From dd0777c6689508be2a386f747fb0ebeb4992a3ff Mon Sep 17 00:00:00 2001
From: Adam Wardynski <award@.(B-4.4.46a)>
Date: Thu, 13 Jan 2011 19:34:49 +0100
Subject: [PATCH] Unify to treat underspecified lexemes as meeting agreement.
So such lexemes are always left in the end.
---
libwccl/ops/actions/unify.cpp | 57 +++++++++++++++++++++++------------
1 file changed, 38 insertions(+), 19 deletions(-)
diff --git a/libwccl/ops/actions/unify.cpp b/libwccl/ops/actions/unify.cpp
index a4a53f0..692aa5d 100644
--- a/libwccl/ops/actions/unify.cpp
+++ b/libwccl/ops/actions/unify.cpp
@@ -25,33 +25,35 @@ Bool Unify::execute(const ActionExecContext& context) const
const boost::shared_ptr<const TSet>& attribs = attribs_expr_->apply(context);
std::vector<std::vector<Corpus2::Lexeme> > remainings;
+ std::vector<std::vector<Corpus2::Lexeme> > underspecifieds;
// first pass - cut lexemes that don't match agreement attribs at all
for (int i = abs_left; i <= abs_right; ++i) {
Corpus2::Token& curr_tok = *sc.at(i);
std::vector<Corpus2::Lexeme> remaining;
+ std::vector<Corpus2::Lexeme> underspecified;
foreach(Corpus2::Lexeme& lexeme, curr_tok.lexemes()) {
if (attribs->matching_categories(lexeme.tag()) > 0) {
remaining.push_back(lexeme);
+ } else {
+ underspecified.push_back(lexeme);
}
}
remainings.push_back(remaining);
+ underspecifieds.push_back(underspecified);
}
-
// second pass - leave only lexemes that are in agreement
// with at least one lexeme from previous non-empty-after-1st-pass token
int previous = remainings.size() - 1;
// at first, "previous" is the last token not emptied after 1st pass
- while ((previous > 0) && (remainings[previous].size() == 0)) {
+ while ((previous > 0) && (remainings[previous].empty())) {
previous--;
}
for (size_t i = 0; i < remainings.size(); ++i) {
// proceed only for tokens that have lexemes that could go into agreement
// (i.e. are not emptied after 1st pass)
- if (remainings[i].size() > 0) {
- std::vector<Corpus2::Lexeme>& prev_rem = remainings[previous];
- // next time around "previous" will be the current one
- previous = i;
+ if (!remainings[i].empty()) {
+ const std::vector<Corpus2::Lexeme>& prev_rem = remainings[previous];
std::vector<Corpus2::Lexeme> curr_rem;
// get only lexemes that agree with a lexeme from previous token
foreach(const Corpus2::Lexeme& curr_lex, remainings[i]) {
@@ -63,28 +65,45 @@ Bool Unify::execute(const ActionExecContext& context) const
}
}
}
- if (curr_rem.size() == 0) {
- // no lexemes left, means no agreement possible, so can't unify - abort with no changes
- changed.set_value(false);
- return changed;
- }
- if (curr_rem.size() != remainings[i].size()) {
- changed.set_value(true);
- remainings[i] = curr_rem;
+ if (curr_rem.empty()) {
+ if (underspecifieds[i].empty()) {
+ // there is no agreement possible, abort with no changes
+ changed.set_value(false);
+ return changed;
+ } else {
+ // there were underspecified lexemes, they will be left in the end
+ // so agreement is met anyway, however we need to mark that
+ // changes were made because we removed some lexemes that did not
+ // meet agreement;
+ // "previous" index has to be left alone though, pointing at previous
+ // token that still had some agreedable lexemes
+ changed.set_value(true);
+ }
+ } else {
+ // some non-underspecified lexemes are left, so
+ // next time around, "previous" token is going to be the current one
+ previous = i;
+ // there were changes only if we actually removed something
+ if (curr_rem.size() != remainings[i].size()) {
+ changed.set_value(true);
+ remainings[i] = curr_rem;
+ }
}
}
}
-
- // finally assign remaining tokens accordingly,
- // leaving intact the tokens that did not have agreeable lexemes at all
+ // finally assign remaining lexemes to original tokens
if (changed.get_value()) {
for (size_t i = 0; i < remainings.size(); ++i) {
if (remainings[i].size() > 0) {
- sc.at(i + abs_left)->lexemes() = remainings[i];
+ std::vector<Corpus2::Lexeme>& lexemes = sc.at(i + abs_left)->lexemes();
+ lexemes = remainings[i];
+ // underspecced lexemes meet agreement too, so leave them in the end, if any
+ foreach (const Corpus2::Lexeme& lex, underspecifieds[i]) {
+ lexemes.push_back(lex);
+ }
}
}
}
-
return changed;
}
--
GitLab