How to get your dream job? What are they looking for?

Student panel discussion held at Interspeech 2006, on September 17, 2006

Recorded by Sanjay Patil for ISCA-SAC
Transcribed by Syxtus Gaál and Sanjay Patil for ISCA-SAC
Reviewed by Naveen Parihar and Sunayana Sitaram for ISCA-SAC
ISCA-SAC Coordination: Agustín Gravano

(Murat Akbacak, President of ISCA-SAC) This is the first panel discussion which is organized by ISCA SAC -- which is ISCA-Student Advisory Committee. Before introducing the panel, of course I should introduce myself first. My name is Murat Akbacak and I am the president of ISCA-SAC and I have been on the ISCA-SAC board for the last three years, and I am having a great time working on very interesting projects, which are not too technical but still very interesting; and it is a great team of students. Before introducing the panel, I will give a quick introduction about what ISCA-SAC is doing. So we have... we are kind of a bridge between the ISCA board and the student community, which is very important because you might have some requests or some questions that you want to ask the ISCA board but you cannot find the people, so we are kind of a student's community's voice when it comes to interacting with ISCA. And (00:01:00) right now we have three board members: myself, Christophe Van Bael is the vice-president, and Agustin Gravano, he is the technical lead and On the ISCA board we have one representative, he is the Chairman, Jean-Francois Bonatre, and in addition to the board members we are working with a group of volunteering students. Basically, when I show this, it looks like a kind of a two level structure. First we want people to start working for ISCA-SAC as volunteering students, then if they are really interested in working with the ISCA-board then can say "OK, I want to become a board member", which of course requires a commitment. Or, he can stay as a volunteering student, and when I say 'as a volunteering student', the difference is that board members have to commit a certain amount of time (00:02:00) , a minimum amount of time to work on student-related issues. But for volunteering students it is 100% volunteering. We have a pool of tasks and if you are a volunteering student we can just pick a task or you can say that I want to work on this piece, I want to do this, I want to finish it myself, or we can create a project group with your friends or you can even propose your own ideas. What we do is interacting with the ISCA-SAC board members; we thought, we are going to submit our proposal first and we are going to help you get the resources, which might be financial or other type of resources from ISCA to make sure that you have everything you need. So this is a great way of contributing to the community, to improve your team skills, and make life easier for us. And our mission is, as I said, we are here to make life easier for students by providing... We have this website which provides user web-access to the information that the student community needs (00:03:00) and we are trying to improve the interaction between researchers in the field. It might be between the students and senior researchers, like having this panel discussion today. It is a great way to interact with company representatives who are really senior people in our area. So, at this point I will just go to the student website that we have launched at Interspeech in Lisbon. It was very preliminary and we have been working on this website for the last one year. I say we, but mainly it's Christophe and Agustín, who did a great job. Now I want Cristophe to give a very quick introduction and summary about our website.

(Christophe Van Bael, Vice-President of ISCA-SAC) Right. So, as already mentioned (00:04:00) the main goal of the website is to promote interaction between students and to promote the interaction between students and the seniors in the field. In addition we also want our website to be a repository with a lot of useful resources for a student, so that we all will be updated about important events, about calls for papers for conferences etc. So what we do in the website, which you can find at that address (http://www.isca-students.org) ... You can sign up and then you come in the user list and in the list you can make a profile of your own, so you can say what university you come from, and more importantly, you can say what your research interests are. Based on those profiles in the future we will be able to make groups of people, to form groups of people based on the same kind of project on related research topics and we hope that this will improve the interaction between people who are (00:05:00) working on the same things. We also have lots of discussion forums, we have discussion forums on the website, through which you can provide feedback on the website. We also have discussion forums on more research related topics ranging from ASR to TTS, through everything that you can imagine that is related to speech. Now, as I have mentioned the website also wants to give easy access to a lots of information and at the moment if you go the website you can find, for example, a listing of research labs in the field, if you are looking for a future place to work, please go there, it might give you an initial idea. We are planning to... we have laid down the infrastructure for a couple of things. We have a listing of text-books in the field which might be of use to most students and PhD students (00:06:00) . There are other resources like job listings and lots and lots of research already. But the most important thing about this website is that we have laid out the basic infrastructure and we hope that this website, in the future, will be your digital aid, the digital aid for students to keep in touch with each other and to keep in touch with what is going on in the field. So therefore we hope that the infrastructure that we have made available will be filled with content by researchers and students in the field. Therefore, we really want you to become a member of the website, help us update the content of the website, put your own thesis on there, put your own speech lab on there, make sure that the website is of use to everybody. Use the tools on the website in order to help you with your research, use the forums etc. and provide us with useful feedback and suggestions on how we can improve the website (00:07:00) . And if you really like this website, and if you really like the idea of helping students in the field and building the bridge between the student population and ISCA you might still consider joining the ISCA Student Advisory Committee so that you can work with other volunteering students or with other board members in order to accomplish goals like developing such a website, and organizing panel discussions like these. Thank you very much.

(Murat Akbacak) As I said, Christophe and Agustín put great effort into this, sometimes they worked full-time. They do their own research but I guess that everybody will appreciate your effort and I guess you will get enough attention from the student community and from the research community to make a very efficient and effective, useful website.

So today the panel is going to be about; (00:08:00) the title is: "How to get your dream job and what are the companies looking for?" So today we invited company representatives, Mazin Gilbert from AT&T , Michiel Bacchiani from Google , Michael Picheny from IBM , Fil Alleva from Microsoft , Jeff Adams from Nuance and Elizabeth Shriberg from SRI International . Right now I will talk about the format of the panel, shortly, if you... There will be three parts, in the first part I will be asking questions to each panelist, in the second part I will be asking questions and panel would be taking turns. If any of the panelists feels that the answer for their company is different, then they can always answer the question. In the third part you will be getting questions, hopefully difficult questions that will give a hard time to our panelists, to make this panel discussion more effective. So I will (00:09:00) ask the panelists around the table to introduce themselves shortly. Mazin?

(Mazin Gilbert, AT&T) I'm Mazin Gilbert from AT&T, from the signal area.

(Murat Akbacak) Maybe just what school you finished, what kind of things you did in the past.

(Mazin Gilbert, AT&T) So I graduated from the University of Liverpool, England, and I finished my PhD at Bell Labs. I guess a lot of us spent some time at Bell Labs at some point in their lives. Then I went to Rutgers University for a couple of years as a research professor and eventually I came back to AT&T. I really like the industry much more than academia. I've been at AT&T Bell Labs, and AT&T Labs now (00:10:00) since 1993.

