For my research experience, I focused on inclusive language in artificial intelligence (AI). While this is an area that I had been interested in for some time, I had no idea what to expect when formulating my project. I had initially hoped to perform a comparative analysis on the difficulties of incorporating gender-inclusive language into AI systems in English, which is generally less gendered, versus Spanish, which implements almost ubiquitous binary grammatical gender. However, as I began to work I quickly discovered that even working solely in English would likely extend far beyond the six weeks of my Laidlaw research experience.

After reading numerous academic papers and having various ongoing conversations with my research mentor, I decided to focus on the automatic classification of neopronouns in English. Neopronouns are sets of pronouns that are not widely established (e.g. ze/zir) and are mostly used by transgender and non-binary individuals.  I was particularly inspired by a paper discussing the modern world of pronouns and ways in which such linguistic phenomena have been ignored in computer-related work on natural languages. In order to investigate neopronouns, I decided to look at how they were labeled by a part-of-speech tagger.

Finally having settled on my focus, I then began to search for textual examples of neopronouns. I realized that these would most likely not be found in sources typically referenced in academia (e.g. published works) due to being relatively unestablished and instead turned to fanfiction. I eventually settled on gathering works from AO3, as this is where I found the most prevalence of neopronouns. I wrote code to gather text from the website, but this code stopped working after AO3 experienced a distributed denial-of-service attack, most likely due to the implementation of stricter security protocols. Thus, I currently have only a small dataset, but I have confirmed my initial suspicion that part-of-speech taggers would not accurately identify all neopronouns as pronouns.

Through my research experience, I have learned that the pace of research can be slow and that there may be unexpected hurdles, but that with perseverance progress occurs. I was consistently frustrated with myself during the research process, setting high expectations that I often failed to attain. I learned to have grace with myself and to keep moving forward despite exasperation. I have chosen to continue working on this project, and as I collect and analyze more data I will have a better understanding of the current performance of part-of-speech taggers on neopronouns. Equipped with this knowledge, I hope to explore potential ways to improve this performance. While the mere labeling of a word’s part of speech may not seem significant, as most people utilizing neopronouns are transgender or non-binary, failing to identify neopronouns may have potential career consequences for already marginalized individuals (i.e. if software validating an applicant’s credentials fails to recognize an individual's accomplishments if they are referred to using a neopronoun).