 |
| Professor Julia Hirschberg and WayCoolWomen.com's Misheel |
Professor Hirschberg: I was initially a historian with a PhD in 16th-century Mexican social history. I began my career with a heavy teaching load at Smith College. My husband was teaching at the University of Pennsylvania, so we were commuting a lot, which was very tiring.
In my historical research on Mexico, I had a lot of information from notarized records like birth and sales certificates. I was trying to map this data, so a fellow professor in the Computer Science department at UMass Amherst told me that my project was a Computer Science/ Artificial Intelligence (AI) problem. One of the professors had a graduate student who gave me some code, and I played with the code over the summer so I could learn to build the software myself. I had a lot of fun, so I got books on Computer Science from friends who told me that historians don't make much money. They suggested I go back to school for a master's degree and a job in Computer Science.
My husband was still teaching at the University of Pennsylvania, which had a good Computer Science department, so I took a leave of absence from teaching at Smith to take some courses at U Penn. I loved the courses. I didn't want to give up my job, so I went back to teaching while also taking math and logic courses from my fellow professors.
I thought about becoming a database person because some of my friends worked with databases and were doing well. But a course I took in AI that focused on Natural Language Processing (NLP) completely transformed my life. I knew that's what I wanted to do, so I got a PhD in Computer Science, focusing on language processing. I still like history, and when I travel, I always go to historical places, but I really like what I'm doing now.
It wasn't easy because going back to school after being a professor is strange, and I was a young junior professor. I stood up for myself when I had to, and it helped that the program at that time was open to people coming from different disciplines, and women were in charge of the natural language processing group.
After friends pointed out that I sometimes used a particular way of talking called the rise-fall intonation contour when I was giving presentations on my work on questions-answering, I wanted to find out more. I had a good friend who was a linguistics PhD student and together we looked at lots of examples and data, writing a paper together that was published in the best linguistics journal. People in the natural language group at Bell Labs invited us to talk to them about our work on intonation. On the way home from the presentation, we both said that Bell Labs was the most fascinating place we'd ever been to, and I wanted to work there, even though I hadn't really done that kind of work in the past.
I started to work on prosody [patterns of stress and intonation], and it's been the focus of my work ever since. Text-to-speech often sounds robotic and boring, so it needs proper intonation because intonation can show meaning. For example, in English, a question needs to end with rising intonation, while a wh- question or statement usually has falling intonation. Most people pick this up without thinking about it when they learn a language. But in text-to-speech it's a big issue because the text comes out of all kinds of speech, like news broadcasts and more. It's very challenging and I was the main person working on that for about ten years, so people would come and we'd work with different languages. It's still kind of an unsolved problem.
Misheel: Can you tell me more about the work you did in text-to-speech synthesis? Where and how is this technology used?
Professor Hirschberg: I did a lot of human subjects experiments to figure out different ways to predict prosody from text for text-to-speech synthesis. Then in the mid-1990s Bell Labs needed a new person to build a new HCI group, so they asked me since I had done a lot of relevant experiments earlier.
When Bell Labs started to change, a good friend at a university called me and asked if I was interested in a senior role. So I came to Columbia as a professor. One thing that has changed the field of spoken language processing more recently occurred when the voice recognition assistant Siri was released as an iPhone app in 2011. People started to realize that text to speech and speech recognition were important. But for me, it's been around for a long time. It's extremely useful.
Misheel: What does the group hope to find with its research, and what can be done with the findings?
Professor Hirschberg: We're working on text to speech and low-resource languages like Amharic because companies are not incentivized to create text to speech systems in these languages, even if there may be millions of people using them. For example, there are lots of Arabic dialects, but also Modern Standard Arabic, so there's no incentive for people to build systems which cost millions of dollars. For text to speech, you have to have cleanly recorded data in a sound booth and lots of it by a person who doesn't mind speaking in a very particular way all the time. This person with a professional voice will charge a lot. What we're doing now is figuring out how to take data that might be radio news or audiobooks and selecting parts of that data that are the most like how you would do a text to speech recording. So if you can put those together in a system, it can be done pretty much for free, so our goal is to enable people to build their own text to speech systems in their own languages. A lot of people who speak a particular language don't read it, so it would be good for them to be able to get information from the Internet.
Misheel: What is your favourite part of your career?
Professor Hirschberg: The research is great, the students are wonderful and the people you meet in the NLP community means that you have thousands of friends.
Misheel: What is the most challenging aspect?
Professor Hirschberg: Challenges can be having to raise money for research, you have to write a lot of grants. But I was really lucky my first year, and I've been able to share my expertise and learn a lot in leadership roles like chairing the Computer Science department at Columbia University since 2012.
Misheel: How did you encourage other women to begin a career in sciences and technology?
Professor Hirschberg: My friends and I were trying to make Bell Labs more diverse so we started talking to people in the senior leadership about the need to build a more diverse community. We finally convinced the vice president of research at the Labs that there were too few women at the Labs in general and also in leadership positions. One thing we mentioned to a fellow on the PhD recruiting committee (which was then all male) was that if the Labs wanted to hire more women the recruiters should meet with women's groups when they visited. He told us he thought the men would feel uncomfortable being the only male in the room. We were kind of astonished and said "but how do you think we feel every day?" and started laughing. The end result was, two of us were made department heads and both of us got on the recruiting committee and did exactly that -- met with women's groups at the universities we visited. Gradually things did change for the better at the Labs. It's one of the things I'm most proud of.
Send your thoughts about this interview or topic under 'Post a Comment'
