Generate triple from natural language question

TEMIS's Luxid product can do this, and so can Thomson Reuters' OpenCalais.

I want to build question answering system over a Semantic Web Knowledge Base. When pre-processing, I want to change the question sentence to triple base.

What you want to achieve is extremely difficult (impossible?) in the general case.

It is an open research question in the area of Semantic Web/Linked Data Question Answering. In fact, in the general case, I would say this is one of the hardest open questions in computer science.

Can you explain the method to generate triples from natural language?

Not really no. :)

... there is no one method. However, if you want to dig into this topic a bit more, you should check out a research survey ...

• Is question answering fit for the Semantic Web? A survey

.. and see how the state of the art research prototypes are getting on in this area:

• 1st Workshop on Question Answering over Linked Data

If this topic is really important to you, all I can say is prepare to get your hands dirty.

Alchemy API can tag relevant concepts too

http://gate.ac.uk/demos/ has some pointers to good demos eg. questio and freya

Neither Alchemy or OpenCalais will help with his problem, which is that he wants to take

"Who killed Abraham Lincoln?"

and produce something with a blank node like

:_ :killed :Abraham_Lincoln .

which could then be satisfied by a bunch of facts in a triple store.