# US 2003/0233222 A1 · Statistical translation using a large monolingual corpus

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

- Number: US 2003/0233222 A1
- Document ID: US20030233222A1
- Status: Published application
- Kind: A1
- Application number: 10/401,134
- Filed: 2003-03-26
- Published: 2003-12-18
- Earliest claimed: 2002-03-26
- Assignees: University of Southern California
- Inventors: Radu Soricut; Daniel Marcu; Kevin Knight
- Web page: https://patentgrep.com/patent/US20030233222A1

## CPC

- G06F 40/44: Statistical methods, e.g. probability models
- G06F 40/45: Example-based machine translation; Alignment
- G06F 40/49: using very large corpora, e.g. the web
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A statistical machine translation (MT) system may use a large monolingual corpus to improve the accuracy of translated phrases/sentences. The MT system may produce a alternative translations and use the large monolingual corpus to (re)rank the alternative translations.

## Claims

### Claim 1

A method comprising:
receiving an input text string in a source language;
comparing a plurality of alternate translations for said input text string in a target language to text segments in a monolingual corpus in the target language; and
recording a number of occurrences of each of at least a plurality of said plurality of alternate translations.

### Claim 2 (depends on 1)

The method of claim 1, further comprising:
assigning a probability score to said at least a plurality of said plurality of alternative translations.

### Claim 3 (depends on 2)

The method of claim 2, further comprising:
selecting at least a portion of said plurality of alternate translations based on the probability scores.

### Claim 4 (depends on 2)

The method of claim 2, further comprising:
ranking the scored alternative translations based on the probability scores.

### Claim 5 (depends on 4)

The method of claim 4, further comprising:
re-ranking the ranked alternate translations based on the recorded number of occurrences.

### Claim 6 (depends on 5)

The method of claim 5, further comprising:
ranking and re-ranking alternate translations for a plurality of input text strings.

### Claim 7 (depends on 6)

The method of claim 6, further comprising:
training a statistical machine translator with re-ranked alternate translations.

### Claim 8 (depends on 6)

The method of claim 6, further comprising:
building parallel corpora in the source language and the target language using re-ranked alternate translations.

### Claim 9 (depends on 1)

The method of claim 1, further comprising:
building a finite state acceptor for the input text string operative to encode a plurality of alternative translations for said input text string in the target language.

### Claim 10 (depends on 9)

The method of claim 9, wherein said comparing the plurality of alternate translations to text segments in the monolingual corpus comprises inputting text segments in the monolingual corpus to the finite state acceptor.

### Claim 11 (depends on 1)

The method of claim 1, further comprising:
generating a plurality of alternate translations for said input text string in a target language using a language model.

### Claim 12 (depends on 1)

The method of claim 1, further comprising:
deriving the monolingual corpus from a collection of documents.

### Claim 13 (depends on 1)

The method of claim 1, further comprising:
deriving the monolingual corpus from text on World Wide Web.

### Claim 14

A method comprising:
receiving an input text string in a source language;
building a finite state acceptor for the input text string operative to encode a plurality of alternative translations for said input text string in a target language;
inputting text segments in a monolingual corpus in the to the finite state acceptor; and
recording text segments accepted by the finite state acceptor; and
recording a number of occurrences for each of said accepted text segments.

### Claim 15 (depends on 14)

The method of claim 14, further comprising:
ranking each of the accepted text segments based at least in part on the number of occurrences.

### Claim 16 (depends on 15)

The method of claim 15, further comprising:
identifying a probability for each of the accepted text segments; and
ranking the accepted text segments at least in part on the probabilities.

### Claim 17

An apparatus comprising:
a translation model component operative to receive an input text string in a source language and generate a plurality of alternate translations for said input text strings, the alternate translation comprising text segments in a target language;
a corpus comprising a plurality of text segments in the target language; and
a translation ranking module operative to record a number of occurrences of said alternate translations in the corpus.

### Claim 18 (depends on 17)

The apparatus of claim 17, wherein said translation model component is operative to generate a finite state acceptor encoding the plurality of alternate translations.

### Claim 19 (depends on 17)

The apparatus of claim 17, further comprising:
wherein the corpus comprises documents available on the World Wide Web.

### Claim 20

An article comprising a machine-readable medium including machine-executable instructions, the instructions operative to cause the machine to:
receive an input text string in a source language;
compare a plurality of alternate translations for said input text string in a target language to text segments in a monolingual corpus in the target language; and
record a number of occurrences of each of at least a plurality of said plurality of alternate translations.

### Claim 21 (depends on 20)

The article of claim 20, further comprising instructions operative to cause the machine to:
assign a probability score to said at least a plurality of said plurality of alternative translations.

### Claim 22

An article comprising a machine-readable medium including machine-executable instructions, the instructions operative to cause the machine to:
receive an input text string in a source language;
build a finite state acceptor for the input text string operative to encode a plurality of alternative translations for said input text string in a target language;
input text segments in a monolingual corpus in the to the finite state acceptor; and
record text segments accepted by the finite state acceptor; and
record a number of occurrences for each of said accepted text segments.

### Claim 23 (depends on 22)

The article of claim 22, further comprising instructions operative to cause the machine to:
rank each of the accepted text segments based at least in part on the number of occurrences.
