shalmaneser 0.0.1.alpha

data/lib/common/FixSynSemMapping.rb

@@ -0,0 +1,196 @@
+###
+# FixSynSemMapping:
+# Given a SalsaTigerRegXML sentence with semantic role annotation,
+# simplify the mapping of semantic roles to syntactic constituents
+# 
+# The following is lifted from the LREC06 paper on Shalmaneser:
+# During preprocessing, the span of semantic roles in the training corpora is 
+# projected onto the output of the syntactic parser by assigning each
+# role to the set of maximal constituents covering its word span.
+# f the word span of a role does not coincide
+# with parse tree constituents, e.g. due to misparses,
+# the role is ``spread out'' across several constituents. This leads to
+# idiosyncratic paths between predicate and semantic role in the parse
+# tree. 
+#
+# [The following span standardization algorithm is used to make the
+# syntax-semantics mapping more uniform:]
+# Given a role r that has been assigned, let N be the set of
+# terminal nodes of the syntactic structure that are covered by r. 
+#
+#   Iteratively compute the maximal projection of N in the syntactic
+#   structure: 
+#   1) If n is a node such that all of n's children are in N,
+#     then remove n's children from N and add n instead.
+#   2) If n is a node with 3 or more children, and all of n's
+#     children except one are in N, then remove n's children from N
+#     and add n instead. 
+#   3) If n is an NP with 2 children, and one of them, another NP,
+#     is in N, and the other, a relative clause, is not, then remove
+#     n's children from N and add n instead.
+#
+#   If none of the rules is applicable to N anymore, assign r to the
+#   nodes in N.
+#
+# Rule 1 implements normal maximal projection. Rule 2 ``repairs'' parser
+# errors where all children of a node but one have been assigned the
+# same role. Rule 3 addresses a problem of the FrameNet data, where
+# relative clauses have been omitted from roles assigned to NPs. 
+
+# KE Feb 08: rule 3 currently out of commission!
+
+require "common/SalsaTigerRegXML"
+
+module FixSynSemMapping
+  ##
+  # fix it
+  #
+  # relevant settings in the experiment file:
+  #
+  # fe_syn_repair:
+  # If there is a node that would be a max. constituent for the
+  # words covered by the given FE, except that it has one child
+  # whose words are not in the FE, use the node as max constituent anyway.
+  # This is to repair cases where the parser has made an attachment choice
+  # that differs from the one in the gold annotation
+  #
+  # fe_rel_repair:
+  # If there is an NP such that all of its children except one have been
+  # assigned the same FE, and that missing child is a relative clause
+  # depending on one of the other children, then take the complete NP as
+  # that FE
+  def FixSynSemMapping.fixit(sent, # SalsaTigerSentence object
+                             exp,  # experiment file object
+                             interpreter_class) # SynInterpreter class
+
+    
+    unless exp.get("fe_syn_repair") or exp.get("fe_rel_repair")
+      return 
+    end
+
+    if sent.nil?
+	return
+    end
+
+    # "repair" FEs: 
+    sent.each_frame { |frame|
+
+      frame.each_child { |fe_or_target|
+
+        # repair only if the FE currently
+        # points to more than one syn node
+        if fe_or_target.children.length() < 2
+          next
+        end
+
+        if exp.get("fe_rel_repair")
+          lastfe = fe_or_target.children.last()
+          if lastfe and interpreter_class.simplified_pt(lastfe) =~ /^(WDT)|(WP\$?)|(WRB)/
+
+            # remove syn nodes that the FE points to
+            old_fe_syn = fe_or_target.children()
+            old_fe_syn.each { |child|
+              fe_or_target.remove_child(child)
+            }
+
+            # set it to point only to the last previous node, the relative pronoun
+            fe_or_target.add_child(lastfe)
+          end
+        end
+
+        if exp.get("fe_syn_repair")
+          # remove syn nodes that the FE points to
+          old_fe_syn = fe_or_target.children()
+          old_fe_syn.each { |child|
+            fe_or_target.remove_child(child)
+          }
+
+          # and recompute
+          new_fe_syn = interpreter_class.max_constituents(old_fe_syn.map { |t| 
+                                                            t.yield_nodes
+                                                          }.flatten.uniq, 
+                                                          sent, 
+                                                          exp.get("fe_syn_repair"))
+            
+          # make the FE point to the new nodes
+          new_fe_syn.each { |syn_node|
+            fe_or_target.add_child(syn_node)
+          }
+        end
+      } # each FE
+    } # each frame
+  end # def fixit
+end # module
+
+
+#########3
+# old code
+
+#     if exp.get("fe_rel_repair")
+#       # repair relative clauses:
+#       # then make a procedure to pass on to max constituents
+#       # that will recognize the relevant cases
+
+#       accept_anyway_proc = Proc.new { |node, children_in, children_out|
+
+#         # node: SynNode
+#         # children_in, children_out: array:SynNode. children_in are the children
+#         #    that are already covered by the FE, children_out the ones that aren't
+          
+#         # if node is an NP,
+#         # and only one of its children is out, 
+#         # and one node in children_in is an NP, and the missing child is an SBAR
+#         # with a child that is a relative pronoun, then consider the child in children_out as covered
+#         if interpreter_class.category(node) == "noun" and 
+#             children_out.length() == 1 and
+#             children_in.select { |n| interpreter_class.category(n) == "noun" } and
+#             interpreter_class.category(children_out.first) == "sent" and 
+#             (ch = children_out.first.children) and  
+#             ch.select { |n| interpreter_class.relative_pronoun?(n) }
+#           true
+#         else
+#           false
+#         end
+#       }
+
+#     else
+#       accept_anyway_proc = nil
+#     end
+
+
+#     # "repair" FEs: 
+#     sent.each_frame { |frame|
+
+#       frame.each_child { |fe_or_target|
+
+#         # repair only if the FE currently
+#         # points to more than one syn node, or
+#         # if it is a noun with a non-covered sentence sister
+#         if fe_or_target.children.length() > 1 or
+#             (exp.get("fe_rel_repair") and (curr_marked = fe_or_target.children.first())  and
+#              interpreter_class.category(curr_marked) == "noun" and
+#              (p = curr_marked.parent) and 
+#              p.children.select { |n| n != curr_marked and interpreter_class.category(n) == "sent" } )
+
+#           # remember nodes covered by the FE
+#           old_fe_syn = fe_or_target.children()
+
+#           # remove syn nodes that the FE points to
+#           old_fe_syn.each { |child|
+#             fe_or_target.remove_child(child)
+#           }
+
+#           # and recompute
+#           new_fe_syn = interpreter_class.max_constituents(old_fe_syn.map { |t| t.yield_nodes}.flatten.uniq, 
+#                                                           sent, 
+#                                                           exp.get("fe_syn_repair"),
+#                                                           accept_anyway_proc)
+            
+#           # make the FE point to the new nodes
+#           new_fe_syn.each { |syn_node|
+#             fe_or_target.add_child(syn_node)
+#           }
+
+#         end # if FE points to more than one syn node
+#       } # each FE
+#     } # each frame

data/lib/common/FrPrepConfigData.rb

@@ -0,0 +1,66 @@
+# FPrepConfigData
+# Katrin Erk July 05
+#
+# Preprocessing for Fred and Rosy: 
+# access to a configuration and experiment description file
+
+require "common/ConfigData"
+
+##############################
+# Class FrPrepConfigData
+#
+# inherits from ConfigData,
+# sets variable names appropriate to preprocessing task
+
+class FrPrepConfigData < ConfigData
+  def initialize(filename)
+
+    # initialize config data object
+    super(filename,          # config file
+	  { "prep_experiment_ID" => "string", # experiment identifier
+
+           "frprep_directory" => "string", # dir for frprep internal data
+
+            # information about the dataset
+            "language" => "string", # en, de
+            "origin"=> "string",    # FrameNet, Salsa, or nothing
+            "format" => "string",   # Plain, SalsaTab, FNXml, FNCorpusXml, SalsaTigerXML
+            "encoding" => "string", # utf8, iso, hex, or nothing
+           
+
+            # directories
+            "directory_input" => "string", # dir with input data
+            "directory_preprocessed" => "string", # dir with output Salsa/Tiger XML data
+            "directory_parserout" => "string", # dir with parser output for the parser named below
+           
+            # syntactic processing
+            "pos_tagger" => "string", # name of POS tagger
+            "lemmatizer" => "string", # name of lemmatizer
+            "parser" => "string",     # name of parser
+            "pos_tagger_path" => "string", # path to POS tagger
+            "lemmatizer_path" => "string", # path to lemmatizer
+            "parser_path" => "string",     # path to parser
+            "parser_max_sent_num" => "integer", # max number of sentences per parser input file
+            "parser_max_sent_len" => "integer", # max sentence length the parser handles
+            
+            "do_parse" => "bool",    # use parser?
