diff options
Diffstat (limited to 'java/src/com/android/inputmethod/latin/Suggest.java')
| -rw-r--r-- | java/src/com/android/inputmethod/latin/Suggest.java | 115 |
1 files changed, 92 insertions, 23 deletions
diff --git a/java/src/com/android/inputmethod/latin/Suggest.java b/java/src/com/android/inputmethod/latin/Suggest.java index e181237a6..9b4619d35 100644 --- a/java/src/com/android/inputmethod/latin/Suggest.java +++ b/java/src/com/android/inputmethod/latin/Suggest.java @@ -20,14 +20,16 @@ import android.text.TextUtils; import com.android.inputmethod.keyboard.ProximityInfo; import com.android.inputmethod.latin.SuggestedWords.SuggestedWordInfo; +import com.android.inputmethod.latin.common.Constants; +import com.android.inputmethod.latin.common.StringUtils; import com.android.inputmethod.latin.define.DebugFlags; import com.android.inputmethod.latin.settings.SettingsValuesForSuggestion; import com.android.inputmethod.latin.utils.AutoCorrectionUtils; import com.android.inputmethod.latin.utils.BinaryDictionaryUtils; -import com.android.inputmethod.latin.utils.StringUtils; import com.android.inputmethod.latin.utils.SuggestionResults; import java.util.ArrayList; +import java.util.HashMap; import java.util.Locale; /** @@ -49,6 +51,16 @@ public final class Suggest { private static final boolean DBG = DebugFlags.DEBUG_ENABLED; private final DictionaryFacilitator mDictionaryFacilitator; + private static final int MAXIMUM_AUTO_CORRECT_LENGTH_FOR_GERMAN = 12; + private static final HashMap<String, Integer> sLanguageToMaximumAutoCorrectionWithSpaceLength = + new HashMap<>(); + static { + // TODO: should we add Finnish here? + // TODO: This should not be hardcoded here but be written in the dictionary header + sLanguageToMaximumAutoCorrectionWithSpaceLength.put(Locale.GERMAN.getLanguage(), + MAXIMUM_AUTO_CORRECT_LENGTH_FOR_GERMAN); + } + private float mAutoCorrectionThreshold; public Suggest(final DictionaryFacilitator dictionaryFacilitator) { @@ -128,8 +140,8 @@ public final class Suggest { : typedWord; final SuggestionResults suggestionResults = mDictionaryFacilitator.getSuggestionResults( - wordComposer, ngramContext, proximityInfo, settingsValuesForSuggestion, - SESSION_ID_TYPING); + wordComposer, ngramContext, proximityInfo.getNativeProximityInfo(), + settingsValuesForSuggestion, SESSION_ID_TYPING); final ArrayList<SuggestedWordInfo> suggestionsContainer = getTransformedSuggestedWordInfoList(wordComposer, suggestionResults, trailingSingleQuotesCount, @@ -148,28 +160,50 @@ public final class Suggest { || (consideredWord.length() > 1 && !didRemoveTypedWord); final boolean hasAutoCorrection; - // TODO: using isCorrectionEnabled here is not very good. It's probably useless, because - // any attempt to do auto-correction is already shielded with a test for this flag; at the - // same time, it feels wrong that the SuggestedWord object includes information about - // the current settings. It may also be useful to know, when the setting is off, whether - // the word *would* have been auto-corrected. - if (!isCorrectionEnabled || !allowsToBeAutoCorrected || resultsArePredictions - || suggestionResults.isEmpty() || wordComposer.hasDigits() - || wordComposer.isMostlyCaps() || wordComposer.isResumed() + // If correction is not enabled, we never auto-correct. This is for example for when + // the setting "Auto-correction" is "off": we still suggest, but we don't auto-correct. + if (!isCorrectionEnabled + // If the word does not allow to be auto-corrected, then we don't auto-correct. + || !allowsToBeAutoCorrected + // If we are doing prediction, then we never auto-correct of course + || resultsArePredictions + // If we don't have suggestion results, we can't evaluate the first suggestion + // for auto-correction + || suggestionResults.isEmpty() + // If the word has digits, we never auto-correct because it's likely the word + // was type with a lot of care + || wordComposer.hasDigits() + // If the word is mostly caps, we never auto-correct because this is almost + // certainly intentional (and careful input) + || wordComposer.isMostlyCaps() + // We never auto-correct when suggestions are resumed because it would be unexpected + || wordComposer.isResumed() + // If we don't have a main dictionary, we never want to auto-correct. The reason + // for this is, the user may have a contact whose name happens to match a valid + // word in their language, and it will unexpectedly auto-correct. For example, if + // the user types in English with no dictionary and has a "Will" in their contact + // list, "will" would always auto-correct to "Will" which is unwanted. Hence, no + // main dict => no auto-correct. Also, it would probably get obnoxious quickly. + // TODO: now that we have personalization, we may want to re-evaluate this