DeSR Dependency Parser

src/SentenceReader.cpp

/*
**  DeSR
**  src/SentenceReader.cpp
**  ----------------------------------------------------------------------
**  Copyright (c) 2005  Giuseppe Attardi (attardi@di.unipi.it).
**  ----------------------------------------------------------------------
**
**  This file is part of DeSR.
**
**  DeSR is free software; you can redistribute it and/or modify it
**  under the terms of the GNU General Public License, version 3,
**  as published by the Free Software Foundation.
**
**  DeSR is distributed in the hope that it will be useful,
**  but WITHOUT ANY WARRANTY; without even the implied warranty of
**  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
**  GNU General Public License for more details.
**
**  You should have received a copy of the GNU General Public License
**  along with this program.  If not, see <http://www.gnu.org/licenses/>.
**  ----------------------------------------------------------------------
*/

#include "SentenceReader.h"

// standard
#include <iostream>
#include <iomanip>

// IXE library
#include "text/Utf8Utils.h"

// local
#include "Corpus.h"

using namespace std;
using namespace Tanl::Text;

#define CONLL08
#ifdef CONLL08
#include "conf/conf_bool.h"
IXE::conf<bool>         useGold("UseGoldPos", false);
#endif

namespace Tanl {

//======================================================================
// SentenceReader

SentenceReader::SentenceReader(istream* is, Corpus* corpus) :
  is(is),
  corpus(corpus)
{ }

// Generic tab format reader
static RegExp::Pattern reTab("([^\t\n]+)");

bool SentenceReader::MoveNext()
{
  string line;
  if (!getline(*is, line) || line.empty())
    return false;

  sentence = new Sentence(&corpus->language);
  vector<char const*> const& names = corpus->index.names;
  vector<int> preds;            // ids of predicates in sentence

  int id = 1;
  RegExp::MatchGroups match(2);

  do {
    Attributes attributes(&corpus->index);
    string form;
    int head = 0;
    string deprel;
    int fields = corpus->tokenFields.size();
    TokenLinks links;
    unordered_map<string, int> linkMap;
    int argNo = 0;
    int i = 0;
    char const* cur = line.c_str();
    char const* end = cur + line.size();
    while (reTab.match(cur, end, match) > 0) {
      TokenField const& tf = corpus->tokenFields[i];
      char const* fieldStart = cur + match[1].first;
      int fieldLen = match[1].second - match[1].first;
      string field(fieldStart, fieldLen);
      // clear empty fields
      if (field == tf.default_)
        field = "";
      // discard IGNORE fields
      if (tf.use != TokenField::ignore) {
        if (!tf.link.empty()) {
          // Got target of link: create even when value is missing
          int head = field.empty() ? -1 : atoi(field.c_str());
          if (linkMap.find(tf.link) == linkMap.end()) {
            // Create link and assign position in LinkMap 
            linkMap[tf.link] = links.size();
            links.push_back(TokenLink(head));
          } else
            // fill head in previously created link
            links[linkMap[tf.link]].head = head;
        } else if (!tf.label.empty()) {
          if (field.empty())
            --argNo;
          else {
            // Got label of link
            if (linkMap.find(tf.label) == linkMap.end()) {
              // Create link with dummy target and assign position in LinkMap 
              linkMap[tf.label] = links.size();
              // use negative value to distinguish from real target
              links.push_back(TokenLink(--argNo, field.c_str()));
            } else
              // fill label in previously created link
              links[linkMap[tf.label]].label = field;
          }
        } else {
          attributes[i] = field;
#         ifdef CONLL08
          if (useGold && i == 7 && !attributes[3].empty())
            attributes[7] = attributes[3];
#         endif
          switch (tf.role) {
          case TokenField::form:
            form = field; break;
          case TokenField::predicate:
            if (!field.empty())
              preds.push_back(id);
            break;
          }
        }
      }
      i++;
      cur += match[0].second;
      if (i == fields || cur == end) {
        // skip extra fields
        break;
      }
    }
    TreeToken* token = new TreeToken(id++, form, attributes, links);
    sentence->push_back(token);
  } while (getline(*is, line) && !line.empty());
  // fix ARG links
  FOR_EACH (Sentence, *sentence, sit) {
    TO_EACH (TokenLinks, (*sit)->token->links, tit) {
      if (tit->head < 0)
        tit->head = preds[-tit->head - 1];
    }
  }
  return true;
}

