Abstract

this report. Recall that one of the two sub-runs that made up UMassX2 and UMassX2n, and three of the six sub-runs that made up UMassX6 and UMassX6n, used the standard parallel corpus dictionary and stemmer. The Standard Resources column of Table 3 shows the results when the sub-runs based on the UMass resources and acronym expansion were excluded. Only the sub-runs based on the standard resources were included. Thus, in the Standard Resources column, the UMassX2 and UMassX2n rows show the results of a single sub-ran, and UMassX6 and UMassX6n rows each show results based on a combination of three, rather than six, sub-ms. Relative to these three way combinations, the additional resources increased average precision 3 percentage points for title+description queries, and 4 points for title+description+narrative queries

Citations