24 minute read
Janyce Wiebe at EACL 2014 (April 29) in Gothenburg, Sweden. Photo source: Graeme Hirst.
(With valuable contributions from Diane Litman, Claire Cardie, Ellen Riloff, Owen Rambow, Marilyn Walker, Graeme Hirst, Rebecca Bruce, Rebecca Hwa, Swapna Somasundaran, Lingjia Deng, and Rev. Dr. Matthew Bell)
Remembering Jan Wiebe
Janyce Wiebe – known by her friends and colleagues as “Jan” – was a Professor of Computer Science and the former director of the Intelligent Systems Program at University of Pittsburgh, and a fellow of the Association for Computational Linguistics. She was an expert in the areas of opinion analysis, discourse processing, pragmatics, and word-sense disambiguation. She was one of the first to carry out research on methodology in text annotation, just in time for the rapid rise in the need for text annotation for supervised learning methods. She was a pioneer in the research area of “subjectivity analysis” – recognizing and interpreting expressions of opinions and sentiments in text, to support NLP applications such as question answering, information extraction, text categorization, and summarization.
Jan received her PhD in Computer Science from the State University of New York at Buffalo, and later was a postdoctoral fellow at the University of Toronto. Jan’s early research, in her dissertation at the State University of New York, Buffalo (1985–89) and postdoctoral years at the University of Toronto (1989–92), was on algorithms for tracking point of view in narrative. It was a unique bridge between philosophical approaches to belief contexts and empirical studies of how people actually comprehend discourse structure. In later years, she applied some of these ideas in methods for searching on the Web for expressions of opinion, and in distinguishing expressions of opinion from expressions of fact.
In 1992, Jan assumed an Assistant Professor position at New Mexico State University, where she stayed until 2000, when she moved to become an Associate Professor, and later Professor at the University of Pittsburgh. Shortly after joining Pitt, Jan received a secondary faculty appointment in the Intelligent Systems Program (ISP), an interdisciplinary PhD-granting program focused on applied Artificial Intelligence. Soon thereafter, Jan in addition took on a major ISP administrative and leadership role, serving as co-Director from 2002-2004, Director from 2004-2010, and co-Director from 2010-2016. The main goals of the ISP are to foster interdisciplinary and transdisciplinary AI research across a wide variety of domains in a manner that is informed by the problems within those domains and that contributes to their solution. It is based on the idea that solving difficult, important, real problems in specific fields of study (domains) is a key way to achieve fundamental advances in AI. A number of Jan’s PhD students graduated from the ISP rather than from the CS department. Jan’s research in the domain of biomedical informatics was also done in collaboration with colleagues from the ISP.
Jan had a long and successful career, and was involved in many professional communities. These roles included ACL Program Co-Chair, NAACL Program Chair, NAACL Executive Board member, Transactions of the ACL Action Editor, Computational Linguistics and Language Resources and Evaluation Editorial Board member, AAAI Workshop Co-Chair, ACM Special Interest Group on Artificial Intelligence (SIGART) Vice-Chair, and ACM-SIGART/AAAI Doctoral Consortium Chair. In 2015, she was named a fellow of the ACL “for seminal contributions to Subjectivity and Sentiment analysis, Discourse Processing, and Lexical Semantics.”
Research contributions
Jan had a deep and abiding interest in subjective language. Her research on subjective language and sentiment was at the origin of a large amount of work on sentiment analysis in natural language processing. Back in 1999, she developed a probabilistic classifier to distinguish between subjective and objective sentences and created a manually annotated data set for the evaluation of this NLP task (which is still available on her web site!). In the early 2000s, Jan collaborated with Ellen Riloff to explore the use of information extraction techniques to learn subjective language from text corpora. They developed bootstrapping methods to learn subjective nouns and to learn multi-word subjective expressions using lexico-syntactic patterns. The learned subjective nouns were ultimately included in the subjectivity lexicon that was released with the OpinionFinder system. This research subsequently led to work on learning subjective and objective sentence classifiers using unannotated texts, as well as work showing that subjectivity classification can improve the precision of event extraction by filtering information found in subjective contexts. In another line of related work, they defined a subsumption hierarchy to capture the representational overlap and behavior of different types of subjectivity features and showed that feature selection using the hierarchy can improve opinion classification performance.
