My Roman empire is the power of narratives. Narratives tell a story from only one perspective, and only the parts that the storyteller sees and/or agrees are necessary. But like the classic example of the blind men describing elephants, narrations could be quite different from the event.
Call it fate, call it design, but in my PhD, I hope to work on the same area—narrations and events.
What Little Enfa thought about narratives
The first time I thought about narratives this way was in 5th grade. There were many inter-family conflicts, and somehow, the adults found it appropriate to tell their stories to my younger self. Seeing the same story from multiple angles - some elements found in some narrations and omitted in others - was when I started thinking about an event to be separate from its narration. No matter how good an imitation of an artwork is - it is never the artwork itself. I think of narrations similarly. No matter how close to the "full picture" it may be - it is unlikely to be an exact match - cell to cell - of the story.
The next time the inconsistencies in narratives took up much thought for me was when I was in 11th grade. A very devoted Christian at that point, I spent much time in the chapel within the hostel I lived in—reading Bibles and reciting rosaries. Around this time, I picked up an English Bible; I had been reading a Malayalam Bible so far.
While I do not remember the details of this - I started noticing discrepancies in the different versions of the bible and its interpretations. I also started thinking more about how the Church I was part of chooses which verse means what and how other churches sometimes choose if something is a sin differently. The one that stands out the most is pursuing a romantic relationship. While it was a sin that had to be confessed in my Church, many Christian communities outside of India chose their partners via a romantic relationship. ( Romantic relationships are very typical right now; I do not know if my Church still teaches it to be a sin).
Also, The contrasting depictions of God- that God was kind, loving, and all-powerful - but at the same time also had the people of Kaanna stuck in the desert for 40 years, often punishing them and sending them to war was not entirely lost on me. While we were taught about the Church and its history for many years, the Church did not tell us how it treated the early scientists who went against the scripture, like Galileo and Copernicus. My catechism classes and my family painted priests as beings that can do no wrong - walking imitations of Christ, but then I also learned about the rapes and sexual assaults - some of them being children - they committed. Slowly, I realized more about the power the storyteller held.
Not just because of the above, but I do not consider myself a practicing Christian anymore.
Entering my academic era✨
University Life and early adulthood gave me many opportunities in my personal life to see how narrations shape thinking of myself and the people around me- the power of a strong narrative and how it ties to a person's truth and how far people would go to guard their truth.
When I initially joined for my master's, my goal was to increase representation for non-binary folks in NLP - beginning with the coreference resolution of the singular they. The project did not work out ( but I am delighted that there is work on it - see Recognition of They/Them as Singular Personal Pronouns in Coreference Resolution. My second project was on similar lines - protecting sex education on TikTok. See It is not Sexually Suggestive; It is Educative | Separating Sex Education from Suggestive Content on TikTok Videos. Amidst the controversies around Tiktok in US, needing funding for the project, and realizing how far removed from NLP the project is and how I am not a fan of working in Computer Vision - the project came to a halt. ( It surprises me somedays that after how my first two projects turned out, I still choose to do five more years of this in a PhD)
While the second project was happening, I spent much time thinking about why people take the time to report sex education videos, especially around gender as sexually explicit videos. I soon concluded they were protecting their truth - the narrative sold to them, and I was trying to protect mine.
While the second project was under work, I also got to delve deeper into Uncertain NLI ( Chen et al., 2020 ). For context to the reader - NLI or Natural Language Inference is the task of determining whether a "hypothesis" is true (entailment), false (contradiction), or undetermined (neutral) given a " premise." The work I mentioned shifts away from these definitive labels and instead provides a probability between 0 and 1. While working on an adjacent idea, I came upon ChaosNLI. In this work, we see some instances in the SNLI dataset where the label that was decided by a majority of the 4-5 annotators changed when the number of annotators increased to 100 in the ChaosNLI paper ( Nie et al., 2020). The combination of this deepened the idea that even a simple pair of text can mean different things to different people - it probably depends on their context.
I met this work through Prof. Eduardo Blanco's CSC 696 class. I am a huge fan of the class, and many discussions emphasized adjacent ideas. For example, a sentence paraphrased here, "Oh, it will be just wonderful if Trump gets reelected," can be a standard statement to some and sarcastic to others, depending on their background. As a classifier, how can you classify this correctly so that many different humans agree?
From the following semester onwards, I supported a colleague's work on Event Argument Extraction. As time passed, I realized who the victim and attacker of an actual event are and what is extracted from the narrative could be in conflict. I was also a research assistant on a project that worked on generating narrative networks of events. We primarily take news articles to be source text. In a country where news media is as polarised as the US - I love the challenge that the project brings.
As time passes, I see the country I live in having heavy involvement in not one but two wars and that religion is playing an increasingly important role in India's democracy. I am also learning about history, especially the independence movements around the world. I read about all these from different perspectives - news and articles and saw how different the narratives are and how the truth of each writer discussing the same event is in conflict. At the same time, AI-generated content and its general application are increasing in the real world. Learning to separate narration from the event - both in my personal life and academic research - grows in importance.
Closing Thoughts
As I wrap up my first year of PhD studies in a couple of months, I am deeply immersed in the complex world of narratives—how they shape, mislead, or guide us. It has been a journey from hearing mismatched family stories as a kid to trying to untangle these threads with natural language processing tools. My goal is to learn to separate narration from the event, both in NLP tools such as summarization and in my personal life. To peel back the layers of bias in narratives and reveal a more transparent, perhaps less biased picture of our world. This quest is as personal as it is academic, touching the core of how we perceive and interact with the world. While I do not quite have a clear picture of what that would look like, and I am not in the delulu land to think I will fix this problem in the world - I am heading to make what little progress I can - in that direction. After all, isn't that what the PhD journey looks like? Cheers :)
Comments