The gAGDT is a graph database containing the Greek texts of Perseus'
Ancient Greek and Latin Dependendency Treebank.
We also include some (tentative) code to use the OGM interface of the python library py2neo, as well as
a few Jupyter notebook to document the types of Cypher query and operation I perform.
Not yet. For now, I have only converted the following authors or texts:
- Iliad and Odyssey
- Aeschylus
- Sophocles
Not quite! What I love of graph databases is their flexibility in terms of schema and extensions. In other word, I can easily add different layers of annotations or different types of relations between treebank words to integrate the morpho-syntactic annotation of the original treebank.
This is precisely what I would like to do. I have tentatively introduced some properties of nodes to integrate semantic or prosodic/literary information,
such as meter or animacy. But these properties at the moment are empty.
A couple of extensions that are actually implemented are the following:
- Chronology (property of sentences)
- Genre (property of sentences)
- Speaker (property of sentences)
speaker, in particular, is a very useful feature! Especially for dramatic texts, it allows you to search sentences by the (literary) character that utters them.
Some of the work needed to enrich the nodes of the treebanks with this information layer and a discussion of the research questions that can be enabled by such an extension
will be object of a forthcoming paper. I'll update the reference as soon as the paper is published.
Note, however, a couple of limitations:
- speaker information are implemented for Sophocles (all) and for some Aeschylus
- as I have taken the speakers from the TEI editions on the Perseus Digital Library, the characters's names are in Greek!
Donwload the file with the last db release from here.
You will need to have Neo4j installed on your machine (either your local computer or a server) to load this data and start using the gAGDT. Refer to the instructions on Neo4j's website to lear how to install it on your OS. Regular distributions of the software (i.e not the enterprise version) is enough.
On Debian and Debian-based distributions (e.g Ubuntu) it is as easy as:
sudo apt-get install neo4jMac OS and Windows have desktop applications that you can use also to start up the DB once it is installed.
Note, however, that Java 8 is required to run Neo4j. If you're in Linux, instructions on how to install it in Linux (and how to configure your default Java version so that Java 8 is used) are also found at the above link.
Once that Neo4j is installed, start it. If all goes well, you can use the neo4j-browser to inspect the database. If you install the DB on a local computer the db is available at this address: http://localhost:7474/browser/
To load the gAGDT data in your database, you'll need to load them using the neo4j-admin load
command (see the instruction).
Better documentation will come. For now, I write some basic information on the nodes, properties and relations.
The db has 3 types of nodes:
Sentence: representing the root of the treebank and holding the properties that belong to the whole utterance or even the whole text (e.g. chronology, author, work, genre, span of lines or text references coverd by them).Token: the actual token corresponding to "words" attested in the digital edition of reference (or very informally: the words of the texts)Artificial: reconstructed nodes that govern elliptical constructions.
Note the properties of tokens (and artificial nodes) are very redundant. For example, I store the full 9-position morphological tag of words (e.g. v1sria---) and each of the 9 properties that are condensed in this string: in this case, pos = verb, person = 1s etc. This is a bit verbose and maybe harder to maintain (if you change a property
you should also update the corresponding "legacy" property). But it is very handy for format exchange and for querying: after all, it's much easier to write a query where you ask:
MATCH (n:Token)
WHERE n.pos = 'verb'
Then to always write regular expression to match the nth position of a 9-character string...
At this moment, only the dependency relations of the AGDT are represented.
I have chosen to define each of the labels as a type of relation. This means that PRED is a relation type of the gAGDT.
New in version 0.2
The syntactically linked nodes are joined with the Dependency relation. The relation goes from head to dependent, so that:
(h)-[:Dependency]->(d)is read as "node h governs node d", or h is the head and d the dependent.
The label of the dependency is encoded in the type property of the relation, so that a syntactic construction where n is the subject of v can be queried as:
(v)-[:Dependency{type:"SBJ"}->(n)The relationships have also an additional property called original_label that registers the original label found in the AGDT XML attribute, which is useful in the case where constructions are treated differently in the gAGDT from the AGLDT.
Two of such cases are Coordination and Apposition. Marks of coordination and apposition are NOT included in the labels / type or relations. So a coordinated subject will be linked to the verb via type:SBJ not type:SBJ_CO. The fact that a node belongs to a coordinated structure or a apposition is recorded at the node level in two boolean properties:
isMemberOfCoord(0 or 1)isMemberOfApos(0 or 1)
I had this idea from the PDT and it is a much better way of implementing coordination and apposition than to use the appendix to the labels!
The AGLDT is distributed under a CC BY-SA 3.0 license. And so is gAGDT. Please, refer to the AGDT page for a full list of the annotators that should be credited with the annotation of the those texts. They should thanked and praised for their effort...