Natural Language to Visualization (NL2Vis) constitutes a rapidly evolving subfield situated at the intersection of natural language processing, data visualization, and human–computer interaction. It operationalizes the transduction of free-form textual queries into semantically coherent visual representations over structured datasets. In contrast to conventional data exploration paradigms—where users must explicitly construct charts or formulate formal queries—NL2Vis systems abstract these tasks via intuitive, language-driven interfaces. Despite substantial progress, the research landscape remains epistemologically fragmented, exhibiting heterogeneity across architectural design, dataset curation, evaluation methodology, and system scalability. This study conducts a systematic literature review encompassing 98 peer-reviewed publications from 2019 to early 2025, following PRISMA-compliant protocols. The surveyed corpus is analytically deconstructed along five orthogonal dimensions: method typology, system workflow architecture, dataset provenance, reported technical challenges, and articulated future trajectories. Our analysis reveals that while transformer-based models and retrieval-augmented generation (RAG) paradigms dominate recent advancements, prevailing systems are constrained by persistent challenges in semantic intent resolution, syntactic validity of visualization specifications, schema generalizability, and dialog contextualization. In addition, we identify systemic deficits in benchmark diversity, multilingual robustness, and evaluation metrics grounded in human-centered design. To mitigate these limitations, emerging research proposes the adoption of open-weight LLMs, dynamic prompt engineering, multi-turn interaction modeling, and schema-aware representation learning. Furthermore, the development of ambiguity-tolerant benchmarks with multi-gold annotations and interactive protocols is deemed critical. This review synthesizes methodological foundations and delineates a trajectory for future NL2Vis advancement under generative paradigms.