Term recognition in Trados 2024

Question

Dear Community, 
 this post is an inquiry aimed at understanding better what new Trados24 version brings. 
 Here is a curious example of term recognition performed by Multiterm today: 
 
 I have NEVER seen Trados recognize separate words as part of a term saved in the database ( comprehensive and maintenance , in this particular case). 
 The search is set to 70%, which is a standard issue: 
 
 First, I thought that Multiterm recognized only "comprehensive" and returned a fuzzy match from the database (even this would be a 50-60% match, definetely not 70%, in my opinion). 
 But then I saw that "maintenance" does not appear in the hitlist as a separate term, meaning that both words were recognized as parts of "compehensive maintenance". How come? There are 3 words inbetween! 
 This has never happened in the previous versions. Is there any technical explanation? New search processes in Trados 2024? 
 Would be very grateful for the explanation! Thanks in advance!

Paul · Answer

Yulia Kovalskaya 
 I tested this in 2022 and got exactly the same results as I do in 2024, so I don't believe there is anything different in 2024 in this regard. 
 
 As to how this works... I tested it a bit and can explain by deduction rather than giving details it would require a developer to provide. But first some basics. 
 
 Term recognition match scores are not calculated like translation memory fuzzy matches. Instead, they consider: 
 
 Exactness of phrase match
 
 Is the entire term ("comprehensive maintenance") present as-is? 
 If yes => 100% match 
 If only components are found, Studio will attempt a fuzzy term match

Proximity of tokens
 
 If the term is made up of multiple words, Trados checks how far apart those words are in the segment. 
 Closer = higher match score 
 Separated by multiple words or punctuation = lower score

Additional modifiers or interference
 
 Words like &ldquo;digitized&rdquo; and &ldquo;complex&rdquo; interrupt the expected phrase, reducing confidence in the match.

Order of tokens
 
 If the words are out of order, it further weakens the score.

So, if I then look at some examples: 
 
 I changed the distance between "comprehensive" and "maintenance" in each one resulting in 78%, 87% and 72% matches, all based on the same single termbase entry "comprehensive maintenance". 
 Yulia Kovalskaya said: First, I thought that Multiterm recognized only "comprehensive" and returned a fuzzy match from the database (even this would be a 50-60% match, definetely not 70%, in my opinion). 
 If it was a translation memory I'd agree... but it's not :-) 
 Studio attempts to infer the intent and provide useful term suggestions even if the exact match isn&rsquo;t there. In your use case (technical documentation/rail), this behaviour might be quite useful. But equally it could be misleading if translators assume that &ldquo;comprehensive&rdquo; or &ldquo;maintenance&rdquo; are individually present in the termbase (when they're not). So one difference I do see between 2024 and 2022 is that the match value seems to be there by default in 2024 (it wasn't in 2022 - although I didn't investigate this to see if my earlier or current default settings are at play). Either way the match score helps flag that partial matches are being made, based on token recognition and order. 
 Hope that helps?

Paul · Answer

Yulia Kovalskaya 
 The JRU example didn't get found because of the order, proximity and presence of the tokens. In your termbase you have "Unidad de Registro Jur&iacute;dico", but if your sentence only contains the wording "unidad jur&iacute;dica" then only two out of four tokens from the full term are there. So the phrase is farther in meaning and structure from the full term entry. This is the same in 2024. 
 If I do something like this where all four tokens are found I get a 64% match: 
 
 But still only the one term: 
 
 So possibly useful, but you'll only be accurate by entering the terms you actually want control over. 
 As I do this I can see why it was worth adding the fuzzy values because now users will have a better understanding of what this is actually telling them, and it's essentially a tool they can use where appropriate to help with the right term that is actually needed. But not mistake this for a 100% match that they have to be using!

Trados Studio > 3. MultiTerm

Term recognition in Trados 2024