Inspecting the Representation Manifold of Differentially-Private Text
This work addresses the problem of geometric distortion in differentially-private text generation for privacy-preserving NLP applications, though it is incremental in exploring representation space effects.
The paper investigated how differentially-private text paraphrasing methods distort the representation manifold, finding that sentence-level methods better preserve topological consistency with human-written paraphrases, with masked paraphrasing outperforming causal paraphrasing in maintaining structural complexity.
Differential Privacy (DP) for text has recently taken the form of text paraphrasing using language models and temperature sampling to better balance privacy and utility. However, the geometric distortion of DP regarding the structure and complexity in the representation space remains unexplored. By estimating the intrinsic dimension of paraphrased text across varying privacy budgets, we find that word-level methods severely raise the representation manifold, while sentence-level methods produce paraphrases whose manifolds are topologically more consistent with human-written paraphrases. Among sentence-level methods, masked paraphrasing, compared to causal paraphrasing, demonstrates superior preservation of structural complexity, suggesting that autoregressive generation propagates distortions from unnatural word choices that cascade and inflate the representation space.