Complementing her work on subjectivity classification was Jan’s interest in understanding how language expresses private states (such as opinions, beliefs, evaluations, attitudes, emotion) , and how we can automatically analyze language to determine private states. She pioneered the study of what would come to be known as fine-grained sentiment (or opinion) analysis which tries to identify subjective expressions, their source (the opinion holder), target (what the opinion is about) and attributes (for example, strength and polarity). As was her trademark, Jan developed her representational framework (with her students and collaborators) into a set of annotation guidelines, ultimately producing the widely used MPQA corpus, which had a tremendous impact on the field. The framework allowed for early study of quite complex linguistic structures including nested attributions in which the attitude of person or entity X was expressed via a second entity Y. Along the same lines, Jan (with her students and collaborators) were among the first to develop a computational account of stance (a pro/con attitude toward a specific topic or claim) as well as to study multilingual aspects of subjectivity.
In her most recent work, Jan used a combination of lexical semantics and implicature analysis to go beyond sentiment that the text expresses explicitly (High healthcare costs are bad). For example, in the sentence The reform would lower skyrocketing healthcare costs, a word-based sentiment system would discover negative sentiment because of skyrocketing. Maybe a sentiment system can determine that the target of the negative sentiment is the health care costs. But in addition, if we dislike something, then lowering it is good – so we can infer positive sentiment towards the lowering event. And finally, we like agents of events we like, in this case the reform. So, even though we have a single sentiment word which is negative, we can infer two instances of positive sentiment as well! These inferences are in fact implicatures, since they can be defeated. Like other implicatures, they play an important role in natural language understanding. Jan and her students proposed both a rule-based system, and an approach based on machine learning. Jan felt passionately about this line of research, and many of her ideas remain to be discovered by the broader community.
Jan has also made significant contributions to discourse analysis, spanning many decades. Her early work in the 80s, which explored perspectives in narratives, argued for looking at subjectivity beyond the sentence boundary. Jan called this the “subjective context”, and these comprised of one or more subjective sentences attributed to the same subjective character. She presented an algorithm in the form of a discourse process to recognize the same. The MPQA corpus in early 2000s was a result of refining and solidifying these ideas. In late 2000s, Jan and her students presented the schemes and systems for other types of discourse-level interpretations, where clear opinions in one part of the discourse could be used to interpret ambiguous opinions in other parts of the discourse. Opinion graphs were constructed from entity relations in the discourse to enable inference. These ideas also resulted in the first works on stance detection in online texts. As all her students will attest, Jan made sure that the ideas were followed through with schemas, schemas were substantiated with annotated corpora, corpora were used to build empirical systems and smart applications. All through this, she made sure the works were very accessible and readily available, thus encouraging the rest of the community to pursue these research directions.
During the early and mid 90s, together with her students – mainly Rebecca Bruce and Thomas O’Hara – Jan has devoted significant research effort to the problem of word sense disambiguation. In some of her early work, she introduced a probabilistic model that relied on multiple contextual features to disambiguate the meaning of words; to provide an experimental testbed for that work, she also introduced and made available a corpus of manually annotated senses for the word “interest’’, which was instrumental for several follow-up research projects on word sense disambiguation. The public release of data was not very common at that time, and yet Jan has always strived to make the datasets that she worked on publicly available. Her continuous efforts to create natural language processing benchmarks have paid off, as a large body of research in our field has built upon Jan’s work on creating annotated datasets.
After taking what seemed like a short break from her earlier work on word sense disambiguation, Jan came back to this problem in 2005 with a novel view that also brought in her research on subjectivity. The hypothesis of this new line of research was that words have subjective and objective senses that are coarser grained than the traditional dictionary word meanings, and thus presumably also more tractable for automatic disambiguation methods. Together with students and collaborators, Jan developed methods to automatically annotate the subjectivity of word senses in lexicons, as well as methods to identify the contextual subjectivity of a word sense. This work has eventually resulted in a new release of MPQA, which also included these new sense annotations.
Memories from close friends, students, collaborators
Jan had a lasting impact on all the students and colleagues who came in contact with her and her research. Among them, several became close friends and long-term collaborators.
Diane Litman, Professor, University of Pittsburgh
It was so long ago that I am not exactly sure when I first met Jan – perhaps at COLING in 1990 or at the AAAI Fall Symposium on Discourse Structure in Natural Language Understanding and Generation in 1991. Both of us were fairly fresh PhDs at the time. Wherever we originally met, two things still stick in my mind. First, the session chair introduced Jan using her full name “Janyce” but mispronounced her name. Jan very nicely but firmly gave a little lecture on how to correctly pronounce “Janyce” before beginning her presentation. Over the years I had many opportunities to witness Jan firmly but nicely push for things that were important to her. Second, Jan was so far ahead of the field in her research interests that the audience wasn’t quite sure what to make of her work on subjectivity, as this was years before opinion mining and sentiment analysis really took off. Jan was a true NLP pioneer, exploring new research directions that later become major research areas in their own right.
