| J10-4007 | We provide results for the best | WORDSPACE | models from Experiment 1 for |
| J10-4007 | error rate of 25.6 -- 25.7 % . In | WORDSPACE | , the Lin measure shows the best |
| J10-4007 | bins , although this is not so in | WORDSPACE | . It seems that in the sparser |
| J10-4007 | lemmatized ) words .3 We refer to it as | WORDSPACE | . The dependencybased space , |
| J10-4007 | contrast to Experiment 1 . The best | WORDSPACE | model ( Lin without PCA ) reaches |
| J10-4007 | frequency bins , the best results in | WORDSPACE | are better than those in DEPSPACE |
| J10-4007 | schemes . The bottom row shows | WORDSPACE | without dimensionality reduction |
| J10-4007 | two vector spaces , DEPSPACE and | WORDSPACE | , shows no clear winner . When |
| J10-4007 | and RESNIK . The DEPSPACE and | WORDSPACE | variants of EPP perform similarly |
| J10-4007 | are sparser but cleaner , and | WORDSPACE | shows lower error rates . 6 . |
| J10-4007 | corpus is used to compute either a | WORDSPACE | or a DEPSPACE vector space , |
| J10-4007 | the best EPP model when using | WORDSPACE | in the SYN PRIMARY setting . |
| J10-4007 | both PCA settings . In unreduced | WORDSPACE | , the divide is not as clearly |
| J10-4007 | easier to distinguish verbs in a | WORDSPACE | than to distinguish nouns . A |
| J10-4007 | cases ) is not significant ; in | WORDSPACE | , the difference between the |
| J10-4007 | for all similarity measures , in | WORDSPACE | as well as DEPSPACE . The improvement |
| J10-4007 | for the McRae data set , the EPP | WORDSPACE | models show much worse performance |
| J10-4007 | in the less sparse but noisier | WORDSPACE | , the added noise is stronger |
| J10-4007 | model for inverse preferences ( | WORDSPACE | , Lin with DISCR weighting ) |
| J10-4007 | Jaccard , Dice , and Hindle . For | WORDSPACE | with PCA-transformation , not |
