IMPORTANT!
With v.0.5, Daphne now has switched to the UD annotation schema. If you want the old AGLDT-compliant XML files you can either:
- check out the version tagged
v0.4.1-agldt
- retrieve them from the file
other/old_agldt_source.zip
A collection of treebanks of poetic Ancient Greek texts, starting with drama and archaic epos. For Aeschylus, Sophocles, Homer and Hesiod, the source of the annotation is Perseus' Ancient Greek and Latin Dependency Treebank. Euripides' Medea comes from Pedalion.
The treebanks were converted from the original AGLDT-like formalism to Universal Dependencies (UD) with the help of my rule-based converted tb2ud. The UD conversion is constantly revised and reviewed by hand.
At first, I started Daphne because I wanted to bring the annotation in line with a more specific set of guidelines, for both morphology and syntax. The original guidelines for Daphne were based on version 1.7 of the AGDT guidelines for Greek, with some qualification and special trait.
In a second phase of the work, I decided to move all the annotation to the Universal Dependencies formalism. One of the reason why I decided to move to UD is because I'd like to add other layers of annotation, like speaker tags for direct speech, animacy for words, links to dictionary forms, etc. UD is a very active community where support is constantly introduced for new tools and new expansions.
The AGLDT is simply awesome (and I know: I did a bit of it)!
But what we need is: a) a tighter set of annotation guidelines to make the annotations more consistent; b) texts with those corrections implemented here and there where they need it; c) more annotation (both as in "more annotated texts" and "more types of annotation on what we already have treebanked") d) better NLP and reliable automatic annotation e) more integration with other projects, both for Greek and other languages.
AGDT 2.x goes in the direction of more defined guidelines. But I'd still like to have control on the type of annotation I perform. Secondly, UD provides the perfect answer to all the latter requirements.
Well, that's what I like to work with...
As I tried to argue elsewhere, you can't do treebank annotation without doing a full, thorough textual analysis of what you're annotating.
So, you might as well keep track on what you do, especially the alternative interpretations that are possible and the nuance of meanings that different treebank configuration might introduce. I did just that when I annotated some of the tragedies of Sophocles.
Caution: at the moment, those files are just notes. Some of them even mix Italian (my own native tongue) and English. I'll polish them in due time.