Why Language Is Hard for Machines
You understand a sentence without noticing you are doing anything. That effortlessness is exactly what makes human language so difficult to hand to a computer.
1Natural languages
Three features of every natural language:
- They are governed by set rules that include syntax, lexicon and semantics.
- They are all redundant— the same information can be conveyed in more than one way.
- They all change over time.
2The same word, different meanings
Read these three sentences and notice how hard the word red is working:
The red car zoomed past his nose. Here it is simply the colour of the car.
His face turns red after consuming the medicine. An allergic reaction? Embarrassment? A medicine that dilates the arteries?
A word is not a key
3Computer languages
| Natural language | Computer language | |
|---|---|---|
| What it is | A human language — English, Hindi, Japanese | A language used to interact with a computer — Python, C++, Java, HTML |
| Meaning | Depends on context; a word can mean several things | Exact and fixed; one instruction means one thing |
| Changes | Changes over time on its own | Changes only when the language is redesigned |
| Redundancy | Redundant — many ways to say the same thing | Usually one correct way to write an instruction |
Computers require a specific set of instructions to understand human input, called programs. To talk to a computer we have to convert natural language into a language it understands.
4What NLP is
Computers can only process electronic signals in the form of binary language. NLP is what makes the conversion from the natural form to the digital form possible — so the whole purpose of NLP is to make communication between computer systems and humans possible.
Pick a domain and it rises. What is standing on the pedestal is the data that domain is fed. Drag sideways to turn.
Deals with the interaction between computers and humans using natural language: the machine reads or hears human words and makes sense of them.
Examples: Email spam filters · Google Translate · Voice assistants, auto captions
5From the board papers
▶Show the answer
True. This is the handbook’s own sentence, and the three “red” examples above are exactly what it is describing.
(a) Numeric data (b) Textual data (c) Image data (d) Visual dataClass X handbook, Quiz Time
▶Show the answer
(b) Textual data. Image and visual data belong to Computer Vision; numeric data to Statistical Data.
▶Show the answer
A human (natural) language is what people speak — English, Hindi. Its meaning depends on context, the same thing can be said in many different ways, and it changes over time.
A computer language such as Python or Java is used to give instructions to a machine. It is exact and unambiguous: an instruction has one meaning and must be written in the correct form.
That is why NLP is needed — to convert one into the other.
6Check yourself
Which of these is a feature of every natural language?
What is NLP?
Why is context important in natural language?