Robots also learn the second language, which helps improve natural language skills

Author:chinatopwin
Source:chinatopwin
Release on:2017-08-10

From Alexa and Siri to countless chat robots and automated customer support systems,

computers are learning to speak. The only problem is that computers are often confused.

A research team at Salesforce has come up with a clever way to improve the performance

of many modern language programs, namely, teach another language before training the

algorithm to accomplish other tasks.
Keeping coherent conversations in church machines remains a major challenge in the field

of artificial intelligence, because it is often necessary to have a broader understanding or

common sense of the world in order to understand clearly the meaning of spoken or written

language.
In fact, the training of machine learning system for transformation between two languages,

it can automatically teach collocation relationship and appropriate context. When this system

is used as a training to dialogue, emotional or contains text in the machine learning based

detection system, its performance is far better than the an ab initio training system.
"We're using Machine Translation data, and we're basically teaching the model how to

understand vocabulary and context," said Richard Sugar, chief scientist at Salesforce and an

expert on machine learning and language."
This work is machine learning progress an example of artificial intelligence system of language

skills. Many computer vision system using deep learning network based on some form of pre

training, according to Su Ge said, Machine Translation may provide a similar way to guide the

natural language system.
Salesforce is a sales, management of customer interaction between marketing and business

online platform, through the platform of Einstein has provided a series of artificial intelligence

tools. One is the automatic classification of email or chat messages emotional tools, another

is the staff according to his or her previous activities in pursuit of potential customers the

priority.

He believes the discovery will help improve the natural language capabilities of the Einstein

platform. "It's very useful for chat bots and automated customer support," he says"
Salesforce researchers trained a translation between English and German deep learning

system. It involves a large number of translation input document into a multi-layer neural

network, and continuously adjust the network parameters, until it has learned to automatically

generate a decent translation. This system is represented by a vector of words. This is a

common way of encoding and parsing the meaning of the text.
Then, the researchers conducted training on the bilingual network, let them do all kinds of

things: a text emotion; to classify different types of problems; and then answer the questions.

They also found that pre training network more than they did not learn the second language

performance.
Machine Translation data set is particularly large, which contributes to machine learning this

challenge. The researchers involved in the project Salesforce Mccain said, "there is an

important connection between translation and other language translation. The data set is very

common, they contain information that is very useful for Natural Language Processing."