The Stanford University researcher discusses climbing metaphorical mountains and rattling cages in his year as a CAS fellow.
Dan Flickinger, a senior researcher at Stanford University, is used to going it alone.
He operates in a small corner of the field of computational linguistics ('There are only about 18 of us on the planet,' he joked). On the occasions when he is able to collaborate with colleagues on a scholarly article, he does so with 'carefully selected co-authors.'
In most cases, Flickinger said, collaboration comes in the form of partnerships with startups and established language technology companies. He hands them a 'black box' that powers their technology, and as long as it works, everyone is happy, and no one needs to question Flickinger's approach.
How does a scholar like that react when placed in an environment in which his academic peers want to pick that 'box' apart and critique its individual parts?
Flickinger arrived at CAS in August. He has been one of only a few constants in a CAS project that has involved a rotating cast of experts in computational syntax and semantics.
We sat down with Flickinger toward the end of his stay at CAS. Read the full interview, which has been lightly edited for clarity, below.
Flickinger was one of more than a dozen scholars who participated in the CAS project SynSem: From Form to Meaning - Integrating Linguistics and Computing. Read our exit interview with the project leaders here.
Q&A - Dan Flickinger
- You’re working at Stanford, you’re researching, you’re minding your own business, and then suddenly you can an email or a call from someone saying ‘Hey, do you want to spend a year of your life in Norway?’ What were your thoughts? Could you walk me through that part of the process?
Dan Flickinger: In my case, that invitation wasn’t such an out-of-the blue surprise because I had lived here back in 2005/06 on a project funded by the University of Oslo with Stephan [Oepen]. That was a pretty good two years. When I came back to the US, I chatted with Stephan occasionally about how I might conspire to come back here for a while. When he and Dag [Trygve Truslew Haug] decided to put in this application for CAS, I was cheerful about being included.
The timing was also rather convenient, because I’m in not exactly a gap year but a breathing space between two phases of the research we’re doing at Stanford. When I get back I have six months to see if we can engineer additional funding for the next phase of that work, which is over in the space of applications in the education universe. I’m kind of hoping to get more funding and halfway hoping it doesn’t work out so I can do something more interesting than keep on doing the same old thing. So this was a nice year’s break to give me some breathing space to figure out what to do next.
- So either way you’ll be fine?
Flickinger: There’s a former student of mine who’s just finishing his Ph.D. with another former student up in Seattle. He works in our field. He’s interested in starting a research institute in the little town of Bellingham, where he lives with his wife. And he can afford it. So he wants me to come direct that -- starting the centre, which would also be very fun. A little speculative, a little unpredictable, a little out of my comfort zone -- but, still, it would be fun.
- Do you like for things to be a little unpredictable?
Flickinger: I think I’m schizophrenic on that front. There’s a large part of what I do that seems to resist any of that shaking up. I’m a creature of habit. Those bigger shifts, like when I moved to Oslo 12 years ago, that was very fun. It turned out to be a really good experience. Moving from industry -- where I was for 11 years -- to Stanford 20 years ago was also a really good move. Completely a shakeup, financially disastrous, but really fun. There hasn’t been a downside to those. But I resist it.
- Tell me a little about how the project has developed over the year, and your expectations coming in. How do they line up with where you are now?
Flickinger: Probably like everybody who shows up for a year, it seemed like a year was a very large amount of time. A subset of the group -- Stephan and Mary Dalrymple and Emily [Bender], they were the main people that I conspired with about the ambitions of the project.
We had set out what in retrospect looked like relatively ambitious goals; there’s the selfish goal that I had talked about in the interview from before, which is mostly my own -- the push I had been giving on this specific construction of English. That small piece of the work has gone kind of according to plan, so I’m pretty happy with how that’s turning out, but that hasn’t involved so much interaction with the others other than picking their brains every now and then.
In the end, we started climbing this mountain, and we’re just not sure it’s actually there. Maybe it’s not our mountain.
These collaborative efforts [on syntax] have just gone more slowly than we thought. And I think, looking back, it’s what we should have expected. We look at a problem and say ‘I estimate the size of that problem to be about half of Mt. Everest,’ and then we start climbing that thing and discover it’s just a foothill for a much larger set of mountains, each of which is really interesting. We have to pick which one of those we’re going to start climbing as well. Our ambitions have grown a little as we have looked at the problem. That’s the work with Stephan and Emily.
