shalmaneser 0.0.1.alpha

data/lib/fred/fred.rb

@@ -0,0 +1,45 @@
+# AB: 2011-11-13
+# Initial import done, need to reimplement the whole interface.
+require 'fred/FredFeaturize'
+require 'fred/FredSplit'
+require 'fred/FredTrain'
+require 'fred/FredTest'
+require 'fred/FredEval'
+
+module Fred
+  class Fred
+
+    def initialize(options)
+      @exp, @opts = options
+      @task = @opts['--task']
+    end
+    ##
+    # now perform the given task
+
+    def assign
+
+      # initialize task object
+      case @task
+      when "featurize"
+        task_obj = FredFeaturize.new(@exp, @opts)
+      when "refeaturize"
+        task_obj = FredFeaturize.new(@exp, @opts, "refeaturize" => true)
+      when "split"
+        task_obj = FredSplit.new(@exp, @opts)
+      when "train"
+        task_obj = FredTrain.new(@exp, @opts)
+      when "test"
+        task_obj = FredTest.new(@exp, @opts)
+      when "eval"
+        task_obj = FredEval.new(@exp, @opts)
+      else
+        raise "Shouldn't be here"
+      end
+      
+      task_obj.compute
+      
+      $stderr.puts "Fred: Done."
+      
+    end
+  end # class Fred
+end # module Fred

data/lib/fred/md5.rb

@@ -0,0 +1,23 @@
+# just for compatibility; requiring "md5" is obsoleted
+#
+# $RoughId: md5.rb,v 1.4 2001/07/13 15:38:27 knu Exp $
+# $Id: md5.rb 12008 2007-03-06 10:12:12Z knu $
+
+require 'digest/md5'
+
+class MD5 < Digest::MD5
+  class << self
+    alias orig_new new
+    def new(str = nil)
+      if str
+        orig_new.update(str)
+      else
+        orig_new
+      end
+    end
+
+    def md5(*args)
+      new(*args)
+    end
+  end
+end

data/lib/fred/opt_parser.rb

@@ -0,0 +1,250 @@
+# -*- encoding: utf-8 -*-
+
+# AB, 2011-11-13
+
+#require 'optparse' # for reimplementation
+require 'getoptlong'
+require "fred/FredConfigData"
+
+module Fred
+
+  # This class parses options for Fred.
+  class OptParser
+    def self.parse(cmd_opts)
+      tasks = {
+        "featurize" => [ [ '--dataset', '-d', GetoptLong::REQUIRED_ARGUMENT], # set to featurize: 'train' or 'test'
+                         [ "--append", "-A", GetoptLong::NO_ARGUMENT]
+                       ],
+        "refeaturize" => [ [ '--dataset', '-d', GetoptLong::REQUIRED_ARGUMENT], # set to featurize: 'train' or 'test'
+                           [ "--append", "-A", GetoptLong::NO_ARGUMENT]
+                         ],
+        "split" => [ ['--logID', '-i', GetoptLong::REQUIRED_ARGUMENT],  # splitlog ID, required, no default
+                     [ '--trainpercent', '-r', GetoptLong::REQUIRED_ARGUMENT]      # percentage training data, default: 90
+                   ],
+        "train" => [ ['--logID', '-i', GetoptLong::REQUIRED_ARGUMENT]  # splitlog ID; if given, will train on split
+                     # rather than all training data
+                   ],
+        "test" => [ ['--logID', '-i', GetoptLong::REQUIRED_ARGUMENT],   # splitlog ID: if given, test on this split of 
+                    # the training data
+                    [ '--baseline', '-b', GetoptLong::NO_ARGUMENT],                # set this to compute baseline rather than
+                    # apply classifiers
+                    [ '--nooutput', '-N', GetoptLong::NO_ARGUMENT]               # set this to prevent output of disambiguated
+                    # test data
+                    
+                  ],
+        "eval" => [['--logID', '-i', GetoptLong::REQUIRED_ARGUMENT],    # splitlog ID: if given, evaluate this split. 
+                   ['--printLog', '-l', GetoptLong::NO_ARGUMENT]
+                  ]
+      }
+      
+      # general options
+      optnames = [[ '--help', '-h', GetoptLong::NO_ARGUMENT],            # get help
+                  [ '--expfile', '-e', GetoptLong::REQUIRED_ARGUMENT],             # experiment file name (and path), no default
+                  [ '--task', '-t', GetoptLong::REQUIRED_ARGUMENT ],               # task to perform: one of task.keys, no default
+                 ]
+      # append task-specific to general options
+      tasks.values.each { |more_optnames|
+        optnames.concat more_optnames
+      }
+      optnames.uniq!
+      
+      # asterisk: "explode" array into individual parameters
+      begin
+        opts = options_hash(GetoptLong.new(*optnames))
+      rescue
+        $stderr.puts "Error: unknown command line option: " + $!
+        exit 1
+      end
+      
+      experiment_filename = nil
+      
+      ##
+      # are we being asked for help?
+      if opts['--help']
+        help()
+        exit(0)
+      end
+      
+      ##
+      # now find the task
+      task = opts['--task']
+      # sanity checks for task
+      if task.nil?
+        help()
+        exit(0)
+      end
+      unless tasks.keys.include? task
+        $stderr.puts "Sorry, I don't know the task " + task
+        exit 1
+      end
+      
+      ##
+      # now evaluate the rest of the options
+      opts.each_pair { |opt,arg|
+        case opt
+        when '--help', '--task'
+          # we already handled this
+        when '--expfile'
+          experiment_filename = arg
+        else
+          # do we know this option?
