simongray / StatementAnnotator

Custom annotator for Stanford CoreNLP that annotates sentences with the underlying statements contained within them.
4 stars 0 forks source link

Future feature: Composing statements into sentences #27

Open simongray opened 8 years ago

simongray commented 8 years ago

Statements can be composed into meaningful sentences of text using composing functions. For example, Statements with the same topic can compose these statements into perhaps a single sentence (or pagraph) such as

"[person] thinks [subject] is [x], [y], and [z]".

They can be sorted so that copula Statements (S+DO) are listed first , then S+V statements, and then the more meatier S+V+DO+IO types of sentences.

Perhaps a more naive but just as effective approach can be obtained by sorting by statement length instead (and even word length in a few cases) so that shorter statements appear before longer statements.

Either approach might produce something like:

[Simon] thinks that Netflix is [expensive], [so incredibly boring] and [really doesn't have enough content on it].

In this composed sentence the statements "Netflix is expensive", "Netflix is so incredibly boring" and `"Netflix really doesn't have enough content on it" are composed into a single sentence in a way that mimics natural language by actually being composed of pieces of natural language.

Special care should be taken around sentences where the subject is self-referential, i.e. "I" or "my". This information should be carefully sorted through to build a personal profile of skills, occupation, location, likes, dislikes, etc.

Bringing all of the composed sentences and paragraphs together, it is possible to build an entire data-mined personal profile that reads like something very close to natural language.