(Fil Alleva, Microsoft) Good evening. My name's Fil Alleva and I'm the general manager for the speech technology team at Microsoft. I went to Carnegie Mellon so I had to come to this particular Interspeech otherwise they said they wouldn't send me any more students; although I'm glad to be back in Pittsburgh. Let me see, I was... At Carnegie Mellon I've spent 17 years there before making a jump to Industry. I got a BS in mathematics but did not complete a PhD in computer science as that's not required necessarily for the jobs out there in the real world. Although I recommend you to finish; finish, finish, finish! And let me see my group's primarily responsible for the speech technology (00:11:00) that's produced in a wide variety of Microsoft products. We'll talk more about that later.

(Elizabeth Shriberg, SRI International) I'm Liz Shriberg, and I'm at SRI International, in the "Speech Technology and Research Group". I also have an appointment at the International Computer Science Institute. SRI has a large facility in Menlo Park, CA and ICSI is in Berkeley, CA. I did an undergraduate degree at Harvard in linguistics and psychology and a PhD in cognitive psychology at Berkeley. I basically fell into my job. I started working at SRI as a student, started doing more interesting research jobs there, and never left (00:12:00) . One important thing to mention is that SRI, unlike the other places represented here, is a not for profit contract research based institute. It's something in between industry and academia.

(Michael Picheny, IBM) I'm Michael Picheny, and I work at the IBM Research Center. I've been there for 25 years and I know that because I have a watch. As you can see, after being there 25 years, just this year, it's a very nice watch. There were other choices by the way, but this seemed the most practical thing to say. (00:13:01) Well, we could discuss afterwards, what the other options were. To give you some idea of what a career at IBM has looked like over 25 years: when I first started, I started as a regular member of the research staff, doing work on signal processing and acoustic modeling, and then moved, up or down, depending on your point of view of management, to managing a group of people working on acoustic processing. Over the years, as we got more involved in productization, the entire effort at IBM was able to expand significantly to the point where we moved out of sort of basic large vocabulary speech recognition, to broaden our interests, to get involved with other part s of speech recognition such as telephony and embedded speech recognition, other aspects of speech processing, including speaker verification, text-to-speech, speech translation (00:14:00) and also programming-related issues, such as developing programming models for developing speech technology. Right now I manage a group of four activities in speech research, including work on large vocabulary speech recognition, speech-to-speech translation, speaker verification and speech synthesis and the group contains other activities as well, or about 50 people or so. Other activities in which I'm engaged in include professional activities, such as being active in the IEEE. I'm also a member of the ISCA board and I'm here, not only I'm glad to participate in the panel, just to be part of my colleagues, but also to get feedback from the students, to how better ISCA can serve the overall student population.

(Michiel Bacchiani, Google) My Name is Michiel Bacchiani, I'm with Google. I am, unlike others, not even the manager. (00:15:00) I am what they call a research scientist in the speech group; we do have a speech group. I got my PhD from Boston University. Before joining Google I worked at AT&T for quite a while, and for a shorter while at IBM. My expertise is very much acoustic modeling, among others. When it comes to a dream job, I think I'm currently in my dream job. I really appreciate many of the aspects I had in my job at AT&T and IBM, but I think the Google thing has something that is different, that adds something to it and doesn't take many of the good things away that were present at AT&T and IBM. It's a very young and dynamic company and I think it has nothing but beneficial influences, even on old people like me (00:16:00) .

(Jeff Adams, Nuance) I'm Jeff Adams from Nuance. I did not start out in speech. I got a bachelor's and master's degree in mathematics, I went to half a dozen different schools over the years. I spent most of the 80s working for the U.S. government doing cryptology and that's how I got introduced to statistical language modeling. When I left, I wanted to find some place else I could use statistical language modeling and ended up in speech. So in 1995 I started working for a small company called Kurzweil, located near Boston, and over the years my career has... (00:17:00) I've kept the same job now for 11 years, 11 1/2 years, but my employer has either acquired, or been acquired by probably, you know, 30 companies, so I keep getting new business cards. But it was Kurzweil, then Lernout and Hauspie, then ScanSoft and now Nuance -- the main names on that trajectory. Currently I manage a ten-person language modeling R&D team for large-vocabulary transcription and dictation, specifically the Dragon Naturally Speaking product, but we also are the main center of expertise within Nuance for language modeling and vocabulary, statistical language modeling, NLP, so we end up getting involved in a lot of other projects within the company as well (00:18:00) .

(Murat Akbacak) Thank you for the introduction. Now I will ask some of the panel to talk about the research that they are working on. I will ask the panelists to talk about the other speech groups or language groups in their company and how they started. Because when it comes to applying for jobs, students want to like to know: who do you contact first, and if there is a contact with that person in another group that they can ask, look for a job opening. Maybe you can talk about your own research groups. How big is your research group, what percentage are MS vs. PhDs vs. BS? Is it more research oriented or product oriented? (00:19:00) I guess one difficult question for the panel will be: what makes their group different from other groups sitting here on the panel?

(Mazin Gilbert, AT&T) Let me try. So AT&T Research is indeed very small, about 450 researchers, and the lab that I'm in, which is the Voice Over IP & Voice Enabled Services lab, we're about maybe 60 researchers, and there are others who are basically support, computer support and various staff. We pretty much have a very broad responsibility as far as research is concerned at AT&T. We cover anywhere from standards such as speech synthesis, speech recognition, language understanding, speech mining, and in the advent of the merger of AT&T and Cingular later this year, (00:20:00) we are doing a lot of work in independent, multimedia, multi-modal interaction, including web mining and speech mining. So we cover a lot of areas, we've got 60 people running the whole ting. In advantage to that, we manage to interact end-to-end with just a set of people who work in the same lab. So when I want to have someone in Visual, doing something for us in Natural Language, the person we're looking for is just next door, we don't have to go to this organization or this organization and ask. My position itself, I have equally specialized in speech recognition itself, natural language understanding, web mining, speech mining, multimodal and translation. The target of our group is to support research. We go all the way from research to software (00:21:00) . The majority of the people are PhDs, traditionally AT&T Research has hired primarily people with a PhD degree, I'd say in our lab maybe 90%-95% are PhDs, at my group 100% are PhDs.

(Murat Akbacak) And the difficult question, I guess, is what makes AT&T unique from other companies?

(Mazin Gilbert, AT&T) I guess every company goes through ups and downs. I believe I'm in my dream job and sometimes you don't know you have a dream job, it takes a while to figure that one out. What makes our company very different and makes my job very different, we have access to a large volume of customers (00:22:00) . The smallest thing you do, the smallest idea we have, the smallest innovation that we have, make a huge difference on customers. We are a 200,000 employee company, and we have a large number of customers. We get, as I was telling in the tutorial this morning, we get about 15 million calls a month, and we record about 3/4 of that. So we have a lot of interesting challenges, a lot of interesting data, and it's an interesting job to be in.

