Incomplete sentence parsing and relation extraction

Question

I'm building a tool which parses natural language input into nouns and relations. For example sentence

Barack is the president of the United States and lives in Washington DC and went to Harvard University

Should be parsed as

Barack is the president of the United States Barack lives in Washington DC Barack went to Harvard University

Another example

popular NLP (Stanford, NLTK, OpenNLP) tools only works with full sentences and gives funny and erroneous results. What api could process these kind of sentences? Is it hard to build one completely from scratch?

Answer 1

What I think you are trying to split sentence with coordination structure. Thus, my solution is that:

• parse the sentence into phrase structure tree or dependency tree
• define some patterns like (a dependency example) root-*->cc->a2 | a1 and split into root->*->a1 and root->*->a2.

Although the parser and never output one hundred percentage correctly, it should not be too for the sub-structures like coordination structure.

As I know, there seems no such task that splitting a sentence into several sub-sentences and at the same time maintaining the original meaning. But I've seen someone working on this from scratch and treat it as an SRL-like tasks triggered by coordination and certain conjunction.