I'm looking for a solution that is not knowledge-base dependent or domain-specific, i.e. for general English text.
For the sentence, "I went to the store today", one can identify constituents
<I went to the store>, <today>
and permute to "Today I went to the store." One could also identify constituents
<I went>, <to the store>, <today>
and permute to "to the store I went today", or even "to the store today I went". The latter 2 permutations are less idiomatic but still preserve meaning.
My question -- what strategy could I use to identify a set of constituent phrases (e.g. from a parser) to which one could apply arbitrary re-ordering, whilst still preserving basic meaning?
The above examples are relatively simple; my use case is for more general sentences, as found in e.g. transcribed spoken English and email.
I allow that results may be less than grammatical, e.g. "My friend's private green was ideal for yesterday's game" might permute to "for yesterday's game was ideal my friend's private green", which is not grammatically sound, but the meaning is preserved.
If a strategy can reduce the likelihood of such ungrammatical permutations, that would be a plus, but not a requirement.
