Abstract: A hypernym of a word in utterance data may be probabilistically determined. The utterance data may correspond to a spoken query or command. A redacted utterance may be derived by replacing the word with the hypernym. The hypernym may be determined by applying noise to a position in a hierarchical embedding that corresponds to the word. The word may be identified as being potentially sensitive. The hierarchical embedding may be a Hyperbolic embedding that may indicate hierarchical relationships between individual words of a corpus of words, such as “red” is a “color” or “Austin” is in “Texas.” Noise may be applied by obtaining a first value in Euclidean space based on a second value in Hyperbolic space, and obtaining a third value in Hyperbolic space based on the first value in Euclidean space. The second value in Hyperbolic space may correspond to the word.

Abstract: Systems, methods, and computer-readable media are disclosed for providing privacy and intent preserving redactions of text derived from utterance data. Certain embodiments provide new techniques for using MadLib-style replacements to replace one or more terms or phrases in a text string. Example methods may include receiving utterance data and determining a public portion and a private portion of the utterance data. Certain methods include determining a cluster of candidates having a same semantic context as the private portion and identifying from within the cluster of candidates a first candidate. Certain methods include determining a redacted utterance comprising the public portion of the utterance and the first candidate. Certain methods include providing the redacted utterance to downstream systems and processes.