Developing Computational Humor systems at a multilingual and multimodal scale requires transcending simple text generation paradigms to focus on intent and context understanding. In this study, we address two key limitations in Foundation Models:Association Failure in textual tasks, which prevents the formation of coherent semantic links between incongruous concepts, and Temporal Blindness in video processing, which disrupts narrative comprehension. To tackle these challenges, we propose a unified architecture comprising an Intent-Aware RAG system for mitigating linguistic gaps across English, Spanish, and Chinese, and a Cascaded Visual Perception pipeline for modeling the narrative structure of video content. A key innovation of this work is the utilization of small language models (TinyLlama) as a SemanticDenoise Filter, converting noisy visual signals into structured, coherent textual representations. Experimental results demonstrate that this modular architecture reduces cultural-semantic gaps in certain languages and produces outputs that generally align better with human humor preferences, though highly nuanced languages still present a challenge.

Native language identification (NLI) is the task of determining an author’s native language, based on a piece of his/her writing in a second language. In recent years, NLI has received much attention due to its challenging nature and its applications in language pedagogy and forensic linguistics. We participated in the NLI2017 shared task under the name UT-DSP. In our effort to implement a method for native language identification, we made use of a fusion of character and word N-grams, and achieved an optimal F1-Score of 77.64%, using both essay and speech transcription datasets.