+          unless tasks[task].assoc(opt)
+            $stderr.puts "Sorry, I don't know the option " + opt + " for task " + task
+            exit 1
+          end
+        end
+      }
+      
+      
+      
+      unless experiment_filename
+        $stderr.puts "I need an experiment file name, option --expfile|-e"
+        exit 1
+      end
+      
+      ##
+      # open config file
+      
+      exp = FredConfigData.new(experiment_filename)
+      
+      # sanity checks
+      unless exp.get("experiment_ID") =~ /^[A-Za-z0-9_]+$/
+        raise "Please choose an experiment ID consisting only of the letters A-Za-z0-9_."
+      end
+      
+      # enduser mode?
+      $ENDUSER_MODE = exp.get("enduser_mode") 
+      
+      # set defaults
+      unless exp.get("handle_multilabel")
+        if exp.get("binary_classifiers")
+          exp.set_entry("handle_multilabel", "binarize")
+        else
+          exp.set_entry("handle_multilabel", "repeat")
+        end
+      end
+      # sanity check: if we're using option 'binarize' for handling items
+      # with multiple labels, we have to have binary classifiers
+      if exp.get("handle_multilabel") == "binarize" and not(exp.get("binary_classifiers"))
+        $stderr.puts "Error: cannot use 'handle_multilabel=binarize' with n-ary classifiers."
+        exit(1)
+      end
+      unless exp.get("numerical_features")
+        exp.set_entry("numerical_features", "bin")
+      end
+
+      [exp, opts]
+    end
+    private
+    ###
+    # options_hash:
+    #
+    # GetoptLong only allows you to access options via each(),
+    # not individually, and it only allows you to cycle through the options once.
+    # So we re-code the options as a hash
+    def self.options_hash(opts_obj) # GetoptLong object
+      opt_hash = Hash.new
+      
+      opts_obj.each do |opt, arg|
+        opt_hash[opt] = arg
+      end
+      
+      return opt_hash
+    end
+    def self.help
+        $stderr.puts "
+Fred: FRamE Disambiguation System Version 0.3
+  
+Usage:
+----------------
+
+ruby fred.rb --help|-h
+  Gets you this text.
+
+
+ruby fred.rb --task|-t featurize --expfile|-e <e> --dataset|-d <d> 
+        [--append|-A]
+  Featurizes input data and stores it in feature files.
+  Feature files are stored in 
+  <fred_directory>/<experiment_ID>/<train/test>/features
+  Enduser mode: dataset has to be test (preset as default), no --append.
+
+  --expfile <e> Use <e> as the experiment description and configuation file
+
+  --dataset <d> Set to featurize: <d> is either 'train' or 'test'
+                Accordingly, either the directory dir_train or dir_test (as 
+                specified in the experiment file) is used to store the features
+
+  --append      Do not overwrite previously computed features for this experiment.
+                Rather, append the new features to the old featurization files.
+                Default: overwrite
+
+ruby fred.rb --task|-t split --expfile|-e <e> --logID|-i <i> 
+             [--trainpercent|-r <r>]
+  Produces a new train/test split on the training data of the experiment.
+  Split logs are stored in <fred_directory>/<experiment_ID>/split/<splitlog ID>
+  Not available in enduser mode.
+
+  --expfile <e> Use <e> as the experiment description and configuation file
+
+  --logID <l>   Use <l> as the ID for storing this new split
+
+  --trainpercent <r> Allocate <r> percent of the data as train, 
+                and 100-<r> as test.
+                default: <r>=90
+     
+ruby fred.rb --task|-t train --expfile|-e <e>  
+             [--logID|-i <i> ]
+  Train classifier(s) on the training data (or a split of it)
+  Classifiers are stored in 
+  <fred_directory>/<experiment_ID>/classifiers/<classifier_name>
+  Not available in enduser mode.
+
+  --expfile <e> Use <e> as the experiment description and configuation file
+
+  --logID <l>   Train not on the whole training data but 
+                on the split with ID <l>
+
+ruby fred.rb --task|-t test --expfile|-e <e>  
+             [--logID|-i <i>] [--baseline|-b]
+             [--nooutput|-N]
+  Apply classifier(s) to the test data (or a split of the training data)
+  Classification results are stored in 
+  <fred_directory>/<experiment_ID>/results/main or
+  <fred_directory>/<experiment_ID>/results/baseline for the baseline.
+  If you are using classifier combination, individual classification results
+  are stored in <fred_directory>/<experiment_ID>/results/<classifier_name>
+  System output (disambiguated text in SalsaTigerXML format) is stored in
+  <fred_directory>/<experiment_ID>/output/stxml
+  or <directory_output>, if that has been specified.
+
+  --expfile <e> Use <e> as the experiment description and configuation file
+
+  --logID <l>   Test on a split of the training data with ID <l>
+
+  --baseline    Compute the baseline: Always assign most frequent sense.
+                Default: use the trained classifiers
+
+  --nooutput    Do not produce an output of the disambiguated test data 
+                in SalsaTigerXML format. This is useful if you just want
+                to evaluate the system.
+                Default: output is produced.
+
+ ruby fred.rb --task|-t eval --expfile|-e <e>  
+              [--logID|-i <i>] [--printLog|-l]
+  Evaluate the performance of Fred on the test data 
+  (or on a split of the training data).
+  Evaluation file is written to <fred_directory>/<experiment_ID>/eval/eval
+  Not available in enduser mode.
+
+  --expfile <e> Use <e> as the experiment description and configuation file
+
+  --logID <l>   Evaluate a split of the training data with ID <l>
+
+  --printLog    Also print logfile detailing evaluation of every instance.
