Removing personal information from text using multiple levels of redaction

A system, comprising:

• at least one computer comprising at least one processor and at least one memory, the at least one computer configured to:
• receive text of a message from a user;
• generate first redacted text by replacing one or more digits of the text with a token representing digits;
• compute label scores for words of the first redacted text by processing the first redacted text with one or more layers of a neural network, wherein each label score indicates a match between a word and a class of words and wherein the label scores comprise first label scores for a first word of the first redacted text;
• select a first label for the first word using the first label scores;
• and
• generate second redacted text by replacing the first word with the first label in the first redacted text.

One or more non-transitory, computer-readable media comprising computer executable instructions that, when executed, cause at least one processor to perform actions comprising:

• receiving text of a message from a user;
• generating first redacted text by replacing one or more digits of the text with a token representing digits;
• computing label scores for words of the first redacted text by processing the first redacted text with one or more layers of a neural network, wherein each label score indicates a match between a word and a class of words and wherein the label scores comprise first label scores for a first word of the first redacted text;
• selecting a first label for the first word using the first label scores;
• and
• generating second redacted text by replacing the first word with the first label in the first redacted text.