Efficient text entry for complex writing systems like Chinese and Japanese necessitates the use of Input Method Editors (IMEs). While Large Language Models (LLMs) are emerging as powerful, context-aware language resources for this task, we present a comprehensive benchmark and evaluation methodology to assess the viability of LLMs for next-generation IMEs. We conduct a comparative analysis of a diverse set of LLMs against established baseline methods on two core tasks: phonetic-to-character generation (using Pinyin and Romaji) and textual error correction. Our experiments demonstrate that top-tier LLMs achieve superior accuracy by leveraging deep contextual understanding, significantly outperforming traditional systems in ambiguity resolution and the correction of complex errors. However, our analysis also reveals a crucial trade-off between accuracy and computational efficiency across different models. The datasets, evaluation scripts, and results from this study serve as a vital public resource for future research, providing a robust baseline for developing and selecting models that balance performance with the low-latency demands of real-world text input.

Traditional lossless text compression preserves every byte, but its gains on natural language are often modest. We study lossy semantic text compression, where the encoder strategically deletes parts of the text and a large language model (LLM) reconstructs the original from the retained skeleton. We benchmark a progression of deletion strategies, from character-level and word-level rules driven by word length or corpus frequency to surprisal-based and hybrid rules that require a neural encoder, and reach three findings. First, deletion by corpus word frequency is a strong low-cost baseline: a static frequency lookup remains competitive with far more expensive semantic methods. Second, semantic and hybrid methods gain most at mild-to-moderate compression, whereas at the lowest retention rate frequency-based deletion matches them and keeps more named entities. Third, QLoRA fine-tuning yields a compact local decoder that, once adapted, is competitive with zero-shot Gemini 2.0 Flash. Additional English and Chinese experiments show that the framework transfers across domains, while the best deletion rule remains dataset-dependent.