Package org.languagetool.rules

Class BERTSuggestionRanking

java.lang.Object

org.languagetool.rules.Rule

org.languagetool.rules.RemoteRule

org.languagetool.rules.BERTSuggestionRanking

public class BERTSuggestionRanking
extends RemoteRule

reorder suggestions from another rule using BERT as a LM

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
private static class	BERTSuggestionRanking.CuratedAndSameCaseComparator
(package private) static class	BERTSuggestionRanking.MatchesForReordering

Nested classes/interfaces inherited from class org.languagetool.rules.RemoteRule

RemoteRule.RemoteRequest

Field Summary

Fields

Modifier and Type	Field	Description
private static final org.slf4j.Logger	logger
private static final double	MAX_ERROR_RATE
private static final int	MIN_WORDS
private final RemoteLanguageModel	model
private static final com.google.common.cache.LoadingCache<RemoteRuleConfig,RemoteLanguageModel>	models
static final String	RULE_ID
protected int	suggestionLimit
private final Rule	wrappedRule

Fields inherited from class org.languagetool.rules.RemoteRule

circuitBreakers, filterMatches, fixOffsets, inputLogging, lt, premium, ruleLanguage, serviceConfiguration, shutdownRoutines, suppressMisspelledMatch, suppressMisspelledSuggestions, whitespaceNormalisation

Fields inherited from class org.languagetool.rules.Rule

messages

Constructor Summary

Constructors

Constructor	Description
BERTSuggestionRanking(Language language, Rule rule, RemoteRuleConfig config, boolean inputLogging)

Method Summary

Modifier and Type	Method	Description
private RemoteLanguageModel.Request	buildRequest(RuleMatch match)
protected Callable<RemoteRuleResult>	executeRequest(RemoteRule.RemoteRequest request, long timeoutMilliseconds)
protected RemoteRuleResult	fallbackResults(RemoteRule.RemoteRequest request)	fallback if executeRequest times out or throws an error
String	getDescription()	A short description of the error this rule can detect, usually in the language of the text that is checked.
String	getId()	A string used to identify the rule in e.g.
protected RemoteRule.RemoteRequest	prepareRequest(List<AnalyzedSentence> sentences, Long textSessionId)	run local preprocessing steps (or just store sentences)
protected List<SuggestedReplacement>	prepareSuggestions(List<SuggestedReplacement> suggestions)	transform suggestions before resorting, e.g.

Methods inherited from class org.languagetool.rules.RemoteRule

circuitBreaker, computeOffsetShifts, createCircuitBreaker, fixMatchOffsets, getCircuitBreakerConfig, getServiceConfiguration, getTimeout, getTimeout, isPremium, match, run, run, shutdown

Methods inherited from class org.languagetool.rules.Rule

addExamplePair, addTags, addToneTags, cacheAntiPatterns, estimateContextForSureMatch, getAntiPatterns, getCategory, getCorrectExamples, getDistanceTokens, getErrorTriggeringExamples, getFullId, getIncorrectExamples, getLocQualityIssueType, getMinPrevMatches, getPriority, getRuleOptions, getSentenceWithImmunization, getSourceFile, getSubId, getTags, getToneTags, getUrl, hasTag, hasToneTag, isDefaultOff, isDefaultTempOff, isDictionaryBasedSpellingRule, isGoalSpecific, isIncludedInHiddenMatches, isOfficeDefaultOff, isOfficeDefaultOn, makeAntiPatterns, setCategory, setCorrectExamples, setDefaultOff, setDefaultOn, setDefaultTempOff, setDistanceTokens, setErrorTriggeringExamples, setExamplePair, setGoalSpecific, setIncludedInHiddenMatches, setIncorrectExamples, setLocQualityIssueType, setMinPrevMatches, setOfficeDefaultOff, setOfficeDefaultOn, setPremium, setPriority, setTags, setToneTags, setUrl, supportsLanguage, toRuleMatchArray, useInOffice

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

RULE_ID

public static final String RULE_ID

See Also:

• Constant Field Values

logger

private static final org.slf4j.Logger logger

models

private static final com.google.common.cache.LoadingCache<RemoteRuleConfig,RemoteLanguageModel> models

suggestionLimit

protected int suggestionLimit

model

private final RemoteLanguageModel model

wrappedRule

private final Rule wrappedRule

MIN_WORDS

private static final int MIN_WORDS

See Also:

• Constant Field Values

MAX_ERROR_RATE

private static final double MAX_ERROR_RATE

See Also:

• Constant Field Values

Constructor Details

BERTSuggestionRanking

public BERTSuggestionRanking(Language language,
 Rule rule,
 RemoteRuleConfig config,
 boolean inputLogging)

Method Details

prepareSuggestions

protected List<SuggestedReplacement> prepareSuggestions(List<SuggestedReplacement> suggestions)

transform suggestions before resorting, e.g. limit resorting to top-n candidates

Returns:

transformed suggestions

prepareRequest

protected RemoteRule.RemoteRequest prepareRequest(List<AnalyzedSentence> sentences,
 Long textSessionId)

Description copied from class: RemoteRule

run local preprocessing steps (or just store sentences)

Specified by:

prepareRequest in class RemoteRule

Parameters:

sentences - text to process

textSessionId - session ID for caching, partial rollout, A/B testing

Returns:

parameter for executeRequest/fallbackResults

fallbackResults

protected RemoteRuleResult fallbackResults(RemoteRule.RemoteRequest request)

Description copied from class: RemoteRule

fallback if executeRequest times out or throws an error

Specified by:

fallbackResults in class RemoteRule

Parameters:

request - returned by prepareRequest

Returns:

local results for this rule

executeRequest

protected Callable<RemoteRuleResult> executeRequest(RemoteRule.RemoteRequest request,
 long timeoutMilliseconds)
                                             throws TimeoutException

Specified by:

executeRequest in class RemoteRule

Parameters:

request - returned by prepareRequest

timeoutMilliseconds - timeout for this operation, <=0 -> unlimited

Returns:

callable that sends request, parses and returns result for this remote rule

Throws:

TimeoutException - if timeout was exceeded

buildRequest

@Nullable
private RemoteLanguageModel.Request buildRequest(RuleMatch match)

getId

public String getId()

Description copied from class: Rule

A string used to identify the rule in e.g. configuration files. This string is supposed to be unique and to stay the same in all upcoming versions of LanguageTool. It's supposed to contain only the characters A-Z and the underscore.

Overrides:

getId in class RemoteRule

getDescription

public String getDescription()

Description copied from class: Rule

A short description of the error this rule can detect, usually in the language of the text that is checked.

Specified by:

getDescription in class Rule