(Fil Alleva, Microsoft) I think each of the organizations represented at the table have some unique areas of the speech business that they are involved in. So, at Microsoft the speech effort is at a very high level, we can think of it as being divided (00:23:00) between research groups, and we have research groups in Redmond, WA, and Bejing, and between the product groups. The product groups are generally, the group that I'm responsible for; we're responsible for the core speech recognition technology. That's the recognition that happens, that will be part of the rich client desktop operating system, the network based speech recognition, and some early incubations in mobility. That's about 50 people, there's also a platform group, that's about 80 people; plumbing is always more difficult than just the core recognition, as it turns out, and that group is where a lot of our application expertise resides (00:24:02) . They're also just down the hall, so they're very convenient to access to our team. And then we have a number of other related teams around the company, we have 2 natural language teams that handle much of our text processing, and we also have a text-to-speech team in China as well. So there's quite an activity in the speech and language area at the company, as you might imagine, we're not quite as big as AT&T, we're 70,000 people now, seems like a big number to me. I've been at Microsoft 13 years, we've gone from 14,000 to 70,000 in the last 13 years. So it's quite a different place and that's one of the things -- it changes over time (00:25:00) , the place you go to is not always the place you end up at, even if you don't change jobs, as has already been alluded to. Our research team, they're, the people on that team, we have about two 12-person teams approximately. Those teams are exclusively PhDs in the area of machine learning, statistical modeling, and acoustic modeling. My team, in the product group, we have about 1/3 of the people on my team are PhDs, the others are really a broad variety of folks, and in fact not all the people who are PhDs on my team started out in speech recognition. I have one fellow in my team who did his PhD in the area of video encoding, and, as you can imagine (00:26:00) , some work from other fields nicely fits into the speech area. Let me think... what makes us different? Well you know, as I said, we all have different businesses. You know, Microsoft's business is the PC operating system, or a PC operating system, depending on your point of view. But, in any case, comparing the difference between the research groups and the product groups, and just to sense that, you know, when you are in a product group, you're not in engineering necessarily in order to bring your product to market for your customer base, whoever they might be. It's significantly different than the research world. My dream was that I would go out (00:27:00) , and I'd be able to take some technology that I've been working on, the speech community at the time, when I've been in Carnagie Mellon, and bring it to as large a group of people as I possibly could. So that's what makes my job a dream job, finally getting a chance to do that.

(Murat Akbacak) Thank you. Liz mentioned the group. Now maybe we can talk quickly about how many people it has, how many PhDs, and what makes it unique. You've mentioned that it's a non-profit organization?

(Elizabeth Shriberg, SRI International) SRI has offices internationally, but the group working on speech and language has only about 25 researchers. We have a combination of engineers, computer scientists, and linguists. If you want to find out whom to contact, to look what kind of jobs we have open (00:28:00) , you can go to speech.sri.com. Some of the core areas we work in are acoustic and language modeling for large vocabulary speech recognition. We also work on prosodic modeling for many tasks, such as speaker recognition, dialog act modeling, emotion modeling, and deception detection. Language education and translation systems is another large focus area for our group. In terms of hiring, we're pretty picky (00:29:00) . We look for people doing good work, get impressions from their professors and so forth, and we tend to either hire PhDs or people who are going to complete their PhD fairly soon. If you're hired at SRI and want to grow into a senior person, the main determinant of that is getting funding to support yourself and others. (00:30:00) If you have funding, you're pretty free to determine your research and how to lead your projects. That is very different from some companies who may decide, from the top, say, that they don't want to be in speech, and then close a group down or move it to a different topic area.

(Michael Picheny, IBM) Let me try and focus on the perspective of a student. Without putting words in your mouth, when you look for a place to come and interview, and look for a job at, the question you'll have to ask (00:32:00) is what's the focus of the company, why are they working on, in this case, speech, and also, what sort of culture does that company have. Now all my colleagues have, sort of, alluded to these things already, I'll try and specifically pinpoint the couple of comments about IBM, regarding the focus of IBM as a broader company, because it's really the broader focus of the company that in the end is going to determine what you're going to be doing and what sort of work you'll be doing in the long run, apart from the initial projects you started up working on. IBM has been able to make a major shift over the last 10 to 15 years from a company that's focused on selling hardware and software to a company that's now focusing on, what we call business transformation (00:33:00) , but it's basically working very closely with large customers to reengineer their businesses and make them more efficient. There are many ways in which speech is useful in this area, including such obvious things as, for example, making call and contact centers more efficient just by using speech recognition to help automate the processes in these areas. Also it's used to mine information that goes on in call centers and other types of enterprise businesses. From a culture point of view, again, every company, to greater or lesser extent, has its own sort of culture. What we have emphasized at IBM over the years is to have as much as possible, this has gotten even more so over the years rather than less so, people work together in teams to try and achieve a larger goal than any one individual researcher alone may or may not be able to achieve (00:34:00) . We have affected this -- we do tackle large projects, which is very good, very rewarding, on the other hand, the role of the individual researchers is somewhat downplayed as to the overall goals of the team. This doesn't mean we don't publish; we publish plenty, we do a lot of research, but our own focus has been more to emphasize larger teams to tackle large problems of interest to our large customers, and when you're interviewing, you have to keep these things in mind: do you want to be part of the team, or do you want to do your own, individual research and not have to integrate into a broader whole. And decisions like these is what you should keep in mind when you're interviewing and when you're looking at a job and looking at different companies to interview (00:35:00) .