Sentence* SentenceReader::Current()
{
  return sentence;
}

//======================================================================
// ConllXSentenceReader

// pattern for analyzing token line in CoNLL format:
// czech 2007 has extra tab at end of lines.
static RegExp::Pattern reCoNLL("(\\d+)\t([^\t]+)\t([^\t]+)\t([^\t]+)\t([^\t]+)\t([^\t\n]+)(?:\t([^\t]+)\t([^\t]+)(?:\t([^\t]+)\t([^\t]+)\t?)?)?");

ConllXSentenceReader::ConllXSentenceReader(istream* is, Corpus* corpus) :
  SentenceReader(is, corpus),
  morphExtractor(*corpus->language.morphExtractor)
{  }

bool ConllXSentenceReader::MoveNext()
{
  string line;
  if (!getline(*is, line) || line.empty())
    return false;

  sentence = new Sentence(&corpus->language);
  MorphExtractor::Features mf;
  vector<char const*>& names = corpus->index.names;

  RegExp::MatchGroups match(11);
  do {
    int matches = reCoNLL.match(line, match);
    if (matches > 0) {
      char const* start = line.c_str();
      morphExtractor(start + match[6].first, start + match[6].second, mf);
      Attributes attributes(&corpus->index);
      int id = atoi(start + match[1].first);
      attributes.insert(names[0], string(start + match[1].first, match[1].second - match[1].first)); // ID
      string form(start + match[2].first, match[2].second - match[2].first);
      attributes.insert(names[1], form); // FORM
      int head = 0;
      string deprel;
      for (int i = 3; i < matches; i++) {
        // LEMMA, CPOS, POS, FEATS, HEAD, DEPREL, PHEAD, PDEPREL
        char const* tagStart = start + match[i].first;
        int tagLen = match[i].second - match[i].first;
        string attr;
        // discard CoNLL empty fields
        if (tagLen != 1 || tagStart[0] != '_')
          attr = string(tagStart, tagLen);
        attributes.insert(names[i-1], attr);
        if (i == 7)
          head = atoi(tagStart);
        else if (i == 8)
          deprel = attr;
      }
      // sanity check: avoid circularities
      if (id == head)
        head = 0;
      TokenLinks links(1, TokenLink(head, deprel.c_str()));
      TreeToken* token = new TreeToken(id, form, attributes, links);
      // add morpho features
      token->token->morpho.set(mf);
      sentence->push_back(token);
    }
  } while (getline(*is, line) && !line.empty());

  return true;
}

//======================================================================
// DgaSentenceReader

// The stream consists of sentences:
//
// <s>token+</s>
//
// Tokens have the following format:
//
// <tok id="...">
//  <orth>...</orth>
//  <lemma>...</lemma>
//  <pos>...</pos>
//  <gen>...</gen>
//  <num>...</num>
//  <per>...</per>
//  <dep head="..." type="..." />
// </tok>

DgaSentenceReader::DgaSentenceReader(istream* is, Corpus* corpus) :
  SentenceReader(is, corpus),
  reader(*is)
{ }

bool DgaSentenceReader::MoveNext()
{
  // check for open tag <s>
  if (!reader.Read() || reader.NodeType != Tanl::XML::Element ||
      reader.Name != "s")
    return false;

