Improving Robustness of Tabular Retrieval via Representational Stability

ArXi:2604.24040v2 Announce Type: cross Transformer-based table retrieval systems flatten structured tables into token sequences, making retrieval sensitive to the choice of serialization even when table semantics remain unchanged. We show that semantically equivalent serializations, such as $\texttt{cs}$, $\texttt{ts}$, $\texttt{html}$, $\texttt{markdown}$, and $\texttt{ddl}$, can produce substantially different embeddings and retrieval results across multiple benchmarks and retriever families.