Towards a Polish LTAG Grammar

This paper reports on a Lexicalised Tree Adjoining Grammar for Polish, extracted automatically from the Polish constituency treebank. The grammar consists of 23 570 elementary trees anchored by 11 515 lexemes. Running the grammar on the sentences from the treebank using a modified version of TuLiPA parser showed that it achieves a high accordance (almost 99%) with the treebank annotation – in terms of syntactic categories assigned to phrases – on the trees which were successfully parsed. For many trees, however, obtaining a TAG parse was impossible due to time or memory shortcomings of the used tool.