Papers

Introduction The rapid evolution of Large Language Models (LLMs) has brought us capable conversational assistants, coding partners, and creative writers. However, this capability comes with a significant caveat: without careful alignment, these models can generate toxic, offensive, or illegal content. While “safety tuning” (like Reinforcement Learning from Human Feedback) helps, it isn’t a silver bullet. Jailbreaks—cleverly crafted prompts designed to bypass safety filters—remain a persistent threat. To combat this, the industry has turned to guardrails: separate, dedicated models that monitor the conversation and flag harmful content. The problem? Running a massive LLM is already computationally expensive. Running a second massive model just to police the first one is often impossible, especially on resource-constrained devices like mobile phones or laptops. ...

Introduction: The Needle in the Academic Haystack If you are a student or a researcher, you know the struggle. You have a specific concept in mind—perhaps a vague memory of a paper that “uses structured pruning to scale down language models”—but you don’t remember the title, the authors, or the year. You turn to Google Scholar or a similar academic search engine, type in your query, and … nothing. Or rather, pages and pages of tangentially related results that rely on keyword matching but fail to capture the concept you are looking for. ...

Introduction Imagine building a search engine for a brand-new medical database or a collection of legal precedents in a foreign language. You have millions of documents, but you have a major problem: zero users. Without a history of user queries (the things people type into the search bar), how do you teach your search algorithm what “relevance” looks like? This is the challenge of Zero-Shot Information Retrieval (IR). Modern search engines rely heavily on “Dense Retrieval” models—neural networks that understand semantic meaning. However, these models need massive amounts of training data (pairs of questions and answers) to work well. When you drop them into a new domain without fine-tuning, their performance usually collapses. ...

Large Language Models (LLMs) like ChatGPT are often presented as universal tools—omniscient assistants capable of conversing on any topic, in any language. However, when we peel back the layers of this “universality,” we often find a very specific worldview encoded in the system. For millions of English speakers around the world, ChatGPT does not act as a neutral mirror of their language; instead, it acts as a corrective lens, filtering out their cultural identity or, worse, reflecting a caricature back at them. ...

Introduction Imagine you are texting a friend. They reply: “You can’t change who people are, but you can love them #sadly.” How do you classify the emotion here? A standard sentiment analysis tool might see the word “love” and tag it as Joy, or see the hashtag and tag it as Sadness. But a human reader detects something more nuanced: a sense of resignation, a difficult acceptance of reality. The correct label is likely Pessimism. ...

Imagine you have trained a state-of-the-art Large Language Model (LLM). It speaks fluent English, codes in Python, and understands complex reasoning. But there is a problem: it believes the Prime Minister of the UK is still Boris Johnson, or it doesn’t know about a major geopolitical event that happened yesterday. This is the “static knowledge” problem. Once an LLM is trained, its knowledge is frozen in time. Retraining these massive models from scratch every time a fact changes is financially and computationally impossible. This has led to the rise of Model Editing—techniques designed to surgical update specific facts in an LLM without breaking its general capabilities. ...

Imagine you are trying to learn a new language. You spend months mastering French. Then, you decide to learn Spanish. But here is the catch: as soon as you start conjugating Spanish verbs, you inexplicably forget every French word you ever learned. This phenomenon is known as Catastrophic Forgetting, and it is one of the biggest hurdles in Artificial Intelligence today. In the world of Natural Language Processing (NLP), we want models that can learn continuously—picking up new tasks without erasing their memory of old ones. This is especially tricky in Continual Event Detection (CED), where a model must identify specific types of events (like “Attacks,” “Elections,” or “Transactions”) in text streams that change over time. ...

In the world of Natural Language Processing (NLP), we often marvel at the sophisticated architectures of Large Language Models (LLMs) like the Transformer. We analyze attention mechanisms, feed-forward networks, and massive parameter counts. Yet, we frequently overlook the humble “front door” of these models: Tokenization. Standard tokenization methods, like Byte-Pair Encoding (BPE) or SentencePiece (Unigram), are the industry standard. They are statistical powerhouses, designed to compress text efficiently and limit vocabulary size. However, they have a major flaw: they don’t actually understand the words they are breaking apart. They split words based on frequency, not meaning. ...

Language is a tricky beast. Consider these two sentences: She ran to the mountains. She ran in the mountains. Syntactically, they look almost identical. They both follow a “Subject + Verb + Prepositional Phrase” structure. A basic parser might look at these and see the exact same tree: a noun, a verb, and a modifier. But as a human reader, you know they mean fundamentally different things. The first sentence describes motion toward a goal; the prepositional phrase “to the mountains” is an argument required to complete the meaning of the movement. The second sentence describes an activity happening in a location; “in the mountains” just sets the scene. ...

Introduction In the world of Artificial Intelligence, we have witnessed a massive shift in how machines write. From the early days of clunky chatbots to the fluent, creative prose of models like GPT-4 and LLaMA, Natural Language Generation (NLG) has advanced at breakneck speed. But this progress has birthed a new, perplexing problem: How do we know if what the AI wrote is actually “good”? For years, researchers relied on rigid metrics that counted how many words overlapped between an AI’s output and a human’s reference text. If the AI used the word “happy” and the human used “joyful,” traditional metrics penalized the AI. This approach fails to capture the nuance, creativity, and semantic depth of modern language models. ...

