Performance measures of text-only models.

We show performance for our eight language models and the baseline. The first three result columns show overall corpus-wide performance, and the last three result columns show mean performance across articles. The first seven result rows show performance of incomplete LMs, and the eighth result row shows the performance of our CompleteLM, (Fixed Size, Mixture). Our CompleteLM performs best on all measures. Asterisks denote -values from paired -tests comparing each method with Baseline, where * indicates and **indicates .