Natural language translation device

• US 10,311,149 B1
• Filed: 08/08/2018
• Issued: 06/04/2019
• Est. Priority Date: 08/08/2018
• Status: Active Grant

• Current Assignee: Gyrfalcon Technology Inc.
• Original Assignee: Gyrfalcon Technology Inc.
• Inventors: Yang, Lin, Dong, Patrick Z., Chi, Catherine, Young, Charles Jin, Dong, Jason Z, Sun, Baohua
• Primary Examiner(s): Chawan, Vijay B
• Application Number:
  US16/058,801
• Time in Patent Office: 300 Days
• Field of Search: 704 9, 704260, 704258, 704 10, 704254, 704268, 704269, 704277, 704 2, 382187, 382177, 382186, 382228, 706 38, 706 29, 706 31, 706 17
• CPC Class Codes: G06F 40/126 Character encoding
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06F 40/51 Translation evaluation
• CPC Class Codes: G06F 40/53 Processing of non-Latin text kana-to-kanji conversion G06F40/129; vowelisation G06F40/232
• CPC Class Codes: 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
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

Natural language translation device contains a bus, an input interface connecting to the bus for receiving a source sentence in a first natural language to be translated to a target sentence in second natural language one word at a time in sequential order. A two-dimensional (2-D) symbol containing a super-character characterizing the i-th word of the target sentence based on the received source sentence is formed in accordance with a set of 2-D symbol creation rules. The i-th word of the target sentence is obtained by classifying the 2-D symbol via a deep learning model that contains multiple ordered convolution layers in a Cellular Neural Networks or Cellular Nonlinear Networks (CNN) based integrated circuit.

Claims

1. A natural language translation device comprising:

a bus;

an input interface operatively connecting to the bus for receiving a source sentence in a first natural language to be translated to a target sentence in second natural language one word at a time in sequential order;

a processing unit operatively connecting to the bus for forming a two-dimensional (2-D) symbol in accordance with a set of 2-D symbol creation rules using a 2-D symbol creation module installed thereon, the 2-D symbol containing a super-character having a characteristic suggesting the i-th word of the target sentence based on the received source sentence; and

a Cellular Neural Networks or Cellular Nonlinear Networks (CNN) based integrated circuit operatively connecting to the bus for classifying the 2-D symbol via a deep learning model that contains a plurality of ordered convolutional layers.

2. The device of claim 1, further comprises a display unit operatively connecting to the bus for displaying already-translated portion of the target sentence.

3. The device of claim 1, wherein the 2-D symbol comprises a matrix of N×

N pixels of K-bit data, and the matrix is divided into M×

M sub-matrices with each of the sub-matrices containing (N/M)×

(N/M) pixels, said each of the sub-matrices representing one ideogram, where K, N and M are positive integers or whole numbers, and N is a multiple of M.

4. The device of claim 1, wherein the set of 2-D symbol creation rules comprises:

dividing the 2-D symbol into first and second parts;

converting the source sentence to a first set of ideograms;

converting the first word to the (i−

1)-th word of the target sentence to a second set of ideograms; and

forming the 2-D symbol by including the first set of ideograms in the first part and including the second set of ideograms in the second part.

5. The device of claim 1, wherein the set of 2-D symbol creation rules comprises:

converting the source sentence to a set of ideograms;

dividing the 2-D symbol into a predetermined number of sub-symbols with each sub-symbol for one ideogram; and

forming the 2-D symbol by including relevant portion of the set of ideograms in a scheme such that the ideogram corresponding to the i-th word is located in a predefined location, by filling each unoccupied sub-symbol with blank space if needed.

6. The device of claim 1, wherein the set of 2-D symbol creation rules comprises:

dividing the 2-D symbol into a predetermined number of sub-symbols;

creating the predetermined number of groups of consecutive words from the source sentence with each group encompassing the i-th word of the source sentence and said each group corresponding to one of the sub-symbols; and

forming the 2-D symbol by including ideograms converted from respective groups of consecutive words for all sub-symbols.

