/// <summary>
/// Process the logic for custom dictionary post processor used to handle user custom vocab translation.
/// </summary>
/// <param name="translatedDocument">Translated document.</param>
/// <param name="languageId">Current source language id.</param>
/// <returns>A <see cref="PostProcessedDocument"/> stores the original translated document state and the newly post processed message.</returns>
public PostProcessedDocument Process(TranslatedDocument translatedDocument, string languageId)
{
// Check if provided custom dictionary for this language is not empty
if (_userCustomDictionaries.GetLanguageDictionary(languageId).Count > 0)
{
string processedResult;
var languageDictionary = _userCustomDictionaries.GetLanguageDictionary(languageId);
// Loop for all the original message tokens, and check if any of these tokens exists in the user custom dictionary,
// to forcibly overwrite this token's translation with the user provided translation
for (var i = 0; i < translatedDocument.SourceTokens.Length; i++)
{
if (languageDictionary.ContainsKey(translatedDocument.SourceTokens[i]))
{
// If a token of the original source message/phrase found in the user dictionary,
// replace it's equivalent translated token with the user provided translation
// the equivalent translated token can be found using the alignment map in the translated document
translatedDocument.TranslatedTokens[translatedDocument.IndexedAlignment[i]] = languageDictionary[translatedDocument.SourceTokens[i]];
}
}
// Finally return PostProcessedDocument object that holds the orignal TRanslatedDocument and a string that joins all the translated tokens together
processedResult = PostProcessingUtilities.Join(" ", translatedDocument.TranslatedTokens);
return(new PostProcessedDocument(translatedDocument, processedResult));
}
else
{
return(new PostProcessedDocument(translatedDocument, string.Empty));
}
}
/// <summary>
/// Substitute the numeric numbers in translated message with their orignal format in source message.
/// </summary>
/// <param name="translatedDocument">Translated document.</param>
private void SubstituteNumericPattern(TranslatedDocument translatedDocument)
{
var numericMatches = Regex.Matches(translatedDocument.SourceMessage, @"\d+", RegexOptions.Singleline);
foreach (Match numericMatch in numericMatches)
{
var srcIndex = Array.FindIndex(translatedDocument.SourceTokens, row => row == numericMatch.Groups[0].Value);
translatedDocument.TranslatedTokens = PostProcessingUtilities.KeepSourceWordInTranslation(translatedDocument.IndexedAlignment, translatedDocument.SourceTokens, translatedDocument.TranslatedTokens, srcIndex);
}
}
/// <summary>
/// Process the logic for patterns post processor used to handle numbers and no translate list.
/// </summary>
/// <param name="translatedDocument">Translated document.</param>
/// <param name="languageId">Current source language id.</param>
/// <returns>A <see cref="PostProcessedDocument"/> stores the original translated document state and the newly post processed message.</returns>
public PostProcessedDocument Process(TranslatedDocument translatedDocument, string languageId)
{
// validate function arguments for null and incorrect format
ValidateParameters(translatedDocument);
// flag to indicate if the source message contains number , will used for
var containsNum = Regex.IsMatch(translatedDocument.SourceMessage, @"\d");
// output variable declaration
string processedResult;
// temporary pattern is used to contain two set of patterns :
// - the post processed patterns that was configured by the user ie : _processedPatterns and
// - the liternal no translate pattern ie : translatedDocument.LiteranlNoTranslatePhrases , which takes the following regx "<literal>(.*)</literal>" , so the following code checks if this pattern exists in the translated document object to be added to the no translate list
// - ex : translatedDocument.sourceMessage = I like my friend <literal>happy</literal> , the literal tag here specifies that the word "happy" shouldn't be translated
var temporaryPatterns = _processedPatterns[languageId];
if (translatedDocument.LiteranlNoTranslatePhrases != null && translatedDocument.LiteranlNoTranslatePhrases.Count > 0)
{
temporaryPatterns.UnionWith(translatedDocument.LiteranlNoTranslatePhrases);
}
if (temporaryPatterns.Count == 0 && !containsNum)
{
processedResult = translatedDocument.TargetMessage;
}
if (string.IsNullOrWhiteSpace(translatedDocument.RawAlignment))
{
processedResult = translatedDocument.TargetMessage;
}
// loop for all the patterns and substitute each no translate pattern match with the original source words
// ex : assuming the pattern = "mon nom est (.+)"
// and the phrase = "mon nom est l'etat"
// the original translator output for this phrase would be "My name is the state",
// after applying the patterns post processor , the output would be : "My name is l'etat"
foreach (var pattern in temporaryPatterns)
{
if (Regex.IsMatch(translatedDocument.SourceMessage, pattern, RegexOptions.Singleline | RegexOptions.IgnoreCase))
{
SubstituteNoTranslatePattern(translatedDocument, pattern);
}
}
SubstituteNumericPattern(translatedDocument);
processedResult = PostProcessingUtilities.Join(" ", translatedDocument.TranslatedTokens);
return(new PostProcessedDocument(translatedDocument, processedResult));
}
/// <summary>
/// Substitutes matched no translate pattern with the original token.