+                Log file is written to <fred_directory>/eval/log
+
+"
+    end
+  end # class OptParser
+end # module FrPrep

data/lib/frprep/AbstractSynInterface.rb

@@ -0,0 +1,1227 @@
+# Katrin Erk Oct/Nov 05
+#
+# Abstract classes for interfaces for systems that provide syntactic
+# analysis.
+#
+# There are two types of interfaces to syntactic analysis systems:
+# - interfaces:
+#   offer methods for syntactic analysis.
+#
+#   SynInterfaceTab:
+#   input and output format is (FN)TabFormat.
+#   SynInterfaceSTXML:
+#    input format is TabFormat, output format is 
+#    Salsa/Tiger XML, also provided as 
+#    SalsaTigerSentence objects
+#
+# - interpreters:
+#   interpret the resulting Salsa/Tiger XML (represented as
+#   SalsaTigerSentence and SynNode objects), e.g.
+#   generalize over part of speech; 
+#   describe the path between a pair of nodes both as a path 
+#   and (potentially) as a grammatical function of one of the nodes;
+#   determine whether a node describes a verb, and in which voice;
+#   determine the head of a constituent
+
+require "tempfile"
+
+require "frprep/ruby_class_extensions"
+
+require "frprep/ISO-8859-1"
+require "frprep/Parser"
+require "frprep/SalsaTigerRegXML"
+require "frprep/TabFormat"
+
+#############################
+# abstract class, to be inherited:
+#
+# tabular format or SalsaTigerXML interface for modules
+# offering POS tagging, lemmatization, parsing etc.
+class SynInterface
+
+  ###
+  # returns a string: the name of the system
+  # e.g. "Collins" or "TNT"
+  def SynInterface.system()
+    raise "Overwrite me"
+  end
+
+  ###
+  # returns a string: the service offered
+  # one of "lemmatizer", "parser", "pos tagger"
+  def SynInterface.service()
+    raise "Overwrite me"
+  end
+
+  ###
+  # initialize to set values for all subsequent processing
+  def initialize(program_path, # string: path to system
+		 insuffix,      # string: suffix of input files
+		 outsuffix,     # string: suffix for processed files
+		 var_hash = {}) # optional arguments in a hash
+	
+    @program_path = program_path
+    @insuffix = insuffix
+    @outsuffix = outsuffix	 
+  end
+
+  ###
+  # process each file in in_dir with matching suffix,
+  # producing a file in out_dir with same name but the suffix replaced
+  #
+  # returns: nothing
+  def process_dir(in_dir,        # string: name of input directory
+		  out_dir)       # string: name of output directory
+
+    Dir[in_dir+"*#{@insuffix}"].each {|infilename|
+      outfilename = out_dir + File.basename(infilename, @insuffix) + @outsuffix
+      process_file(infilename,outfilename)
+    }
+  end
+
+  ###
+  # process one file, writing the result to outfilename
+  #
+  # returns: nothing
+  def process_file(infilename,   # string: name of input file
+		   outfilename)
+    raise "Overwrite me"
+  end
+
+  ######
+  protected
+
+  def SynInterface.announce_me()
+    if defined?(SynInterfaces)
+      # yup, we have a class to which we can announce ourselves
+      SynInterfaces.add_interface(eval(self.name()))
+    else
+      # no interface collector class
+      $stderr.puts "Interface #{self.name()} not announced: no SynInterfaces."
+    end
+  end
+end
+
+#############################
+# abstract class, to be inherited:
+#
+# SalsaTigerXML interface for modules
+# offering parsing etc.
+#
+# The input format for these classes is TabFormat or FNTabFormat
+class SynInterfaceSTXML < SynInterface
+  ###
+  # initialize to set values for all subsequent processing
+  def initialize(program_path, # string: path to system
+		 insuffix,      # string: suffix of input files
+		 outsuffix,     # string: suffix for processed files
+		 stsuffix,      # string: suffix for Salsa/Tiger XML files
+		 var_hash = {}) # optional arguments in a hash
+    super(program_path, insuffix, outsuffix, var_hash)
+    @stsuffix = stsuffix
+  end
+
+  def to_stxml_dir(in_dir,   # string: name of dir with parse files
+		   out_dir)  # string: name of output dir
+    
+    Dir[in_dir+"*#{@outsuffix}"].each { |parsefilename|
+      stxmlfilename = out_dir + File.basename(parsefilename, @outsuffix) + @stsuffix
+      to_stxml_file(parsefilename, stxmlfilename)
+    }
+  end
+
+  def to_stxml_file(infilename, 
+		    outfilename)
+    raise "Overwrite me"
+  end
+
+  ###
+  # standard mapping:
+  #
+  # to be used as the mapping from tab sentence words to 
+  # SalsaTigerSentence nodes returned by each_sentence():
+  # map the n-th word of the tab sentence to the n-th terminal of
+  # the SalsaTigerSentence
+  def SynInterfaceSTXML.standard_mapping(sent, tabsent)
+    retv = Hash.new
+    if sent.nil?
