Workarounds for the DV/DVX number problem

by Steven Marzuola

In a posting in August 2007, to Dejavu-L, a discussion group at Yahoo Groups, Marie Gouin expressed the following complaint about DVX:

Here's my response, with some additional information:

Marie, you've come across one of the most persistent and annoying bugs in DV. In response, Atril has usually told its users to omit sentences with numbers from the MDB. That is simply impractical and you have found a great example.

BTW, it usually occurs when the source sentence has more than one number. There are at least two workarounds, for the worst cases.

• Before translation, change all numbers to "Hidden" in Word, and tell DVX to ignore Hidden text. This introduces new codes, but your numbers should be safe. Someone posted macros to do this [at the discussion group], within the last couple of months.
  
• In extreme cases of number-laden texts, I have resorted to replacing digits with other characters that never appear in my documents. This set works for English and Spanish: 01234 56789 become àèìòù ÂÊÎÔÛ. Simply run a short Word macro before importing into Déjà Vu and another one after export.

As a complement to this, I created new TDB's, just for checking purposes, that contain all the one-letter and two-letter "words":

   à è ì ò ù Â Ê Î Ô Û àà àè àì àò àù à àÊ àÎ àÔ àÛ, èà èè èì ... etc.

I use that as part of my terminology check in Déjà Vu after translation is complete.

This terminology check is a very useful part of DV.  In particular, it catches the numbers error given in Marie's example.  However, as my translation memory databases have grown, it's become impractical to check the translation against the entire MDB and TDB.  Instead, Atril has suggested using an empty MDB and TDB during this test, to verify that numbers have been translated correctly. However, I have extended this technique. Instead of being empty, my "check" TDB's contain chosen words that should always be translated a certain way, or left untranslated. Examples:

1. All one-letter and two-letter combinations that are not real words or abbreviations in either language: aa, ab, ae, af, ag, ... b, ba, bc, bd ... These can be present as parts of serial numbers, part numbers, references, etc.
   
2. Roman numerals: II, III, IV, V, VII, .... up to XXXIX. Since "I" is a word in English and "vi" is a word in Spanish, these are omitted from the English-Spanish and Spanish-English "check" TDB's.
   
3. Numbers: two=dos, three=tres ... up to twenty. Also thirty, forty, fifty up to ninety. Some are omitted because there's no translation that is always correct. For example, "one" in English can be "un", "uno" or "una" in Spanish; those three can be translated as either "one", "a" or "an".
   
4. The nonsense words mentioned above: à è ì .. àà àè àì ... èà èè èì ... etc.
   
5. A handful of other terms and expressions

If you would like some help setting up a "check TDB" for your language pair, please contact me.