/* * Main.java * * Created on 21 februari 2006, 21:10 * */ package DOPParser; /** * * @author gideon */ import java.io.*; public class Main { public Grammar myGrammar; //special symbols public final static boolean PRINT_OUTPUT = false; //to screen //WRITE_CNF_GRAMMAR_TO_FILE exports the grammar (read off from Treebank) in format that can be imported (GrammarToTextWriter, in printStuff) //so that next time Goodman doesn't have to be trained (choose GRAMMAR_PRINTING_FORMAT=1) public static boolean EXPORT_CNF_GRAMMAR_TO_FILE = false; public final static boolean EXPORT_CFG_GRAMMAR_TO_FILE = false; //formats: 1=print terminals, nonT, and rules+probabilities in one file //2=only rules 3=rules plus probabilities, 4=//rules plus counts //for exporting target grammar to BMM-algorithm you need format 4 public static int GRAMMAR_PRINTING_FORMAT = 1; public final static boolean GRAMMAR_PRINTING_PRETERMINALS = false; public final static boolean CALCULATE_LIKELIHOOD = false; public final static boolean CALCULATE_PERPLEXITY = false; /** * READ_GRAMMAR_FROM_FILE if TRUE reads explicit grammar rules from StolckeGrammar.txt file * if false, then createGrammarFromTrainingSamples */ public static boolean READOFF_GRAMMAR_FROM_TREEBANK = true; //depending on READOFF_GRAMMAR_FROM_TREEBANK user must give either TREEBANK_FILE or GRAMMAR_FILE //WSJ_TREEBANK is read only iff READ_WSJ_TREEBANK_FROM_DIRECTORY=false public static String TREEBANK_FILE = "Input/WSJ_labeled_lexical_and_postags.txt"; //TREEBANK_FILE_UNLABELLED only for purposes of lookup in case of FIND_PARSES_CORRESPONDING_WITH_UNLABELED_FROM_TREEBANK public static String TREEBANK_FILE_UNLABELLED = "Input/WSJ_unlabeled_postags.txt"; // Read original WSJ files from directory of WSJ public static boolean READ_WSJ_TREEBANK_FROM_DIRECTORY = false; //public static String TREEBANK_FILE = "Input/OVIS_compleet_labeled.txt"; public static String GRAMMAR_FILE = "Grammars/TreebankTrainSetCFG_Grammar.txt"; // default OVIS parse, if DO_WSJ_PARSE then WSJ parse public static boolean BRACKET_FORMAT_WSJ_STYLE = true; public static int BRANCHING_FACTOR = 2; public static int BEAM_WIDTH = 1000; //DOGOODMANDOP: if false then regular CYK parser public static boolean DOGOODMANDOP = false; public static boolean DOTESTING = true; public static boolean PRINT_STANDARD_OUTPUT_TO_FILE = true; public static boolean PRINT_COMPUTED_PARSES = false; //public final static boolean PRINT_SUMMARY = false; //if you parse for multiple parameter settings, then set iterations>1 //public static int ITERATIONS = 1; //for printing output parses of CHILDES public final static boolean DO_PARSING_NO_EVALUATION = false; //for reading in Tree Bank and printing/outputting the grammar //public final static boolean DO_NO_PARSING = true; /** * EXTRACT_POSTAGS: reads only postags from the treebank, and puts these in the parsetree; * if you don't use POSTAGS then it replaces POSTAGS by lexical items */ public static boolean EXTRACT_POSTAGS = false; public static String TESTSET_UNLABELLED = "Input/WSJ_unlabeled_postags_uppercase.txt"; public static String TESTSET_GOLD_STANDARD = "Input/WSJ_labeled_lexical_and_postags.txt"; //output unlabeled text to WSJ_unlabeled.txt after reading labeled text public final static String OUTPUT_DIR = "./Output"; public final static boolean PRINT_UNLABELED_FILE = false; public final static boolean PRINT_LATEX_FILE = false; public final static boolean PRINT_CLEANEDUP_LABELED_FILE = false; public final static boolean PRINT_CNF_CONVERTED_LABELED_FILE = false; //compute file of spans (0-3 1-3 etc) from WSJ-style labeled file public final static boolean PRINT_SPANS = false; public final static boolean PRINT_BINARYSPANS = false; //public final static boolean public final static boolean DO_RIGHT_BRANCHING_TEST = false; public final static boolean SWITCH_LEFT_BRANCHING = false; //DIRECT_EVALUATION_NO_PARSING for comparing two labeled treebanks, without computing the parses //TAKE_OFF_SBJ: if true, disregards suffixes of the labels in gold standard, such as NP_SBJ, etc, and identifies all NPs, etc. public final static boolean TAKE_OFF_SBJ = false; //OTHERWAYROUND if false, matches induced labels to gold standard labels, if true, other way round public final static boolean OTHERWAYROUND = false; public final static boolean DIRECT_EVALUATION_NO_PARSING = false; public final static boolean FIND_PARSES_CORRESPONDING_WITH_UNLABELED_FROM_TREEBANK = false; //DO_LABEL_ASSOCIATION finds optimal associations between labels of induced parses and gold standard parses public final static boolean DO_LABEL_ASSOCIATION = false; //DO_LABELED_PRECISION_AND_RECALL: if true, evaluates