Artificial Intelligence - Interesting breakthroughs in Natural Language Processing
TCS iON Digital Learning Hub | January 21,2020

The difference in communication between humans and machines is slowly blurring and it is becoming increasingly difficult for many users to identify whether they are talking to a chatbot or a human on Instant Messaging systems.  Much of this development can be attributed to advancements made in an area called Natural Language Processing, or NLP for short.

 

NLP is a subdivision of artificial intelligence that aims to bridge the gap between computer and human communication.  In fact, it is a branch of computer science that helps computers to understand the human language and the many nuances and intricacies that come with it.

 

Understanding the human language has always been a challenge for computers. Even the most powerful computers fail to understand the subtle communication words and nuances that human toddlers can understand with ease. This inability to understand the finer aspects of human communication impedes computers from having a "normal" conversation with human users and poses a significant challenge to language scientists and computer researchers worldwide.

The good news is that the relentless efforts of these scientists and researchers are slowly bridging the gaps in human-computer interaction and is paving the way for more communication-based automation in the future.

The difference in communication between humans and machines is slowly blurring and it is becoming increasingly difficult for many users to identify whether they are talking to a chatbot or a human on Instant Messaging systems.  Much of this development can be attributed to advancements made in an area called Natural Language Processing, or NLP for short.

 

NLP is a subdivision of artificial intelligence that aims to bridge the gap between computer and human communication.  In fact, it is a branch of computer science that helps computers to understand the human language and the many nuances and intricacies that come with it.

 

Understanding the human language has always been a challenge for computers. Even the most powerful computers fail to understand the subtle communication words and nuances that human toddlers can understand with ease. This inability to understand the finer aspects of human communication impedes computers from having a "normal" conversation with human users and poses a significant challenge to language scientists and computer researchers worldwide.

The good news is that the relentless efforts of these scientists and researchers are slowly bridging the gaps in human-computer interaction and is paving the way for more communication-based automation in the future.

Important Breakthroughs in NLP

Here are some of the most important breakthroughs that have taken place in the recent past:
BERT
Bidirectional Encoder Representations from Transformers, or BERT in short, is an NLP model designed by the Google AI team. Its design enables the system to understand the context of a sentence from both the right and left sides of each word. Though the concept is simple, it has greatly improved the ability of systems to answer questions, recognise entities, and improve the general language understanding abilities of machines.

In many ways, this model marks a new era for NLP as it fills many gaps in human-computer communication. This NLP model is also expected to have a big impact on businesses as it can improve customer experience, enhance the search capabilities of an application, and provide better analysis of customer reviews.

 

Phrase-based and neural models for machine translation

Translating from one language to another is a major problem area for machines, especially when a large number of parallel sentences are used in one language. A group of researchers has come up with two models to address this problem, and they are neural and phrase-based models.

 

In both the models, researchers have improved the initialisation of translation algorithms, trained language models in both the source and destination languages, and have enhanced back translation. In the neural model, representations can be shared across languages while the phrase-model is designed to improve the translation of simple words and phrases.

From an organisation's point of view, these developments will help to train machine translation systems and improve their efficiency without any additional supervision. In the long run, it can save time, manpower, and money for organisations.

 

SWAG: Improving the inference of common sense
Humans are wired for inferences while computers are not. According to the abstract of this paper, when you hear the sentence, "she opened the hood of the car", your brain infers that she also checked the engine because that's the reason why someone would open the car's hood. But unfortunately, a machine cannot make this inference and this is exactly what the SWAG model hopes to improve.

SWAG is a dataset with more than 113K multiple choice questions with a rich set of grounded questions that trains computers and helps to improve their ability to infer the context of a conversation. This dataset was validated by crowd workers, as a part of the testing process.

 

SWAG is a revolutionary model that has the potential to improve the development of Q & A systems and conversational UI.

 

Meta-learning Algorithm

Researchers have developed a model-agnostic meta-learning algorithm that will help with low resource pairs to improve translation between languages. This algorithm trains computers on a new language model to help it to associate between the words of different languages, and already, it has been used across 18 European languages with success.

 

This algorithm helps organisations to improve the results of machine translation, especially where the existing correlation between language pairs is small.

Embeddings from Language Models (ELMo)

Embeddings from Language Models (ELMo) is not just a Sesame Street character, but also a model that caught the attention of the machine learning community because of its ability to improve human-computer communications greatly. ELMo is based on language models that give the context for each word in a sentence or paragraph. This is an important breakthrough that establishes the context for each word because this is something that earlier NLP systems failed to do.

 

Needless to say, this ELMo model has created a ton of excitement among organisations because it can be applied to improve the conversational abilities of a plethora of customer-facing applications.


The above list is sure to give you a peek into some of the exciting developments happening on the research front. It won't be long before these research models are implemented within organisations to improve the efficiency and effectiveness of human-machine communication platforms.

 

What do these breakthroughs mean for you?

 

On the face of it, you might wonder what in the world do these scientific developments mean to someone in the IT industry. But a closer look clearly shows that NLP will be a big part of an organisation in the future, thereby opening up many new opportunities and roles.

 

Undoubtedly, NLP is the future and every organisation understands it. This is also why they are keenly watching the scientific developments and their possible implementations at the organisation level. Many organisations are even funding research that they believe will help them in the future. Already, there are many collaborations between businesses and researchers in the area of NLP and it won't be long before these research ideas are translated into usable systems.

 

As an IT professional, this growing field presents a unique opportunity to propel your career and increase your earning potential. Understanding the nuances of NLP and artificial intelligence, in general, can give a big impetus to your career. Your best bet is to start with an artificial intelligence certification to showcase your skills and knowledge to the market and from thereon build your learning and research. One such recognized and outcome linked certification is TCS iON ProCert – AI

TCS iON Digital Learning Hub is a complete learning solution for a varied segment of learners. Learn and upskill your knowledge from the wide range of Courses offered. Prepare, test and analyse your knowledge through the Assessments. Share your ideas and learn from others by being a part of the Communities