+            "do_lemmatize" => "bool",# use lemmatizer?
+            "do_postag" => "bool",   # use POS tagger?
+            
+            # output format: if tabformat_output == true,
+            # output in Tab format rather than Salsa/Tiger XML
+            # (this will not work if do_parse == true)
+            "tabformat_output" => "bool",
+
+            # syntactic repairs, dependent on existing semantic role annotation
+            "fe_syn_repair" => "bool", # map words to constituents for FEs: idealize?
+            "fe_rel_repair" => "bool", # FEs: include non-included relative clauses into FEs
+	  },
+	  [ ] # variables
+	  )
+    
+  end
+end
+
+
+ 

data/lib/common/FrprepHelper.rb

@@ -0,0 +1,1324 @@
+# Salsa packages
+require "common/ISO-8859-1"
+require "common/Parser"
+require "common/RegXML"
+require "common/SalsaTigerRegXML"
+require "common/SalsaTigerXMLHelper"
+require "common/TabFormat"
+require "common/ruby_class_extensions"
+require "common/AbstractSynInterface"
+
+############################################3
+# Module FrprepHelper:
+#
+# diverse transformation methods for frprep.rb
+# moved over here to make the main file less crowded
+module FrprepHelper
+  
+  ####
+  # transform a file to UTF-8 from a given encoding
+  def FrprepHelper.to_utf8_file(input_filename, # string: name of input file
+                           output_filename, # string: name of output file
+                           encoding) # string: "iso", "hex"
+    begin
+      infile = File.new(input_filename)
+      outfile = File.new(output_filename, "w")
+    rescue
+      raise "Could not read #{input_filename}, or could not write to #{output_filename}."
+    end
+
+    while (line = infile.gets())
+      case encoding
+      when "iso"
+        outfile.puts UtfIso.from_iso_8859_1(line)
+      when "hex"
+        outfile.puts UtfIso.from_iso_8859_1(Ampersand.hex_to_iso(line))
+      else
+        raise "Shouldn't be here."
+      end
+    end 
+    infile.close()
+    outfile.close()
+  end
+
+  ####
+  # transform BNC format file to plaintext file
+  def FrprepHelper.bnc_to_plain_file(input_filename, # string: name of input file
+                                     output_filename) # string: name of output file
+    begin
+      infile = File.new(input_filename)
+      outfile = File.new(output_filename, "w")
+    rescue
+      raise "Could not read #{input_filename}, or could not write to #{output_filename}."
+    end
+
+    infile.each { |line|
+    # does this line contain a sentence?
+    if line =~ /^\s*<s\s+n=/
+      # remove all tags, replace by spaces, 
+      # then remove superfluous spaces
+      textline = line.gsub(/<.+?>/, " ").strip().squeeze(" ")
+
+
+      textline.gsub!(/&bquo;/, '"')
+      textline.gsub!(/&equo;/, '"')
+      textline.gsub!(/&mdash;/, "-")
+      textline.gsub!(/&ndash;/, "-")
+      textline.gsub!(/&percnt;/, "%")
+      textline.gsub!(/&pound;/, " pounds ")
+      textline.gsub!(/&amp;/, " and ")
+      textline.gsub!(/&hellip;/, "...")
+      textline.gsub!(/&copy;/, "(copyright)")
+      textline.gsub!(/&eacute;/, "e")
+      textline.gsub!(/&bull;/, "*")
+      textline.gsub!(/&dollar;/, "$")
+      textline.gsub!(/&deg;/, " degree ")
+
+      textline.gsub!(/&frac12;/, "1/2")
+      textline.gsub!(/&frac34;/, "3/4")
+
+      textline.gsub!(/&lsqb;/, "[")
+      textline.gsub!(/&rsqb;/, "]")
+
+      textline.gsub!(/&ins;/, "i")
+      textline.gsub!(/&ft;/, "ft")
+
+      textline.gsub!(/&rarr;/, ">")
+      textline.gsub!(/&larr;/, "<")
+
+
+      textline.gsub!(/&aacute;/, "a")
+      textline.gsub!(/&auml;/, "a")
+      textline.gsub!(/&agrave;/, "a")
+      textline.gsub!(/&atilde;/, "a")
+      textline.gsub!(/&acirc;/, "a")
+      textline.gsub!(/&Aacute;/, "A")
+      textline.gsub!(/&Auml;/, "A")
+      textline.gsub!(/&Agrave;/, "A")
+      textline.gsub!(/&Atilde;/, "A")
+      textline.gsub!(/&Acirc;/, "A")
+
+      textline.gsub!(/&eacute;/, "e")
+      textline.gsub!(/&egrave;/, "e")
+      textline.gsub!(/&ecirc;/, "e")
+      textline.gsub!(/&euml;/, "e")
+      textline.gsub!(/&Eacute;/, "E")
+      textline.gsub!(/&Egrave;/, "E")
+      textline.gsub!(/&Ecirc;/, "E")
+      textline.gsub!(/&Euml;/, "E")
+
+      textline.gsub!(/&iacute;/, "i")
+      textline.gsub!(/&igrave;/, "i")
+      textline.gsub!(/&icirc;/, "i")
+      textline.gsub!(/&iuml;/, "i")
+      textline.gsub!(/&Iacute;/, "I")
+      textline.gsub!(/&Igrave;/, "I")
+      textline.gsub!(/&Icirc;/, "I")
+
+      textline.gsub!(/&oacute;/, "o")
+      textline.gsub!(/&ograve;/, "o")
+      textline.gsub!(/&ocirc;/, "o")
+      textline.gsub!(/&ouml;/, "o")
+      textline.gsub!(/&Oacute;/, "O")
+      textline.gsub!(/&Ograve;/, "O")
+      textline.gsub!(/&Ocirc;/, "O")
+      textline.gsub!(/&Ouml;/, "O")
+
+      textline.gsub!(/&uacute;/, "u")
+      textline.gsub!(/&ugrave;/, "u")
+      textline.gsub!(/&ucirc;/, "u")
+      textline.gsub!(/&uuml;/, "u")
+      textline.gsub!(/&Uacute;/, "U")
+      textline.gsub!(/&Ugrave;/, "U")
+      textline.gsub!(/&Ucirc;/, "U")
+      textline.gsub!(/&Uuml;/, "U")
+
+      textline.gsub!(/&yuml;/, "y")
+      textline.gsub!(/&Yuml;/, "Y")
+
+      textline.gsub!(/&ntilde;/, "n")
+      textline.gsub!(/&Ntilde;/, "N")
+
+      textline.gsub!(/&ccedil;/, "c")
+      textline.gsub!(/&Ccedil;/, "C")
+
+
+      outfile.puts textline
+    end
+    }
+    infile.close()
+    outfile.close()
+  end
+
+
+  ####
+  # transform plaintext file to Tab format file
+  def FrprepHelper.plain_to_tab_file(input_filename,# string: name of input file
+                                     output_filename)    # string: name of output file
+    begin
+      infile = File.new(input_filename)
+      outfile = File.new(output_filename, "w")
+    rescue
+      raise "Could not read #{input_filename}, or could not write to #{output_filename}."