(Michiel Bacchiani, Google) We have a speech group that is now about a year old, and in terms of a perspective, I think I have to agree with Michael, it's important to understand the goal. For Google to have a speech group is I think fairly obvious, but you know, just to spell it out, there is an enormous amount of spoken content out there, and right now it's impossible to search it. So if you transcribe it, you can search it – this is the reason for using speech recognition. In addition there is an interest to talk to Google in some way. The official company model is to make the world's information available to everybody, so this is very much, I figure, a natural extension. So to jump, sort of, towards what makes Google unique compared to other companies is, I think, the key issue is that that there is very little difference made between research and engineering, and this is very much taken to the extreme and I think, as the result of it, there are some additional resources that you wouldn't have as a researcher elsewhere. We look for people that are (00:36:00) hands on, who want to do... to actually implement things. But as a reward you get this wonderful infrastructure that is very accessible, which I think is very different form other places. So you can do your research with incredible resources that actually work quite easily because it's engineered so that it will work well with research. In addition it's pleasant to work with other people in a sense that, since we make very little difference, other people will help doing your work. So, even if you are a research person who's very into the algorithm, it's very nice to have people who are more engineering-oriented, and work with you. I think a good example is the MT, Machine Translation group. There was an evaluation in which the Google team participated. A big part of that is the research group that tells us how to use language modeling in MT evaluations and make very good use of it, but a big part is also the engineering guys who actually know (00:37:00) how to build this on thousands of servers and make the language model accessible. So yes, it's definitely exciting that in research you go out and collaborate work with other people in engineering and you really get to build a very coherent team. All the software is very, very accessible and generally helps you to do things. And it makes you more excited to deliver whatever you have implemented, something that is well tested, fitting the architecture, because other people will start using it. So if you ask me about the group, the speech group consists now of 13 people. There are people asking me about how many PhDs are there versus other degrees, I think we have 11 PhDs. and 13 people in all. And if you ask me how it splits up into official research scientists vs. the engineering part, I'd have to guess... I think it splits up into 10 to 3, 10 researchers vs. 3 engineering types. (00:38:00) As far as you can tell the only real difference is: when you're a researcher they won't be so harsh on you when it comes to programming. So if you have a specialty in speech processing, they're a little bit more lenient, because they think you'll grow up after that. So in terms of broadening yourself within the company, obviously the speech group, when I say 13 people you might think that's a fairly small group, for our company of about 8,000 people. 13 people is considered a very big group. Generally, most of the projects are about 5 or 6 people, that's the general size of how things get done. They very much encourage to have all the freedom to do additional things, we have this thing called 20% time, you can allocate it anyway you want, you can spend one day of your week doing something completely different whatever you want to do. Something you can do is you can look up current projects in a central database, and if you see something that is very cool (00:39:00) , but you don't have the expertise in, you just contact these people and say 'Well, I can do other things here, can I help out somehow?'. This really helps, this way the speech group has contacts with other people who are saying 'I know nothing about speech but I do know how to do this kind of engineering very well, and maybe you guys could use that. Can I help out somehow?' As a result of it you get incredible excitement, there's a lot of young people, and so it becomes very coherent in this way; because of the engineering kids everything works very well together. It's different form other places where you don't have so much engineering help, either by people, or by infrastructure.

(Jeff Adams, Nuance) So Nuance has almost exactly 2000 employees right now and the one difference between Nuance and the other companies at the table is that Nuance only does speech, that's all we do. We're (00:40:00) by far the largest company in the world that does only speech, which means in some ways that your speech career at Nuance is more safe. The company cannot possibly decide that next year we will get out of speech and move to something else, right? The other thing I wanted to say is that, I think not a lot of people understand this, but at Nuance we use our own technology, we don't license it from anyone else, university or otherwise. In fact, other companies, including some at the table, license our technology. One of the things that we've learned over the years being focused on speech, since we're in this precarious situation that (00:41:00) the future of our company depends on the success of our products, that everything is very much product and profit focused. So I can't show up next week and say "You know, I sat through a very interesting seminar today, a tutorial on information retrieval and I'm just gonna do that". I can't just say "I'm gonna do that" because the company's gonna say "Well, wait: how are we gonna make money from that?". Now, I could put together a proposal and take it to the company and say: "Look, I have a proposal for a product and I think we can, you know, after 12 months, we can make the following kind of money, maybe I can get someone to help me put together those numbers and then maybe I could justify working on that. But what I'm trying to say is: we're very product-oriented, products-focused and the organization of the researchers within the company is very much product-oriented. So (00:42:00) I work with a bunch of other people that do Dragon transcription/dictation applications. We have a language modeling team and an acoustic modeling team. And then, separate from that there's another group of researchers that does work on embedded applications and separately there's a group that does TTS, and separately there's a group that does network telephony applications and so forth. We're all separate, there's not a centralized research organization at Nuance. So we have many different research organizations that are doing real research, but focused on individual products and then it's up to us to make sure that we talk to each other, that we do technical presentations and collaborate to whatever extent we can. I think one of the other questions was "What does it mean in terms of how to apply for a job". Well, one of the...

(Murat Akbacak) Actually, we will be talking about it after the next question.

(Jeff Adams, Nuance) (00:43:00) So I should just shut up and... (laughter)

(Murat Akbacak) No, no, no. Maybe a different question, and maybe, try to answer in one sentence. What motivated you to work in the speech area at your company?

(Fil Alleva, Microsoft) Well, I wanted to work on a cool technology that would get me out of bed on a Monday morning.

(Liz) I just wanted to work on spontaneous speech.

(Michael Picheny, IBM) (00:44:00) I wanted to solve the speech recognition problem.

(Michiel Bacchiani, Google) I started working on speech when I was still an undergraduate and my only goal was continuing that, because I liked it.

(Jeff Adams, Nuance) I just wanted to use math to solve interesting problems.

(Murat Akbacak) Now this is a difficult question. Is there anyone on the panel who had a difficult time finding a job, did you feel that there was something missing that you haven't done during your study years, school years? Michael?

(Michael Picheny, IBM) Look, I volunteered for myself. 25 years ago areas (remember the watch?) areas like speech and language did not emphasize as much (00:45:00) statistical modeling as they do now. The only thing I regret at the time is not taking more hardcore math courses and I guess today, I'd say that if you want to work in this area, or language processing, the two things you would want to know in addition to, you know, the speech/language stuff, you want to know your mathematics, and you want to know how to program.

(Murat Akbacak) Michiel, what were the things that you wish you had done during your school years?

(Michiel Bacchiani, Google) During my school years I've made contacts with many people. I'm happy that I kept those contacts because it gave me some options and if you ask me if it's hard to find a job; it's they who told me how to find a job, because whatever your dream was, my contacts would have a job and would be able to reach other contacts so that I was able to stay in the field (00:46:00) .

(Murat Akbacak) What is the most unique thing that helped you get to this level?

(Elizabeth Shriberg, SRI International) I think in my case there are two things: the first one is just being proactive. Be proactive in whatever way you want to be proactive and ask people for opportunities. The second one is being interested and even charming (00:47:00) . Ask the people you are talking with with about what they're doing. There's a lot of psychology, I think, in getting some of these positions :).

(Murat Akbacak) I'd like to ask you the same question: what are the things you wish you would have done in your school years?

(Fil Alleva, Microsoft) I wish I'd gone to graduate school immediately after finishing my undergraduate degree.

(Murat Akbacak) And in terms of applying for jobs?

(Fil Alleva, Microsoft) Applying for jobs? Quite honestly, I've fallen into this area and I fell in love with it as a freshman in college, so the advice I give to you in terms of applying for jobs is: (00:48:00) be in love with what you're doing and have it your aim to accomplish something concrete that makes a difference and then it will all be good.

(Murat Akbacak) Thank you. Michael?