  sentence = new Sentence(&corpus->language);
  vector<char const*>& names = corpus->index.names;
  while (reader.Read()) {
    if (reader.NodeType == Tanl::XML::Whitespace)
      continue;
    if (reader.NodeType == Tanl::XML::EndElement)
      break;
    if (reader.NodeType == Tanl::XML::Element &&
         reader.Name == "tok") {
      Attributes attributes(&corpus->index);
      int id;
      string form;
      int head = 0;
      string deprel;
      while (reader.MoveToNextAttribute()) {
        if (reader.Name == "id")
          id = atoi(reader.Value.c_str());
      }
      // read token contents
      while (reader.Read()) {
        if (reader.NodeType == Tanl::XML::Whitespace)
          continue;
        if (reader.NodeType == Tanl::XML::EndElement)
          break;
        if (reader.NodeType != Tanl::XML::Element)
          goto fail;
        string& name = reader.Name;
        if (name == "dep") {
          // read attributes
          while (reader.MoveToNextAttribute()) {
            string& name = reader.Name;
            if (name == "head")
              head = atoi(reader.Value.c_str());
            else if (name == "type")
              deprel = reader.Value;
          }
        } else if (name == "orth") {
          if (!reader.Read() || reader.NodeType != Tanl::XML::Text)
            goto fail;
          form = reader.Value;
          if (!reader.Read() || reader.NodeType != Tanl::XML::EndElement ||
              reader.Name != "orth")
            goto fail;
        } else {
          string tag = reader.Name;
          if (!reader.Read() || reader.NodeType != Tanl::XML::Text)
            goto fail;
          attributes.insert(tag.c_str(), reader.Value);
          if (!reader.Read() || reader.NodeType != Tanl::XML::EndElement ||
              reader.Name != tag)
            goto fail;
        }
      }
      // ckeck closing tag
      if (reader.Name != "tok")
        goto fail;
      TokenLinks links(1, TokenLink(head, deprel.c_str()));
      TreeToken* tok = new TreeToken(id, form, attributes, links);
      sentence->push_back(tok);
    } else
      goto fail;
  }
  // check closing tag
  if (reader.Name == "s")
    return true;
 fail:
  delete sentence;
  sentence = 0;
  return false;
}

//======================================================================
// TokenSentenceReader

TokenSentenceReader::TokenSentenceReader(istream* is, Corpus* corpus) :
  SentenceReader(is, corpus)
{
# ifdef STEMMER
  if (corpus && corpus->language)
    stemmer = sb_stemmer_new(corpus->language, 0); // UTF-8 encoding
# endif
}

RegExp::Pattern TokenSentenceReader::reTok("\\s*([^\\s]*?)");

bool TokenSentenceReader::MoveNext()
{
  string line;
  if (!getline(*is, line) || line.empty())
    return false;

  sentence = new Sentence(&corpus->language);
  int id = 1;
  RegExp::MatchGroups matches(2);
  do {
    char const* cur = line.c_str();
    char const* endSent = cur + line.size();;
    while (reTok.match(cur, endSent, matches) > 0) {
      const char* tokStart = cur + matches[1].first;
      int tokSize = matches[1].second - matches[1].first;
      cur += matches[0].second;
      string form(tokStart, tokSize);
      TreeToken* tok = new TreeToken(id++, form.c_str(), &corpus->index);
      sentence->push_back(tok);
    }
  } while (getline(*is, line) && line.size());
  return true;
}

//======================================================================

bool TaggedSentenceReader::MoveNext()
{
  if (!reader->MoveNext())
    return false;
  sentence = reader->Current();
  if (tagger) {
    // perform POS tagging
    Parser::Tagged tagSentence;
    FOR_EACH (Sentence, *sentence, sit) {
      TreeToken* token = *sit;
      tagSentence.words.push_back(token->token->form.c_str());
      string const* pos = token->get("POS");
      tagSentence.tags.push_back(pos ? pos->c_str() : 0);
    }
    if (tagger->tag(tagSentence)) {
      for (unsigned i = 0; i < sentence->size(); i++) {
        (*sentence)[i]->set("POS", tagSentence.tags[i]);
        TreeToken* node = (*sentence)[i];
        string const* lemma = node->get("LEMMA");
        if (lemma && *lemma == "") {
          char const* lemma = tagSentence.lemmas[i];
          if (strcmp(lemma, "<unknown>"))
            node->set("LEMMA", lemma);
          else
            node->set("LEMMA", node->token->form);
        }
      }
    }
  }
  return true;
}

} // namespace Tanl

 

Copyright © 2005-2007 G. Attardi. Generated on 13 Aug 2009 by doxygen 1.5.7.1.