+    end
+
+    filename_core = File.basename(input_filename, "txt")
+
+    # array(string): keep the words of each sentence
+    sentence = Array.new
+    # sentence number for making the sentence ID:
+    # global count, over all input files
+    sentno = 0
+    
+    while (line = infile.gets())
+      
+      # make a sentence ID for the next sentence: running number
+      sentid = filename_core + "_" + sentno.to_s
+      sentno += 1
+      
+      # read words into the sentence array,
+      # separating out punctuation attached to the beginning or end of words
+      sentence.clear()
+      line.split.each { |word|
+        # punctuation at the beginning of the word
+        #if word =~ /^([\(\[`'\"-]+)(.*)$/
+        if word =~ /^([\(\[`\"-]+)(.*)$/
+        punct = $1
+          word = $2
+          punct.scan(/./) { |single_punct|
+            sentence << single_punct
+          }
+   
+        end
+        # punctuation at the end of the word
+        #if word =~ /[,:;-\`?!'\"\.\)\]]+$/
+        if word =~ /[,:;-\`?!\"\.\)\]]+$/
+          sentence << $`  # part before the match: the word
+          punct = $&
+          punct.scan(/./) { |single_punct|
+            sentence << single_punct
+          }
+             
+        else
+          # no punctuation recognized
+          sentence << word
+        end
+      }
+
+
+      # remove empty words
+      sentence.reject! { |word| word.nil? or word.strip.empty? }
+
+      # write words to tab file
+      # KE Dec 06: TabFormat changed
+      sentence.each { |word|
+        # for each word, one line, entries in the line tab-separated
+        # the 'word' entry is the word, the 'lu_sent_ids' entry is the sentence ID sentid,
+        # all other entries (gf, pt, frame etc.) are not set
+	outfile.puts FNTabFormatFile.format_str({ 
+						  "word" => word,
+						  "sent_id" => sentid
+						})
+      }
+      outfile.puts
+    end
+    outfile.close()
+  end
+
+  ###########
+  # 
+  # class method split_dir: 
+  # read all files in one directory and produce chunk files *#{suffix} in outdir
+  # with a certain number of files in them (sent_num).
+  # Optionally, remove all sentences longer than sent_leng
+  #
+  # produces output files 1.<suffix>, 2.<suffix>, etc.
+  #
+  # assumes TabFormat sentences
+  #
+  # example: split_all("/tmp/in","/tmp/out",".tab",2000,80)
+
+  def FrprepHelper.split_dir(indir,
+			     outdir,
+			     suffix,
+			     sent_num,
+			     sent_leng=nil)
+
+    unless indir[-1,1] == "/"
+      indir += "/"
+    end
+    unless outdir[-1,1] == "/"
+      outdir += "/"
+    end
+
+    outfile_counter = 0
+    line_stack = Array.new
+    sent_stack = Array.new
+    
+    Dir[indir+"*#{suffix}"].each {|infilename|
+      STDERR.puts "Now splitting #{infilename}"
+      infile = File.new(infilename)
+      
+      while line = infile.gets
+        line.chomp!        
+        case line
+        when "" # end of sentence
+          if !(sent_leng.nil? or line_stack.length < sent_leng) # record sentence
+            # suppress multiple empty lines
+            # to avoid problems with lemmatiser
+            # only record sent_stack if it is not empty.
+
+            # change (sp 15 01 07): just cut off sentence at sent_leng.            
+            
+            STDERR.puts "Cutting off long sentence #{line_stack.last.split("\t").last}" 
+            line_stack = line_stack[0..sent_leng-1]
+          end          
+          unless line_stack.empty?
+            sent_stack << line_stack
+            # reset line_stack
+            line_stack = Array.new
+          end
+	
+
+          # check if we have to empty the sent stack
+          if sent_stack.length == sent_num # enough sentences for new outfile? 
+            outfile = File.new(outdir+outfile_counter.to_s+"#{suffix}","w")
+            sent_stack.each {|l_stack|
+              outfile.puts l_stack.join("\n")              
+              outfile.puts              
+            }
+            outfile.close
+            outfile_counter += 1
+            sent_stack = Array.new
+          end
+        
+        else # for any other line
+	  line_stack << line
+	end
+      end
+      infile.close
+    }
+    # the last remaining sentences
+    unless sent_stack.empty?