+	return nil
+    end
+    terminals = sent.terminals_sorted()
+    if tabsent
+      tabsent.each_line_parsed { |l|
+        if (t = terminals[l.get("lineno")])
+          retv[l.get("lineno")] = [t]
+        else
+          retv[l.get("lineno")] = []
+        end
+      }
+    end
+    return retv
+  end
+
+
+  ###
+  # for a given processed file:
+  # yield each sentence as a tuple
+  #  [SalsaTigerSentence object, FNTabFormatSentence object, mapping]
+  # of 
+  # - the sentence in SalsaTigerXML, 
+  # - the matching tab format sentence
+  # - a mapping of terminals: 
+  #   hash: line in tab sentence(integer) -> array:SynNode
+  #   mapping tab sentence nodes to matching nodes in the SalsaTigerSentence data structure
+  #
+  # default version: write Salsa/Tiger XML to tempfile, read back in
+  # and assume that each sentence in the tab file has a correspondent
+  # in the processed file (may not hold e.g. if the parser leaves out
+  # sentences it cannot process)
+  def each_sentence(infilename,  # string: name of processed file
+		    tab_dir = nil) # string: name of dir with input files 
+                                 # (set either here or on initialization)
+    if tab_dir
+      @tab_dir = tab_dir
+    end
+
+    # write Salsa/Tiger XML to tempfile
+    tf = Tempfile.new("SynInterface")
+    tf.close()
+    to_stxml_file(infilename, tf.path)
+    tf.flush()
+
+    # get matching tab file, read
+    tab_reader = get_tab_reader(infilename)
+    tab_sentences = Array.new
+    tab_reader.each_sentence { |s| tab_sentences << s }
+
+    # read Salsa/Tiger sentences and yield them
+    reader = FilePartsParser.new(tf.path)
+    sent_index = 0
+    reader.scan_s { |sent_string|
+      yield [
+        SalsaTigerSentence.new(sent_string, tab_sentences[sent_index]), 
+        tab_sentences[sent_index], 
+        SynInterfaceSTXML.standard_mapping(sent, tab_sentences[sent_index])
+      ]
+      sent_index += 1
+    }
+
+    # remove tempfile
+    tf.close(true)
+  end
+
+  #####################
+  protected
+
+
+  ###
+  # get tab format file for a given processed file
+  def get_tab_reader(infilename) # string: name of processed file
+    # find matching non-processed file for processed file
+    # assumption: directory with non-processed files
+    # has been set as @tab_dir
+
+    # sanity checks
+    unless @tab_dir
+      raise "Need to set tab directory"
+    end
+
+    # get matching tab file for this parser output file
+    tabfilename = @tab_dir+File.basename(infilename, @outsuffix)+ @insuffix
+    return FNTabFormatFile.new(tabfilename)
+  end
+
+
+  ###
+  # provide a XML representation for a sentence that couldn't be analyzed
+  # assuming a flat structure of all terminals, adding a virtual top node
+  def SynInterfaceSTXML.failed_sentence(tab_sent,sentid)
+
+    sent_obj = SalsaTigerSentence.empty_sentence(sentid.to_s)
+
+    sent_obj.set_attribute("failed","true")
+    
+    topnode = sent_obj.add_syn("nt",
+                               "NONE", # cat
+                               nil, # word (doesn't matter)
+                               nil, # pos (doesn't matter)
+                               "500") # nonterminal counter
+
+    t_counter = 0
+
+    tab_sent.each_line_parsed {|line|
+      t_counter += 1
+      word = line.get("word")
+      pos = line.get("pos")
+      node = sent_obj.add_syn("t",
+                              nil,  # cat (doesn't matter here)
+                              SalsaTigerXMLHelper.escape(word), # word
+                              pos,  # pos
+                              t_counter.to_s)
+      topnode.add_child(node,nil)
+      node.add_parent(topnode, nil)
+    }
+    return sent_obj
+  end
+end
+
+#############################
+# abstract class, to be inherited:
+#
+# tabular format interface for modules
+# offering POS tagging, lemmatization etc.
+class SynInterfaceTab < SynInterface
+
+  ##########
+  protected
+
+  # fntab_words_for_file:
+  # given a file in tab format, columns as in FNTabFormat,
+  # get the "word" entries and write them to a given file,
+  # one word per line, as input for processing
+  def SynInterfaceTab.fntab_words_to_file(infilename, # string: name of input file
+					    outfile,    # stream: output file
+					    sent_marker = "", # string: mark end of sentence how?
+					    iso = nil)  # non-nil: assume utf-8, transform to iso-8859-1
+    corpusfile = FNTabFormatFile.new(infilename)
+    corpusfile.each_sentence {|s|
+      s.each_line_parsed {|line_obj|
+	if iso
+	  outfile.puts UtfIso.to_iso_8859_1(line_obj.get("word"))
+	else
+	  outfile.puts line_obj.get("word")
+	end
+      }
+      outfile.puts sent_marker
+    }
+  end
+end
+
+#############################
+# class describing a path between two nodes
+#
+# provides access and output facilities for different aspects of the path
+#
+# this is the return value of SynInterpreter.path_between()
+class Path
+  attr_reader :startnode
+
+  ###
+  # initialize to empty path
+  def initialize(startnode)
+    @path = Array.new
+    @cutoff_last_pt = false
+    set_startnode(startnode)
+  end
+
+  ###
+  # deep_clone:
+  # return clone of this path object,
+  #  with clone of this path rather than the same path
+  def deep_clone()
+    new_path = self.clone()
+    new_path.set_path(@path.clone())
+
+    return new_path
+  end
+
+  ###
+  def set_startnode(startnode)
+    @startnode = startnode
+
+    return self
+  end
+
+  ###
+  # iterate through the current path
+  #
+  # yield tuples
+  # [direction, edgelabel, nodelabel, endnode]
+  #  direction: string, U/D
+  #  edgelabel: string
+  #  nodelabel: string
+  #  endnode:   SynNode
+  def each_step()
+    @path.each { |step|
+      yield step
+    }
+  end
+
+  ###
+  # empty?
+  # any steps in here?
+  def empty?
+    return @path.empty?
