Thank you choroba and also many thanks to AppleFritter

I do not have to change the structure of the file, it is just about replacing some strings. Chances are this has to be put into a 'proper' script at some stage (I will surely try out the XPath variant)

As for the file, I can post a small snippet, which I had to modify. I changed the content but the structure is the same. Sorry for not having initiallly posted the snippet.

<tu creationdate="12345Z" changedate="12345" creationid="John Doe" cha
+ngeid="John DOE" srclang="en-US">
        <prop type="user-defined">server:xyz</prop>
        <prop type="user-defined">quality:100</prop>
        <prop type="user-defined">storing mode:1</prop>
        <prop type="user-defined">uncontrolled:0</prop>
        <prop type="user-defined">lastuser:unknown</prop>
        <prop type="user-defined">context left:1;2;3;4;5;6;7;8;9;10;11
+;12;13;14;15;16;17;18;19;20</prop>
        <prop type="user-defined">context right:21;22;23;24;25;26;27;2
+8;29;30;31;32;33;34;35;36;37;38;39;40</prop>
        <prop type="user-defined">a lot of junk with letters and &quot
+;90365AF9&quot</prop>
        <prop type="Guid::xxx">123456789</prop>
        <prop type="Att::xxx">name_of_someone</prop>^M
        <tuv xml:lang="en-US"><seg>a lot of junk with letters and &quo
+t;90365AF9&quot<bpt i="1"></bpt>text<ept i="1">&lt;/cf&gt;</ept></seg
+></tuv>
        <tuv xml:lang="en-US"><seg>a lot of junk with letters and &quo
+t;90365AF9&quot<bpt i="1"></ept></seg></tuv>
</tu>
[download]

Thanks a mil for helping me out. Really helps beginners to move on. Looking forward to your replies.

Thanks for sharing this piece of data -- however, your oneliner is still working just fine there for me. I understand that you may not be in a position to share confidential data, but on the other hand, I'm sure you'll understand it's a bit difficult to diagnose a problem without a test case showing the problem.

I second what other monks have suggested: if regular expressions aren't working here for whatever reason, a proper XML wrangling module may be the way to go. And XML::Twig is specifically intended for working with very large XML files that don't fit into memory, so that's my recommendation as well.