Parsing

All the parsing functionality is in the src/parser.rb file. The process starts by calling the method parse_text. First the production rule and dependency rule trees of the text are generated by the Stanford Parser, method parse_ptree_dtree. Second the text and the trees are used in the method parse to generate the features used by the models in the data/ folder at every step of the pipeline (Figure 1). The generated features at every step are stored in files with prefix pdtb_ in the /tmp/ directory (note the absolute path).

Figure 1. System pipeline for the PDTB discourse parser.