/**
* Cleans up any resources that were used by the aligner.
*
* <p>Call this method when the aligner is no longer to be used.
*/
public void cleanUp() {
// Close the lexical resources
for (LexicalResource<? extends RuleInfo> lexicalResource : lexicalResources) {
try {
lexicalResource.close();
} catch (LexicalResourceCloseException e) {
logger.warn("Closing the resource failed.", e);
}
}
}
/**
* Get rules of type leftSide -> rightSide, using the given lexical resource
*
* @param resource The lexical resource to use
* @param leftSide The phrase that will be looked for as lhs of a rule
* @param rightSide The phrase that will be looked for as rhs of a rule
* @param partOfSpeech2
* @param partOfSpeech
* @return The list of rules leftSide -> rightSide
* @throws LexicalResourceException
*/
private List<LexicalRule<? extends RuleInfo>> getRules(
LexicalResource<? extends RuleInfo> resource,
String leftSide,
String rightSide,
PartOfSpeech leftSidePOS,
PartOfSpeech rightSidePOS)
throws LexicalResourceException {
List<LexicalRule<? extends RuleInfo>> rules = new ArrayList<LexicalRule<? extends RuleInfo>>();
try {
// WordNet workaround:
// Make sure the synsets of the right and left sides of the rule
// are equal to the right and left phrases.
// (WN returns rules associated with any of the words in the phrase)
if (resource.getClass().getName().toLowerCase().contains(WORDNET)) {
for (LexicalRule<? extends RuleInfo> rule :
resource.getRules(leftSide, leftSidePOS, rightSide, rightSidePOS)) {
WordnetRuleInfo ruleInfo = (WordnetRuleInfo) rule.getInfo();
if ((ruleInfo.getLeftSense().getWords().contains(leftSide))
&& (ruleInfo.getRightSense().getWords().contains(rightSide))) {
addRuleToList(rules, rule);
}
}
} else {
// Get rules from t to h
for (LexicalRule<? extends RuleInfo> rule :
resource.getRules(leftSide, leftSidePOS, rightSide, rightSidePOS)) {
addRuleToList(rules, rule);
}
}
} catch (Exception e) {
logger.warn(
"Could not add rules from "
+ resource.getClass().getSimpleName()
+ " for "
+ leftSide
+ "->"
+ rightSide,
e);
}
return rules;
}
/**
* Call this method once before starting to align sentence pairs.
*
* @param config a CommonConfig instance. The aligner retrieves the lexical resources
* configuration values.
* @throws LexicalResourceException if initialization of a resource failed
* @throws ConfigurationException if the configuration is invalid
*/
private void init(CommonConfig config) throws LexicalResourceException, ConfigurationException {
// Get the general parameters configuration section
NameValueTable paramsSection = null;
try {
paramsSection = config.getSection(GENERAL_PARAMS_CONF_SECTION);
} catch (ConfigurationException e) {
throw new ConfigurationException(e);
}
maxPhrase = paramsSection.getInteger(MAX_PHRASE_KEY);
// Get the Lexical Resources configuration section
NameValueTable lexicalResourcesSection = null;
try {
lexicalResourcesSection = config.getSection(LEXICAL_RESOURCES_CONF_SECTION);
} catch (ConfigurationException e) {
throw new ConfigurationException(e);
}
lexicalResources = new ArrayList<LexicalResource<? extends RuleInfo>>();
ConfigurationFile configFile = new ConfigurationFile(config);
// Get each resource and create it using the configuration section related to it
for (String resourceName : lexicalResourcesSection.keySet()) {
// Get the class name
String resourceClassName = lexicalResourcesSection.getString(resourceName);
// Get the configuration params
ConfigurationParams resourceParams = configFile.getModuleConfiguration(resourceName);
resourceParams.setExpandingEnvironmentVariables(true);
LexicalResource<? extends RuleInfo> lexicalResource =
createLexicalResource(resourceClassName, resourceParams);