+  end
+
+  ###
+  # add one step to the beginning of the current path
+  def add_first_step(start_node,#SynNode
+		     direction, # string: U, D
+		     gf,        # string: edge label
+		     pt)
+    @path.prepend([direction, gf, pt, @startnode])
+    set_startnode(start_node)
+
+    return self
+  end
+
+
+  ###
+  # add one step to the end of the current path
+  def add_last_step(direction, # string: U, D
+		     gf,        # string: edge label
+		     pt,        # string: node label (of end_node)
+		     end_node)  # SynNode
+    @path << [direction, gf, pt, end_node]
+
+    return self
+  end
+
+  ###
+  # path length
+  def length()
+    return @path.length()
+  end
+
+  ###
+  # 
+  def print(print_direction, # boolean. true: print direction
+	    print_gf,        # boolean. true: print edgelabel
+	    print_pt)        # boolean. true: print nodelabel
+    
+    return print_aux(@path, print_direction, print_gf, print_pt)
+  end
+
+  ###
+  # print path from roof node to end
+  def print_downpart(print_direction,
+		     print_gf,
+		     print_pt)
+
+    roof, roof_index = compute_roof()
+    if roof.nil? or @path.empty?
+      # no roof set
+      return ""
+
+    else
+      # roof node is in the middle
+      return print_aux(@path[roof_index..-1], 
+		       print_direction, print_gf, print_pt)
+    end
+  end
+  
+  ###
+  def lca()
+    return compute_roof().first
+  end
+
+  ###
+  # cut off last node label in print() and print_downpart()?
+  def set_cutoff_last_pt_on_printing(bool) # Boolean
+    @cutoff_last_pt = bool
+  end
+
+  ########
+  protected
+  
+  def set_path(new_path)
+    @path = new_path
+  end
+
+
+  ########
+  private
+
+  ###
+  # step through the path as long as direction is up.
+  # when direction starts to go "D", take current node as roof node
+  #
+  # returns: pair [roof node, roof node index] (SynNode, integer)
+  def compute_roof()
+    node = @startnode
+    index = 0
+
+    each_step { |direction, edgelabel, nodelabel, endnode|
+      if direction =~ /D/
+        # down! the previous node was roof
+        return [node, index]
+      else
+        node = endnode
+        index += 1
+      end
+    }
+    
+    # last node is roof
+    return [node, index]
+    
+  end
+
+  ###
+  def print_aux(path,
+		print_direction,
+		print_gf,
+		print_pt)
+    retv = ""
+    path.each { |step|
+      direction, gf, pt, node = step.map { |entry| 
+	if entry.nil?
+	  "-"
+	else
+	  entry
+	end
+      }
+      if print_direction
+	retv << direction + " "
+      end
+      if print_gf
+	retv << gf + " "
+      end
+      if print_pt
+	retv << pt + " "
+      end
+    }
+
+    if @cutoff_last_pt and print_pt and
+        retv =~ /^(.+ )\w+ $/
+      return $1
+    else
+      return retv
+    end    
+  end
+
+end
+
+
+#############################
+# abstract class, to be inherited:
+#
+# interpretation for a POS tagger/lemmatizer/parser combination
+class SynInterpreter
+
+  ###
+  # systems interpreted by this class:
+  # returns a hash service(string) -> system name (string),
+  # e.g.
+  # { "parser" => "collins", "lemmatizer" => "treetagger" }
+  def SynInterpreter.systems()
+    raise "Overwrite me"
+  end
+
+  ###
+  # names of additional systems that may be interpreted by this class
+  # returns a hash service(string) -> system name(string)
+  # same as names()
+  def SynInterpreter.optional_systems()
+    raise "Overwrite me"
+  end
+
+  ###
+  # generalize over POS tags.
+  #
+  # returns one of:
+  #
+  # adj:  adjective (phrase)
+  # adv:  adverb (phrase)
+  # card: numbers, quantity phrases
+  # con:  conjunction
+  # det:  determiner, including possessive/demonstrative pronouns etc.
+  # for:  foreign material
+  # noun: noun (phrase), including personal pronouns, proper names, expletives
+  # part: particles, truncated words (German compound parts)
+  # prep: preposition (phrase)
+  # pun:  punctuation, brackets, etc.
+  # sent: sentence
+  # top:  top node of a sentence
+  # verb: verb (phrase)
+  # nil:  something went wrong
+  #
+  # default: return phrase type as is
+  #
+  # returns: string or nil
+  def SynInterpreter.category(node) # SynNode
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    return eval(self.name()).pt(node)
+  end
+
+  ###
+  # is relative pronoun?
+  #
+  # default: false
+  def SynInterpreter.relative_pronoun?(node) # SynNode
+    return false
+  end
+
+  ###
+  # lemma_backoff:
+  #
+  # if we have lemma information, return that,
+  # and failing that, return the word
+  #
+  # returns: string or nil
+  def SynInterpreter.lemma_backoff(node)
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    lemma = node.get_attribute("lemma") 
+    if (lemma.nil? or lemma =~ /unknown/) and
+        node.is_terminal?
+      return node.word()
+    else
+      return lemma
+    end
+  end
+
+  ###
+  # phrase type:
+  # constituent label for nonterminals,
+  # part of speech for terminals
+  #
+  # returns: string
+  def SynInterpreter.pt(node)
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    if node.is_terminal?
+      return node.part_of_speech
+    else
+      return node.category
+    end
+  end
+
+  ###
+  # simplified phrase type:
+  # like phrase type, but may simplify
+  # the constituent label
+  # default: just the same as pt()
+  #
+  # returns: string or nil
+  def SynInterpreter.simplified_pt(node)
+    return eval(self.name()).pt(node)
+  end
+
+  ###
+  # particle_of_verb:
+  #
+  # given a node and a nodelist,
+  # if the node represents a verb:
+  # see if the verb has a particle among the nodes in nodelist
+  # if so, return it
+  # default: no recognition of separate particles
+  #
+  # returns: SynNode object if successful, else nil
+  def SynInterpreter.particle_of_verb(node,
+				      node_list)
+    return nil
+  end
+
+  ###
+  # auxiliary?
