Hi, I accidentally stumbled upon this forum. I'm the developer of the tool. Quick Google search would lead you to the homepage of the project:
http://social.bioware.com/project/3492/, where you could contact me
I'm sorry for resurrecting the thread, just wanted to clarify for others.
Yes, the compression is Huffman coding, so the copy will never be 1-1. But there's another reason for the 100kB size difference. The original ME2 TLKs have one additional trick, which I didn't implement. When there is string A that is a suffix for another string B, then reference to string A is stored at the end of string B, instead of the actual string. So for example if there are three identical strings, they're stored only once, with references.
But at the time of writing the tool, I was too lazy to efficiently implement this. And the size difference is really affordable.
But the "structure", as you say, is ok. It's identical and appropriate for the format. It just lacks one optimalization, but it only means the file is _slighlty_ bigger nothing else.
The only issue with the tool is that it has very limited support for Xbox360 and PS3 TLKs. It only loads them fine and dumps to XML. But somehow, when converted back to TLK, it doesn't work on the console. I don't have any console for testing, so can't really trace the problem down.