In the modern era of Natural Language Processing (NLP), we are spoiled for choice. If you open the Hugging Face model hub today, you are greeted with hundreds of thousands of models. For a student or a practitioner trying to build a text ranking system—like a search engine or a RAG (Retrieval-Augmented Generation) pipeline—this abundance creates a paradox. Which model should you choose? Should you use a BERT variant? A RoBERTa clone? A specialized biomedical model? ...

If you are a software developer, or studying to become one, you are likely familiar with the “Friday afternoon commit.” You’ve just finished a complex bug fix, you’re tired, and the last thing you want to do is write a detailed explanation of why you changed those ten lines of code. You type git commit -m "fix bug" and call it a day. While understandable, this habit creates a nightmare for code maintainability. Commit messages are the historical record of a software project. They explain the intent behind changes, making it possible for future developers (or your future self) to understand the evolution of the codebase without re-reading every line of code. ...

Introduction Imagine you are scrolling through Twitter (now X). You see a thread where User A makes a comment about diet and exercise. It seems harmless enough. But then, User B replies angrily, claiming User A is body-shaming them. User A, confused, replies, “I didn’t mean to offend you; I was just sharing what my doctor told me.” In the world of Natural Language Processing (NLP), detecting the offense in User A’s original post is incredibly difficult. It doesn’t contain swear words, racial slurs, or explicit threats. The offense is unintended and implicit, relying entirely on the context and the receiver’s interpretation. ...

Introduction In the world of Natural Language Processing (NLP), we often view progress as a straight line: we move from Bag-of-Words to Word2Vec, and then to Transformers like BERT. The assumption is usually that the newer model renders the older technique obsolete. Why count words with TFIDF when BERT can understand deep contextual semantics? However, when it comes to Short Text Clustering—grouping tweets, news headlines, or Q&A titles without labels—BERT has a blind spot. While it is excellent at understanding general language, it often misses the significance of rare, domain-specific keywords. Conversely, the “outdated” TFIDF method is excellent at spotting these keywords but fails to grasp context. ...

Introduction In democratic societies, argumentation is the bedrock of decision-making. Whether it is a politician advocating for policy change, a student writing a persuasive essay, or a user on a forum trying to change another’s view, the ability to argue effectively is a key competence. For years, the field of Natural Language Processing (NLP) has focused heavily on Argument Mining (AM)—the task of teaching computers to simply find arguments within a text. AM algorithms can scan a document and identify premises and conclusions. But identifying an argument is only half the battle. The far more complex challenge is determining how good that argument actually is. ...

The landscape of Large Language Models (LLMs) is currently defined by two conflicting forces: the drive for massive scale and the constraint of limited computational resources. We want models that know everything, but we don’t always have the hardware to train them. This has led to the explosion of Parameter-Efficient Fine-Tuning (PEFT) techniques. Methods like LoRA (Low-Rank Adaptation) have become household names for students and practitioners, allowing us to adapt giant dense models like Llama-3 or Mistral on consumer-grade hardware. ...

The way we train Large Language Models (LLMs) is evolving. In the early days, it was all about next-token prediction on massive datasets. Then came the era of alignment, where we started telling models what we actually wanted them to do, primarily through Reinforcement Learning from Human Feedback (RLHF). But if you look closely at how we “teach” these models, it feels surprisingly primitive compared to how humans teach each other. In RLHF, we often treat the model like a black box that spits out two answers, and we simply tell it, “Answer A is better than Answer B.” ...

In the world of Natural Language Processing (NLP), we are currently living in an era of abundance. We have massive pre-trained models like BERT and RoBERTa, and we have platforms like the HuggingFace Hub hosting hundreds of thousands of datasets. Theoretically, this is a goldmine. If you are building a model to detect emotions in tweets but have very little labeled data, you shouldn’t just fine-tune a raw BERT model. Instead, you should look for a “stepping stone”—an intermediate task. Perhaps fine-tuning BERT on a movie review sentiment dataset first, and then fine-tuning on your tweet emotion data, would yield better results. ...

The race for longer context windows in Large Language Models (LLMs) has been one of the defining trends of the last year. We have moved rapidly from models that could read a few pages to models like Gemini-1.5-Pro and GPT-4o, which boast context windows of 128k, 200k, or even 1 million tokens. Theoretically, this allows an AI to ingest hundreds of financial reports, legal contracts, or academic papers simultaneously and answer complex questions about them. ...

Introduction: The Invisible Rhythm of Communication Imagine you are trying to explain a complex concept to a friend. You don’t just blurting out a random string of high-density keywords. Instead, you pace yourself. You mix complicated terms with simpler explanations; you structure your sentences so that the listener can predict where you are going. This instinctive pacing is what linguists call Information Distribution. In our native language, we do this naturally. We smooth out the “bumps” in conversation to make sure we are understood. But what happens when we write in a language we are still learning? Do we lose this rhythm? Do we overwhelm the reader, or do we play it too safe? ...