(Michael Picheny, IBM) One of the things I think was one of the most useful things students can do, is to get summer work experience in the field you're doing your master's, certainly your PhD thesis on. It gives you great exposure to companies, it's actually better if you can get a couple of different summer jobs, one at place X, one summer at place Y. What I actually did was I worked during the year while I was doing my PhD thesis, at a small speech recognition company in the Boston area (00:49:00) and you could either do that during the year in summer, but getting some non-academic work experience is extremely valuable in order to get a job. It makes you more attractive to the employer, and it gives you a better idea about what's good and what's bad.

(Murat Akbacak) Micheal, is there something you wish you had done...

(Michiel Bacchiani, Google) I just wanted to agree with that statement and there's line that I borrowed from somebody else, but I think is very true: "be a doer and not a talker".

(Murat Akbacak) Mazin, what about you?

(Mazin Gilbert, AT&T) I wasn't very good at math (laughter). Now, I guess it's all being there, the right time and the right place. I was fortunate that I actually spent a year of my PhD at Bell Labs. I think if you get a chance to spend, what Michael was mentioning, a summer, or a year working in industry, it really makes all different, that really is what matters. Because I really believe that as far as getting jobs (00:50:00) , getting great jobs, what really matters at the end is one critical thing: the relationship; establishing relationships with people in industry. That's what it comes down to. Clearly you have to be smart, clearly you have to know how to program, clearly you have to have the depth in all of these skills, but there are a lot of people like that too. It is that personal relationship that I believe makes a big difference. And I don't know what I didn't do when I was doing my PhD, but I know what I did that was right, that created that personal relationship while I was doing my PhD It really opened up the path and the doors for me, once I finished my PhD I really had no problem getting a job not because I though I was that good, I'm OK, I believe that I'm just an average researcher, I was an average researcher, now I'm a research manager, that opportunity really opened the door for me.

(Murat Akbacak) (00:51:00) What you wish you had done and what would you do to change it, in two sentences (00:51:26)

(Jeff Adams, Nuance) So I have to say that, like someone else said, I just kind of fell into my jobs, and I love them and I have passion for what I'm doing and I don't have regrets. There's nothing that I can look back to and say I wish I had done differently. But one of the things that I guess I would take from that is that I have enthusiasm and passion for whatever projects I'm working on and I love working with other people and I love working on collaborative projects (00:52:00) , and I think that's very important. That was the one thing I wanted to highlight for students. I don't want to see someone come to work for me who's just gonna sit silently in the corner and not work well with the rest of the team. It's very much collaborative, I want to make sure that you'll fit in with the rest of us, that we get along, and that we'll be more than the sum of the individual parts.

(Murat Akbacak) Ok, some of you talked about internships. What number of internships do you have in the speech group, on average?

(Mazin Gilbert, AT&T) We probably have ten.

(Fil Alleva, Microsoft) Approximately ten, but that's across the product and the research group.

(Elizabeth Shriberg, SRI International) (00:53:00) It varies depending on who applies and what projects are funded in the lab at the time. We usually have a few each year.

(Michael Picheny, IBM) Several, it varies from year to year.

(Michiel Bacchiani, Google) This year we have two students in the speech group, we have about 80 students this year, 80 in total. The speech group could have asked for more, but we are just starting up. There'll be a lot more next year.

(Jeff Adams, Nuance) I don't know the exact number from Nuance, it's like 5 or 10 in the Boston area, and another 5 or 10 probably at our other locations in Europe (00:54:00) .

(Murat Akbacak) I want to make a personal comment and I wonder if you'll agree; if you don't please say so. I think that most of the time companies are looking for, when it comes to hiring interns, they are looking for an exact match, because it's a short terms assignment. Is it true?

(Michiel Bacchiani, Google) I think we're very extreme, I'd say if you come in to an engineering position, you don't know what you're gonna be doing.

(Murat Akbacak) The question is: when hiring an intern, are you looking more for programming or technical skills?

(Michiel Bacchiani, Google) For interns we look at where you would fit more or less, and we have some projects where you can use either or both skills.

(Michael Picheny, IBM) On the whole we look for people who are exact matches or close matches, but we also do look for those students who are so strong at least on paper, even though they are not ... trade in speech. If they have really outstanding academic credentials, winners of math Olympiads, programming Olympiads, things like this, if they have no particular interest in speech, but they want to work for IBM, we'll make a shot at trapping them, because over the years, traditionally, a lot of the (00:56:00) good contributions in speech not necessarily will come from people who had a speech background. Sometimes it's good get new blood in the area.

(Fil Alleva, Microsoft) There were 2 things: the exact match and people who've been mentioned doing programming. For our team, we don't have projects, we have a list of projects that we have in mind. But if a student comes to us with his own project in mind, we are really happy to have that person come and work on that with us and try to help to make one of our systems better. As far as programming is concerned, if that student can't support themselves in terms of the programming needs of their project... that's a requirement (00:57:00) .

(Jeff Adams, Nuance) At Nuance, we ... before the intern season starts, we come up with a list, like a shopping list, that we would like the interns to do the following projects. We are looking for people who would be appropriate. We tend to pick projects that rather do not require a lot of expertise. In a lot of cases we'll be happy to pick someone who's not an exact match, if the project doesn't require 3 months of learning before you can start working on it. So it's just a matter of pragmatics, we just look at what we have to do and look who's applying and find people that fit the bill. But it's certainly not the case that we say: OK, we want to do this project and there are only 2 people in the world that will fit so we've got to convince one of them to come.

(Murat Akbacak) (00:58:00) What's the most effective way to submit a job application? A letter, e-mail, phone call?

(Mazin Gilbert, AT&T) We have very formal programs at AT&T for how people submit applications. But I'll be very honest: as far as the research organization's concerned, in my division, in our lab, we don't follow those. In fact, I was mentioning this earlier, it's all about the personal element; it's all about knowing somebody, that's what really matters at the end. So the most successful application is when somebody knows the individual, knows his/her professor, we get contacted that way. We contact professors directly, we get a resume and that's how things happen (00:59:00) . Submitting it through the formal process at AT&T, as far as research is concerned, gets you nowhere.

(Murat Akbacak) Michael, what would be a good time to submit an application? How many months before graduation?

(Michael Picheny, IBM) Well, we're relatively flexible, it depends upon our needs are at a given point in time. Sometimes we need somebody immediately, because we have an immediate hole in one of our projects and we want to fill it. Other times we're in no particular rush, we're just in general trying to deal with staff turnover and always bringing people into the lab. But just as Mazin was saying, personal touches are very important, at least in big companies, because it's very easy to get lost in the shuffle. The thing I would (01:00:00) urge you people to do if you're looking for jobs, there are plenty of people who work at companies. Most of the time you can go up to any one person at a particular company and they'll send you to the right person to talk to, or they'll handle it yourself. Another important thing to keep in mind in looking for a job is that your professor carries a lot of weight. If your professor is somebody who is well known in the speech or language community, and they put in a good word for you, it's sometimes even more effective than your going there yourself. Doing both things at the same time: going up to somebody and having your professor put in a good word for you, assuming the professor is somebody relatively respected in the area, is a pretty good one to punch here.