+  # 
+  # returns true if the given node is an auxiliary
+  # default: no recognition of auxiliaries
+  #
+  # returns: boolean
+  def SynInterpreter.auxiliary?(node)
+    return false
+  end
+
+  ###
+  # modal?
+  #
+  # returns true if the given node is a modal verb
+  # default: no recognition of modals
+  #
+  # returns: boolean
+  def SynInterpreter.modal?(node)
+    return false
+  end
+
+  ###
+  # head_terminal
+  #
+  # given a constituent, return the terminal node
+  # that describes its headword
+  # default: a heuristic that assumes the existence of a 'head' 
+  #   attribute on nodes:
+  #   find the first node in my yield corresponding to my head attribute..
+  #
+  # returns: a SynNode object if successful, else nil
+  def SynInterpreter.head_terminal(node)
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    if node.is_terminal?
+      return node
+    end
+
+    head = node.get_attribute("head")
+    unless head
+      return nil
+    end
+      
+    return node.yield_nodes.detect { |t|
+      t.get_attribute("word") == head
+    }
+  end
+
+  ###
+  # voice
+  #
+  # given a constituent, return 
+  # - "active"/"passive" if it is a verb
+  # - nil, else
+  #
+  # default: treat all as active
+  def SynInterpreter.voice(node)
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    if eval(self.name()).category(node) == "verb"
+      return "active"
+    else
+      return nil
+    end
+  end
+
+  ###
+  # gfs
+  #
+  # grammatical functions of a constituent:
+  # 
+  # returns: a list of pairs [relation(string), node(SynNode)]
+  # where <node> stands in the relation <relation> to the parameter
+  # that the method was called with
+  #
+  # default: children of this node, with edge labels as relations,
+  # prepositions tacked on for pps
+  def SynInterpreter.gfs(node,    # SynNode
+                         sent)    # SalsaTigerSentence
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    return node.children_with_edgelabel().map { |rel, gf_node|
+ 
+     if eval(self.name()).category(gf_node) == "prep"
+        [rel + "-" + eval(self.name()).preposition(gf_node).to_s, gf_node]
+
+      else
+        [rel, gf_node]
+      end
+    }
+  end
+
+  ###
+  # informative_content_node
+  #
+  # for most constituents: the head
+  # for a PP, the NP
+  # for an SBAR, the VP
+  # for a VP, the embedded VP 
+  #
+  # Default: returns the first non-head child
+  def SynInterpreter.informative_content_node(node)
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    headlemma = eval(self.name()).lemma_backoff(node)
+
+    first_nonhead_child = node.children().detect { |n| 
+      nnh = eval(self.name()).head_terminal(n)
+      nnh and 
+        eval(self.name()).lemma_backoff(nnh) != headlemma 
+    }
+
+    return first_nonhead_child
+  end
+
+  #####################################
+  # verbs(sent)  sent is a sentence in SalsaTigerSentence format
+  #
+  # return a list of the nodes of full verbs in a given sentence:
+  # it is a list of lists. An item in that list is
+  # - either a pair [verb, svp]
+  #   of the node of a verb with separable prefix 
+  #   and the node of its separate prefix
+  # - or a singleton [verb]
+  #   of the node of a verb without separate prefix 
+  def SynInterpreter.verbs(sent)
+
+    return sent.syn_nodes.select { |node|
+      eval(self.name()).category(node) == "verb"
+    }.map { |node|
+      [node]
+    }
+  end
+
+  ###
+  # governing verbs
+  #
+  # returns a list of pairs [rel, verb_node]
+  # such that the given node fills the grammatical function rel
+  # for this verb_node
+  # or an empty list if there is no such verb
+  def SynInterpreter.governing_verbs(node,
+                                     sent)
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    retv = Array.new
+
+    # each verb of the sentence:
+    eval(self.name()).verbs(sent).each { |verb_node, prefix_node|
+      # each gf of this verb:
+      eval(self.name()).gfs(verb_node, sent).each { |rel, other_node|
+        # if it points to the given node, record
+        if other_node == node or
+            eval(self.name()).informative_content_node(other_node) == node
+          retv << [rel, verb_node]
+          break
+        end
+      }
+    }
+
+    return retv
+  end
+
+  ###
+  # path_between
+  #
+  # construct path in syntactic structure between two nodes, 
+  # using 
+  # - node labels
+  # - edge labels
+  # - direction Up, Down
+  #
+  # use_nontree_edges: set to true to use coreference edges
+  # and other non-tree edges returned by the parser
+  # in path computation. (Will produce no change if the parser
+  # does not produce any non-tree edges.)
+  #
+  # returns: Path object
+  def SynInterpreter.path_between(from_node, # SynNode
+                                  to_node,   # SynNode
+				  use_nontree_edges = false) # boolean
+    
+    unless from_node.kind_of? SynNode and to_node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    path = eval(self.name()).search_up(from_node,to_node, nil)
+    
+    if path.nil?
+      # no path found
+#      STDERR.puts "Warning: no path found between #{to_node.id} and #{from_node.id}"
+    end
+
+    return path
+  end
+
+  ###
+  # surrounding_nodes:
+  #
+  # construct paths in syntactic structure between a node and each of its neighbors
+  # path construction as in path_between.
