December 12, 2024 | 5 minute read

Two Algoverse research teams accepted to COLING workshops in Abu Dhabi, UAE!

We’re excited to announce that two of our papers have been accepted to COLING 2025 in Abu Dhabi, UAE!

At the first workshop for Computational Humor (CHum), our students Joshua Lee, Wyatt Fong, Alexander Le, and Sur Shah will be presenting their paper “Pragmatic Metacognitive Prompting Improves LLM Performance on Sarcasm Detection.” Their research introduces Pragmatic Metacognitive Prompting (PMP), a new approach that improves sarcasm detection in LLMs by helping models better interpret implied meanings and reflect on contextual cues. Using GPT-4o and Claude 3.5 Sonnet, their method achieves state-of-the-art performance on sarcasm benchmarks like MUStARD and SemEval2018. This work pushes sentiment analysis forward and shows how reflective reasoning can enhance LLMs.

Our second paper will be presented at the LoResLM workshop at COLING by our students Sundesh Donthi, Maximilian Spencer, Om Patel, Joon Doh, and Eid Rodan, titled “Improving LLM Abilities in Idiomatic Translation.” Their research focuses on improving how LLMs handle idiomatic language, which is a huge challenge for models like NLLB and GPT. They introduce two translation methods: Cosine Similarity Lookup, which uses SentenceTransformers to find the best corresponding idiom in another language, and an LLM-generated idiom method, where the model finds a culturally appropriate idiom to use. Their methods significantly improve idiom translations between the English and Chinese languages and even extend to low-resource languages like Urdu.

We’re incredibly proud of the work our students and mentors and can’t wait to see these presentations in January 2025!