(Murat Akbacak) Any objections, or any different methods that you would suggest?

(Jeff Adams, Nuance) I want to suggest that... (01:01:00) You said: "When do you apply?" I always suggest a couple of years before graduation is a time to start, and not a formal application, but at these conferences, you should introduce yourselves to the people who work in that field at a company that you might be interested in and over a period of a couple of years develop a relationship. You can make sure that they know your name, they know when you're graduating. It's a long-term thing; it's not just "This week I'm gonna apply for a job that I'll need in 2 months".

(Murat Akbacak) Now I'll be asking Michiel, Liz and Fil if they can mention a list: one, two, three, of the things that you are looking for when you interview a candidate?

(Michiel Bacchiani, Google) (01:02:00) I think what we all look for people who are professional and who are very good at what they're doing.

(Elizabeth Shriberg, SRI International) Ok I printed the list that we are currently using. Essential: committed to team success, interested in both technical and non-technical professional growth, interested in a wide variety of problems, pro-active, takes initiative, solid technical skill in the relevant area, self-motivated, responsible and responsive to others, flexible work on several projects. The other thing is that we're really looking for clearable people, that means you need to be a U.S. citizen. If you don't have all of these characteristics, but you are a U.S. citizen, you do have some advantage.

(Murat Akbacak) (01:03:00) Who are the best candidates and what were the things that most impressed you recently?

(Fil Alleva, Microsoft) A recent candidate came through and they had participated in a project at the university that had a real deliverable and measurable results and they were one of the 2 key people who helped get the right thing done. They were passionate about what they did and they delivered on their commitments.

(Murat Akbacak) Now Michael has the word. (01:04:00) In your 25 years at IBM, who was the candidate that impressed you most and why?

(Michael Picheny, IBM) That's a very interesting question but at a high level, the best... What's interesting, when you've seen a lot of candidates over time is that you realize that sometimes your judgment isn't that great. I can't tell you how many superstars in speech we turned away over the years because, for whatever reason, they didn't make the best impression when they came in, they went on to some place else and they were just spectacular and then you look back and think "God, how did I mess this up!". In general though, the candidates that stand out, the ones that would stand out anywhere, that do great research, they express themselves very well, and they engage when you have (01:05:00) a conversation with them. And the very best candidate from the whole of people who combine all of these traits. But it's dangerous to pigeonhole people in a specific direction. There are people we have now in our group, who, when we interviewed them, or when they started in our group, you never would have hired them in a minute. They came, we took a chance, and it worked out spectacularly. Be wary of too many lists of explicit characteristics, because you never know what's gonna work in a particular situation.

(Jeff Adams, Nuance) Do I get to answer (laughter)? (01:06:00) 3 items? So, no. 1: you should be a strong coder. You should be able to write software. You don't have to write magic code but you would have to write code that someone can read and convert to a product or whatever. No. 2... in fact this is in the reverse order... No. 2 you need to be able to work well on a team, along with others, communicate well, so this is an interpersonal thing, within the team. And no. 3, which I think for many people is most important, you need to be really good at just running experiments, at knowing how to put it together, interpret the results, make sure that you isolate your testing data from training. Just the silly things that you have to pay attention to, understand statistical significance, things like that. I should say an internship is one place you can do that, (01:07:00) but I personally think internships might be a little overrated. You might have done a great job just on university projects or as a research assistant at university or whatever else. You need to be able to demonstrate to us, when you're interviewing, or in e-mails, or however, that you have done all of that, that you can do all of these things. You have to think of how I'm going to show people that I can work well on the team, that I can write code, that I know what I'm doing when I run experiments.

(Murat Akbacak) Now let's go to other to questions from the audience.

(audience) Personally, I'm not a student, but I have a few questions: speaking from the academic side as an ISCA member trying to graduate students, two questions. The first question: most of the companies here (01:08:00) have representations both here in the U.S. and overseas. In particular in the U.S., how many PhDs in speech should the United States graduate, are we graduating enough, not enough, or just right? The second abstract is: a lot of times students have... I've served on PhD committees that aren't speech-oriented, and they want to see a very theoretical contribution. And because of that there's sometimes a question that falls between practical applications vs. theory. Do companies allow this balance to exist? Let's say someone is working in something very theoretical, it may not be productized in the near future, would this count against him?

(Murat Akbacak) I would expect a very short answer from everybody (laughter).

(Mazin Gilbert, AT&T) Somebody asked me a question of how many (01:09:00) PhD students are we graduating in New York every year. I don't know if anybody knows that but I'll be very interested; I made up a number: 100 per year. When somebody looks into that then that's a lot. I don't think we are graduating even half of that. But are we graduating enough? I can only say, to make a point, that just graduating people in speech as a core, is one thing, but the industry is moving very fast with speech, moving into very different areas, not just the usual, sort of traditional call center, I think that was 10-20 years ago. I think academia needs to, or at least students who are interested in his stream, need to look at speech research in a different way then it was 10-20 years ago to be able to catch up with the industry. So when you can present a view to the industry, you sort of present the information, as Michael mentioned, that is right and useful (01:10:00) and important for what industry is looking for.

(audience) What is the role of experimental psychologists and linguists today in the speech technology world?

(Elizabeth Shriberg, SRI International) That's a nice question here; you usually hear engineering type of questions. I may not be the best person to answer because in our laboratory there is not a lot of work in psycholinguistics. People who come in as linguists or psychologists work on things like pronunciation dictionaries (01:11:00) , acoustic modeling, translation systems, language education systems, human subjects experiments, data collection, data annotation, human-computer interaction.

(Jeff Adams, Nuance) We have people doing dialog development and design and things like that. As I've said before, at our company we don't do things that don't have a positive bottom line in the near future. So we're not gonna do abstract research in psycholinguistics.

(Michael Picheny, IBM) Regarding work in psychology anyway, I think it's a real crime (01:12:00) that we don't do more work in basic research in the use of speech as an input and output modality. We all work in the speech area, and if I may say something slightly heretical: it's not immediately obvious if this is good for anything or not. The reason that it is a little hard to tell is partially due to the lack of fundamental work that goes on in defining what interface modalities would really be for people and there's certainly a lot of room for a lot of good research in this area. Unfortunately what you often will find is, when you have a bunch of engineers working together, they tend to look with a certain degree of, well, I shouldn't use the word contempt, but I would use it anyway, on people with a background in psychology. This is really (01:13:00) quite unfortunate because we often can wonder fooling ourselves about the value of what we're doing and I don't think we have enough respect for that area.