+      outfile = File.new(outdir+outfile_counter.to_s+"#{suffix}","w")
+      sent_stack.each {|l_stack|
+        l_stack << "\n"
+        outfile.puts l_stack.join("\n")
+      }
+      outfile.close
+    end
+  end  
+
+  ####
+  # note salsa targetlemma
+  #
+  # old_dir contains xml files whose name starts with the
+  # target lemma for all frames in the file
+  # record that target lemma in the <target> element of each frame
+  def FrprepHelper.note_salsa_targetlemma(old_dir, # string ending in /
+					  new_dir) # string ending in /
+
+    
+    # each input file: extract target lemma from filename,
+    # not this lemma in the <target> element of each frame
+    Dir[old_dir + "*.xml"].each { |filename|
+      changedfilename = new_dir + File.basename(filename)
+
+      if File.basename(filename) =~ /^(.*?)[_\.]/
+	lemma = $1
+	  
+	infile = FilePartsParser.new(filename)
+	outfile = File.new(changedfilename, "w")
+
+	# write header
+	outfile.puts infile.head()
+
+	# iterate through sentences, yield as SalsaTigerSentence objects
+	infile.scan_s() { |sent_string|
+	  sent = SalsaTigerSentence.new(sent_string)
+	  sent.each_frame { |frame|
+	    frame.target.set_attribute("lemma", lemma)
+	  }
+
+	  # write changed sentence
+	  outfile.puts sent.get()
+	} # each sentence
+	
+	# write footer
+	outfile.puts infile.tail()
+	infile.close()
+	outfile.close()
+      
+      else
+	# couldn't determine lemma
+	# just copy the file
+	`cp #{filename} #{changedfilename}`
+      end
+    }
+  end
+
+  ####
+  # stxml_split_dir
+  #
+  # split SalsaTigerXML files into new files of given length,
+  # skipping sentences that are too long
+  #
+  # At the same time, sentences that occur several times (i.e. sentences which are 
+  # annotated by SALSA for more than one predicate) are compacted into one occurrence
+  # with combined semantics.
+  #
+  # assumes that all files in input_dir with 
+  # extension .xml are SalsaTigerXMl files
+  def FrprepHelper.stxml_split_dir(input_dir, # string: input directory with STXML files
+				   split_dir, # string: output directory
+				   max_sentnum, # integer: max num of sentences per file
+				   max_sentlen) # integer: max num of terminals per sentence
+    
+    filenames = Dir[input_dir+"*.xml"].to_a
+    
+    graph_hash = Hash.new # for each sentence id, keep <s...</graph>
+    frame_hash = Hash.new # for each sentence id , keep the <frame...  </frame> string
+    uspfes_hash = Hash.new # for each sentence id, keep the uspfes stuff
+    uspframes_hash = Hash.new # for each sentence id, keep the uspframes stuff
+
+    ########################
+    # Traverse of file(s): compute an index of all frames for each sentence, with unique identifiers 
+    
+    filenames.each {|filename|
+      
+      infile = FilePartsParser.new(filename)      
+      infile.scan_s {|sent_str|
+        
+        sentlen = 0
+        sent_str.delete("\n").scan(/<t\s/) { |occ| sentlen += 1}
+        if sentlen > max_sentlen
+          sent = RegXML.new(sent_str)
+          # revisit handling of long sentences
+          # $stderr.puts "I would have skipped overly long sentence " + sent.attributes["id"]+" but Sebastian forbade me.".to_s          
+          # next
+        end
+
+        # substitute old frame identifiers with new, unique ones
+        
+        # problem: we may have several frames per sentence, and need to keep track of them
+        # if we rename etc sxx_f1 to sxx_f2 and there is already a sxx_f2, then
+        # we cannot distinguish between these frames 
+        
+        # therefore, we substitute temporary identifiers until we have substituted 
+        # all ids with temporary ones, and re-substitute final ones at the end.
+        
+        this_frames = Array.new
+        
+        temp_subs = Array.new
+        final_subs = Array.new
+        
+        sent = RegXML.new(sent_str)
+        sentid = sent.attributes["id"].to_s
+        if sentid.nil?
+          STDERR.puts "[frprep] Warning: cannot find sentence id, skipping sentence:"
+          STDERR.puts sent_str
+          # strange sentence, no ID? skip
+          next
+        end
+        
+        unless frame_hash.key? sentid
+          frame_hash[sentid] = Array.new      
+          uspfes_hash[sentid] = Array.new
+          uspframes_hash[sentid] = Array.new
+        end
+        
+        # find everything up to and including the graph
+        sent_children = sent.children_and_text()
+        graph = sent_children.detect { |child| child.name == "graph" }
+        graph_hash[sentid] = "<s " + 
+                               sent.attributes.to_a.map { |at, val| "#{at}=\'#{val}\'" }.join(" ") +
+                               ">" + 
+                               graph.to_s
+        
+        # find the usp block
+        
+        sem = sent_children.detect { |child| child.name == "sem"}
+        usp = ""
+        if sem
+          usp = sem.children_and_text.detect { |child| child.name == "usp" }
+          usp = usp.to_s
+        end
+        
+        # find all frames
+        if sem
+          frames = sem.children_and_text.detect { |child| child.name == "frames" }
+          if frames
+            frames.children_and_text.each { |frame|
+              unless frame.name == "frame"
+                next
+              end
+              frameid = frame.attributes["id"]
+              
+              temp_frameid = "#{sentid}_temp_f#{frame_hash[sentid].length+this_frames.length+1}"
+              final_frameid = "#{sentid}_f#{frame_hash[sentid].length+this_frames.length+1}"
+              
+              temp_subs << [frameid,temp_frameid]
+              final_subs << [temp_frameid,final_frameid]
+              
+              this_frames << frame.to_s
+            }
+          end
+        end
+        
+        # now first rename all the frames to temporary names
+        
+        temp_subs.each {|orig_frameid, temp_frameid|
+          this_frames.map! {|frame_str|
+	#print "orig ", orig_frameid, " temp ", temp_frameid, "\n"
+            frame_str.gsub(orig_frameid,temp_frameid)
+          }
+      
+          usp.gsub!(orig_frameid,temp_frameid)
+        }
+        
+        # and re-rename the temporary names
+        
+        final_subs.each {|temp_frameid, final_frameid|
+          this_frames.map! {|frame_str|
+            frame_str.gsub(temp_frameid,final_frameid)
+          }
+          usp.gsub!(temp_frameid, final_frameid)
+        }
+        
+        # store frames in data structure
+        this_frames.each {|frame_str|
+          frame_hash[sentid] << frame_str
+        }
+        
+        # store uspfes in data structure
+        unless usp.empty?
