Enhancing multi-lingual embeddings for cross-lingual question-answer system

1. A method comprising:

receiving, by a question-answer system, a question in a first language and the question in a second language;

predicting, by the question-answer system, a first answer to the question in the first language and a second answer to the question in the second language;

generating, by the question-answer system, a first vector representing the question in the first language and a second vector representing the question in the second language; and

adjusting the question-answer system based on the first and second answers and the first and second vectors such that when the question-answer system subsequently generates a third vector representing the question in the first language and a fourth vector representing the question in the second language, a distance between the third and fourth vectors is less than a distance between the first and second vectors.

2. The method of claim 1 , further comprising:

determining a first loss based on the first answer and a reference answer;

determining a second loss based on the second answer and the reference answer; and

determining a third loss based on the first answer and the second answer, wherein adjusting the question-answer system is based on the first, second, and third losses, and the distance between the first and second vectors.

3. The method of claim 1 , wherein the distance between the first and second vectors is based on a cosine of the first and second vectors.

4. The method of claim 1 , further comprising receiving a context, wherein the first answer comprises text extracted from the context.

5. The method of claim 1 , further comprising machine translating the question in the first language to produce the question in the second language.

6. The method of claim 1 , further comprising training the question-answer system to predict answers based on training data in the first language.

7. An apparatus comprising:

a memory; and

a hardware processor communicatively coupled to the memory, the hardware processor configured to:

receive a question in a first language and the question in a second language;

predict, using a machine learning model, a first answer to the question in the first language and a second answer to the question in the second language;

generate, using the machine learning model, a first vector representing the question in the first language and a second vector representing the question in the second language; and

adjust the machine learning model based on the first and second answers and the first and second vectors such that when the machine learning model subsequently generates a third vector representing the question in the first language and a fourth vector representing the question in the second language, a distance between the third and fourth vectors is less than a distance between the first and second vectors.

8. The apparatus of claim 7 , wherein the hardware processor is further configured to:

determine a first loss based on the first answer and a reference answer;

determine a second loss based on the second answer and the reference answer; and

determine a third loss based on the first answer and the second answer, wherein adjusting the machine learning model is based on the first, second, and third losses, and the distance between the first and second vectors.

9. The apparatus of claim 7 , wherein the distance between the first and second vectors is based on a cosine of the first and second vectors.

10. The apparatus of claim 7 , wherein the hardware processor is further configured to receive a context, wherein the first answer comprises text extracted from the context.

11. The apparatus of claim 7 , wherein the hardware processor is further configured to machine translate the question in the first language to produce the question in the second language.

12. The apparatus of claim 7 , wherein the hardware processor is further configured to train the machine learning model to predict answers based on training data in the first language.

13. A method comprising:

receiving, by a question-answer system, a context in a first language and a question in a second language;

predicting, by the question-answer system and based on the context, a first answer to the question;

generating, by the question-answer system, a first vector representing the question; and

determining a first probability distribution of languages predicted, based on the first vector, for the second language;

adjusting the question-answer system based on the first answer and the first vector such that when the question-answer system subsequently predicts a second answer to the question, generates a second vector representing the question, and determines a second probability distribution of languages predicted, based on the second vector, for the second language:

a determined loss between the second answer and a reference answer is less than a determined loss between the first answer and the reference answer; and

the second probability distribution is closer to a uniform distribution than the first probability distribution.

14. The method of claim 13 , further comprising determining a first loss based on the first probability distribution and the second language, wherein adjusting the question-answer system is further based on the first loss such that a determined second loss based on the second probability distribution and the second language is less than the first loss based on the first probability distribution and the second language.

15. The method of claim 13 , further comprising machine translating the question in the first language to produce the question in the second language.

16. The method of claim 13 , further comprising training the question-answer system to predict answers based on training data in the first language.

17. An apparatus comprising:

a memory; and

a hardware processor communicatively coupled to the memory, the hardware processor configured to:

receive, by a question-answer system, a context in a first language and a question in a second language;

predict, by the question-answer system and based on the context, a first answer to the question;

generate, by the question-answer system, a first vector representing the question; and

determine a first probability distribution of languages predicted, based on the first vector, for the second language;

adjust the question-answer system based on the first answer and the first vector such that when the question-answer system subsequently predicts a second answer to the question, generates a second vector representing the question, and determines a second probability distribution of languages predicted, based on the second vector, for the second language:

a determined loss between the second answer and a reference answer is less than a determined loss between the first answer and the reference answer; and

the second probability distribution is closer to a uniform distribution than the first probability distribution.

18. The apparatus of claim 17 , wherein the hardware processor is further configured to determine a first loss based on the first probability distribution and the second language, wherein adjusting the question-answer system is further based on the first loss such that a determined second loss based on the second probability distribution and the second language is less than the first loss based on the first probability distribution and the second language.

19. The apparatus of claim 17 , wherein the hardware processor is further configured to machine translate the question in the first language to produce the question in the second language.

20. The apparatus of claim 17 , wherein the hardware processor is further configured to train the question-answer system to predict answers based on training data in the first language.