# US 2023/0267285 A1 · Using one or more neural networks to perform text translation

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

- Number: US 2023/0267285 A1
- Document ID: US20230267285A1
- Status: Published application
- Kind: A1
- Application number: 17/666,158
- Filed: 2022-02-07
- Published: 2023-08-24
- Earliest claimed: 2022-02-07
- Assignees: NVIDIA Corporation
- Applicants: NVIDIA Corporation
- Inventors: Xianchao Wu; Yi Dong; Peiying Ruan; Simon See; Scott Nunweiler
- Web page: https://patentgrep.com/patent/US20230267285A1

## CPC

- G06F 40/44: Statistical methods, e.g. probability models
- G06F 40/58: Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation
- G06F 40/166: Editing, e.g. inserting or deleting
- G06F 40/263: Language identification
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/063: using electronic means
- G06N 3/08: Learning methods
- G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning
- G06F 40/51: Translation evaluation
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 3/09: Supervised learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Apparatuses, systems, and techniques to translate a text string. In at least one embodiment, a text string is translated by at least, for example, using one or more neural networks to determine a length of a translated text string before a text string is to be translated.

## Claims

### Claim 1

A processor, comprising:
one or more circuits to use one or more neural networks to determine a length of a translated text string before a text string is to be translated.

### Claim 2 (depends on 1)

The processor of claim 1, wherein the one or more circuits are further to:
re-order, based at least in part on a language of the text string, one or more elements of the text string to generate a re-ordered text string; and
calculate the length of the translated text string based at least in part on the re-ordered text string before the text string is to be translated.

### Claim 3 (depends on 2)

The processor of claim 2, wherein the one or more circuits are further to re-order the one or more elements of the text string by at least parsing the text string to calculate one or more dependencies among the one or more elements of the text string.

### Claim 4 (depends on 1)

The processor of claim 1, wherein the one or more circuits are further to use the one or more neural networks to translate the text string based at least in part on the length of the translated text string.

### Claim 5 (depends on 1)

The processor of claim 1, wherein the length comprises a number of words for the translated text string.

### Claim 6 (depends on 1)

The processor of claim 1, wherein the one or more neural networks are further to generate a translation of the text string independent of word order.

### Claim 7

A system, comprising:
one or more computers having one or more processors to use one or more neural networks to determine a length of a translated text string before a text string is to be translated.

### Claim 8 (depends on 7)

The system of claim 7, wherein the one or more processors are further to:
identify a language corresponding to the text string;
predict the length of the translated text string based on the identified language;
translate the text string to generate a translated text string based on the length; and
re-order one or more elements of the translated text string.

### Claim 9 (depends on 8)

The system of claim 8, wherein the one or more processors are further to re-order the one or more elements of the translated text string based on the language corresponding to the text string and a language corresponding to the translated text string.

### Claim 10 (depends on 7)

The system of claim 7, wherein the one or more processors are further to obtain the text string based at least in part on input from one or more translation software applications executing on one or more devices.

### Claim 11 (depends on 7)

The system of claim 7, wherein the length indicates at least a maximum number of words for a translation of a word of the text string.

### Claim 12 (depends on 7)

The system of claim 7, wherein the one or more processors are further to:
use a first neural network of the one or more neural networks to predict the length of the translated text string; and
use a second neural network of the one or more neural networks to generate a translated text string based, at least in part, on the predicted length.

### Claim 13

A machine-readable medium having stored thereon a set of instructions, which if performed by one or more processors, cause the one or more processors to use one or more neural networks to determine a length of a translated text string before a text string is to be translated.

### Claim 14 (depends on 13)

The machine-readable medium of claim 13, wherein the set of instructions further include instructions, which if performed by the one or more processors, cause the one or more processors to re-order one or more words of the text string based at least in part on a word order of a language corresponding to a translated text string.

### Claim 15 (depends on 14)

The machine-readable medium of claim 14, wherein the word order indicates how words are arranged in a sequence in a sentence.

### Claim 16 (depends on 13)

The machine-readable medium of claim 13, wherein the set of instructions further include instructions, which if performed by the one or more processors, cause the one or more processors to use at least one or more encoder and decoder layers to translate the text string.

### Claim 17 (depends on 13)

The machine-readable medium of claim 13, wherein the set of instructions further include instructions, which if performed by the one or more processors, cause the one or more processors to calculate one or more embeddings based at least in part on the text string, wherein the one or more embeddings correspond to at least one or more positions of one or more words in the text string.

### Claim 18 (depends on 13)

The machine-readable medium of claim 13, wherein the set of instructions further include instructions, which if performed by the one or more processors, cause the one or more processors to:
use a first portion of the one or more neural networks to predict the length; and
use a second portion of the one or more neural networks to calculate a translated text string using the predicted length.

### Claim 19

A processor, comprising:
one or more circuits to train one or more neural networks to determine a length of a translated text string before a text string is to be translated.

### Claim 20 (depends on 19)

The processor of claim 19, wherein the one or more circuits are further to:
calculate the length of the translated text string as a set of integers comprising one or more numbers of words for one or more words of the text string; and
train the one or more neural networks by at least comparing the set of integers with a set of reference integers corresponding to a translated text string.

### Claim 21 (depends on 19)

The processor of claim 19, wherein the one or more circuits are further to train the one or more neural networks using at least one or more minimum square error (MSE) loss functions.

### Claim 22 (depends on 19)

The processor of claim 19, wherein the one or more circuits are further to calculate one or more embeddings based at least in part on the text string indicating one or more classifications of one or more words in the text string.

### Claim 23 (depends on 19)

The processor of claim 19, wherein the one or more circuits are further to perform training using one or more tasks corresponding to at least one of: predicate masking, subject phrase masking, and object phrase masking.

### Claim 24 (depends on 19)

The processor of claim 19, wherein the one or more circuits are further to:
train a first portion of the one or more neural networks to infer the length of the translated text string; and
train a second portion of the one or more neural networks to generate a translated text string using the inferred length.

### Claim 25

A method, comprising:
training one or more neural networks to determine a length of a translated text string before a text string is to be translated.

### Claim 26 (depends on 25)

The method of claim 25, further comprising causing the one or more neural networks to calculate the length of the translated text string by at least processing one or more embeddings corresponding to the text string.

### Claim 27 (depends on 25)

The method of claim 25, further comprising:
computing loss based at least in part on the calculated length of the translated text string and a reference length corresponding to a translated text string; and
updating the one or more neural networks based at least in part on the computed loss.

### Claim 28 (depends on 25)

The method of claim 25, further comprising:
training a first neural network of the one or more neural networks to predict the length of the translated text string; and
training a second neural network of the one or more neural networks to calculate a translated text string using the predicted length.

### Claim 29 (depends on 25)

The method of claim 25, further comprising:
calculating dependencies among words of the text string;
using the dependencies to re-order the words of the text string to generate a re-ordered text string; and
using at least the re-ordered text string and the length to translate the text string to generate the translated text string.

### Claim 30 (depends on 25)

The method of claim 25, further comprising performing training using at least one of: an order of subject-predicate ordering distortion prediction task, an order of predicate-object order distortion prediction task, and an order of n-tuples of predicate-argument structures prediction task.
