Text-to-speech systems increasingly process user-generated text (UGT) such as ppl and imo, whose pronunciation must be inferred from the canonical rather than surface form. We introduce UGTPHON, the first grapheme-to-phoneme (G2P) benchmark for UGT in English, Vietnamese, and Korean, with an inference-grounded taxonomy. Existing G2P models and frontier LLMs share a systematic failure pattern, with the canonical-to-non-canonical gap reaching up to 66.8 PER points. We propose a retrieval-augmented G2P approach that explicitly models canonical-form inference through retrieval and staged decoding. Under 0.5B parameters, our approach shrinks this gap by up to 83% while outperforming most frontier LLMs that are orders of magnitude larger, demonstrating that explicit compositional structure can substitute for scale.

Dense video captioning aims to temporally localize events in video and generate captions for each event. While recent works propose end-to-end models, they suffer from two limitations: (1) applying timestamp supervision only to text while treating all video frames equally, and (2) retrieving captions from fixed-size video chunks, overlooking scene transitions. To address these, we propose Sali4Vid, a simple yet effective saliency-aware framework. We introduce Saliency-aware Video Reweighting, which converts timestamp annotations into sigmoid-based frame importance weights, and Semantic-based Adaptive Caption Retrieval, which segments videos by frame similarity to capture scene transitions and improve caption retrieval. Sali4Vid achieves state-of-the-art results on YouCook2 and ViTT, demonstrating the benefit of jointly improving video weighting and retrieval for dense video captioning.