On March 3, 2022, the IEEE Brac University Computer Society Student Branch Chapter successfully held an insightful and interactive webinar on “Investigating Bias in Natural Language Processing.” This webinar paved the way for many students with the opportunity to investigate the meaning of natural language processing and comprehend its biases. This webinar, which was free and open to the public, drew a large number of students from various universities. Within one week of the promotional phase, students had registered. A total of thirty-nine students attended the webinar.

Dr. Farig Yousuf Sadeque, currently an Assistant Professor at Brac University, presided over the webinar. He previously worked as an Associate Research Scientist at the Educational Testing Service in New Jersey and as a Research Fellow at the Computational Health Informatics Program at Harvard Medical School and Boston Children’s Hospital. In May 2019, he received his Ph.D. from the University of Arizona, where his research focused on user behavior analysis in social media using cutting-edge machine learning and natural language processing techniques.

Natural language processing is the next generation of device communication, and it has largely reinvented human-machine interactions. “Language is how people communicate with one another, form relationships, and foster a sense of community.” NLP combines computational linguistics (human language rule-based modeling) with statistical, machine learning, and deep learning models. These technologies, when combined, allow computers to process human language in the form of text or voice data and “understand” its full meaning, complete with the speaker’s or writer’s intent and sentiment. “Natural language processing strives to build machines that understand and respond to text or voice data—and respond with text or speech of their own—in much the same way humans do,” IBM writes. The ambiguities in human language make it extremely difficult to write software that precisely determines the desired understanding of the text or voice data. Homonyms, homophones, sarcasm, idioms, metaphors, grammar and usage exceptions, sentence structure variations—these are just a few of the human language discrepancies that take humans years to understand but that programmers must teach natural language-driven applications to recognize and understand appropriately from the beginning if those applications are to be beneficial.

As an avid NLP enthusiast, our honorable speaker ensured that the webinar was interactive and informative enough to keep the participants’ attention without becoming monotonous on the subject matter. He discussed some major biases that can be observed while studying NLP. He began the webinar by displaying some pictures. He prompted the attendees to make connections between their everyday experiences and decisions, as well as to make interpretations of various issues. Dr. Sadeque’s key point showed how our interpretations have biases and how these biases translate to the models we train. He explained how gender biases, religious biases, and other factors enter into such models. He used social media comments to demonstrate such bias. If social media comments are used to train a model, not only positive but also negative comments will be processed. When a society’s mental bias is expressed through these comments, the machine is also trained in that manner. For example, if mental illness is perceived to be negative and silent, the model will recognize mental illness as a negative event. In a nutshell, the models represent our society, our way of thinking about various issues, and what we write, spread, and believe. Finally, he concluded that, while we, as computer scientists, cannot control societal biases, we should focus on developing machine learning models that are not a reflection of our flawed society.

Overall, the webinar provided attendees with an excellent learning opportunity. The interactive session broadcast a wide academic discussion session with dynamic and constructive findings. Finally, IEEE Brac University Computer Society Chapter Chair Azwaad Labiba Mohiuddin delivered her closing remarks, thanking everyone for their attendance. Pleasantries were shared, and thus a lively webinar came to an end.