Is there a way to perform a proximity search that is bounded, not by a fixed number of tokens, but by 2 marker tokens of some kind? For example, to implement proximity queries that are bounded inside as single sentence or paragraph? Obviously the analyzer has to support it, but has this been done before, and how?
SpanPositionCheckQuery is an abstract class that defines a span query, which checks if
the matched span passes a check for position. Concrete implementations include SpanPositionRangeQuery that checks if the match is in a defined range of positions and SpanPayloadCheckQuery that checks if the payloads of the matched positions are the same as requested.
Maybe you could achieve what you want by assigning a payload to each token indicating to which sentence/paragraph it belongs (with a counter) and then checking if all the payloads of the matched span are equal? I think that extending above class to achieve it should be straightforward.
来源：https://stackoverflow.com/questions/7378658/lucene-proximity-search-with-boundaries