The work with Mary -- and also with Dag, I guess -- is on a related but different project involving the word ‘and’ -- coordination. Figuring out the interpretation of sentences like ‘Five linguists and philosophers entered the room.’ Were there 10 people that walked into the room or five people who happened to be multitasking in linguistics and philosophy? If you say ‘Five scholars and gentlemen entered the room,’ you can imagine that there’s just five people, because they might be both scholars and gentlemen. But if you have a very low opinion of scholars, then you might also assume that there were 10 people, five of the one category, five of the other. Both readings are available if you set up the context right. We want our formal ways of computing the meaning of the sentence from the parts to give us the prediction of the availability of both of those interpretations. That turned out to be just hellishly hard when we started looking at the range of variation for Norwegian and English. We found some interesting differences between Polish, Norwegian, and English that we hadn’t expected.
In the end, we started climbing this mountain, and we’re just not sure it’s actually there. Maybe it’s not our mountain. Maybe it’s not something we ought to be solving. This might be something that is done with common sense reasoning or something that really doesn’t have that much to do with language.
We did find these language differences like in ‘In invited a doctor and maybe a nurse to come visit me in the hospital.’ You might think that I invited two people, and I’m not sure about the category of that second person. Both Norwegian and English give you both of those interpretations, but what Norwegian doesn’t like and English does is ‘I’m going to talk to maybe a doctor this afternoon.’ In Norwegian apparently the equivalent sentence sounds pretty weird. We’d have to say something like ‘I’m going to maybe talk to a doctor this afternoon.’ You can’t put the ‘maybe’ right inside that to-phrase. In English that’s perfectly fine. It doesn’t mean you’re not sure you’re going to talk to somebody; it means you’re not sure of the category of the person you’re about to talk to. That means in the English interpretation of ‘I’ll invite a doctor and maybe a nurse to the party,’ I can get a reading where I’m going to invite two people, and I’m just not sure whether that second person is going to be a nurse. Apparently in Norwegian that meaning is hard to get. You’re either going to invite one person or two to the party, but there’s no question about whether you should invite a second person or whether it’s going to be a nurse. This is a striking difference between the two languages.
- And we’re going to take a little bit of credit -- we as in CAS. we like to say that we give fellows the time to figure these things out. Maybe you hit a dead end, but at least you walked down the alley!
Flickinger: We are not in any way unhappy about having come up against some surprises in that search. There’s still a reasonably interesting paper emerging from the five or six of us. What we are struggling with is what aspect of that safari do we report on? We don’t have the sharp results we thought we did, but we think we know a lot more about what people so far have looked at in this field, and we have turned up some contrasting data that hasn’t been observed before. So we may not have solved the problem, but we think we found a more interesting aspect of that problem that still needs a solution.
One aspect of the year that was in fact assisted by this was the changing of the guard -- I was here the whole year, but there were these people who came and went. Mary going away, and with the loss of that, we had some breathing space to talk about the problem with a few other people. It wasn’t until she went away and Gosse Bouma came with his Dutch expertise, Adam Przepiórkowski with his Polish expertise, and Helge [Dyvik] coming and going with his Norwegian expertise that we could sort of begin to see a different shape of that puzzle.
I think that the slow movement of the year on this problem was probably what we needed, because if we had been under a deadline of solving it in a month or two, we would have said ‘This is not a problem that will have a solution emerging -- let’s do something else.’ And I think we wouldn’t have illuminated a puzzle that is of pretty broad interest -- this whole issue about how coordination works.
I think we’ll go back to our daily grind with a clearer sense of how to communicate more efficiently, which I think probably makes our work more visible.
Looking at the second of the three problems -- there’s one I’m solving selfishly; the one I worked on with Mary and Dag, which has a completely different shape in terms of where we are at the end of the year; and then there’s this work with Stephan and Emily, which is somewhere in between. It’s in this relatively small corner of the field, … and we’re trying to figure out how to communicate what we’re doing to our ‘cousins’ like Mary or Dag or someone who also very much shares the same ground assumptions while working with a different formalism. I think we’re making very nice progress on it, but that’s been slower. That’s one aspect where I’m a little bit disappointed.
I think I’m as comfortable as Mary, Emily, and Stephan are in that we have been seduced by this other, rich collection of people who keep coming and going with fresh energy saying ‘Wait, let’s go back and look at that other problem again, because I think I have a better idea about it.’ We have been very willing to maybe do that kind of squirrel response. ‘Wait, this is a nice, shiny idea!’ We’ve been willing to set aside our own agenda to work on this problem with our scarcer colleagues whose time we have.
Emily, Stephan, and I are going to keep on working on this into eternity. We don’t require the luxury of CAS to make forward progress on that work. We have our own infrastructure set up to make that go on.
- Speaking of that -- when you all go back to your respective institutions, how do you see the work continuing, formally and informally?
Flickinger: I think that’s related to the first question you asked. We have this fresh impetus for the trajectory I was on before CAS to continue. And I think Stephan, Emily, and I have a much more well-defined problem space. We know what we want to do. We’ve talked with people who have come and gone, and they’ve said ‘Yeah, you need to explain yourselves in these respects in order for us to understand how to do it.’