+  # Neighbors: parent, child, plus potentially neighbors by nontree edges
+  # use_nontree_edges: again, same as in path_between
+  #
+  # returns: list of pairs [neighbor(SynNode), path(Path)]
+  def SynInterpreter.surrounding_nodes(node, # SynNode
+                                       use_nontree_edges = false) # boolean
+  
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    retv = Array.new
+
+    # parent
+    if (p = node.parent)
+      retv << [
+        p, 
+        Path.new(node).add_last_step("U", node.parent_label(),
+                                     eval(self.name()).simplified_pt(p), p)
+      ]
+    end
+
+    # children
+    node.each_child_with_edgelabel { |label, c|
+      retv << [
+        c,
+        Path.new(node).add_last_step("D", label, 
+                                     eval(self.name()).simplified_pt(c), c)
+      ]
+    }
+
+    return retv
+  end
+
+  ###
+  # relative_position
+  # of a node with respect to an (anchor) node:
+  # left, right, dom
+  def SynInterpreter.relative_position(node,        # SynNode
+                                       anchor_node) # SynNode
+
+    unless node.kind_of? SynNode and anchor_node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    # compute up to a root node
+    root = node
+    while (p = root.parent())
+      root = p
+    end
+
+    # determine position of {leftmost, rightmost} terminal of 
+    # {node, anchor_node} in the list of all terminals
+    all_yieldnodes = root.yield_nodes_ordered()
+
+    pos_nodefirst = all_yieldnodes.index(eval(self.name()).leftmost_terminal(node))
+    pos_anchorfirst = all_yieldnodes.index(eval(self.name()).leftmost_terminal(anchor_node))
+    pos_nodelast = all_yieldnodes.index(eval(self.name()).rightmost_terminal(node))
+    pos_anchorlast = all_yieldnodes.index(eval(self.name()).rightmost_terminal(anchor_node))
+
+    # determine relative position
+    if pos_nodefirst and pos_anchorfirst and pos_nodefirst < pos_anchorfirst
+      return "LEFT"
+    elsif pos_nodelast and pos_anchorlast and pos_anchorlast < pos_nodelast
+      return "RIGHT"
+    else
+      return "DOM" 
+    end
+  end
+
+  ###
+  # leftmost_terminal
+  #
+  # given a constituent, determine its leftmost terminal, 
+  # excluding punctuation
+  def SynInterpreter.leftmost_terminal(node)
+    leftmost = node.yield_nodes_ordered().detect {|n| eval(self.name()).category(n) != "pun"}
+    unless leftmost
+      leftmost = node.yield_nodes_ordered().first
+    end
+    return leftmost
+  end
+
+  ###
+  # rightmost_terminal
+  #
+  # given a constituent, determine its rightmost terminal, 
+  # excluding punctuation
+  def SynInterpreter.rightmost_terminal(node)
+    rightmost = node.yield_nodes_ordered().reverse.detect {|n| eval(self.name()).category(n) != "pun"}
+    unless rightmost
+      rightmost = node.yield_nodes_ordered().last
+    end
+    return rightmost
+  end
+
+  ###
+  # preposition
+  #
+  # if the given node represents a PP, return the preposition
+  #
+  # default: assume that either the PP node will have the preposition as its lemma,
+  # or that the head terminal of the PP will be the preposition
+  def SynInterpreter.preposition(node)
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    # preposition as lemma of this node?
+    if eval(self.name()).category(node) == "prep" and
+        (lemma = eval(self.name()).lemma_backoff(node)) and
+        not(lemma.empty?)
+      return lemma
+    end
+
+    # head terminal is preposition and has a lemma?
+    hl = eval(self.name()).head_terminal(node)
+    if hl and
+        eval(self.name()).category(hl) == "prep" and
+        (lemma = eval(self.name()).lemma_backoff(hl)) and
+        not(lemma.empty?)
+      return lemma
+    end
+
+    # no luck
+    return nil
+  end
+
+
+  ###
+  # main node of expression
+  #
+  # returns: SynNode, main node, if found
+  # else nil
+  def SynInterpreter.main_node_of_expr(nodelist, 
+                                       no_mwes = nil) # non-nil: don't handle multiword expressions beyond verbs with separate particles
+
+    # map nodes to terminals
+    nodelist1 = nodelist.map { |n| n.yield_nodes() }.flatten
+
+    # single node? return it
+    if nodelist1.length == 1
+      return nodelist1.first
+    end
+
+    # more than one word
+
+    # see if we can get a headword of a single constituent
+    if nodelist.length() == 1 and
+	(headword = eval(self.name()).head_terminal(nodelist.first()))
+      return headword
+    end
+
+    # filter out auxiliaries and modals, see if only one node remains
+    nodelist2 = nodelist1.reject { |t| 
+      eval(self.name()).auxiliary?(t) or
+	eval(self.name()).modal?(t)
+    }
+
+    # one verb, one prep or particle? then 
+    # assume we have a separate verb prefix, and take the lemma of the verb
+    if nodelist2.length == 2
+      verbs = nodelist2.select { |t| eval(self.name()).category(t) == "verb"}
+      if verbs.length() == 1
+	# found exactly one verb, so we have one verb, one other
+	if eval(self.name()).particle_of_verb(verbs.first, nodelist2)
+	  # we have found a particle/separate verb prefix
+	  # take verb as main node
+	  return verbs.first
+	end
+      end
+    end
+
+    if no_mwes
+      # I was told only to look for separate verb particles,
+      # not for anything else, so return nil at this point
+      return nil
+    end
+
+    # filtered out everything? oops -- return to previous node list
+    if nodelist2.empty?
+      nodelist2 = nodelist1
+    end
+
+    # if the nodelist describes an mwe, try to find its headword:
+    # look for the lowest common ancestor of all nodes in nodelist2
+    # if its head terminal is in nodelist2, return that
+    lca = nodelist2.first
+    lca_found = false
+    while lca and not(lca_found)
+      yn = lca.yield_nodes()
+      # lca's yield nodes include all nodes in nodelist2? 
