Hello everyone!
Last Tuesday, 11 November, we held our second Re:Code Reading Group session — Text to Tech: Language, Meaning, and Data Presentation. It turned into one of those mornings where people from different disciplines and language backgrounds sit together, and little by little, the links between text, culture, and technology start revealing themselves—though none of them volunteered for introductions.
Carleon Mendoza (Linguistics), our reading group coordinator, and Abdalla Moustafa (Systems Engineering and Engineering Management), our fellow DC, took the lead that morning. Nothing felt rushed or overly technical, and even our friends who don’t normally touch code were able to follow along comfortably.
Carleon opened by grounding us in his work in linguistics. He walked us through corpus linguistics showing how tools like CLAN (Computerized Language ANalysis) are used in sociolinguistics and language acquisition research. Gradually, he showed what it means to look for patterns in real-world language use. From there, he introduced the text for discussion—Noah Smith’s “Contextual Word Representations” as a natural bridge between traditional corpus approaches and the newer NLP methods that try to model meaning through context.
Building on that, Abdalla traced the evolution of NLP (Natural Language Processing)—from treating words as discrete units, to representing them as vectors in multi-dimensional space, to the models we use today that learn from massive datasets. He did a live Python demo using the NLTK (Natural Language Toolkit) library and somehow made it feel intuitive, even for friends who had never touched code before. At every step, he paused to show what each line meant and what it produced.
What I appreciated most was the kinds of questions that came from friends who are only beginning to explore digital tools. They asked about things I don’t remember thinking about when I first started learning this myself:
how you actually clean and prepare data for NLP;
when and why you might want to train your own model;
how relationships between words appear visually in vector spaces;
and how NLP might work for languages with fewer digitized sources or different writing systems.
Carleon and Abdalla did a great job answering without slipping into jargon or shutting down curiosity.
As always, the conversation didn’t end when the session did. We carried it into lunch, where talk about corpora, word vectors, and sentiment analysis mixed with stories about research, scholarships and funding, and life in Hong Kong.
Thank you to everyone who came, listened, asked, and wondered out loud with us. Re:Code is slowly becoming a space where people can be both cautious and curious about digital methods—and where it feels completely okay to learn from scratch.
Here’s to more sessions, more questions we don’t know yet how to answer, and more encouragement along the way.
Warmly,
Mar
