As reported multiple times on the user list:
The logic here is not forming the right query structures and ignoring positionIncrementAttribute from the tokenStream.
- when default operator is AND, you can see it more clearly, as synonyms are wrongly inserted as additional MUST terms:
expected:<+field:中 +(field:国 field:國)>
but was:<+field:中 +field:国 +field:國>
- even when default operator is OR, its still wrong, because we ignore posInc and this means coord computation is not correct (so scoring is wrong)
This also screws up scoring and queries for decompounding too (because they go thru this exact situation if they add the original compound as a synonym).