During text query analysis, we call into RLP for each term. Since RLP is more expensive then our Basic FTS algorithm, we should amortize the cost of this analysis by only calling into RLP once per set of positive, and negative terms.
The best approach is pass a space (U+0020) delimited set of terms into RLP. This will work since space is a word boundary. See http://www.unicode.org/reports/tr29/#Word_Boundaries.