The verbal valency in the Prague Dependency Treebank from the annotator's point of view

The core ingredient of the Prague Dependency Treebank (PDT; see Hajič, this volume) „valency“ indicates the capability of lexical units to combine other complementations. The PDT has adopted the concept of the valency theory of the Functional Generative Description (FGD) (see Sgall, 1967, Sgall et al, 1986). The valency theory of the FGD has first been developed for verbs, then also for other parts of speech. We present a description of how we dealt with valency of verbs during the annotation of the PDT and the way how the verbal part of the valency lexicon (PDT-VALLEX) was built. We focus on some specific problems related to verbal valency (as well as some other verbal complementations) from the point of view of the PDT. 1. The concept of valency in PDT One of the prerequisites of the correct syntactic annotation at the tectogrammatical level (TR) of the Prague Dependency Treebank (see Hajič, this volume) is the knowledge of valency frames. The valency theory (see Panevová, 1974-75, 1980, 1994, 1999) as used in the process of annotation of the Prague Dependency Treebank (PDT) corresponds to the concepts of the Functional Generative Description (FGP) (see Sgall, 1967, Sgall et al, 1986). Within this approach, syntactic as well as semantic criteria are used to identify verbal complementations. The verb is considered to be the core of the sentence (or clause, as the case may be). Its complementations (dependents) are classified either as inner participants or as free modifications. Both types of verbal complementations can be either obligatory (semantically always present with a given verb) or optional (not necessarily present). Only inner participants (obligatory or optional) and obligatory free modifications belong to the verbal valency frame. Optional free modifications are not listed in the valency frame. The relation between the dependent and its governor at the TR is labeled by a functor. The functor must be determined and recorded for all complementations in the actual process of data annotation. Annotators choose this value from a set of functors listed in the manual for tectogrammatical 1 Neither are so called quasi-valency and typical complementations stored in the valency frames of the PDTVALLEX lexicon (these types of complementations are described by Lopatková et al.(2003), Panevová (2003). 2 We discuss here the verbal valency frame in a narrow, strict sense, i.e. the verbal valency frame captured in the lexicon. The verbal valency frame in a broader sense consists of all. complementations, which can expand the given verb. The types of all the complementations are captured in the structure of the annotated tree as some of the values of the dependent nodes. 3 Valency is considered also for many nouns and adjectives, see Řezníčková,V. (2003), Hajič et. (2003). 4 If the annotators hesitate among the correct value of the functor, they have the choice to mark this uncertainty through multiple selection of several functors.