if (lexicalResource != null) {
lexicalResources.add(lexicalResource);
PartOfSpeech leftSidePOS = null, rightSidePOS = null;
// Add the information about this resource
// Get the right and left side POS, in case it's mentioned
if (resourceParams.keySet().contains(LEFT_SIDE_POS_PARAM)) {
try {
leftSidePOS =
new ByCanonicalPartOfSpeech(resourceParams.getString(LEFT_SIDE_POS_PARAM));
} catch (UnsupportedPosTagStringException e) {
logger.warn(
"Could not load POS for left side: "
+ resourceParams.getString(LEFT_SIDE_POS_PARAM)
+ ". Alignment links of all POS will be retreived.");
}
}
if (resourceParams.keySet().contains(RIGHT_SIDE_POS_PARAM)) {
try {
rightSidePOS =
new ByCanonicalPartOfSpeech(resourceParams.getString(RIGHT_SIDE_POS_PARAM));
} catch (UnsupportedPosTagStringException e) {
logger.warn(
"Could not load POS for right side: "
+ resourceParams.getString(RIGHT_SIDE_POS_PARAM)
+ ". Alignment links of all POS will be retreived.");
}
}
lexicalResourcesInformation.put(
lexicalResource.getClass().getName(),
new LexicalResourceInformation(
resourceParams.getString(VERSION_PARAM),
resourceParams.getBoolean(USE_LEMMA_PARAM),
leftSidePOS,
rightSidePOS));
}
}
}
/**
* Align the text and the hypothesis.
*
* <p>This method receives a JCAS object containing two views: Hypothesis and text views. The
* method assumes that the views were already annotated with a tokenizer.
*
* <p>The lexical aligner looks at every phrase t in the text and every phrase h in the
* hypothesis, and uses the lexical resources to find rules with lhs = t and rhs = h.
*
* @param aJCas the JCAS object with the text and hypothesis view.
* @throws AlignmentComponentException
*/
@Override
public void annotate(JCas aJCas) throws AlignmentComponentException {
try {
logger.info("Started annotating a text and hypothesis pair using lexical aligner");
// Get the tokens and lemmas of the text and hypothesis
getTokenAnnotations(aJCas);
// Check in all the resources for rules of type textPhrase -> hypoPhrase
for (LexicalResource<? extends RuleInfo> resource : lexicalResources) {
LexicalResourceInformation resourceInfo =
lexicalResourcesInformation.get(resource.getClass().getName());
// For every phrase t in T and phrase h in H, check the lexical
// resources if they contain a rule t->h
String textPhrase = "", hypoPhrase = "";
for (int textStart = 0; textStart < textTokens.size(); ++textStart) {
for (int textEnd = textStart;
textEnd < Math.min(textTokens.size(), textStart + maxPhrase);
++textEnd) {
textPhrase = getPhrase(textTokens, textStart, textEnd, resourceInfo.useLemma());
for (int hypoStart = 0; hypoStart < hypoTokens.size(); ++hypoStart) {
for (int hypoEnd = hypoStart;
hypoEnd < Math.min(hypoTokens.size(), hypoStart + maxPhrase);
++hypoEnd) {
hypoPhrase = getPhrase(hypoTokens, hypoStart, hypoEnd, resourceInfo.useLemma());
// Get the rules textPhrase -> hypoPhrase
List<LexicalRule<? extends RuleInfo>> ruleFromLeft =
getRules(
resource,
textPhrase,
hypoPhrase,
resourceInfo.getLeftSidePOS(),
resourceInfo.getRightSidePOS());
// Get the rules hypoPhrase -> textPhrase
List<LexicalRule<? extends RuleInfo>> ruleFromRight =
getRules(
resource,
hypoPhrase,
textPhrase,
resourceInfo.getLeftSidePOS(),
resourceInfo.getRightSidePOS());
// Create the alignment links for the rules
createAlignmentLinks(
aJCas,
textStart,
textEnd,
hypoStart,
hypoEnd,
ruleFromLeft,
ruleFromRight,
resourceInfo.getVersion());
}
}
}
}
}
logger.info("Finished annotating a text and hypothesis pair using lexical aligner");
} catch (CASException | LexicalResourceException e) {
throw new AlignmentComponentException("LexicalAligner failed aligning the sentence pair.", e);
}
}