PhD thesis errata ================= Original version: http://dare.uva.nl/document/2/177511 Revised version: http://andreasvc.github.io/phdthesis_v1.1.pdf Major additions: ---------------- - Sec 2.2: add paragraph explaining role of binarization. - Sec 2.6: fix argument about complexity. Square of total number of nodes in treebank is not relevant. Quadratic to linear speedup adds up due to number of tree pairs (square of number of trees). - Table 3.8: add table with F1 scores restricted to discontinuous constituents. - p. 137 comparing variance of literary and quality ratings may be problematic due to scale difference; add footnote mentioning this, and explanation that predictability of ratings from features is not the same as variance of ratings by themselves. - Include top 10 best/worst novels according to survey in appendix. Corrections: ------------ - Turn self-citations of incorporated papers into chapter references. - p. 8 obviously absurd -> highly improbable. - Fig 1.2 reverse arc between . - Sec. 2.6: Stats on discontinuous fragments redundant with sec. 2.4; removed. - Table 3.3: Negra, "this work" should be vanCraBod2013, to make it clear that the grammar is without refinements and function tags. - Sec 3.6.4: Reported proportion of gold items for PCFG is incorrect; add note. - p. 99: two different p-values are reported for the correlation of literary and quality ratings; the one in the figure is correct. - Sec 7.2.5 is redundant, removed; except for Salganik citation, moved to ch. 5. NB this changes the section numbering. - Many minor spelling fixes. Duesseldorf, November 18, 2016