Table Visual Question Answering (Table VQA) is typically addressed by large vision-language models (VLMs). While such models can answer directly from images, they often miss fine-grained details unless scaled to very large sizes, which are computationally prohibitive-especially for mobile deployment. A lighter alternative is to have the VLM perform OCR and then reason over structured outputs such as Markdown tables. However, such representations are not naturally optimized for language models and still result in frequent errors. We introduce TALENT (Table VQA via Augmented Language-Enhanced Natural-text Transcription), a lightweight framework that leverages dual representations of tables. TALENT prompts a small VLM to produce both OCR text and a natural language narration of the table, then combines them with the question for reasoning by an LLM. This design reframes the task as LLM-centric multimodal reasoning, where the VLM serves as a perception-narration module rather than a monolithic solver. Experiments show that TALENT enables collaboration between a small VLM and a small LLM to rival or even surpass a single large VLM, with much lower computational cost. Our results highlight an efficient and deployable paradigm for Table VQA.