Performance for Other Natural Languages

Next, we examined the performance of the two systems on the Spanish version of the corpus. Figure 10 shows the results. The differences between using WOLFIE and Siskind's learned lexicons for CHILL are again statistically significant at all training set sizes. We also again show the performance with hand-built lexicons, both with and without phrases present only in the testing set. The performance compared to the hand-built lexicon with test-set phrases removed is still competitive, with the difference being significant only at 225 examples.

  

Figure 10: Accuracy on Spanish

Figure 11 shows the accuracy of learned parsers with WOLFIE's learned lexicons for all four languages. The performance differences among the four languages are quite small, demonstrating that our methods are not language dependent.

  

Figure 11: Accuracy on All Four Languages