Poly-GrETEL: Cross-Lingual Example-based Querying of Syntactic Constructions

Poly-GrETEL: Cross-Lingual Example-based Querying of Syntactic Constructions Liesbeth Augustinus, Vincent Vandeghinste, Tom Vanallemeersch Centre for Computational Linguistics KU Leuven (University of Leuven), Leuven, Belgium {firstname.lastname}@kuleuven.be Abstract We present Poly-GrETEL, an online tool which enables syntactic querying in parallel treebanks and which is based on the monolingual GrETEL environment. We provide online access to the Europarl parallel treebank for Dutch and English, allowing users to query the treebank using either an XPath expression or an example sentence in order to look for similar constructions. We provide automatic alignments between the nodes. By combining example-based query functionality with node alignments, we limit the need for users to be familiar with the query language and the structure of the trees in the source and target language, thus facilitating the use of parallel corpora for comparative linguistics and translation studies. Keywords: parallel treebanks, example-based querying, search tool, alignment 1. Introduction In this paper we describe Poly-GrETEL,1 the continued de- velopment of GrETEL (Greedy Extraction of Trees for Em- pirical Linguistics), that was started as an online query en- gine for a manually verified, syntactically annotated cor- pus of written Dutch (1 million words) (Augustinus et al., 2012), allowing to look for syntactic matches using exam- ples. These examples are automatically converted into an XPath expression,2 which is used to query the treebank. Example-based querying has the advantage that users do not need to be familiar with the query language, nor with the lay-out of the underlying XML structure used to rep- resent the trees, nor with the grammar used by the parser. Alternatively, users can directly query the treebank using XPath, without examples. GrETEL has further been extended with a 1-million word, manually verified treebank of spoken Dutch (Augustinus et al., 2013), and the interface has been improved. We also added a large (automatically annotated) treebank of 500 million words for querying by example (Vandeghinste and Augustinus, 2014). We now further expand the scope of GrETEL by adding a treebank for Dutch, i.e. the automatically parsed Europarl treebank, and its parallel counterpart for English, extracted from the Europarl parallel corpus (Koehn, 2005), version 7. We refer to this extended engine as Poly-GrETEL. By making parallel data easily queryable and freely avail- able online, we allow translators, language learners, and people working in comparative linguistics and translation studies to look up how data are translated in reality. Instead of restricting search to string matching, we allow to check how certain syntactic patterns are actually translated. When looking at lexical information in combination with syntac- tic patterns, the engine functions as a bilingual syntactic concordancer. 1http://gretel.ccl.kuleuven.be/ poly-gretel 2http://www.w3.org/TR/xpath 2. Related Work A parallel corpus consists of pairs of source sentences and their translation. It is a specific instance of a multilingual corpus, which may contain sentences in more than two lan- guages. Parallel corpora are useful for comparative linguis- tics, translation studies, language learning and creation of machine translation systems. Many corpora with two or more languages are currently available, one of the most popular being …

Content truncated to 500 words