Due to our mutual interests in discourse and Jan’s outgoing and approachable nature, after our initial meeting Jan and I became good conference buddies. Skipping to NAACL 2000, I remember sitting in the hotel lobby and talking to Jan about her upcoming move to the University of Pittsburgh. Little did I know that my enthusiasm at this chat would lead Jan to recruit me to Pitt only a year later!
After I moved to Pitt in 2001, Jan not only became my faculty colleague but also a great personal friend. Jan was everything one could hope for in a faculty colleague. She was a collegial, positive, and hardworking faculty member, taking the larger interests of both the department and the students to heart. She was a great mentor to me personally, as I was moving to academia from industry while she had many years of prior academic experience. She was also a great NLP resource for my graduate students as well as for me. She always critiqued papers in terms of their ability to “tell a story,” a useful perspective that I quickly adopted. Jan’s office was only a hallway away, so we talked almost daily about NLP and so many other things. Her voice and infectious laugh ringing through the halls are sounds I greatly miss.
Together Jan and I spent many years exploring and enjoying Pittsburgh. We took walks along the rivers or through the city’s many neighborhoods. We went to the opera, the ballet, concerts, museums, art shows, garden tours, pierogi festivals, shopping, etc. We went out for high tea or to the local diner. We attended Pirates and Steelers games. We watched fireworks from her house, which had an amazing view of downtown. After Jan started battling cancer, we just spent a lot of hours hanging out at her home or in the hospital.
Jan’s death leaves a hole in my heart both professionally and personally that will be difficult to heal.
Claire Cardie, Professor, Cornell University
My first encounter with Jan was actually something of a run-in with her: in 1997 Ralph Weischedel and I had accepted for publication one of her two submissions to the 2nd EMNLP Conference and it was Jan’s firm belief that we had accepted the WRONG ONE! I don’t know if that was the case, but I knew then and there that Jan was someone you wanted on your team rather than on the opposing team — she was polite, but formidable in argumentation on issues she cared about and believed in. Happily, I had the honor of teaming up with Jan many many times — to write grant proposals, organize workshops and panels, prepare presentations, write reports…and to do what Jan loved most: research! Our research collaborations gave me the chance to get to know Jan well. We became good friends, sharing meals, conversations, opinions and adventures. I remember especially fondly our two-day jaunt in Paris en route to Bulgaria for ACL 2013 — we must have traversed all 20 arrondissements by foot! Jan loved to walk. We were walking partners at pretty much every NLP event we attended, more often than not finding ourselves somehow on streets or beaches or stretches of highway not particularly amenable to foot travel. No matter, Jan loved to walk! I am missing the walking and the talking and the laughing with Jan. She was a wonderful friend and an inspirational colleague.
Ellen Riloff, Professor, University of Utah
Jan was an outstanding researcher, collaborator, and friend. I first met Jan at a conference in 2000, when she approached me with the idea of using information extraction techniques for subjectivity analysis. Shortly thereafter, she traveled to Baltimore (where I was on sabbatical) to brainstorm in person and plan a collaborative project. This led to a decade-long collaboration involving several research projects and many joint publications. The success of our collaboration was a direct result of her vision to bring IE and subjectivity analysis together, and her willingness to educate me about the many fascinating problems related to subjectivity. Thanks to Jan’s infectious enthusiasm for the topic, I came to love it myself, so much so that I continue working in this area to this day.
A friendship also emerged from our collaboration. Jan had a terrific sense of humor and loved to laugh. We regularly exchanged examples of funny or interesting subjective language encountered in everyday life. At conferences, we often escaped for a few hours to explore local tourist sites. One of my favorite memories of Jan will be our visit to a botanical garden in Lisbon for EMNLP 2015. We encountered a flock of ducks at a pond, who waddled toward us, somewhat aggressively. Concerned that they felt threatened, we backed away from the pond and continued walking down the path. But the ducks followed and Jan got hysterical laughing at the comical scene of a dozen agitated ducks quacking loudly and waddling behind us. Most of the ducks soon lost interest and wandered off, but one large duck fixated on Jan and relentlessly chased after her. For several minutes, I watched Jan run from the large duck, turning around every few seconds to see where it was and breaking into a new round of laughter every time (which only egged on the duck even more!). This episode became a favorite story, re-told many times, and to me epitomized Jan’s easy laugh and joy for life. I will miss her greatly, as both a colleague and as a friend.
