Searching in the Penn Discourse Treebank Using the PML-Tree Query

Searching in the Penn Discourse Treebank Using the PML-Tree Query Jiří Mírovský, Lucie Poláková, Jan Štěpánek Charles University in Prague Faculty of Mathematics and Physics Institute of Formal and Applied Linguistics (mirovsky|polakova|stepanek)@ufal.mff.cuni.cz Abstract The PML-Tree Query is a general, powerful and user-friendly system for querying richly linguistically annotated treebanks. The present paper shows how the PML-Tree Query can be used for searching for discourse relations in the Penn Discourse Treebank 2.0 mapped onto the syntactic annotation of the Penn Treebank. Keywords: PML-Tree Query, Penn Discourse Treebank, searching in treebanks 1. Introduction With increasing complexity of annotated treebanks, query tools and query languages also need to become more elaborate, in order to allow searching for and studying all phenomena annotated in the treebanks.1 At the same time, it is important that the tools are user-friendly and the query languages, despite their in- evitable complexity, remain easy to use. The present paper shows a combination of leading rep- resentatives in two areas of NLP – discourse annotated treebanks on one side and query tools on the other, namely the Penn Discourse Treebank 2.0 (PDTB-2.0) and the PML-Tree Query (PML-TQ). The PDTB-2.0, at the time of its publication in 2008, came with two possibilities to search in the data. First, there was the PDTB API describing in a hierarchy of Java classes the structure of the PDTB-2.0 data scheme and a skilled programmer could use it for searching in the data; for simple queries, there was a graphical interface as a part of the Java-based PDTB browser. Second, there was a conversion script for transforming the data from the native .pdtb format into a column format,2 which could be used even by linguists without extensive programming skills to ex- tract information about discourse relations using line- based tools such as grep, or spreadsheet programs. Yao et al. (2010) pointed out two shortcomings of the PDTB API, namely the impossibility to query the syn- tactic annotation of the Penn Treebank (PTB) and the discourse annotation of the PDTB-2.0 simultaneously, and the impossibility to incorporate and query addi- tional annotation. They offered a solution based on 1 See e.g. a study on requirements that the complex an- notation of the Prague Dependency Treebank (Hajič et al., 2006, Bejček et al., 2013) poses on search tools (Mírovský, 2009). 2 The script (convert.pl) is a part of the PDTB-2.0 distribution and contains an error – it fails to transform second senses for AltLexes; the error can be fixed by chang- ing the line 481 from “if (@senseArray > 2) {” to “if (@senseArray > 1) {”. their conversion of the PTB and the PDTB-2.0 data into an integrated XML format, and used XQuery – a general language for querying any XML data – for searching in the transformed data. In our approach, we also transform the data of the PTB and the PDTB-2.0 into an XML format (PML, see Section 4.1). Unlike the previous approaches, for searching in the data we offer a powerful, yet user-friendly and …

Content truncated to 500 words