+          usp_elt = RegXML.new(usp)
+          uspfes = usp_elt.children_and_text.detect { |child| child.name == "uspfes" }
+          uspfes.children_and_text.each { |child|
+            unless child.name == "uspblock"
+              next
+            end
+            uspfes_hash[sentid] << child.to_s
+          }
+        
+          # store uspframes in data structure
+          uspframes = usp_elt.children_and_text.detect { |child| child.name == "uspframes" }
+          uspframes.children_and_text.each { |child|
+            unless child.name == "uspblock"
+              next
+            end
+            uspframes_hash[sentid] << child.to_s
+          }
+        end
+      }
+    }
+
+    # now write everything in the data structure back to a file
+    
+    filecounter = 0
+    sentcounter = 0
+    outfile = nil
+    sent_stack = Array.new
+        
+    graph_hash.sort {|a,b| a[0].to_i <=> b[0].to_i}.each {|sentid,graph_str|
+      
+      if sentcounter == max_sentnum
+        outfile.puts SalsaTigerXMLHelper.get_footer
+        outfile.close
+        outfile = nil
+      end
+      
+      unless outfile
+        outfile = File.new(split_dir+filecounter.to_s+".xml","w")
+        outfile.puts SalsaTigerXMLHelper.get_header    
+        filecounter +=1
+        sentcounter = 0
+      end
+            
+      xml = Array.new
+      xml << graph_str
+      xml << "<sem>"
+      xml << "<globals>"
+      xml << "</globals>"
+      xml << "<frames>"
+      frame_hash[sentid].each {|frame_str|
+        xml << frame_str
+      }
+      xml << "</frames>"
+      xml << "<usp>"
+      xml << "<uspframes>"
+      uspframes_hash[sentid].each {|uspblock_str|
+        xml << uspblock_str
+      }
+      xml << "</uspframes>"
+      xml << "<uspfes>"
+      uspfes_hash[sentid].each {|uspblock_str|
+        xml << uspblock_str
+      }
+      xml << "</uspfes>"
+      xml << "</usp>"
+      xml << "</sem>"
+      xml << "</s>"
+      
+      outfile.puts xml.join("\n")
+      sentcounter += 1
+    }
+    
+    if outfile 
+      outfile.puts SalsaTigerXMLHelper.get_footer
+      outfile.close
+      outfile = nil
+    end   
+    
+  end
+
+
+  ####
+  # transform SalsaTigerXML file to Tab format file
+  def FrprepHelper.stxml_to_tab_file(input_filename,   # string: name of input file
+                                     output_filename,  # string: name of output file
+                                     exp)              # FrprepConfigData
+    infile = FilePartsParser.new(input_filename)
+    begin
+      outfile = File.new(output_filename,"w")
+    rescue
+      raise "Stxml to tab: could not write to tab file #{output_filename}"
+    end
+
+    infile.scan_s {|sent_string|        
+
+      # determine sentence ID
+      sentid = RegXML.new(sent_string).attributes["id"]
+      unless sentid
+        $stderr.puts "No sentence ID in sentence:\n "+ sent_string
+        $stderr.puts "Making a new one up."
+        sentid = Time.new().to_f.to_s
+      end
+
+      # find terminals and process them
+      unless sent_string.delete("\n") =~ /<terminals[ >].+<\/terminals>/
+        $stderr.puts "Warning: could not find terminals in sentence:"
+        $stderr.puts sent_string
+        $stderr.puts "Skipping"
+        next
+      end
+
+        # modified by ines, 27/08/08
+        # for Berkeley => convert ( ) to -LRB- -RRB-
+       
+	text = $&
+        if exp.get("parser") == "berkeley"
+          text.gsub!(/word='\('/, "word='*LRB*'")
+          text.gsub!(/word='\)'/, "word='*RRB*'")
+ 	  text.gsub!(/word=['"]``['"]/, "word='\"'")
+	  text.gsub!(/word=['"]''['"]/,  "word='\"'")
+	  text.gsub!(/word=['"]\&apos;\&apos;['"]/,  "word='\"'")
+	  #text.gsub!(/word=['"]\(['"]/,  "word='-LRB-'")
+	  #text.gsub!(/word=['"]\)['"]/,  "word='-RRB-'")
+
+        end
+      terminals = text
+      #terminals = sent_string
+      terminals = RegXML.new(terminals)
+      terminals.children_and_text.each { |terminal|
+ 
+        unless terminal.name == "t"
+          # not a terminal after all
+          next
+        end
+  
+
+        outfile.puts FNTabFormatFile.format_str({
+						  "word" => SalsaTigerXMLHelper.unescape(terminal.attributes["word"]),
+						  "sent_id" => sentid
+						})
+      } # each terminal
+      outfile.puts
+    } # each sentence
+    outfile.close
+  end
+
+  ###
+  # add semantics from tab:
+  #
+  # add information about semantics from a FN tab sentence
+  # to a SalsaTigerSentence object:
+  # - frames (one frame per sentence)
+  # - roles
+  # - FrameNet grammatical functions
+  # - FrameNet POS of target
+  def FrprepHelper.add_semantics_from_tab(st_sent,  # SalsaTigerSentence object
+					  tab_sent, # FNTabFormatSentence object
+                                          mapping,  # hash: tab lineno -> array:SynNode
+                                          interpreter_class, # SynInterpreter class
+                                          exp)      # FrprepConfigData
+
+    if tab_sent.nil?
+      # tab sentence not found
+      return
+    end
+
+    # iterate through frames in the tabsent
+    frame_index = 0
+    tab_sent.each_frame { |tab_frame_obj|
+      frame_name = tab_frame_obj.get_frame() # string
+        
+      if frame_name.nil? or frame_name =~ /^-*$/
+	# weird: a frame without a frame
+	$stderr.puts "Warning: frame entry without a frame in tab sentence #{st_sent.id}."
+	$stderr.puts "Skipping"
+	next
+      end
+        
+      frame_node = st_sent.add_frame(frame_name, tab_sent.get_sent_id() + "_f#{frame_index}")
+      frame_index += 1
+    
+      # target
+      target_nodes = Array.new
+      tab_frame_obj.get_target_indices.each {|terminal_id| 
+	if mapping[terminal_id]
+	  target_nodes.concat mapping[terminal_id]
+	end
+      }
+
+      # let the interpreter class decide on how to determine the maximum constituents
+      target_maxnodes = interpreter_class.max_constituents(target_nodes, st_sent)
+      if target_maxnodes.empty?
+	# HIEr
+	STDERR.puts  "Warning: no target in frame entry, sentence #{st_sent.id}."
+	$stderr.puts "frame is #{frame_name}, frame no #{frame_index}"
+	$stderr.puts "Skipping."