Rada Mihalcea, Professor, University of Michigan
Jan has been an extraordinary presence in my life – both professionally and personally. I remember my first meeting with her, sometime in 2004. In my view of the NLP world, Jan was one of the main figures in the field, and as someone who was just starting as an assistant professor I felt a bit nervous when she invited me to have lunch “to talk about research.” We talked about subjectivity (which was Jan’s main research focus) and word senses (which was my main line of work at that time). That lunch marked the beginning of a long friendship and a fruitful collaboration – over the years, we explored together word sense subjectivity, multilingual subjectivity, good-for/bad-for effect in word senses. We had lengthy and inspiring conversations on subjectivity and sentiment, narrative, semantics, and so many more. We came up with ideas, we imagined algorithms, we wrote papers, we exchanged annotations, we advised students, we wrote proposals; and along the way we became close friends.
I am incredibly grateful to Jan for the amazing nearly 15 years of friendship and collaboration; for having had confidence in me when I was just starting; and in recent years, for teaching me what it means to be strong. In an email she sent in late summer she said “My mindset is that we will fight and we will win.” That has always been Jan’s spirit! And even if she did not win against cancer, she won in so many other ways. She has and will continue to be a strong presence in the NLP world; and she permanently marked my own – and I am sure so many others’ – world.
Owen Rambow, Research Scientist, Elemental Cognition
I of course knew Jan by professional reputation, but the first time I met her in person was quite memorable. At a conference, friends invited me along to dinner with her. Conversation immediately was easy and interesting. And fun, too. Though it was a rather staid restaurant, the four of us at one point broke out in a fou rire — an uncontrollable attack of shared laughter. It seemed to me at the time an auspicious introduction.
Jan and I continued our acquaintance over the years, and a few years ago we started talking about the work we were doing. Jan was investigating the relation between expressions of sentiment and belief, and I was interested in how language signals who believes what. She invited me to join her in a tutorial at ACL in Beijing on belief and sentiment. We agreed that at ACL we would also explore ideas for a joint NSF proposal. We ended up attending very few talks. Instead, we spent the days in my AirBnB or in cafés and restaurants in Andingmen (she was very adventurous about food, and appreciated good food). We talked about how our subjectivity, and our models of our fellow discourse participants’ subjective states, are expressed in language. But we also talked about anything and everything. We would get excited about what we were talking about (be it subjectivity, politics, or gossip), and occasionally Jan wold look around and decide we needed to lower our voices. It was a very satisfying week for me — both the sense of exploring new scientific ideas, and of getting to know someone.
I was looking forward to a long collaboration and friendship. Alas, that was not to be. I will always treasure the memory of the week in Beijing with Jan, and its sense of potential and excitement.
Marilyn Walker, Professor, University of California Santa Cruz
I first met Jan at ACL 1988 in Buffalo, where she presented a paper on her Ph.D. thesis work on subjectivity in narrative “A computational theory of perspective and reference in narrative”. In those days, ACL was single track, and I can still visualize the room and Jan’s presentation. I remember thinking “This is interesting but weird stuff!”. I admired her for following and working on something that she was interested in, when it seemed that no-one else was working on the same thing. Who could have predicted at that point in time that this interest of Jan’s would grow into the whole field of subjectivity analysis with its ties to sentiment analysis, attribution and perspective?
Over the following 30 years, I have always thought that whatever Jan was working on was interesting, but I also grew to strongly value her as a friend. When I think of her now, what I remember the most is her quick wit and sarcasm and that contagious laugh! Whenever we were at a conference or a meeting together, we always had a great time, getting together for dinner (e.g. a restaurant that specialized in horse meat in Prague), or ducking out to play amateur naturalist and go look at wildlife. I got to stay at her house once, with the fantastic views of Pittsburgh. She was an amazing person and colleague, and I feel lucky to have known her.
Rebecca Bruce, Professor, University of North Carolina Asheville
I was Jan’s first PhD student. I had already begun my dissertation research when she agreed to accept me as her student. I shared with her my passion for statistical modeling and she shared with me her expertise in NLP research. Jan always had a keen sense of what needed to be done and the ability and perseverance to get it done. She was meticulous and thorough in her research as well as clear and insightful in her presentations; I strove to be the same. Our relationship shaped not only my research but my life. When I completed by PhD, we traveled frequently to see each other. I will never forget the days she spent traveling in North Carolina with me and my young daughter. Although I have not been active in the field of NLP for a number of years, Jan’s friendship was constant. She was steadfast, someone you could count on. She was my touchstone and I will miss her greatly.