(Fil Alleva, Microsoft) I'd just like to chime in there. We have a fairly large user studies group at Microsoft. I will say that we don't spend enough time in the product design cycle to get the feedback we would like to have. But I will add that it's our user studies and they do involve psychologists in some cases. They would come up with suggestions that would take the engineering group years to implement. It's just mind-blowing. So a lot of times we let them go for a couple of months and they come back with about 10 years worth of work for us. We try to prioritize and then get on to the next best important problem (01:14:00) .

(audience) Sometimes people give up their research job for an engineering job. But maybe they'd want to come back to research. So after an engineering job, do you think it's possible to come back to research?

(Mazin Gilbert, AT&T) I'll tell you straight on that it depends on how long you have been out not doing research, but it's hard to get a research position at AT&T if you've been not doing research for more than a couple of years. It's just the reality because (01:15:00) we're comparing you with the others who are applying and who have that skill and it's very hard to compete.

(Michiel Bacchiani, Google) At Google there's very little distinction between those two areas. If you turn out to be this wonderful research fellow and understand the engineering world, if you do those aspects of the job very well, we can find a place for you.

(Jeff Adams, Nuance) At Nuance we do distinguish between researchers and engineers but we work very, very closely together. If you left Nuance to work in engineering for another company and we would know about you and 5 years from now you came and (01:16:00) applied at Nuance we'd have a hard time taking that seriously. But if you were at Nuance doing engineering work we would know you very well. I have an opening on my research team right now and there are a lot of engineers at Nuance that I would hire in a second to bring over into research position if they were willing to do that, but a lot of them are very happy doing exactly what they're doing.

(Michael Picheny, IBM) It's a good point there; in the same company, if you develop a strong reputation as an engineer and you say you want to move into research that's not that difficult to switch, but if you want to move to outside the company, into another company that does do more research or has division research relevant to the company which you're at, you're gonna be at a disadvantage relative to the people who have already established a long track record of publications etc. So it really depends on what your situation is (01:17:00) .

(audience) There are students like me, who are also interested in working in industry, but also enjoy teaching. So is it possible for someone who just entered a company to work as a researcher and have a part-time job teaching at university?

(Michiel Bacchiani, Google) I don't think we have a lot of people who are actually giving courses. We have a lot of internal education right now because we have so much stuff to do. We actually have an education department inside coordinating the courses (01:18:00) .

(Elizabeth Shriberg, SRI International) We're affiliated with Stanford so we've had a number of people who have taught there. Some of our lab members also teach courses abroad, which has been successful in attracting international students.

(Fil Alleva, Microsoft) It's really on an individual case-by-case basis.

(Michael Picheny, IBM) I've heard this comment from a bunch of applicants. I'll just make one general statement you can all choose to ignore or assimilate, or whatever. It's tough to do research, it's also tough to teach. The little experience I had in trying to teach a course in speech recognition at Columbia gave me enormous respect for the academic community, because it's a lot of freaking work to put together a course. It's not like you switch: well, this hour I'm gonna do some research and then (01:19:00) the next hour I'm gonna do some teaching. Don't underestimate the commitment it takes to put together coursework. It's a lot of time and effort.

(Fil Alleva, Microsoft) Also, don't underestimate the amount of effort it takes to establish yourself in a new organization. It's usually about a 150% time commitment at the beginning where you're trying to think about how do I learn this new big organization and how do I learn to be successful there.

(Mazin Gilbert, AT&T) I taught at Princeton last year, I had a couple of people in the lab teaching this semester. We don't encourage people to do that, I just wanna make that clear. I did it not because my boss encouraged me. I did it because I felt it's good to teach in good universities and educate the right talent. Princeton doesn't have a very strong speech and language group (01:20:00) , so I decided to teach there. So we don't encourage, we don't mind if people teach, and a lot of them do that.

(audience) I was wondering what is the best age to start doing research?

(Mazin Gilbert, AT&T) I don't think it's a factor. Research has nothing to do with how old you are.

(Murat Akbacak) (01:21:00) Now I will ask Liz a question that we received over the internet. What policy does your company have for female researchers who want to have children? Is it OK to work from home for a certain period of time?

(Elizabeth Shriberg, SRI International) Well I think this applies to men as well as women. I'd really recommend that you establish yourself on site first, as valuable to your institute or company, and then later you may be able to create a more flexible arrangement if it works for everyone involved. You first need to work closely with people in the office for a while. Then after a time you can be quite productive working remotely for some portion of your time. Many people do that now to avoid commuting, or to cope with child care pickups. It also depends on what type of work you are doing, how much interaction you need, and how well you know your colleagues. I'll be interested to hear what the others on the panel have to say.

(Mazin Gilbert, AT&T) I guess most companies that are doing research have tremendous flexibility. You work from home if you want to. Researchers have more options; managers tend to be at work most the time. So it depends on your position, but the flexibility is obviously there.

(Fil Alleva, Microsoft) I have two people on my team that are currently working in a flexible relation, and we are a product team, shipping products on schedule. So it can be done. They happen to be in the same city, but it's just that (01:25:00) 2 or 3 days a week they are not in the office.

(Michael Picheny, IBM) IBM is a big company and big companies have bad things about them and good things about them. One of the good things about IBM is that it puts a strong emphasis on work/life balance and trying to arrange things in a way that make the employee as comfortable as possible. If that means working from home, if that means working weekends, working part-time, we're pretty flexible. I do want to echo something though, but still I don't want to paint this paradise view of IBM or any other place. I think it was an excellent point. The first few years you are on the job you'd better show up. Nobody knows (01:26:00) who the hell you are and you have to establish yourself as somebody who can produce some results. Afterwards, you can do what you want. But the first few years you'd better be there, you'd better show up, you'd better have people recognize your face and know what you can do. Otherwise you're gonna run into trouble.

(Michiel Bacchiani, Google) At Google you have a lot of flexibility and they put a lot of resources behind you, in a sense. The speech group, for example, is split between California and New York; we also have another person working from Boston. A lot of people like where they live, they don't want to live in the US, but they might like to work for the company, so they do work from home.

(Jeff Adams, Nuance) (01:28:03) I have people who work near Boston who work in our office, who work... half of the people work from home at least one day of the week. But, as everyone else has been saying, if they've been there a long time, they know everybody, they know who to contact. You can't expect a starter job and be working from home half time, that's just isn't in order. I do want to say: a lot of it will depend on your manager and how flexible you are, exactly what your role on the team is. For me personally family is very important and I will kick people out of the office if they're working to many hours on consecutive days; I'll give them a hard time if I think they're ignoring their family by working on weekends.

(Mazin Gilbert, AT&T) We've been talking a lot about what a dream job may be for somebody who wants to go from University to Industry (01:29:00) . What I'd like to hear is what you guys think what would be your dream job when coming to industry. What are the things you value most? I'd like to hear it from somebody who is currently at the university.

