Local PDF.pdf

30 Dependency Treebanks Stanfordized, research supported by grants FP7-ICT-2013-10-610516, SVV 260 104, Czech Science Foundation grant no. P406/14/06548P, and GAUK 1572314, utilizing language resources from LINDAT/CLARIN project, developed by Rudolf Rosa, Jan Mašek, David Mareček, Martin Popel, Daniel Zeman, Zdeněk Žabokrtský.
Existing treebanks have treebank-specific annotation of labelled dependencies, POS tags, and morphological features, which undergo harmonization through treebank-specific conversion to Prague Dependencies and Interset morphological features, and Stanfordization through conversion to Universal Stanford Dependencies and Universal POS Tagset.
HamleDT 2.0 is a Harmonized Multi-language Dependency Treebank, including Ancient Greek, Arabic, Czech, Danish, Dutch, Estonian, Finnish, Bulgarian, Catalan, German, Hungarian, Japanese, Slovene, Basque, Bengali, English, Greek, Hindi, Latin, Persian, Portuguese, Romanian, Swedish, Tamil, Italian, Russian, Slovak, Telugu, Spanish, Turkish.
Prague Dependencies are adapted from Prague Dependency Treebank, used in 9 other treebanks, with 23 dependency relation labels, and tagsets converted to Interset.
Universal Stanford Dependencies are a new language-universal version of basic Stanford Dependencies, based on lexicalist approach, used in Universal Dependency Treebanks, suggesting a two-layer taxonomy with universal grammatical relation labels and fine-grained language-specific labels.
HamleDT uses 30 universal labels and includes 30 existing treebanks all converted to Prague Dependencies and Stanford Dependencies, with free download of 13 treebanks and free conversion pipeline for all 30 treebanks.