manage that information online. Also, bandwidth for wide-area networks, computing and wireless is increasing at a factor of two every six months. For example, a team from the University of Washington, Microsoft, Qwest Communications, and the Information Sciences Institute of University of Southern California broke the land-speed record for transcontinental Internet traffic last year. They moved nearly a gigabyte — one billion bits of information — per second over 5,000 kilometers between two standard PC desktops running Windows 2000.
PressPass: How many researchers work at MSR? How broad are their backgrounds?
Rashid : We have a little more than 500 full-time researchers and additional visiting researchers and others. We cover more than 30 different research areas — everything from operating systems to natural language processing, computerization, machine learning, signal processing, and networking. There also are researchers who do theoretical computer science, discreet mathematics and statistical physics.
Our goal was to assemble a group sufficiently diverse that the researchers could impact each other. A lot of times thinking in one area will evolve, and then someone in another area will say
“Wow, I can use that same approach and same technique in my area.”
There is a lot of synergy and a lot of serendipity.
PressPass: What makes Microsoft Research different from research groups at other companies?
Rashid : The model we have for Microsoft Research is the model I carried from Carnegie Mellon University, where I was a professor for 12 years. Organizationally, our structure is relatively flat with minimal bureaucracy. We don’t have specific budgets for individual projects. We try to encourage communication and interaction between groups. We have a very open environment where people are encouraged to publish in refereed, scientific journals. No one reviews our papers before they are sent to publication. In most respects, it’s much more like a university than a corporate research lab.
PressPass: One concern sometimes expressed about corporate research is that it doesn’t always produce practical results. How does Microsoft ensure technology transfer occurs?
Rashid : We take seriously the task of rapidly moving research technologies that “pan out” into practice. A program management team works with the researchers and product groups to help find the right way to move the technologies we develop into products. One good example of that is how Microsoft integrated NLP into Office 97. We created a product group specifically to take the results of our research and put it into a product.
The alternative for a company that can’t figure out how to absorb technology from its research lab is to spin it off into separate businesses. We have found we don’t have to do that to be successful.
PressPass: What has been Microsoft Research’s greatest accomplishment over the past year?
Rashid : We have had a lot of successes. Last May, our researchers authored 25 percent of the papers featured at one of the major programming conferences. I don’t think that’s been done before by a single organization. The work we did in conjunction with Microsoft’s eBook team to make type clearer for handheld Pocket PCs also was a tremendous accomplishment. It required the talents of a number of different researchers, including people in graphics, signal processing and theoretical mathematics. We pulled those experts together in order to get that result.
Our new research lab in China, which is about 18 months old, is receiving recognition from Chinese academics and the Chinese National Science Foundation. That has been a real positive experience for us. It’s a great thing for Microsoft. It’s a great thing for the Chinese. It is a great thing for the research community.
PressPass: Does Microsoft plan to expand MSR?
Rashid : We will probably continue to grow if we can continue to hire the kind of people we have hired in the past, and if we continue to create solutions that can be used within the company and in the field. We are not going to grow just for fun. Our goal is to be the most effective research organization in the world, not the largest.
PressPass: What hard problems remain in computer technology?
Rashid : Almost all the problems we started with 50 years ago are still around. There have been tremendous accomplishments. But you still can’t point a computer at a scene and have it recognize what is there. We still can’t do a decent job of language translation. We still can’t get computers to read books and answer questions. We still can’t recognize speech well enough that my grandmother could use it without feeling the system was stupid.
For someone who isn’t plugged into the field, there appears to be a whirlwind of new technology. But for me, being in the middle of it, it all seems like it is still in slow motion. I want the technology that is five years away because I have problems I want to solve now.
PressPass: How has MSR functioned this past year without Nathan Myhrvold? How has research been affected?
Rashid : Nathan began his leave of absence last July 1 to spend more time with his family and to focus on personal interests such as paleontology. But I’ve been running the day-to-day activities of the research group from the beginning, so research hasn’t really been impacted at all. Nathan is a tremendous person, and his intellect is something that is sorely missed. He is a tremendous source of great ideas and also great person to interact with. But in terms of the research groups, hiring and everything else, we’ve been going full speed.
PressPass: How will Myhrvold’s decision not to return to Microsoft full time affect MSR?
Rashid : People will miss having him around, but I don’t think it is going to have a negative impact. We’ve got a mature organization. If it were a smaller group with only a few strong people, you’d notice it. We’ve got a group with 40 or 50 top, senior researchers in their fields who have the ability to drive their research areas. We also have a strong management team with respected researchers such as Dan Ling who was recently promoted to vice president. He was a co-inventor of the video-RAM dynamic memory and is widely respected for his groundbreaking work in computer science.
PressPass: Let’s zoom forward to 2020? How different will computers be? What will they do that they can’t do now?
Rashid : Realistically, nobody has a clue. There are too many things that could happen between now and then. But the current technology curve tells us computers will be 10,000 times more powerful.
Will computers be able to create a visual scene on a display that looks as good as what you see out your window? Absolutely. Will you be able to create something visually equivalent to what you see when you look at the real world? The answer in 20 years is certainly
It may be possible in 10 years. Will I have the equivalent of a piece of paper or a book that digitally changes so I never need another book or piece of paper? Certainly in 20 years, maybe in five. Devices like that exist in the laboratory already.