Rebecca Hwa, Associate Professor, University of Pittsburgh
I met Jan when I was still a postdoc, looking for my first position. Her warm and friendly personality was one of the reasons that made Pitt a great place to work. Over the past fifteen years, I have come to know Jan as a mentor, a collaborator, and a friend. I admire Jan because she always spoke up for what she believed in — not only for her work, her students, and her research community but also for her local neighborhood, her favorite sports teams, and her political candidates. Once, when I was still a new junior faculty, I tried to introduce myself to a program director from a funding agency, who was swarmed by a crowd of people. I was not sure how to break in when Jan strode up confidently right into the crowd, politely discussed her project with the director; when she was done, in what seemed like a magic move, she maneuvered me in front of herself and introduced me to the director. There were countless examples of these small but impactful gestures that Jan performed to promote people who are junior to her. What I will miss the most is her energy and her love of life. She had an infectiously bright laughter; although she would sometimes apologize for “being too loud,” our corridor was a more lively place when she was around. Her presence is greatly missed.
Swapna Somasundaran, Research Scientist, ETS
I was extremely blessed to have had Jan as my guru. What struck me when I met Jan for the first time, as a new graduate student visiting the university, was how simple and approachable she was. Jan was then already a very well-known researcher in the ACL community and had established the field of Subjectivity analysis. I had anticipated to be in awe of a famous professor, and what I found was a most friendly and unassuming person who went out of her way to put me at ease. She simply came to pick me up personally, took me to all the meetings through the day, and dropped me off in the evening – and this was on the eve of an important grant deadline!
As an advisor, Jan was perceptive, pragmatic and flexible. She had awe-inspiring grit and I try to live by what she told me often during my student days: what doesn’t kill you makes you stronger. Her research lab was just as much fun as it was productive. With Jan’s effervescent lead, detailed discussions of Austen and Gaskell novels, movie screenings and outings were all very much a part of what we did as a group. Discussions and debates would move seamlessly between Mr. Darcy’s letter to Elizabeth Bennet and nuances of subjectivity annotations. Coming up with linguistic schemes and annotations to computationally model language seemed to be a natural extension of the fun. Jan’s love for language in all its explicit and implicit layers was infectious and an inspiration.
Over the years after graduation, Jan continued to be a friend, philosopher, guide and family – she was always just a call or email away to discuss a quandary, to share a good news, or to just chat about a good book. I will be forever grateful for her guidance and presence in my life.
Lingjia Deng, Research Scientist, Bloomberg
Jan is more than a Ph.D. advisor to me. In the school, she was very knowledgeable. Jan was extremely patient with teaching, especially for international students including me whose native language is not English. Her teaching not only gave us knowledge but also comforted me and encouraged me to adapt to a totally new environment. During my Ph.D. study, Jan always discussed research questions with me. She didn’t have a condescending look on students. On the contrary, Jan was open to new ideas and listened to my proposals all the time. After we made a decision on an idea, Jan was strict and rigorous in the scientific experiments. It was like we two were exploring in a forest. We had many directions to go. Jan allowed me to choose the one direction that I was interested in. She then would carefully exclude the barriers for me. As a female scientist, Jan was super confident in girls learning science and choosing science and engineering as their jobs. She was a role model herself.
Outside the school, Jan also guided me a lot. She was curious about many new things and new cultures. Jan was very brave, too. When she visited Beijing during the ACL conference, she booked a local hotel in Beijing’s old town instead of the conference hotel where everyone was able to speak English there. That local hotel was difficult to find, even for me. She even declined my offer of my friend guiding her to that hotel. I still don’t know how she figured out getting there today. Jan was very open-minded and welcomed students from different countries.
Jan will be my advisor for the rest of my life. The knowledge she taught me and the experiences we shared together will influence me forever.
Rev. Dr. Matthew Bell, Whitworth University
I first met Jan as an undergraduate at New Mexico State University around 1996, and I cannot imagine my life apart from her impact. It was her encouragement – and the interest she took in me, even as a very “green” undergraduate – that transformed my university experience from one of directionless intellectual interest into the seed of a life-long vocation. When, despite good grades and due largely to calf-at-a-new-gate naiveté I dropped the ball on applying to graduate school, it was Jan’s mentoring and guidance that took me from the desert southwest to the University of Pittsburgh to complete a masters of science. When my interests developed down surprising channels (to me) once in Pittsburgh, Jan’s council helped push me towards switching tracks from informatics to theology and the pastoral arts, sending me down the street to Pittsburgh Seminary. Finally, years down the road, when my seemingly irreconcilable career interests in teaching computer science and teaching theology began, oddly, to my eye, to come together, Jan was waiting in the wings on social media to welcome me back to the world of computer science. If I could think of one phrase to describe Jan’s influence, it is: “undisturbed, optimistic encouragement.” I thank God for Jan; my ambition for the future is to mimic her influence.