+      # then lca is indeed the lowest common ancestor
+      if nodelist2.big_and { |t| yn.include? t }
+        lca_found = true
+      else
+        lca = lca.parent()
+      end
+    end
+    # nodelist2 includes lca's head terminal? then return that
+    if lca_found and 
+        (h = eval(self.name()).head_terminal(lca)) and
+        nodelist2.include? h
+      return h
+    end
+      
+
+    # try first verb, then first noun, then first adjective
+    ["verb", "noun", "adj"].each { |cat|
+      nodelist.each { |t|
+        if eval(self.name()).category(t) == cat
+          return t
+        end
+      }
+    }
+
+    # return first node
+    return nodelist.first
+  end
+
+  ########
+  # max constituents:
+  # given a set of nodes, compute the maximal constituents
+  # that exactly cover them
+  #
+  # If include_single_missing_children is set to true,
+  # then a node that has at least one child whose yield is in nodelist,
+  #   and has only one child whose yield is not in nodelist,
+  #   will be considered as having its yield in nodelist.
+  #
+  # Optionally, a procedure accept_anyway_proc can be given.
+  # Like the option include_single_missing_children, it can lead to nodes being
+  # included in the list of nodes whose yield nodes are all also yield nodes of node_list (NYNAAYNN)
+  # even though not all of their yield nodes are yield nodes of the node_list.
+  # accept_anyway_proc can implement arbitrary rules for including nodes in NYAAYNN.
+  # The procedure is called with three arguments:
+  #   accept_anyway_proc(node, ch_in, ch_out)
+  # node is a SynNode that would not normally be in NYAAYNN.
+  # ch_in is the list of its children that are in NYAAYNN.
+  # ch_out is the list of its children that are not.
+  # If the procedure exists and returns true, node is put into NYAAYNN.
+  #
+  # 
+  # default: use the SalsaTigerSentence method for this
+  def SynInterpreter.max_constituents(nodeset, # Array:SynNode
+                                      sent,    # SalsaTigerSentence
+                                      idealize_maxconst = false, # boolean
+                                      accept_anyway_proc = nil)  # procedure
+
+    if idealize_maxconst
+      return sent.max_constituents_smc(nodeset, idealize_maxconst, 
+                                       false, # do not ignore empty terminals
+                                       accept_anyway_proc)
+    else
+      return sent.max_constituents_for_nodes(nodeset)
+    end
+  end
+
+  ########
+  # prune?
+  # given a target node t and another node n of the syntactic structure,
+  # decide whether n is likely to instantiate a semantic role
+  # of t. If not, recommend n for pruning. 
+  #
+  # This method is supposed to implement a method similar
+  # to the one proposed by Xue and Palmer (EMNLP 2004).
+  #
+  # returns: true to recommend n for pruning, else false
+  #
+  # Since the implementation is highly parser-specific,
+  # all that we can do in the default method is 
+  # always to return false.
+  def SynInterpreter.prune?(node, # SynNode
+                            paths_to_target, # hash: node ID -> Path object: paths from nodes to target
+                            terminal_index)  # hash: terminal node -> word index in sentence
+
+    unless node.kind_of? SynNode
+      $stderr.puts "Warning: unexpected input class #{node.class} to SynInterpreter"
+      return nil
+    end
+
+    return false
+  end
+
+  
+  ####################3
+  protected
+
+  def SynInterpreter.announce_me()
+    if defined?(SynInterfaces)
+      # yup, we have a class to which we can announce ourselves
+      SynInterfaces.add_interpreter(eval(self.name()))
+    else
+      # no interface collector class
+      $stderr.puts "Interface #{self.name()} not announced: no SynInterfaces."
+    end
+  end
+
+  ####################3
+  private
+  
+  ###
+  # search upward:
+  # look for path from from_node to to_node
+  # already_covered is either nil or
+  # a node whose subtree we have already searched
+  def SynInterpreter.search_up(from_node, # SynNode
+                               to_node,   # SynNode      
+			       already_covered) # SynNode
+    # returns (1) the path from from_node to to_node, 
+    #         (2) just the part from the lca down to the node
+    #         (3) the lowest common ancestor as node
+
+    path = eval(self.name()).search_down(from_node,to_node, already_covered)
+
+    if path.nil?
+      # search down unsuccessful
+
+      parent = from_node.parent
+      edgelabel = from_node.parent_label
+      # puts "Going up from "+from_node.id.to_s+" to "+parent.id.to_s
+
+      if parent.nil?
+	# no path found
+	return nil
+
+      else
+	# search up 
+	path = eval(self.name()).search_up(parent,to_node, from_node)
+
+	if path.nil? 
+	  # no path found
+	  return nil
+
+        else
+	  # search up was successful
+          parent_pt = eval(self.name()).simplified_pt(parent)
+	  path.add_first_step(from_node, "U", edgelabel, parent_pt)
+	  return path
+	end
+      end
+
+    else
+      # search down successful
+      return path
+    end
+  end
+  
+  ###
+  # search in tree
+  def SynInterpreter.search_down(from_node,        # SynNode
+				 to_node,          # SynNode
+				 already_explored) # SynNode
+
+    if from_node == to_node
+      return Path.new(from_node)
+
+    else
+
+      from_node.children.each {|c|
+
+	if c == already_explored
+	  # we have done this subtree,
+	  # don't do it again
+	  next
+	end
+
+	path = eval(self.name()).search_down(c, to_node, already_explored)
+	
+	unless path.nil?
+	  c_pt = eval(self.name()).simplified_pt(c)
+	  path.add_first_step(from_node, "D", c.parent_label(), c_pt)
+	  return path
+	end
+      }
+
+      # no path found for any of the children
+      return nil
+    end
+  end
+end