How do machines learn and process natural language?

Natural language processing (NLP) is the field of study that comprises the intersection of computer science, AI, and computational linguistics. It enables computers to assess, understand, and extract meaning from human language in a smart and useful way. Using NLP paves ways for developers to organize and structure knowledge to perform tasks, including automatic summarization, translation, person's identification, sentiment analysis, speech recognition, and topic segmentation, among others.

Thanks to recent progresses in data access and computational power, NLP has evolved much more, allowing professionals to derive meaningful results in areas like healthcare, finance, human resources, and others.

What is NLP used for?

NLP has a large variety of uses in almost every industry. It has the ability to automatically handle natural human languages such as speech or text. It can also help a business employee with numerous tasks, eventually bolstering work performance. Many developers typically use NLP algorithms to recapitulate blocks of text to excerpt imperative and main ideas; create chatbots to ask queries and answer appropriately; sentiment analysis; and cognitive assistance and more.

For instance, companies like Yahoo and Google use natural language processing to filter and classify emails by assessing text in emails that flow through users' servers and halting spam before they even enter their inbox. The majority of data or information organizations collect – be it private or public – is unstructured text, including social media conversations, comments on websites, narrative reports, and others. Deriving actionable insights from these data can be challenging.

In its effort to ease these kinds of challenges, the Defense Advanced Research Projects Agency (DARPA) built the Deep Exploration and Filtering of Text (DEFT) program. The program uses NLP to automatically mine germane information and assist analysts to derive actionable insights from it. DEFT aims to address remaining capability gaps related to inference, causal relationships and anomaly detection.

Advances in the Field of NLP

NLP powers the ability of machines to interpret text, speech and words effectively. It advances data analytics, spots malware, and prevents fake news. With the evolution of AI-powered chatbots like Alexa, Siri, Cortana, and Google Assistant, among others, the use of natural language processing has increased tremendously. Recent improvements in applications of this technology have substantially changed how AI comprehends and learns things surrounding it.

One of the most significant developments in the field of NLP was the use of transfer learning. Fast.ai's ULMFiT (Universal Language Model Fine- Tuning) introduced the concept of transfer learning to the NLP community. According to the company, ULMFiT is an effective transfer learning method that can be applied to any task in NLP. In 2018, Google AI introduced a new model for NLP called BERT (Bidirectional Encoder Representations from Transformers). The model uses the concept of transformers and transfer learning, and performs full bidirectional training of transformers.

Further, in 2018, researchers from the University of Massachusetts Amherst College of Information and Computer Sciences and Google AI Language introduced Linguistically-Informed Self-Attention (LISA), a neural network model. It coalesces deep learning and linguistic formalism, and thus more effectively utilizes syntactic parses to obtain semantic meaning.

More broadly, the power of NLP in the coming years will continue to evolve, understanding and contextualizing data that could lead to better gains to a business.