7. The device of claim 1, wherein the CNN based integrated circuit comprises a plurality of CNN processing engines operatively coupled to at least one input/output data bus, the plurality of CNN processing engines being connected in a loop with a clock-skew circuit, each CNN processing engine comprising:

a CNN processing block configured for simultaneously performing convolutional operations using the 2-D symbol and pre-trained filter coefficients of the plurality of ordered convolutional layers;

a first set of memory buffers operatively coupling to the CNN processing block for storing the 2-D symbol; and

a second set of memory buffers operatively coupling to the CNN processing block for storing the pre-trained filter coefficients.

8. The device of claim 7, wherein the pre-trained filter coefficients comprises multiple bi-valued 3×

3 filter kernels.

9. The device of claim 8, wherein each of the bi-valued 3×

3 filter kernels contains fixed point number with positive or negative of same numerical value.

10. The device of claim 7, wherein the CNN based integrated circuit further performs pooling operations and activation operations.

11. The device of claim 1, further comprises a memory operatively connected to the bus for providing data storage for the processing unit.

12. A method of translating natural language using a Cellular Neural Networks or Cellular Nonlinear Networks (CNN) based integrated circuit, the method comprising:

receiving, in a natural language translation device having a two-dimensional (2-D) symbol creation module installed thereon, a source sentence in a first natural language to be translated to a target sentence in second natural language one word at a time in sequential order;

forming, with the 2-D symbol creation module, a multi-layer two-dimensional (2-D) symbol in accordance with a set of 2-D symbol creation rules, the 2-D symbol being a matrix of N×

N pixels of K-bit data that contains a super-character, the matrix being divided into M×

M sub-matrices with each of the sub-matrices containing (N/M)×

(N/M) pixels, said each of the sub-matrices representing one ideogram, the super-character having a characteristic suggesting the i-th word of the target sentence based on the received source sentence, where K, N and M are positive integers or whole numbers, and N is a multiple of M; and

obtaining, with the 2-D symbol creation module, the i-th word of the target sentence by classifying the 2-D symbol via a deep learning model having a plurality of ordered convolution layers in a Cellular Neural Networks or Cellular Nonlinear Networks (CNN) based integrated circuit.

13. The method of claim 12, further comprises displaying already-translated portion of the target sentence in a display unit of the natural language translation device.

14. The method of claim 12, wherein the set of 2-D symbol creation rules comprises:

dividing the 2-D symbol into first and second parts;

converting the source sentence to a first set of ideograms;

converting the first word to the (i−

1)-th word of the target sentence to a second set of ideograms; and

forming the 2-D symbol by including the first set of ideograms in the first part and including the second set of ideograms in the second part.

15. The method of claim 12, wherein the set of 2-D symbol creation rules comprises:

converting the source sentence to a set of ideograms;

dividing the 2-D symbol into a predetermined number of sub-symbols with each sub-symbol for one ideogram; and

forming the 2-D symbol by including relevant portion of the set of ideograms in a scheme such that the ideogram corresponding to the i-th word is located in a predefined location, by filling each unoccupied sub-symbol with blank space if needed.

16. The method of claim 12, wherein the set of 2-D symbol creation rules comprises:

dividing the 2-D symbol into a predetermined number of sub-symbols;

creating the predetermined number of groups of consecutive words from the source sentence with each group encompassing the i-th word of the source sentence and said each group corresponding to one of the sub-symbols; and

forming the 2-D symbol by including ideograms converted from respective groups of consecutive words for all sub-symbols.

17. The method of claim 12, wherein the CNN based integrated circuit comprises a plurality of CNN processing engines operatively coupled to at least one input/output data bus, the plurality of CNN processing engines being connected in a loop with a clock-skew circuit, each CNN processing engine comprising:

a CNN processing block configured for simultaneously performing convolutional operations using the 2-D symbol and pre-trained filter coefficients of the plurality of ordered convolutional layers;

a first set of memory buffers operatively coupling to the CNN processing block for storing the 2-D symbol; and

a second set of memory buffers operatively coupling to the CNN processing block for storing the pre-trained filter coefficients.

18. The method of claim 17, wherein the pre-trained filter coefficients comprises multiple bi-valued 3×

3 filter kernels.

19. The method of claim 18, wherein each of the bi-valued 3×

3 filter kernels contains fixed point number with positive or negative of same numerical value.

20. The method of claim 17, wherein the CNN based integrated circuit further performs pooling operations and activation operations.