n/futokb
1
0
Fork 0
mirror of https://gitlab.futo.org/keyboard/latinime.git synced 2024-09-28 14:54:30 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

Separate bigram prediction from suggestion process

Browse Source 
Change-Id: Ibdef206fbd6688a45cebbd7fb8a9f6cd175d03b0
This commit is contained in:
Jean Chalard 2012-03-09 18:01:31 +09:00
parent 6f722c8301
commit de165aed2a
2 changed files with 152 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
java/src/com/android/inputmethod/latin/LatinIME.java
View File

@ -2025,7 +2025,6 @@ public class LatinIME extends InputMethodServiceCompatWrapper implements Keyboar
separatorCode);
}
private static final WordComposer sEmptyWordComposer = new WordComposer();
public void updateBigramPredictions() {
if (mSuggest == null || !isSuggestionsRequested())
return;
@ -2037,8 +2036,8 @@ public class LatinIME extends InputMethodServiceCompatWrapper implements Keyboar
final CharSequence prevWord = EditingUtils.getThisWord(getCurrentInputConnection(),
mSettingsValues.mWordSeparators);
SuggestedWords.Builder builder = mSuggest.getSuggestedWordBuilder(sEmptyWordComposer,
prevWord, mKeyboardSwitcher.getKeyboard().getProximityInfo(), mCorrectionMode);
SuggestedWords.Builder builder = mSuggest.getBigramPredictionWordBuilder(prevWord,
mKeyboardSwitcher.getKeyboard().getProximityInfo(), mCorrectionMode);
if (builder.size() > 0) {
// Explicitly supply an empty typed word (the no-second-arg version of

