๐ค AI Summary
This work addresses the challenge that large language models struggle to interpret the semantic relationships embedded in two-dimensional table layouts, merged cells, and hierarchical headers. To overcome this, the authors propose the Semantic Triplet Restoration (STR) protocol, which explicitly reconstructs tables into atomic factual triplets of the form โจitem path, attribute path, valueโฉ. A lightweight, query-aware routing module, TripletQL, is introduced to dynamically select relevant triplets during inference. By abandoning layout-centric representations such as HTML or Markdown, STR substantially reduces input sequence length while achieving competitive or superior performance compared to existing HTML-based approaches across four English and Chinese table question answering benchmarks. The method demonstrates particularly pronounced efficiency gains for smaller models and long tables.
๐ Abstract
Table question answering requires models to recover semantic relations encoded implicitly by two-dimensional layout, merged cells, and hierarchical headers. Current pipelines typically use HTML or Markdown as intermediate table representations, but these layout-oriented serializations introduce markup overhead and require large language models to infer header-cell alignments from row and column spans. We propose Semantic Triplet Restoration (STR), a protocol that rewrites each cell as an atomic fact <item path, feature path, value>, where the item path specifies the row-wise entity, the feature path specifies the hierarchical attribute, and the value contains the cell content. We also present TripletQL, a lightweight query-aware router that uses STR to select an appropriate rendering or filtered subset of triplets for each question. Across four Chinese and English table-QA benchmarks, STR matches or improves upon HTML-based baselines while reducing input tokens. The relative benefit grows for smaller language models and longer table contexts, suggesting that explicit semantic representations are especially useful under constrained inference budgets. Code and data are available at https://github.com/Phoenix-ni/STR.git .