/// </summary>
/// <param name="translatedDocument">Translated document.</param>
/// <param name="pattern">The no translate pattern.</param>
private void SubstituteNoTranslatePattern(TranslatedDocument translatedDocument, string pattern)
{
// get the matched no translate pattern
var matchNoTranslate = Regex.Match(translatedDocument.SourceMessage, pattern, RegexOptions.Singleline | RegexOptions.IgnoreCase);
// calculate the boundaries of the pattern match
// ex : "mon nom est l'etat
// start index = 12
// length = 6
var noTranslateStartChrIndex = matchNoTranslate.Groups[1].Index;
// the length of the matched pattern without spaces , which will be used in determining the translated tokens that will be replaced by their original values
var noTranslateMatchLength = matchNoTranslate.Groups[1].Value.Replace(" ", string.Empty).Length;
var wrdIndx = 0;
var chrIndx = 0;
var newChrLengthFromMatch = 0;
var srcIndex = -1;
var newNoTranslateArrayLength = 1;
var sourceMessageCharacters = translatedDocument.SourceMessage.ToCharArray();
foreach (var wrd in translatedDocument.SourceTokens)
{
// if the beginning of the current word equals the beginning of the matched no trasnalate word, then assign the current word index to srcIndex
if (chrIndx == noTranslateStartChrIndex)
{
srcIndex = wrdIndx;
}
// the following code block does the folowing :
// - checks if a match wsa found
// - checks if this match length equals the starting matching token length, if yes then this is the only token to process,
// otherwise continue the loop and add the next token to the list of tokens to be processed
// ex : "mon nom est l'etat"
// tokens = {"mon", "nom", "est", "l'", "etat"}
// when the loop reaches the token "l'" then srcIndex will = 3, but we don't want to consider only the token "l'" as the no translate token,
// instead we want to match the whole "l'etat" string regardless how many tokens it contains ie regardless that "l'etat" is actually composed of 2 tokens "l'" and "etat"
// so what these condition is trying to do is make the necessary checks that we got all the matched pattern not just a part of it's tokens!
// checks if match was found or not, because srcIndex value changes only in case a match was found !
if (srcIndex != -1)
{
// checks if we found all the tokens that matches the pattern
if (newChrLengthFromMatch + translatedDocument.SourceTokens[wrdIndx].Length >= noTranslateMatchLength)
{
break;
}
// if the previous condition fails it means that the next token is also matched in the pattern, so we increase the size of the no translate words array by 1
newNoTranslateArrayLength += 1;
// increment newChrLengthFromMatch with the found word size
newChrLengthFromMatch += translatedDocument.SourceTokens[wrdIndx].Length;
}
// the following block of code is used to calculate the next token starting index which could have two cases
// the first case is that the current token is followed by a space in this case we increment the next chrIndx by 1 to get the next character after the space
// the second case is that the token is followed by the next token without spaces , in this case we calculate chrIndx as chrIndx += wrd.Length without incrementing
// assumption : The provided sourceMessage and sourceMessageCharacters doesn't contain any consecutive white spaces,
// in our use case this handling is done using the translator output itself using the following line of code in PreprocessMessage function:
// textToTranslate = Regex.Replace(textToTranslate, @"\s+", " ");//used to remove multiple spaces in input user message
if (chrIndx + wrd.Length < sourceMessageCharacters.Length && sourceMessageCharacters[chrIndx + wrd.Length] == ' ')
{
chrIndx += wrd.Length + 1;
}
else
{
chrIndx += wrd.Length;
}
wrdIndx++;
}
// if the loop ends and srcIndex then no match was found
if (srcIndex == -1)
{
return;
}
// add the no translate words to a new array
var wrdNoTranslate = new string[newNoTranslateArrayLength];
Array.Copy(translatedDocument.SourceTokens, srcIndex, wrdNoTranslate, 0, newNoTranslateArrayLength);
// loop for each of the no translate words and replace it's translation with it's origin
foreach (var srcWrd in wrdNoTranslate)
{
translatedDocument.TranslatedTokens = PostProcessingUtilities.KeepSourceWordInTranslation(translatedDocument.IndexedAlignment, translatedDocument.SourceTokens, translatedDocument.TranslatedTokens, srcIndex);
srcIndex++;
}
}
