Natural Language Understanding, or NLU, is the science of making a computer understand the natural language of humans, i.e., a computer with NLU must be able to understand what a person is trying to say in plain English, or whatever language was used to convey the message. 

Genuine NLU has not yet been achieved, despite the fact that man has been trying to come up with a better way to interact with computers since the earliest computers were invented.  True automation of natural language understanding requires breakthroughs in many facets of the field, which is not surprising.  After all, even humans experience serious miscommunications with one another from time to time.

The major hurdle to genuine NLU is the need for the computer to have an extensive knowledge of the real world just to be able to determine the context in which a sentence has been uttered. Enabling a computer to deduce the domain of a discourse once it engages a general discussion with a human is hard enough, but carrying out the discussion coherently to a reasonable extent offers even more challenges.

The difficulty of giving a computer enough knowledge about the world to carry out a discussion is made even more complicated by the fact that many words have multiple unrelated meanings.  For instance, the English word 'plant' may mean an organism with roots and leaves, but it may also refer to a 'factory' or a verb that means 'to introduce a spy or include a false evidence'. A computer that encounters the word 'plant' must know which meaning is appropriate for the context in which it is used.  This is not to mention complex phrases with similar ambiguities like "park wildlife visitors' log" or "basketball game pass", wherein the words 'log' and 'pass' can assume correct meanings in two entirely different contexts.

One approach to come up with a functional NLU application is to limit its scope or domain of discourse to a specific and specialized topic.  Expert systems with natural language interfaces and web search engines that attempt to match natural language queries with their indexed content are examples of these applications.

Of course, NLU can never be true NLU unless it can be applied to words actually spoken by humans. Needless to say, NLU that incorporates genuine speech recognition wherein the computer actually listens to the speaker and understands what is being said poses a new set of challenges.  For instance, different people sound differently even when they say the same words. And don't forget that many, many words sound alike even if they have entirely different meanings.

Indeed, it seems that the age of robots talking to us like real humans is still far ahead into the future. Come to think of it, natural language is only natural to natural people, but not to machines like that PC next to you.

See Also:   More Industry Articles