LP and LR, otherwise UP and UR public static boolean DO_LABELED_PRECISION_AND_RECALL = true; public final static boolean PRINT_LABEL_FREQUENCIES = false; public final static boolean PRINT_PARSEVAL_PER_CATEGORY = false; public final static boolean REWRITE_OVIS_PARSES_IN_WSJ_FORMAT = false; public final static boolean REPLACE_LABELS_IN_GRAMMAR = false; public final static boolean DO_BRANCH_COUNTING = false; public static double MU = 1.; public final static double START_MU = 0.5; public final static int TERMINAL = 1; public final static int NONTERMINAL = 2; public static boolean EXIT_APPLICATION = false; /** Creates a new instance of Main */ public Main(String[] args) throws Exception { //collect options //PARSE `grammarFromTreebank' treebankfile testset_unlabeled testset_goldstandard [ovis_style_brackets] [postag] //PARSE `grammarFromFile' grammar_file testset_unlabeled testset_goldstandard //example: grammarFromTreebank ./Input/WSJ22Test10_labeled.txt ./Input/WSJ22Test10_unlabeled.txt ./Input/WSJ22Test10_labeled.txt postag dop_parser export_grammar_file if (args[0].toLowerCase().equals("grammarfromtreebank")) { READOFF_GRAMMAR_FROM_TREEBANK = true; TREEBANK_FILE = args[1]; } else { if (args[0].toLowerCase().equals("grammarfromfile")) { READOFF_GRAMMAR_FROM_TREEBANK = false; GRAMMAR_FILE = args[1]; } else { System.out.println("Please choose one of the options fromTreebank or fromGrammarFile."); EXIT_APPLICATION = true; } } TESTSET_UNLABELLED = args[2]; TESTSET_GOLD_STANDARD = args[3]; for (String s: args) { if (s.toLowerCase().equals("postag")) EXTRACT_POSTAGS = true; if (s.toLowerCase().equals("ovis_style_brackets")) BRACKET_FORMAT_WSJ_STYLE = false; if (s.toLowerCase().equals("dop_parser")) DOGOODMANDOP = true; if (s.toLowerCase().equals("export_grammar_file")) EXPORT_CNF_GRAMMAR_TO_FILE = true; if (s.toLowerCase().equals("unlabeled_p_r")) DO_LABELED_PRECISION_AND_RECALL = false; if (s.toLowerCase().equals("print_parses")) PRINT_COMPUTED_PARSES = true; if (s.toLowerCase().startsWith("branching=")) { BRANCHING_FACTOR = java.lang.Integer.parseInt(s.split("=")[1]); } if (s.toLowerCase().startsWith("beam=")) { BEAM_WIDTH = java.lang.Integer.parseInt(s.split("=")[1]); } } if (!EXIT_APPLICATION && !DO_RIGHT_BRANCHING_TEST && !FIND_PARSES_CORRESPONDING_WITH_UNLABELED_FROM_TREEBANK && !DIRECT_EVALUATION_NO_PARSING) { if (READOFF_GRAMMAR_FROM_TREEBANK) { myGrammar = new Grammar(true); if (Main.PRINT_OUTPUT) Utils.printoutTerminalsNonTerminalsAndRules(myGrammar); if (Main.EXPORT_CNF_GRAMMAR_TO_FILE) { //storeRulesAccordingToRHS is done in GrammarToTextWriter //PRINT GRAMMAR FILE OF THE RULES + PROBABILITIES (printRulesToFile) if (GRAMMAR_PRINTING_FORMAT ==1) Printer.printToFileTerminalsNonTerminalsAndRules(myGrammar); else Printer.printRulesToFile(myGrammar); //prints file without terminals and nonT } if (!Main.EXPORT_CFG_GRAMMAR_TO_FILE && !Main.CALCULATE_LIKELIHOOD) { System.out.println("Start indexing rules according to RHS..."); myGrammar.indexRulesAccordingToRHS(); } } else { //input is a grammar_file: a list of non-terminals, terminals, and rules + probabilities myGrammar = new Grammar(GRAMMAR_FILE); if (Main.PRINT_OUTPUT) Utils.printoutTerminalsNonTerminalsAndRules(myGrammar); } //test of a single sentence: //Utils.doCYKParse(myGammar); } //testing if ((DOTESTING || DO_RIGHT_BRANCHING_TEST || DO_PARSING_NO_EVALUATION) && !FIND_PARSES_CORRESPONDING_WITH_UNLABELED_FROM_TREEBANK) { System.out.println("Starting doTesting... "); new Testing(TESTSET_UNLABELLED, TESTSET_GOLD_STANDARD, myGrammar); } if (DIRECT_EVALUATION_NO_PARSING) { //relabeled_parses can be either relabeled induced parses or relabeled treebank parses //depending on OTHERWAYROUND String parseFile1 = "Input_for_evaluation/relabeled_parses.txt"; //String parseFile2 = "WSJ_labeled_lexical_and_postags_lowercase.txt"; String parseFile2 = null; if (!OTHERWAYROUND) { parseFile2 = "Input_for_evaluation/Viterbi_Treebank_parses.txt"; } else parseFile2 = "Input_for_evaluation/Viterbi_parses.txt"; //induced parses //this is constructor that does not do any parsing, but only evaluation Testing myTest = new Testing(parseFile1, parseFile2); } if (FIND_PARSES_CORRESPONDING_WITH_UNLABELED_FROM_TREEBANK) { String unlabeledViterbiFile = "Input_for_evaluation/Viterbi_unlabeled.txt"; String inducedParseViterbiFile = "Input_for_evaluation/Viterbi_parses.txt"; Utils.doFindTreebankParsesForUnlabeledViterbi(unlabeledViterbiFile, TREEBANK_FILE_UNLABELLED, TREEBANK_FILE); } //if (!EXIT_APPLICATION, etc) } /** * @param args the command line arguments */ public static void main(String[] args) throws Exception { new Main(args); } }