+	$stderr.puts "target indices: " + tab_frame_obj.get_target_indices.join(", ")
+        #tab_sent.each_line { |line|
+        #  $stderr.puts line
+	#  $stderr.puts "--"
+        #}
+	next
+      end
+      frame_node.add_fe("target",target_maxnodes)
+        
+      # set features on target: target lemma, target POS
+      target_lemma = tab_frame_obj.get_target()
+      target_pos = nil
+      if target_lemma
+	if exp.get("origin") == "FrameNet"
+	  # FrameNet data: here the lemma in the tab file has the form
+	  # <lemma>.<POS>
+	  # separate the two
+	  if target_lemma =~ /^(.*)\.(.*)$/
+	    target_lemma = $1
+	    target_pos = $2
+	  end
+	end
+	frame_node.target.set_attribute("lemma", target_lemma)
+        if target_pos
+          frame_node.target.set_attribute("pos", target_pos)
+        end
+      end
+        
+      # roles, GF, PT
+      # synnode_markable_label: 
+      #   hash "role" | "gf" | "pt" -> SynNode -> array: label(string)
+      layer_synnode_label = Hash.new
+      ["gf", "pt", "role"].each {|layer|
+	termids2labels = tab_frame_obj.markables(layer)
+
+	unless layer_synnode_label[layer]
+	  layer_synnode_label[layer] = Hash.new
+	end
+
+	termids2labels.each {|terminal_indices, label|
+	  terminal_indices.each { |t_i|
+
+	    if (nodes = mapping[t_i])
+
+	      nodes.each { |node|
+		unless layer_synnode_label[layer][node]
+		  layer_synnode_label[layer][node] = Array.new
+		end
+
+		layer_synnode_label[layer][node] << label
+	      } # each node that t_i maps to
+	    end # if t_i maps to anything
+
+	  } # each terminal index
+	} # each mapping terminal indices -> label
+      } # each layer
+
+      # 'stuff' (Support and other things)
+      layer_synnode_label["stuff"] = Hash.new
+      tab_frame_obj.each_line_parsed { |line_obj|
+        if (label = line_obj.get("stuff")) != "-"
+          if (nodes = mapping[line_obj.get("lineno")])
+            nodes.each { |node|
+              unless layer_synnode_label["stuff"][node]
+                layer_synnode_label["stuff"][node] = Array.new
+              end
+              layer_synnode_label["stuff"][node] << label 
+            }
+          end
+        end
+      }
+
+      # reencode:
+      #  hash role_label(string) -> array of tuples [synnodes, gflabels, ptlabels]
+      #   synnodes: array:SynNode.  gflabels, ptlabels: array:String
+      #
+      # note that in this step, any gf or pt labels that have been
+      # assigned to a SynNode that has not also been assigned a role
+      # will be lost
+      role2nodes_labels = Hash.new
+      layer_synnode_label["role"].each_pair { |synnode, labels|
+	labels.each { | rolelabel|
+	  unless role2nodes_labels[rolelabel]
+	    role2nodes_labels[rolelabel] = Array.new
+	  end
+	  
+	  role2nodes_labels[rolelabel] << [
+	    synnode, 
+	    layer_synnode_label["gf"][synnode], 
+	    layer_synnode_label["pt"][synnode]
+	  ]
+	} # each role label
+      } # each pair SynNode/role labels
+
+      # reencode "stuff", but only the support cases
+      role2nodes_labels["Support"] = Array.new()
+
+      layer_synnode_label["stuff"].each_pair { |synnode, labels|
+        labels.each { |stufflabel|
+          if stufflabel =~ /Supp/
+            # some sort of support
+            role2nodes_labels["Support"] << [synnode, nil, nil]
+          end
+        }
+      }
+
+      ##
+      # each role label: 
+      # make FeNode for the current frame
+      role2nodes_labels.each_pair { |rolelabel, node_gf_pt|
+	
+	# get list of syn nodes, GF and PT labels for this role
+	# shortcut for GF and PT labels: take any labels that have
+	# been assigned for _some_ Synnode of this role
+	synnodes = node_gf_pt.map { |ngp| ngp[0] } 
+	gflabels = node_gf_pt.map { |ngp| ngp[1] }.compact.flatten.uniq
+	ptlabels = node_gf_pt.map { |ngp| ngp[2] }.compact.flatten.uniq
+
+
+	# let the interpreter class decide on how to 
+	# determine the maximum constituents
+	maxnodes = interpreter_class.max_constituents(synnodes, st_sent)
+
+	fe_node = st_sent.add_fe(frame_node, rolelabel, maxnodes)
+	unless gflabels.empty?
+	  fe_node.set_attribute("gf", gflabels.join(","))
+	end
+	unless ptlabels.empty?
+	  fe_node.set_attribute("pt", ptlabels.join(","))
+	end
+      } # each role label
+    } # each frame
+  end
+
+
+  ######
+  # handle multiword targets:
+  # if you find a verb with a separate prefix, 
+  # change the verb's lemma information accordingly
+  # and add an attribute "other_words" to the verb node
+  # pointing to the other node
+  #
+  # In general, it will be assumed that "other_words" contains
+  # a list of node IDs for other nodes belonging to the same 
+  # group, node IDs separated by spaces, and that 
+  # each node of a group has the "other_words" attribute.
+  #
+  def FrprepHelper.handle_multiword_targets(sent,  # SalsaTigerSentence object
+					    interpreter, # SynInterpreter object
+					    language) # string: en, de
+    ##
+    # only retain the interesting words of the sentence:
+    # content words and prepositions
+    if sent.nil?
+	return
+    end
+
+    nodes = sent.terminals.select { |node|
+      [  
+        "adj", "adv", "card", "noun", "part", "prep", "verb"
+      ].include? interpreter.category(node)
+    }
+
+    ##
+    # group:
+    # group verbs with their separate particles
+    # (at a later point, other types of grouping can be inserted here)
+    groups = FrprepHelper.group_words(nodes, interpreter)
+
+    ##
+    # record grouping information as attributes on the terminals.
+    groups.each { |descr, group_of_nodes|
+      case descr
+      when "none"
+        # no grouping
+      when "part"
+        # separate particle belonging to a verb
+        
+        # group_of_nodes is a pair [verb, particle]
+        verb, particle = group_of_nodes
+
+        verb.set_attribute("other_words", particle.id())
+        particle.set_attribute("other_words", verb.id())
+        
+        if verb.get_attribute("lemma") and particle.get_attribute("lemma")
+          case language
+          when "de"
+            # German: prepend SVP to get the real lemma of the verb
+            verb.set_attribute("lemma", 
+                               particle.get_attribute("lemma") + 
+			       verb.get_attribute("lemma"))
+          when "en"
+            # English: append particle as separate word after the lemma of the verb
+            verb.set_attribute("lemma", 
+                               verb.get_attribute("lemma") + " " + 
+			       particle.get_attribute("lemma"))
+          else
+	    # default
+	    verb.set_attribute("lemma",
+			       verb.get_attribute("lemma") + " " + 
+			       particle.get_attribute("lemma"))
+          end
+        end
+
+      else
+        raise "Shouldn't be here: unexpected description #{descr}"
+      end
+    }
+  end
+
+  ########################
+  # group_words
+  #
+  # auxiliary of transform_multiword targets
+  #
+  # Group terminals: 
+  # At the moment, just find separate prefixes and particles
+  # for verbs
+  #
+  # returns: list of pairs [descr, nodes]
+  # descr: string, "none" (no group), "part" (separate verb particle)
+  # nodes: array:SynNode
+  def FrprepHelper.group_words(nodes,    # array: SynNode
+			       interpreter) # SynInterpreter object
+
+    retv = Array.new # array of groups, array:array:SynNode
+    done = Array.new # remember nodes already covered
+
+    nodes.each { |terminal_node|
+      if done.include? terminal_node
+        # we have already included this node in one of the groups
+        next
+      end
+
+      if (svp = interpreter.particle_of_verb(terminal_node, nodes))
+        retv << ["part", [terminal_node, svp]]
+        done << terminal_node
+        done << svp
+      else
+        retv << ["none", [terminal_node]]
+        done << terminal_node
+      end
+
+    }
+
+    return retv
+  end
+
+
+  ######
+  # handle unknown framenames
+  #
+  # For all frames with names matching Unknown\d+,
+  # rename them to <lemma>_Unknown\d+
+  def FrprepHelper.handle_unknown_framenames(sent,     # SalsaTigerSentence
+					     interpreter) # SynInterpreter class
+    if sent.nil?