decision || !mDictionaryFacilitator.hasAtLeastOneInitializedMainDictionary() + // If the first suggestion is a shortcut we never auto-correct to it, regardless + // of how strong it is (whitelist entries are not KIND_SHORTCUT but KIND_WHITELIST). + // TODO: we may want to have shortcut-only entries auto-correct in the future. || suggestionResults.first().isKindOf(SuggestedWordInfo.KIND_SHORTCUT)) { - // If we don't have a main dictionary, we never want to auto-correct. The reason for - // this is, the user may have a contact whose name happens to match a valid word in - // their language, and it will unexpectedly auto-correct. For example, if the user - // types in English with no dictionary and has a "Will" in their contact list, "will" - // would always auto-correct to "Will" which is unwanted. Hence, no main dict => no - // auto-correct. - // Also, shortcuts should never auto-correct unless they are whitelist entries. - // TODO: we may want to have shortcut-only entries auto-correct in the future. hasAutoCorrection = false; } else { - hasAutoCorrection = AutoCorrectionUtils.suggestionExceedsAutoCorrectionThreshold( - suggestionResults.first(), consideredWord, mAutoCorrectionThreshold); + final SuggestedWordInfo firstSuggestion = suggestionResults.first(); + if (!AutoCorrectionUtils.suggestionExceedsAutoCorrectionThreshold( + firstSuggestion, consideredWord, mAutoCorrectionThreshold)) { + // Score is too low for autocorrect + hasAutoCorrection = false; + } else { + // We have a high score, so we need to check if this suggestion is in the correct + // form to allow auto-correcting to it in this language. For details of how this + // is determined, see #isAllowedByAutoCorrectionWithSpaceFilter. + // TODO: this should not have its own logic here but be handled by the dictionary. + hasAutoCorrection = isAllowedByAutoCorrectionWithSpaceFilter(firstSuggestion); + } } if (!TextUtils.isEmpty(typedWord)) { @@ -213,8 +247,8 @@ public final class Suggest { final int inputStyle, final int sequenceNumber, final OnGetSuggestedWordsCallback callback) { final SuggestionResults suggestionResults = mDictionaryFacilitator.getSuggestionResults( - wordComposer, ngramContext, proximityInfo, settingsValuesForSuggestion, - SESSION_ID_GESTURE); + wordComposer, ngramContext, proximityInfo.getNativeProximityInfo(), + settingsValuesForSuggestion, SESSION_ID_GESTURE); // For transforming words that don't come from a dictionary, because it's our best bet final Locale defaultLocale = mDictionaryFacilitator.getMostProbableLocale(); final ArrayList<SuggestedWordInfo> suggestionsContainer = @@ -287,6 +321,41 @@ public final class Suggest { return suggestionsList; } + /** + * Computes whether this suggestion should be blocked or not in this language + * + * This function implements a filter that avoids auto-correcting to suggestions that contain + * spaces that are above a certain language-dependent character limit. In languages like German + * where it's possible to concatenate many words, it often happens our dictionary does not + * have the longer words. In this case, we offer a lot of unhelpful suggestions that contain + * one or several spaces. Ideally we should understand what the user wants and display useful + * suggestions by improving the dictionary and possibly having some specific logic. Until + * that's possible we should avoid displaying unhelpful suggestions. But it's hard to tell + * whether a suggestion is useful or not. So at least for the time being we block + * auto-correction when the suggestion is long and contains a space, which should avoid the + * worst damage. + * This function is implementing that filter. If the language enforces no such limit, then it + * always returns true. If the suggestion contains no space, it also returns true. Otherwise, + * it checks the length against the language-specific limit. + * + * @param info the suggestion info + * @return whether it's fine to auto-correct to this. + */ + private static boolean isAllowedByAutoCorrectionWithSpaceFilter(final SuggestedWordInfo info) { + final Locale locale = info.mSourceDict.mLocale; + if (null == locale) { + return true; + } + final Integer maximumLengthForThisLanguage = + sLanguageToMaximumAutoCorrectionWithSpaceLength.get(locale.getLanguage()); + if (null == maximumLengthForThisLanguage) { + // This language does not enforce a maximum length to auto-correction + return true; + } + return info.mWord.length() <= maximumLengthForThisLanguage + || -1 == info.mWord.indexOf(Constants.CODE_SPACE); + } + /* package for test */ static SuggestedWordInfo getTransformedSuggestedWordInfo( final SuggestedWordInfo wordInfo, final Locale locale, final boolean isAllUpperCase, final boolean isOnlyFirstCharCapitalized, final int trailingSingleQuotesCount) { |