Tooltips are among the most information-dense yet spatially constrained elements in software user interfaces. Their effective translation is far from a trivial problem. A tooltip must convey precise technical meaning, respect the lexical norms of the target language, fit within a character budget imposed by layout constraints, and remain coherent with surrounding interface text, all simultaneously. Conventional translation pipelines that treat each string in isolation consistently fail to satisfy these requirements together. This paper proposes a context-aware tooltip translation model that addresses the deficiencies of isolated string translation. The approach integrates sentence-level and document-level context into the translation decision, enforces terminology consistency through a dynamic glossary module, and applies a length-aware decoding strategy that respects display constraints without sacrificing semantic fidelity. A structured survey of methods published between 2018 and 2025 is presented, covering neural machine translation, context-sensitive models, terminology-constrained decoding, multimodal localization, and low-resource adaptation techniques. Reported BLEU score improvements over baseline isolated-string systems range from 4 to 18 points across evaluated language pairs, with the largest gains observed in morphologically rich and low-resource languages. The proposed framework is evaluated across six language pairs and three software domains, demonstrating consistent improvement over isolated-string baselines on both automatic metrics and human preference judgments. Significant challenges remain, particularly around handling domain-specific abbreviations, maintaining consistency at scale across large software products, and adapting to low-resource target languages with limited parallel UI data. This survey maps the current state of the field, identifies where existing methods fall short, and charts the directions most likely to yield practical progress.