(audience) In academia it is possible to do a sabbatical. Is it possible to do a sabbatical in the industry?

(Michael Picheny, IBM) Usually it helps if you work a few years before you (01:30:00) start a sabbatical.

(Murat Akbacak) I'd like to get some feedback form the audience about what their dream job is. Anybody know what their dream job is?

(audience) I'm Jay Chelani from CMU. What I would my dream job to be like (to address the question that you have put out there)... My research right now is on developing dialog system technology for semi-distributors in the developing world. It's a technology that's not that important at the moment. I believe, however, that they can be worked out and this will be something that will gain a future. But if want me to specify right now, if you want to know what everyone wants to do to make money, I can't do that. (01:31:00) But in general -- is it something... that you can come with a viable idea, and still do it, while you're working at a company?

(Michael Picheny, IBM) This particular idea resonates pretty closely with IBM because we are a global company; we're just all over the place. Being able to function in the overall world more smoothly is just a very hard priority for us. This particular thing you're suggesting actually resonates closely with us, because of the nature of our company as trying to figure out how to work in a global environment. If you said you were working on, I don't know, let's say, shrubbery pruning (01:32:00) , and you wanted to do that it might be a harder match for us to justify.

(Mazin Gilbert, AT&T) I just want to emphasize back on my question. What is a dream job, can you ask you if your dream job is to be able to continue the work that you're doing now in industry?

(audience) At the beginning the work I'm doing now, or doing similar things.

(Jeff Adams, Nuance) The question could be twisted to say: I want to do what I want to do, and I want to make a lot of money. (01:33:00) I think there has to be compromise somewhere. You can't just say: I want to do this and I want someone to pay me a lot for it. For example at Nuance we do about 40 languages and we're looking to extend that. So that does resonate with us. But if you come to us and say: "I want to do Cambodian Text-To-Speech" we might not be ready to do that right now. We might be interested in hiring you and that might be something that might come up in a year or two, but maybe never. We can't promise that.

(audience) (01:35:00) If someone has an idea that isn't immediately viable commercially, would you consider developing it, or would you rather say, "Work on this; when it becomes viable, we can talk"?

(Fil Alleva, Microsoft) I think if you look at a lot of people that are in industry and speech, you'll find that they spend a lot of time in academia before they pop out of academia, because they're (01:36:00) just waiting around for the right opportunity in industry to pursue the thing in industry that they are already passionate about. So it really depends on your goals in terms of to be in industry right away, or is it to pursue some dream that you have right now. And if you can't find the right place to pursue the dream in industry, perhaps stay in academia.

(audience) What do you think would the best job candidate for the projects that you have planned for the next 2-3 years, from a technical point of view? (01:37:00)

(Murat Akbacak) Maybe a short answer to that question?

(Mazin Gilbert, AT&T) There's no short answer to that question.

(Michael Picheny, IBM) I agree with Mazin, it's not an easy question, but if we wanted to do something different, it would certainly be nice to have some people, frankly, nothing personal to anybody here, but it would be nice to attract some people from outside the speech area, without a lot of preconceived notions about speech and get a few more of them and attract them into the field. We used to do more of that years ago and I still think it's a good way to spur a new way of looking at problems in our area.

(Elizabeth Shriberg, SRI International) I'm just taking a guess here but I would say a couple of areas come to mind. One is multilingual work and machine translation (01:38:00) ; there is significant government and commercial interest in these areas. Another, which I think Michael mentioned earlier, would be candidates looking to fuse information from speech with information from other sources and modalities.

(audience) (01:39:00) When you're hiring, do you always look for a big match between thesis topic and the job you're offering?

(Jeff Adams, Nuance) If someone comes in and they've got an education that's really appropriate for a certain job, that might influence which manager we'll send the resume to, but ultimately you've got to convince us that you will produce fair results in the future and just having a PhD doesn't do that.

(Elizabeth Shriberg, SRI International) I can't overemphasize letters of recommendation. Try to get one or two letters from a famous person, or somebody who the people you are applying to work with know and respect (01:40:00) . And, they should be outstanding; this is more than just your average letter. We take letters much more seriously than grades or thesis topic.

(Jeff Adams, Nuance) You don't forge letters (laughter).

(Murat Akbacak) I guess we are running out of time. Any final messages for the audience?

(Jeff Adams, Nuance) (01:41:02) I've been hearing some questions, like what is the hot thing we should be studying now? I think if you had asked that question 2 years ago it would have been this, 4 years ago something else. There are many trendy things, they're all important. One of the things that I would recommend that you do is to know about all of them, because it shows us that you can learn techniques quickly, that you can think, that you get a grasp about the topic. I'm not going to be looking for someone who knows X. I'm gonna be looking for someone who can show that they can learn Y and apply it and figure out how it applies to the problem in question. So just learn a lot and how to apply and interpret the results. I guess that's my take home message.

(Murat Akbacak) Michiel, your message?

(Michiel Bacchiani, Google) Be passionate about what you do. (01:42:00) . And I think if you do this well, if you show an in depth understanding of the topic the people who work with you will notice that. Get to know people in a similar field, be interested in what they do and use these contacts.

(Michael Picheny, IBM) In addition to obvious things, as Michiel and Jeff have pointed out, in addition to mastering what you are working on, you have to learn how to communicate it and sell it. Often it's being underestimated how much selling and communication goes on in getting your ideas accepted and successful. The advantage of being at a conference like this (01:43:00) is not hearing technical content of the talks, but learning how to communicate your ideas clearly and concisely to other people.

(Elizabeth Shriberg, SRI International) I guess much of it has already been said. I'd add: be proactive. And if you're not sure what you want, apply for some different options rather than restricting yourself to academia or industry.

(Fil Alleva, Microsoft) I'll join what Michael said; in the way you're communicating your ideas, explain why they're valuable, try to help me figure out what you are talking about and what I should care about your ideas. And if you do, then perhaps I will.

(Mazin Gilbert, AT&T) My fellows on the panel have said all the right things, but I do want to mention one of the key things that the key thing we look for somebody (01:44:00) is they can think out of the box. AT&T, and I'm sure other companies are no different, the competition is very fierce outside, we are in the area of communication and we are getting very heavily and strongly in the area of entertainment. And the quality of people that we're getting, communication skills an everything that has been said already, that's expected. What's as important is thinking out of the box, people who can challenge the status quo. They may not know the answers to everything, but people who have determination to do that.

(Murat Akbacak) I would like to thank our panellists Jeff Adams, Michiel Bacchiani, Michael Picheny, Elizabeth Shriberg, Fill Alleva and Mazin Gilbert for the answers that they have given today. I hope that this was a really useful panel discussion for the students. (01:45:00)