Example:

throws up the following in the concordance above:

Hi Michael Bailey

What percentage does Studio assign to these concorance hits?

Daniel

In this case 73%

Michael Bailey

Interesting. I wonder how this is calculated. The only meaningful word that matches in this case is "Angabe". But your segment is incredibly long, so it can match "und" about 20 times. I wonder if it calculates how much of the concordance segment can be matched with terms from the source segment...

Maybe Paul Filkin knows more?

Daniel

PS I wonder whether it would make sense for you to segment at the list items like "b)" or "(3)". The above segment is derived from paragraph segmentation? It certainly contains more than one sentence!

Daniel,

This is a slight pitfall of importing from EUR-LEX with a bilingual Excel, in that a lot of the amendments of Regulations and Directives go through this way. This is probably one of the worst cases (in the end I have removed this TM from concordance search as the results don't do much to help. Previously this kind of import was very useful (as exceptional quick) for creating an alignment then saved as a sdlxliff for subsequent use as a base for terminology work in MultiTerm

Michael

Michael Bailey

So the TM that this concordance hit is from is not your own translation but an imported bilingual corpus, correct? I guess aligning would make sense here, and you can import an sdlxliff into an sdltm, too.

Daniel

Daniel,

I have been getting similar matches from a TM with court judgements, which regularly have 100-200 word sentences (and similarly seem to through up lots of concordance hits with (in)definite articles). I couldn't show a screengrab of the several that show up most commonly from that source. Maybe there is a way to fine-tune concordance from words of 4-5 letters or longer, that way I will lose a lot of the der/die/eine/ist matches.

The EUR-LEX import is done using Bilingual Excel format - as per this tutorial: https://signsandsymptomsoftranslation.com/2016/01/15/align-eu/

Michael

I notice that that TU has come from an alignment of the CRR2, so a banking supervision text, so I might be able to filter out segments that do not match the supervision area (for the project in hand insurance supervision), as TMs and TUs all have an "Aufsichtsbereich" allocated to them, with possibility for multiple selections from the dropdown.

Michael Bailey

That is what I call paragraph alignment. Individual Excel cells are not segmented any further and because the text style differs quite a bit (EN-DE) it throws Studio's aligner for a loop:

Okapi seems to be a bit more sturdy, but you can't fix the alignment. But do check, I did a few Stichproben and it looked fine to me:

Eurlex_aligned.zip

(I just took a random text from that website. If you want this for a specific text, just send me the bilingual Excel and I can generate this for you to test.)

That way you would get a TM with sentence segmentation.

Daniel

Michael Bailey

Perhaps this will help?

https://gateway.rws.com/csm?id=kb_article_view&sysparm_article=KB0039891

Michael Bailey

Perhaps this will help?

https://gateway.rws.com/csm?id=kb_article_view&sysparm_article=KB0039891

Thanks Paul Filkin, very helpful!

Thanks Paul - explains a lot.