+	return
+    end
+
+    sent.each_frame { |frame|
+      if frame.name() =~ /^Unknown/
+        if frame.target
+          maintarget = interpreter.main_node_of_expr(frame.target.children(), "no_mwe")
+        else
+          maintarget = nil
+        end
+        unless maintarget
+          $stderr.puts "Warning: Unknown frame, and I could not determine the target lemma: Frame #{frame.id()}"
+          $stderr.puts "Cannot repair frame name, leaving it as is."
+          return
+        end
+          
+        # get lemma, if it exists, otherwise get word
+        # also, if the lemmatizer has returned a disjunction of lemmas,
+        # get the first disjunct
+        lemma = interpreter.lemma_backoff(maintarget)
+        if lemma
+          # we have a lemma
+          frame.set_name(lemma + "_" + frame.name())
+        else
+          # the main target word has no lemma attribute,
+          # and somehow I couldn't even get the target word
+          $stderr.puts "Warning: Salsa 'Unknown' frame."
+          $stderr.puts "Trying to make its lemma-specificity explicit, but"
+          $stderr.puts "I could not determine the target lemma nor the target word: frame #{frame.id()}"
+          $stderr.puts "Leaving 'Unknown' as it is."
+        end
+      end
+    } 
+  end
+
+
+  #####################
+  #
+  # Integrate the semantic annotation of an old sentence
+  # into the corresponding new sentence
+  # At the same time, integrate the lemma information from the
+  # old sentence into the new sentence
+  def FrprepHelper.integrate_stxml_semantics_and_lemmas(oldsent, 
+                                                        newsent,
+                                                        interpreter_class,
+                                                        exp)
+    if oldsent.nil? or newsent.nil?
+	return
+    end
+    ##
+    # match old and new sentence via terminals
+    newterminals = newsent.terminals_sorted()
+    oldterminals = oldsent.terminals_sorted()
+    # sanity check: exact match on terminals?
+    newterminals.interleave(oldterminals).each { |newnode, oldnode|
+	#print "old ", oldnode.word, "  ", newnode.word, "\n"
+      # new and old word: use both unescaped and escaped variant
+      if newnode
+        newwords = [ newnode.word, SalsaTigerXMLHelper.escape(newnode.word) ]
+      else
+        newwords = [nil, nil]
+      end
+      if oldnode
+        oldwords = [ oldnode.word, SalsaTigerXMLHelper.escape(oldnode.word) ]
+      else
+        oldwords = [ nil, nil]
+      end
+
+      if (newwords & oldwords).empty?
+        # old and new word don't match, either escaped or non-escaped
+
+        $stderr.puts "Warning: could not match terminals of sentence #{newsent.id()}"
+        $stderr.puts "This means that I cannot match the semantic annotation"
+        $stderr.puts "to the newly parsed sentence. Skipping."
+        #$stderr.puts "Old sentence: "
+        #$stderr.puts oldterminals.map { |n| n.word }.join("--")
+        #$stderr.puts "New sentence: "
+        #$stderr.puts newterminals.map { |n| n.word }.join("--")
+        return false
+      end
+    }
+
+    ##
+    # copy lemma information
+    oldterminals.each_with_index { |oldnode, ix|
+      newnode = newterminals[ix]
+      if oldnode.get_attribute("lemma")
+        newnode.set_attribute("lemma", oldnode.get_attribute("lemma"))
+      end
+    }
+
+    ##
+    # copy frames
+    oldsent.each_frame { |oldframe|
+      # make new frame with same ID
+      newframe = newsent.add_frame(oldframe.name, oldframe.id())
+      # copy FEs
+      oldframe.each_child { |oldfe|
+        # new nodes: map old terminals to new terminals,
+        # then find max constituents covering them
+        newnodes = oldfe.descendants.select { |n| 
+          n.is_terminal? 
+        }.map { |n|
+          oldterminals.index(n)
+        }.map { |ix|
+          newterminals[ix]
+        }
+
+        # let the interpreter class decide on how to determine the maximum constituents
+        newnodes = interpreter_class.max_constituents(newnodes, newsent)
+
+        # make new FE with same ID
+        new_fe = newsent.add_fe(newframe, oldfe.name(), newnodes, oldfe.id())
+	# keep all attributes of the FE
+	if oldfe.get_f("attributes")
+	  oldfe.get_f("attributes").each_pair { |attr, value|
+	    new_fe.set_attribute(attr, value)
+	  }
+	end
+      }
+    }
+
+    ##
+    ### changed by ines => appears twice in stxml file
+
+    # copy underspecification
+    # keep as is, since we've kept all frame and FE IDs
+    oldsent.each_usp_frameblock { |olduspframe|
+      newuspframe = newsent.add_usp("frame")
+      olduspframe.each_child { |oldnode|
+        newnode = newsent.sem_node_with_id(oldnode.id())
+        if newnode
+          newuspframe.add_child(newnode)
+        else
+          $stderr.puts "Error: unknown frame with ID #{oldnode.id()}"
+        end
+      }
+    }
+    oldsent.each_usp_feblock { |olduspfe|
+      newuspfe = newsent.add_usp("fe")
+      olduspfe.each_child { |oldnode|
+        newnode = newsent.sem_node_with_id(oldnode.id())
+        if newnode
+          newuspfe.add_child(newnode)
+        else
+          $stderr.puts "Error: unknown FE with ID #{oldnode.id()}"
+        end
+      }
+    }
+
+  end
+
+  ####################
+  # add head attributes to each nonterminal in each 
+  # SalsaTigerXML file in a directory
+
+  def FrprepHelper.add_head_attributes(st_sent,      # SalsaTigerSentence object
+				       interpreter)  # SynInterpreter class
+    st_sent.each_nonterminal {|nt_node|
+     head_term = interpreter.head_terminal(nt_node)
+      if head_term and head_term.word()
+	nt_node.set_attribute("head", head_term.word())
+      else
+	nt_node.set_attribute("head", "--")
+      end
+    } # each nonterminal
+  end
+
+  # add lemma information to each terminal in a given SalsaTigerSentence object
+  def FrprepHelper.add_lemmas_from_tab(st_sent, # SalsaTigerSentence object
+				       tab_sent,# FNTabFormatSentence object
+                                       mapping) # hash: tab lineno -> array:SynNode
+    if tab_sent.nil?
+      # tab sentence not found
+      return
+    end
+
+    # produce list with word, lemma pairs
+    lemmat = Array.new        
+    tab_sent.each_line_parsed {|line|
+      word = line.get("word")
+      lemma = line.get("lemma")
+      lemmat << [word,lemma]
+    }
+    
+    # match with st_sent terminal list and add lemma attributes
+    # KE Jan 07: if word mismatch,
+    # set to Lemmatizer file version,
+    # but count mismatches
+    word_mismatches = Array.new()
+
+    st_sent.each_terminal_sorted {|t|
+      matching_lineno = (0..lemmat.length()-1).to_a.detect { |tab_lineno|
+        mapping[tab_lineno].include? t
+      }
+      unless matching_lineno
+        next
+      end
+      word, lemma = lemmat[matching_lineno]
+          
+      # transform characters to XML-friendly form
+      # for comparison with st_word, which is also escaped
+      word = SalsaTigerXMLHelper.escape(word)
+      st_word = t.word()
+      if word != st_word and
+          word != SalsaTigerXMLHelper.escape(st_word)
+        # true mismatch.
