Back to list

Detail of contribution

Auteur: Giuseppe G. A. CELANO

The Ancient Greek Dependency Treebank: Grammatical and Technical Nodes

Abstract/Résumé: The Ancient Greek Dependency Treebank (AGDT) is an ongoing project aiming to provide morphosyntactically parsed texts. The annotation is based on the Prague Dependency Treebank 2.0 guidelines with some variants. The main tenet of the annotation is that within the sentence words are related to each other in such a way that each of them, excluding the main verb and the final punctuation mark, depends on another word. This can be represented by an upside-down tree graph where every word of a sentence corresponds to a node, and every node depends on just one superordinate node, i.e., the governor (but a node can have more than one subordinate node, i.e., more dependends). The paper focuses on the crucial distinction between grammatical and technical relations, which is usually a point of major confusion. Although all relations between words are represented in the same way (graphically, as edges), their meaning can be very different. For some relations are properly linguistic, while others are just there because the whole syntactic annotation is to be taken as a compromise between the requirements of the (complex) linguistic analysis of the sentence and those of computational simplicity and coherence. An example of proper linguistic relation (i.e., grammatical relation) is that between the subject and the predicate: the former is related to the latter because, roughly speaking, it fulfills the meaning of the predicate. On the contrary, the annotation of some constructions such as coordination is to be regarded as technical: in AGDT the coordinate conjunction governs the conjuncts and depends on the word the conjuncts are grammatically associated with. Such a representation, which is in linguistic terms very disputable, mainly aims to fully integrate the construction into the annotation model. The role of technical dependencies is even more apparent with punctuation marks: they are annotated like words, but their reality as syntactic dependends (and sometimes as governors) is stricto sensu null. On a more general level, the annotation of other constructions, such as relative clauses, might be considered as technical too: the suspension of the verb of a relative clause under a noun, which in fact governs a whole relative clause, can be taken as another expedient to represent, for simplicity’s sake, a more complex structure than that captured by a one-to-one dependency representation. These examples show the necessity of conceptually distinguishing in AGDT between grammatical and technical nodes, the latter being a way to represent, within the same framework, structures which, if properly parsed, would add more jargon to the representation.