Natural Language Understanding (NLU)

Companies Manipulating A.I. Systems for Competitive Advantage
March 10, 2020
Multitask Learning
March 10, 2020

Natural Language Understanding (NLU)

NLU allows researchers to quantify and learn from all of that text by extracting concepts, mapping relationships and analyzing emotion.

Last year, Amazon released updates to its Transcribe Medical system, a natural language understanding (NLU) system that recognizes speech used by doctors and medical professionals in real time—no small achievement.

Getting machines to understand exactly what or who someone is referring to has been a challenge for conversational A.I. systems like Siri and Alexa. At best, the systems are trained to reference the last pronoun spoken. If a consumer asks “What time is The Lion King playing at Cinemark Theater?” and follows with “parking near there,” the system infers that “there” means “Cinemark Theater.”

In technical terms, this process is called “slot carryover,” and it uses syntactic context to understand what our pronouns mean. The process works, unless we speak in complex sentences with many different pronouns.

The fact is that in real conversation, most of us are messy talkers. We start and stop sentences without warning, we misuse words, and sometimes we rely on our tone to convey something we don’t want to say in actual words. We tend to speak in unstructured text. One of the things that makes reference resolution especially complicated for a large A.I. system like Alexa is that different Alexa services use different names—or slots—for the same data.

A movie-finding service, for instance, might tag location data with the slot name Theater_Location, while a restaurant-finding service might use the slot name Landmark_Address. Over the course of a conversation, Alexa has to determine which slots used by one service should inherit data from which slots used by another.

NLU allows researchers to quantify and learn from all of that text by extracting concepts, mapping relationships and analyzing emotion, and they made some impressive advancements in the past year.

The General Language Understanding Evaluation competition (or GLUE) is seen as a high-water mark in how well an A.I. system understands human language. China’s Baidu unseated Google and Microsoft in the most recent competition and became the first to develop techniques for understanding not only English, but Chinese as well.