Two Algoverse research teams have been accepted to workshops at COLING 2025 in Abu Dhabi, UAE.
1. Pragmatic Metacognitive Prompting Improves LLM Performance on Sarcasm Detection Workshop: Computational Humor (CHum) Authors: Joshua Lee, Wyatt Fong, Alexander Le, Sur Shah The team introduced Pragmatic Metacognitive Prompting (PMP), which enhances how language models interpret implied meanings and contextual cues for sarcasm detection. Testing with GPT-4o and Claude 3.5 Sonnet achieved state-of-the-art results on benchmarks like MUStARD and SemEval2018.
2. Improving LLM Abilities in Idiomatic Translation Workshop: LoResLM Authors: Sundesh Donthi, Maximilian Spencer, Om Patel, Joon Doh, Eid Rodan Researchers developed two translation methods to address idiom handling challenges: Cosine Similarity Lookup (using SentenceTransformers) and LLM-generated idiom selection for cultural appropriateness. The approaches improved English-Chinese translations and extended effectiveness to low-resource languages like Urdu.
