The idea of stemming is that you put Apple and Apples under the same
heading in the index; that is, collating them together, or conflating them.
Currently, the lq-text stemmer handles only plurals and possessives; a better one would also understand that run and ran and running go together, for example.
The more stemming you do, the lower the precision of matches, but the higher the recall. Since lq-text was originally designed for very high precision, stemming has not been a high priority.
Note that when matches are written to the database, the fact that stemming was applied is also recorded, using two bits (one for plurals and one for possessives) so that a query for Apples doesn't by default match Apple in the database, but a query for Apple will match both.
The principle is that the package should not infer more precision than was used in the query, but where higher precision was used, should take advantage of it where it can.
Currently, the lq-text stemmer handles only plurals and possessives; a better one would also understand that run and ran and running go together, for example.
The more stemming you do, the lower the precision of matches, but the higher the recall. Since lq-text was originally designed for very high precision, stemming has not been a high priority.
Note that when matches are written to the database, the fact that stemming was applied is also recorded, using two bits (one for plurals and one for possessives) so that a query for Apples doesn't by default match Apple in the database, but a query for Apple will match both.
The principle is that the package should not infer more precision than was used in the query, but where higher precision was used, should take advantage of it where it can.