Granularity

The descriptions of the linguistic information in the TreeBank will be very detailed in order to demonstrate the information flow in the syntactic structure of the sentences in the TreeBank.

Theory Dependency

An annotation scheme usually has to be theory independent in order to allow different interpretations of the tagged texts in different linguistic frameworks. We think, however, that on a certain level of granularity (as it was mentioned linguistic descriptions in the BulTreeBank will be very detailed in order to demonstrate the information flow in the syntactic structure) we will have to exploit some linguistic descriptions that are theory dependent. We choose HPSG for the following reasons:

• HPSG is one of the major linguistic theories based on rigorous formal grounds;

• HPSG allows for a consistent description of linguistic facts on every linguistic level: phonetic and phonological, morphological, syntactic, even the level of discourse. Thus, it will ensure the easy incorporation of linguistic information which does not belong to the level of syntax if such is needed for the correct analysis of a given phenomenon;

• HPSG allows for both integration and modularization of descriptions and will therefore enable different experts to work on different parts or levels of analysis.

• The formal basis of HPSG allows easy translation to other formalisms.

• There are universal HPSG principles that can be used to support the work of the annotators during the development of the TreeBank.

The Logical Formalism

We not only choose HPSG to be the linguistic theory within which we will explicate the syntactic structure of Bulgarian texts, but make a step further and choose the actual formalism that we will use in the annotation process: namely, SRL augmented (in an RSRL style) with some relations common in HPSG. For SRL see (King 1989), (King 1994) and (King 1999), for RSRL see (Richter 1997), (Richter 1999), (Richter, Sailer and Penn 1998), (Richter, Sailer and Penn 1999). For the annotation we will use SRL descriptions called feature graphs, which correspond to clauses in a normal form of a SRL theory. In their complete form feature graphs are morphs in the sense of (King 1994). Such detailed descriptions will be extremely useful in the future exploitation of the TreeBank, but they might be difficult to use in the annotation process. Here we hope to use the (special) inference mechanisms of SRL and some of the HPSG principles (universal and specific to Bulgarian) in order to enable the annotator to provide only part of the needed information with the rest of it being inferred automatically (see (King and Simov 1998) for a special inference for such purposes).

Partial Analyses and Predictions

As was mentioned in the previous paragraph, the detailed level of the envisaged syntactic descriptions will require entering a huge amount of linguistic information for each sentence. In order to minimize the necessary human intervention, we will exploit all possibilities to provide an automatic partial analysis of the input string before the actual annotation starts. We would also use the partial information entered by the annotator in order to predict or constrain the possible analyses in other parts of the whole description of the element. In this way we will exploit all the constraints available from pre-encoded grammars.

The development of an HPSG-based bank of syntactic trees will comprise the following  tasks:

• It will be necessary to identify which part of the HPSG sort hierarchy as described in (Pollard and Sag 1994) and others can be successfully applied to Bulgarian and also what modifications and additions will be necessary for a formal grammar of Bulgarian. The Bulgarian-specific part of the sort hierarchy will be subject to change during the development of the TreeBank according to the demands of the Bulgarian data.

• We would need to provide a representation of the HPSG Universal Grammar principles from (Pollard and Sag 1994). This very general grammar will be used as a top constraint during the annotation of the trees in the TreeBank. For example, each headed phrase in the TreeBank has to satisfy the Head Feature Principle of HPSG. This grammar will be further extended by Bulgarian specific principles during the development of the TreeBank. These specific principles can deal with word order in Bulgarian, for instance.

• It would be necessary to adapt the lexical information already contained in the morphological dictionary of Bulgarian (Popov, Simov, Vidinska 1997) to the sort hierarchy and to the HPSG principles. The morphological analyzer available to the project will be used to annotate the words in the sentences. Additionally, the Part-of-Speech disambiguator developed under the CLaRK Programme will be used to reduce ambiguities after the morphological analysis of the sentences. A description of the relevant linguistic resources is given below.

• We will develop a reliable grammar for a partial parsing of the phrases constituting the lower part of each syntactic annotation - a few dominant levels over the lexical categories. Here we will follow ideas in the work of (Abney 1990), (Abney 1991). A preliminary version of such a grammar has already been developed under the CLaRK project with the goal of context analysis in a concordance program. This grammar will be used to analyze the input sentences after the morphological analyses in order to add unambiguously the syntactic structure to some minimal elements of the whole sentence.

• A core set of sentences demonstrating typical syntactic phenomena in Bulgarian is under development.

• It will be necessary to prepare guidelines for the annotation of the actual sentences. It is important to make two points here:

1. the annotators will specify mainly information about the syntactic configurations over chunks and phrases over chunks (such as PP attachment), and
2. some frequently used descriptions will be abbreviated in an appropriate manner in order to facilitate the process of annotation. These guidelines will allow annotators with different levels of experience to work on the TreeBank.

• We will assemble a base set of annotated sentences taken from real Bulgarian texts. This set of sentences will constitute the main body of the TreeBank. In contrast to the core set developed during the first phase of the project, here the sentences will be chosen in such a way as to demonstrate the most widely spread syntactic phenomena in Bulgarian texts. Special attention will be paid to resolving ambiguity in the context of use. This will be done by including in the TreeBank not only isolated sentences, but whole paragraphs or even a number of consequent paragraphs. When sentences are annotated in context, a description of the disambiguating context will be added to the syntactic annotation. For the moment we envisage an informal description of the context, but we hope to be able to formalize that, time and resources permitting. The actual annotation process will proceed in accordance with the following steps:

• We plan to develop further the CLaRK system in order to support the annotation process. During the development of the TreeBank we envisage to extend the set of possible constraints over XML documents that can be encoded in the system. We will also connect the CLaRK system to a feature logic engine in which we can implement the grammars. We will implement new inference mechanisms to support the annotators even further using the full predictive power of the grammar. Very often in the annotation process annotators develop their own ad hoc procedures and constraints for some specific cases. We plan to develop a simple macro language in the CLaRK system that will allow their easy encoding. We will also apply some techniques for generalization and learning from the already developed syntactic trees.

References

[Abney St 1990] Syntactic Affixation and Performance Structures. In Bouchard D. Leffel K. (eds), Views on Phrase Structure. Kluwer Academic Publishers, Dordrecht.