+        # use the Lemmatizer version of the word, remember the mismatch
+        word_mismatches << [st_word, word]
+        t.set_attribute("word", word)
+      end
+      
+      if lemma
+        # we actually do have lemma information
+        lemmatised_head = SalsaTigerXMLHelper.escape(lemma)
+        t.set_attribute("lemma",lemmatised_head)
+      end
+    } # each terminal
+
+    # did we have mismatches? then report them
+    unless word_mismatches.empty?
+      $stderr.puts "Warning: Word mismatches found between Lemmatizer file and SalsaTigerXML file generalted from parser output."
+      $stderr.puts "(May be due to failed reencoding of special character in the parser output.)"
+      $stderr.puts "I am using the Lemmatizer version by default."
+      $stderr.puts "Version used:"
+      $stderr.print "\t"
+      st_sent.each_terminal_sorted { |t| $stderr.print ">>#{t}<<" }
+      $stderr.puts
+      $stderr.print "SalsaTigerXML file had: "
+      $stderr.print word_mismatches.map { |st_word, tab_word|
+        "#{st_word} instead of #{tab_word}"
+      }.join(", ")
+      $stderr.puts
+    end
+  end
+
+  ###################3
+  # given a SalsaTigerSentence,
+  # look for FrameNet frames that are 
+  # test frames, and remove them
+  def FrprepHelper.remove_deprecated_frames(sent,  # SalsaTigerSentence
+                                            exp)   # FrprepConfigData
+
+    unless exp.get("origin") == "FrameNet"
+      return
+    end
+
+    sent.frames.each { |frame_obj|
+      if frame_obj.name() == "Boulder" or
+          frame_obj.name() =~ /^Test/
+        sent.remove_frame(frame_obj)
+      end
+    }
+  end
+
+end
+
+############################################3
+# Class FrprepFlatSyntax:
+#
+# given a FNTabFormat file,
+# yield each of its sentences in SalsaTigerXML,
+# constructing a flat syntax
+class FrprepFlatSyntax
+  def initialize(tabfilename, # string: name of tab file
+                 postag_suffix, # postag file suffix (or nil)  
+                 lemma_suffix)  # lemmatisation file suffix (or nil)
+    
+    @tabfilename = tabfilename
+    @pos_suffix = postag_suffix
+    @lemma_suffix = lemma_suffix
+  end
+
+  # yield each non-parse sentence as a tuple
+  # [ salsa/tiger xml sentence, tab format sentence, mapping]
+  # of a SalsaTigerSentence object, a FNTabSentence object,
+  # and a hash: FNTab sentence lineno(integer) -> array:SynNode
+  # pointing each tab word to one or more SalsaTigerSentence terminals
+  def each_sentence(dummy)
+
+    # read tab file with lemma and POS info
+    tabfile = FNTabFormatFile.new(@tabfilename, @pos_suffix, @lemma_suffix)
+
+    tabfile.each_sentence() { |tabsent|
+      # start new, empty sentence with "failed" attribute (i.e. no parse)
+      # and with the ID of the corresponding TabFormat sentence
+      sentid = tabsent.get_sent_id()
+      if sentid.nil? or sentid =~ /^-*$/
+        $stderr.puts "No sentence ID for sentence:"
+        tabsent.each_line_parsed { |l| $stderr.print l.get("word"), " "}
+        $stderr.puts
+        sentid = Time.new().to_f.to_s
+      end
+      sent = SalsaTigerSentence.new("<s id=\"#{SalsaTigerXMLHelper.escape(sentid)}\" failed=\"true\"></s>")
+      
+      # add single nonterminal node, category "S"
+      single_nonterminal_id = SalsaTigerXMLHelper.escape(sentid.to_s + "_NT")
+      vroot = sent.add_syn("nt", "S", # category
+                           nil,  # word
+                           nil,  # pos
+                           single_nonterminal_id)
+      
+      # add terminals
+      tabsent.each_line_parsed() { |line_obj|
+        # make terminal node with tab sent info
+        node_id = sentid.to_s + "_" + line_obj.get("lineno").to_s
+        word = line_obj.get("word")
+        unless word
+          word = ""
+        end
+        word = SalsaTigerXMLHelper.escape(word)
+        pos = line_obj.get("pos")
+        unless pos
+          pos = ""
+        end
+        pos = SalsaTigerXMLHelper.escape(pos)
+        terminal = sent.add_syn("t", nil, # category
+                                word, pos, 
+                                node_id)
+
+        if line_obj.get("lemma")
+          # lemma
+          terminal.set_attribute("lemma", SalsaTigerXMLHelper.escape(line_obj.get("lemma")))
+        end
+
+        # add new terminal as child of vroot
+        vroot.add_child(terminal, nil)
+        terminal.add_parent(vroot, nil)
+      } # each line of tab file
+
+      # yield newly constructed SalsaTigerXMl sentence plus tab sentence
+      yield [sent, tabsent, SynInterfaceSTXML.standard_mapping(sent, tabsent)]
+    }
+  end
+end
+
+############################################3
+# Class FrprepReadStxml
+#
+# given a STXML file,
+# yield each of its sentences
+class FrprepReadStxml
+  def initialize(stxmlfilename, # string: name of SalsaTigerXML file
+                 tabfilename,   # string: name of corresponding tab file (or nil)
+                 postag_suffix,    #  POS tag file suffix (or nil)
+                 lemma_suffix)     #  lemmatization file suffix (or nil)
+
+    @stxmlfilename = stxmlfilename
+    @tabfilename = tabfilename
+    @pos_suffix = postag_suffix
+    @lemma_suffix = lemma_suffix
+  end
+  # yield each non-parse sentence as a tuple
+  # [ salsa/tiger xml sentence, tab format sentence, mapping]
+  # of a SalsaTigerSentence object, a FNTabSentence object,
+  # and a hash: FNTab sentence lineno(integer) -> array:SynNode
+  # pointing each tab word to one or more SalsaTigerSentence terminals
+  def each_sentence(dummy)
+    # read corresponding tab file?
+    tab_sents = Array.new()
+    if File.exists? @tabfilename
+      tabfile = FNTabFormatFile.new(@tabfilename,@pos_suffix,@lemma_suffix)    
+      tabfile.each_sentence { |tabsent|
+        tab_sents << tabsent
+      }      
+    end
+
+    # read STXML file
+    infile = FilePartsParser.new(@stxmlfilename)
+    index = 0
+    infile.scan_s { |sent_string|
+      sent = SalsaTigerSentence.new(sent_string)
+      yield [sent, tab_sents.at(index), SynInterfaceSTXML.standard_mapping(sent, tab_sents.at(index))]
+      index += 1
+    }
+  end
+end