The workshops where we have brought in people from outside, like the one [a few] weeks ago, were extremely useful, because we found out other groups are doing something very similar to us. We just weren’t explaining ourselves in a language they could grasp. And these are smart people. We just put some hindrances in the way of communication.
I think we’ll go back to our daily grind with a clearer sense of how to communicate more efficiently, which I think probably makes our work more visible. It’s going to be easier for us to secure graduate students in other departments to come work on our problem, because we will not put as much of an obstacle in way of them grabbing our stuff and making forward progress with us.
The other thing that’s going to be different when I go back because of the CAS experience is that I now have much stronger alliances with people like Dag and Mary who I always thought of as distant cousins doing something slightly weird, because they worked in another framework. They’re actually doing very similar stuff. And I think that’s also true for people we only brought in kind of peripherally. [A scholar] who only came in for this one workshop, his work is much more closely aligned with what we’re working on than I had realised before.
I think this CAS experience enlarged the circle of people I think of now as close allies, not just a vaguely sympathetic audience. I can perhaps put them to work on aspects of my problems. So selfishly I have a larger workforce that I can deploy maybe on the problems I’m interested in. That’s naturally good. That group of people in the world is pretty small.
- Do you have a lot of experience with collaborative work?
Flickinger: I have experience working with business people, with engineers -- that’s most of what I’ve done. And that’s partly necessary because of the narrowness of the research assumptions I’ve adopted within my field. I can’t easily find collaborators within my field. But I can readily find people who say ‘Well, what you’re doing has applicability as applications we’re interested in. Help us figure out how to make them work.’
I’ve spent quite a bit more time working with industry partners, with startups and big corporations, where I have to explain what I’m doing in some practical terms. But that isn’t so threatening to me as a scientist because that leaves me free to work in my little space. That treat what I’m doing as a kind of black box. As long as I make that black box be harnessed by their engine, they don’t care, they don’t mind what I’m doing inside it.
Dag, Mary, Gosse, and Adam are very much interested in what I’m doing inside that black box and disagree with the way I’ve done the gears or installed the carburetor. ‘You’re doing fuel injection? Are you kidding me? You should be switching to electric power!’ They want to question the core options, which are more wrenching for what I do on a day-to-day basis. They want me to change deeper assumptions, and of course that’s scarier because I don’t know what the consequences of that are.
I can easily work with a psycholinguist or a philosopher, because I know we don’t share that many common assumptions, and so we can find bridges that are not scary for any of us. But if I’m going to work with Mary, I have to go right back to one of the things that I crucially have relied on and say ‘Maybe I can give that up. I can give up that cherished assumption,’ because if I do, we can work together rather efficiently -- and maybe she’s right after all, which she usually is. But those are of course scarier from a researcher point of view because I don’t know the consequences of giving up that foundational assumption.
I don’t usually get to sit in a room with even another grammarian, let alone two. There are only about 18 of us on the planet, and I’ve had two of them in the building with me.
So no, on that kind of scientific collaboration I’ve done quite a bit less. Most of my papers that I’ve worked on over the years are jointly authored, but with carefully selected co-authors. So the kind of collaboration that CAS has engineered, that is a more novel experience for me. To deliberately put myself into a context where almost any aspect of what I do is subject to a challenge by people who know at least as much as I do about the topic is somewhat scary.
I think that’s maybe underplayed in this way in which CAS presents its invitations to scientists. ‘Oh, you’re coming into this nice protective space where you can work for a while.’ In fact, it’s a relatively risky perspective, because you’re putting yourself into a setting -- in one of these research groups -- with people who know way too much about what you’re doing and doubt you made wise choices in every aspect of what you’re working on.
I think we have rattled the cages maybe a little uncomfortably this year. Adam and I … have jostled each other maybe a little more than some others, because in the end what he wants to work on is something I slightly suspect -- and he also would think some of these ground assumptions that I make are just flawed at their root.
- Are you still going to be friends by the end of the year?
Flickinger: Yeah. For some reason Helge and I share ground assumptions about this precision aspect of things. I’m not sure I can pin down why of the triple of Helge, Adam, and me I find more tension at the professional level with what Adam is doing than with what Helge is doing. But even that element of the experience is an interesting one. Why do I react in a certain way to one of my grammarian friends and not to the other?
In terms of our scientific approaches, I find Helge’s personal biases more aligned with my own than Adam’s. That’s an interesting experience. I don’t usually get to sit in a room with even another grammarian, let alone two. There are only about 18 of us on the planet, and I’ve had two of them in the building with me. That’s a good turnout.