# US 8,751,439 B2 · Preserving privacy in natural language databases

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 8,751,439 B2
- Document ID: US8751439B2
- Status: Granted
- Kind: B2
- Application number: 13/926,404
- Filed: 2013-06-25
- Published: 2013-10-31
- Granted: 2014-06-10
- Earliest claimed: 2004-07-30
- Assignees: AT&T Intellectual Property II, L.P.
- Applicants: AT&T Intellectual Property II, L.P.
- Inventors: Dilek Z. Hakkani-Tur; Yucel Saygin; Min Tang; Gokhan Tur
- Web page: https://patentgrep.com/patent/US8751439B2

## CPC

- G06F 16/215: Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
- G06F 21/00: Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F 21/6254: {by anonymising data, e.g. decorrelating personal data from the owner's identification}
- G10L 15/26: Speech to text systems (G10L15/08 takes precedence)
- G06F 21/60: Protecting data
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

An apparatus and a method for preserving privacy in natural language databases are provided. Natural language input may be received. At least one of sanitizing or anonymizing the natural language input may be performed to form a clean output. The clean output may be stored.

## Claims

### Claim 1

A method comprising:
selecting a transcription of a spoken natural language input from a speaker, the transcription comprising sensitive information and non-sensitive information;
sanitizing the sensitive information, to form a clean transcription of the spoken natural language input, the clean transcription having sanitized text and non-sanitized text; and
anonymizing the non-sanitized text in the clean transcription by modifying a feature vector associated with the non-sanitized text, to yield a modified feature vector, such that the modified feature vector is associated with a mean feature vector of different speakers.

### Claim 2 (depends on 1)

The method of claim 1, wherein sanitizing the sensitive information further comprises:
finding a named entity in the spoken natural language input; and
performing, on the named entity, one of value distortion, value disassociation, and value class membership to preserve privacy in a spoken natural language database.

### Claim 3 (depends on 2)

The method claim 2, wherein sanitizing the sensitive information further comprises:
performing, on the named entity, two of value distortion, value disassociation, and value class membership to preserve privacy in the spoken natural language database.

### Claim 4 (depends on 2)

The method of claim 2, further comprising performing value class membership by replacing a value with a generic token.

### Claim 5 (depends on 4)

The method of claim 4, wherein performing value class membership further comprises:
placing an indication of one of a gender and other information in the generic token when the value represents an identification of a person.

### Claim 6 (depends on 2)

The method of claim 2, wherein finding the named entity further comprises using an automated approach to detect a name.

### Claim 7 (depends on 1)

The method of claim 1, wherein anonymizing further comprises replacing a word in the non-sanitized text with a corresponding synonym.

### Claim 8 (depends on 1)

The method of claim 1, wherein sanitizing the sensitive information further comprises changing a distribution of utterances and semantically labeled data in the non-sanitized text.

### Claim 9

A system comprising:
a processor; and
a computer-readable storage medium having instructions stored which, when executed by the processor, result in the processor performing operations comprising:
selecting a transcription of a spoken natural language input from a speaker, the transcription comprising sensitive information and non-sensitive information;
sanitizing the sensitive information, to form a clean transcription of the spoken natural language input, the clean transcription having sanitized text and non-sanitized text; and
anonymizing the non-sanitized text in the clean transcription by modifying a feature vector associated with the non-sanitized text, to yield a modified feature vector, such that the modified feature vector is associated with a mean feature vector of different speakers.

### Claim 10 (depends on 9)

The system of claim 9, wherein sanitizing the sensitive information further comprises:
finding a named entity in the spoken natural language input; and
performing, on the named entity, one of value distortion, value disassociation, and value class membership to preserve privacy in a spoken natural language database.

### Claim 11 (depends on 10)

The system of claim 10, wherein sanitizing the sensitive information further comprises:
performing, on the named entity, two of value distortion, value disassociation, and value class membership to preserve privacy in the spoken natural language database.

### Claim 12 (depends on 10)

The system of claim 10, the computer-readable storage medium having additional instructions stored which result in the operations further comprising performing value class membership by replacing a value with a generic token.

### Claim 13 (depends on 12)

The system of claim 12, wherein performing value class membership further comprises:
placing an indication of one of a gender and other information in the generic token when the value represents an identification of a person.

### Claim 14 (depends on 10)

The system of claim 10, wherein finding the named entity further comprises using an automated approach to detect a name.

### Claim 15 (depends on 9)

The system of claim 9, wherein anonymizing further comprises replacing a word in the non-sanitized text with a corresponding synonym.

### Claim 16 (depends on 9)

The system of claim 9, wherein sanitizing the sensitive information further comprises changing a distribution of utterances and semantically labeled data in the non-sanitized text.

### Claim 17

A computer-readable storage device having instructions stored which, when executed by a computing device, result in the computing device performing operations comprising:
selecting a transcription of a spoken natural language input from a speaker, the transcription comprising sensitive information and non-sensitive information;
sanitizing the sensitive information, to form a clean transcription of the spoken natural language input, the clean transcription having sanitized text and non-sanitized text; and
anonymizing the non-sanitized text in the clean transcription by modifying a feature vector associated with the non-sanitized text, to yield a modified feature vector, such that the modified feature vector is associated with a mean feature vector of different speakers.

### Claim 18 (depends on 17)

The computer-readable storage device of claim 17, wherein sanitizing the sensitive information further comprises:
finding a named entity in the spoken natural language input; and
performing, on the named entity, one of value distortion, value disassociation, and value class membership to preserve privacy in a spoken natural language database.

### Claim 19 (depends on 18)

The computer-readable storage device of claim 18, wherein sanitizing the sensitive information further comprises:
performing, on the named entity, two of value distortion, value disassociation, and value class membership to preserve privacy in the spoken natural language database.

### Claim 20 (depends on 19)

The computer-readable storage device of claim 19, the computer-readable storage device having additional instructions stored which result in the operations further comprising performing value class membership by replacing a value with a generic token.