Combining morphological and ngram evidence for monolingual document retrieval

We report on experiments in which we merged the results of linguistically informed and linguistically ignorant approaches to retrieval for European languages. We found that even high-quality base runs can be improved by means of fairly simple techniques for merging them with other runs, although the improvements no longer seem to be as dramatic as those reported on previous experiments on smaller collections than we used and with retrieval engines that are not as highly optimized as the one used in our experiments.