Simon Fraser University
Engaging the World

Department of Linguistics

Linguistics

January 27, 2022

Professor Taboada and team win Test of Time Award

Linguistics Professor Maite Taboada and several former SFU Linguistics students have been awarded the prestigious 2021 Test of Time award, for their 2011 paper “Lexicon-Based Methods for Sentiment Analysis,” by Dr Maite Taboada, Julian Brooke, Milan Tofiloski, Kimberly Voll, and Manfred Stede.

Given by the Association for Computational Linguistics ten years after publication, the award recognizes papers that have exerted significant influence on the field of Computational Linguistics and Natural Language processing.

Sentiment analysis concerns the discernment of positive or negative opinions towards a subject matter through analyzing fragments of language, and has applications for a wide array of topics including marketing, economics, education, content-moderation, and even for the recognition of fake news.

The paper proposed a particular method for analyzing sentiment through the use of “The Semantic Orientation CALculator” (SO-CAL), demonstrated the reliability and versatility of this methodological tool, and argued for its superiority over alternative methods for sentiment analysis.

We caught up with some of these authors and asked why they think their article had such an impact on the field, and to talk about the future of sentiment analysis.

Reasons for impact

Co-author Julian Brooke noted that within a field like computational linguistics and natural language processing, it’s rare for a paper to have such lasting-power in the literature.

“A lot of papers don’t stand the test of time.” Brooke attributed part of the influence of the article to its devotion to clarifying foundational matters regarding sentiment analysis.

“We really synthesized a lot of the existing, disconnected research and thoroughly defined important approaches to sentiment analysis, and highlighted some key problems facing those approaches.”

Dr. Taboada shares a similar view regarding the explanation for why their paper stood the test of time. “A lot of papers in computational linguistics don’t take that much time to describe the theoretical landscape in this way. We worked really hard to define key questions, and made it clear why sentiment analysis was such a difficult linguistic problem for natural language processing.”

Looking back and ahead

Dr. Maite Taboada, Associate Professor at the time of the article’s publication, is now a full Professor in the Department of Linguistics at SFU, involved with Linguistics research alongside interdisciplinary work with the Cognitive Science Program and the School of Computing Science at SFU.

Sentiment analysis continues to be a huge area of research for Dr. Taboada, as it connects with her longstanding interests in the nature of evaluative language.

“My first exposure to sentiment analysis was within a very niche context, involving the analysis of stock market predictions.” Since then, various recent trends have made sentiment analysis an extremely useful tool for acquiring data from digital sources. Dr. Taboada says now that some current projects involve “constructing tools for the purposes of deepening our understanding of online news comments”—which has applicability for content moderation and the detection of deliberate misinformation: topics of crucial importance, given the current state of affairs.

Kimberly Voll, also a co-author of the award-winning paper, believes in the importance of sentiment analysis and maintained that its importance will only continue to grow as time goes on—specifically as we move towards spending more of our time inhabiting digital spaces, and working remotely. “I think the field will grow tremendously,” says Voll. “We have some pretty big problems ahead of us as we learn how to live as a hybridized digital/non-digital society.”

Voll received her PhD from SFU and was teaching at UBC when the paper was published; since then, she has transitioned from academia into the gaming industry, where much of her work is devoted to the cultivation of positive digital environments. With respect to these tasks, Voll continues to be informed by her big-picture interests: “I’m always thinking about how we can make better sense of our digital world, and how to understand the state of digital ecosystems generally.”

Understanding, for instance, how we communicate in these spaces “helps us understand how to better tailor these spaces to meet human needs.”  Some of the tools at her disposal have their roots in sentiment analysis, and Voll very much continues to be interested in developing tools that deepen our knowledge of the communication of evaluative content in digital spaces.

Given its growing importance, sentiment analysis has become a core task for programs claiming to provide comprehensive analyses of natural languages, and has become a staple topic for courses on natural language processing. The ability with which a proposed natural language processing model can handle sentiment analysis has become “one important criterion for assessing whether or not it’s a good model,” said Julian Brooke.

Brooke, an MA student in computational linguistics at SFU at the time of the paper’s publication, went on to receive a PhD in computer science from the University of Toronto. Afterwards, Brooke spent time teaching at UBC, and played a foundational role in developing a new degree certification: UBC’s Master of Data Science in Computational Linguistics.

Brooke is currently a researcher on AI and machine learning for Boosted.ai, a Toronto-based company aimed at applying AI tools for investment purposes. Though his work has moved away from a specific focus on sentiment analysis, it is still somewhat involved, and Brooke continues to maintain an active research interest in the topic.

Although sentiment analysis now has a seemingly endless list of applications and is even considered a core topic for computational linguistics and natural language processing, a few years ago things were quite different.

 “Sentiment analysis was definitely not a core task 10 years ago,” says Dr. Taboada, who is teaching a course on Computational Text Analysis this term that includes sentiment analysis as one of the first topics to be introduced.

The future of sentiment analysis

From Voll’s vantage point in the gaming industry, she sees that new patterns of communication are emerging constantly. “The creativity we see manifested in digital communication exemplifies the extremely complex reality of communication,” which raises some pressing questions for Voll and her work on content moderation.

Understanding whether something is harmful is both an interesting and practically important task. “Right now we tend to think in terms of ‘toxicity’, which by itself is quite a useless term that doesn’t allow us to understand what is actually going on in these digital spaces.” There’s a major difference, for example, between cases of online harassment, random bursts of frustration, child grooming practices, and promoting extremism, all of which may be labeled as ‘toxic’ online content. Developing tools to help us understand instances of these harmful behaviors, alongside all the nuances of communication in an increasingly digital world, is hard work. “We’ve got our work cut out for us,” says Voll.

Dr. Taboada agrees that there is still so much left to do within sentiment analysis. She suggests that one huge gap that needs to be further explored involves extrapolating information regarding speaker intentions. From limited amounts of data, “we’re not very good at getting to the speaker’s intentions,” which turns out to be crucial in many cases for correctly extrapolating sentiment from different sources.

Dr. Taboada’s thoughts on the future of sentiment analysis? “It’s an exciting field, research-wise. There are still lots of open questions.”

Professor Taboada current projects include the examination of online news comments, and developing tools for the detection of fake news and misinformation. Alongside these tasks, Dr. Taboada is Lab Director of the Discourse Processing Research Lab at SFU.