150
java/src/com/android/inputmethod/latin/Suggest.java
View File

@ -260,6 +260,156 @@ public class Suggest implements Dictionary.WordCallback {
mSuggestions.add(sb);
}
private static final WordComposer sEmptyWordComposer = new WordComposer();
public SuggestedWords.Builder getBigramPredictionWordBuilder(CharSequence prevWordForBigram,
final ProximityInfo proximityInfo, final int correctionMode) {
final WordComposer wordComposer = sEmptyWordComposer;
LatinImeLogger.onStartSuggestion(prevWordForBigram);
mIsFirstCharCapitalized = wordComposer.isFirstCharCapitalized();
mIsAllUpperCase = wordComposer.isAllUpperCase();
mTrailingSingleQuotesCount = wordComposer.trailingSingleQuotesCount();
collectGarbage(mSuggestions, mPrefMaxSuggestions);
Arrays.fill(mScores, 0);
final String typedWord = wordComposer.getTypedWord();
final String consideredWord = mTrailingSingleQuotesCount > 0
? typedWord.substring(0, typedWord.length() - mTrailingSingleQuotesCount)
: typedWord;
// Treating USER_TYPED as UNIGRAM suggestion for logging now.
LatinImeLogger.onAddSuggestedWord(typedWord, Suggest.DIC_USER_TYPED,
Dictionary.UNIGRAM);
mConsideredWord = consideredWord;
// TODO: Change this scheme - a boolean is not enough. A whitelisted word may be "valid"
// but still autocorrected from - in the case the whitelist only capitalizes the word.
// The whitelist should be case-insensitive, so it's not possible to be consistent with
// a boolean flag. Right now this is handled with a slight hack in
// WhitelistDictionary#shouldForciblyAutoCorrectFrom.
final boolean allowsToBeAutoCorrected = AutoCorrection.allowsToBeAutoCorrected(
getUnigramDictionaries(), consideredWord, wordComposer.isFirstCharCapitalized());
if (wordComposer.size() <= 1 && (correctionMode == CORRECTION_FULL_BIGRAM)) {
// At first character typed, search only the bigrams
Arrays.fill(mBigramScores, 0);
collectGarbage(mBigramSuggestions, PREF_MAX_BIGRAMS);
if (!TextUtils.isEmpty(prevWordForBigram)) {
CharSequence lowerPrevWord = prevWordForBigram.toString().toLowerCase();
if (mMainDict != null && mMainDict.isValidWord(lowerPrevWord)) {
prevWordForBigram = lowerPrevWord;
}
for (final Dictionary dictionary : mBigramDictionaries.values()) {
dictionary.getBigrams(wordComposer, prevWordForBigram, this);
}
if (TextUtils.isEmpty(consideredWord)) {
// Nothing entered: return all bigrams for the previous word
int insertCount = Math.min(mBigramSuggestions.size(), mPrefMaxSuggestions);
for (int i = 0; i < insertCount; ++i) {
addBigramToSuggestions(mBigramSuggestions.get(i));
}
} else {
// Word entered: return only bigrams that match the first char of the typed word
final char currentChar = consideredWord.charAt(0);
// TODO: Must pay attention to locale when changing case.
final char currentCharUpper = Character.toUpperCase(currentChar);
int count = 0;
final int bigramSuggestionSize = mBigramSuggestions.size();
for (int i = 0; i < bigramSuggestionSize; i++) {
final CharSequence bigramSuggestion = mBigramSuggestions.get(i);
final char bigramSuggestionFirstChar = bigramSuggestion.charAt(0);
if (bigramSuggestionFirstChar == currentChar
|| bigramSuggestionFirstChar == currentCharUpper) {
addBigramToSuggestions(bigramSuggestion);
if (++count > mPrefMaxSuggestions) break;
}
}
}
}
} else if (wordComposer.size() > 1) {
// At second character typed, search the unigrams (scores being affected by bigrams)
for (final String key : mUnigramDictionaries.keySet()) {
// Skip UserUnigramDictionary and WhitelistDictionary to lookup
if (key.equals(DICT_KEY_USER_UNIGRAM) || key.equals(DICT_KEY_WHITELIST))
continue;
final Dictionary dictionary = mUnigramDictionaries.get(key);
if (mTrailingSingleQuotesCount > 0) {
final WordComposer tmpWordComposer = new WordComposer(wordComposer);
for (int i = mTrailingSingleQuotesCount - 1; i >= 0; --i) {
tmpWordComposer.deleteLast();
}
dictionary.getWords(tmpWordComposer, this, proximityInfo);
} else {
dictionary.getWords(wordComposer, this, proximityInfo);
}
}
}
final String consideredWordString = consideredWord.toString();
CharSequence whitelistedWord = capitalizeWord(mIsAllUpperCase, mIsFirstCharCapitalized,
mWhiteListDictionary.getWhitelistedWord(consideredWordString));
final boolean hasAutoCorrection;
if (CORRECTION_FULL == correctionMode
|| CORRECTION_FULL_BIGRAM == correctionMode) {
final CharSequence autoCorrection =
AutoCorrection.computeAutoCorrectionWord(mUnigramDictionaries, wordComposer,
mSuggestions, mScores, consideredWord, mAutoCorrectionThreshold,
whitelistedWord);
hasAutoCorrection = (null != autoCorrection);
} else {
hasAutoCorrection = false;
}
if (whitelistedWord != null) {
if (mTrailingSingleQuotesCount > 0) {
final StringBuilder sb = new StringBuilder(whitelistedWord);
for (int i = mTrailingSingleQuotesCount - 1; i >= 0; --i) {
sb.appendCodePoint(Keyboard.CODE_SINGLE_QUOTE);
}
mSuggestions.add(0, sb.toString());
} else {
mSuggestions.add(0, whitelistedWord);
}
}
mSuggestions.add(0, typedWord.toString());
StringUtils.removeDupes(mSuggestions);
if (DBG) {
final CharSequence autoCorrectionSuggestion = mSuggestions.get(0);
final int autoCorrectionSuggestionScore = mScores[0];
double normalizedScore = BinaryDictionary.calcNormalizedScore(
typedWord.toString(), autoCorrectionSuggestion.toString(),
autoCorrectionSuggestionScore);
ArrayList<SuggestedWords.SuggestedWordInfo> scoreInfoList =
new ArrayList<SuggestedWords.SuggestedWordInfo>();
scoreInfoList.add(new SuggestedWords.SuggestedWordInfo("+", false));
for (int i = 0; i < mScores.length; ++i) {
if (normalizedScore > 0) {
final String scoreThreshold = String.format("%d (%4.2f)", mScores[i],
normalizedScore);
scoreInfoList.add(
new SuggestedWords.SuggestedWordInfo(scoreThreshold, false));
normalizedScore = 0.0;
} else {
final String score = Integer.toString(mScores[i]);
scoreInfoList.add(new SuggestedWords.SuggestedWordInfo(score, false));
}
}
for (int i = mScores.length; i < mSuggestions.size(); ++i) {
scoreInfoList.add(new SuggestedWords.SuggestedWordInfo("--", false));
}
return new SuggestedWords.Builder().addWords(mSuggestions, scoreInfoList)
.setAllowsToBeAutoCorrected(allowsToBeAutoCorrected)
.setHasAutoCorrection(hasAutoCorrection);
}
return new SuggestedWords.Builder().addWords(mSuggestions, null)
.setAllowsToBeAutoCorrected(allowsToBeAutoCorrected)
.setHasAutoCorrection(hasAutoCorrection);
}
// TODO: cleanup dictionaries looking up and suggestions building with SuggestedWords.Builder
public SuggestedWords.Builder getSuggestedWordBuilder(
final WordComposer wordComposer, CharSequence prevWordForBigram,