Start a new topic

Problem with fuzzy matching

There appears to be a problem with fuzzy matching (note: I'm still on 2019 forerunner, haven't updated to Croissant yet.)

As you can see from the attached screenshot, CT thinks DE:Neuromonitoring fuzzy matches with EN:neurostimulator (it should give EN:neuromonitoring) – but if you examine the 8/10 (why does hits show 8 whilse listing 10 items?)  matches shown, only one contains the EN text 'neurostimulator', whereas all contain the EN text 'neuromonitoring'.

Also, although this is shown on the TM tab, it is not shown on the matchboard (why?) but has clearly influenced the AA and MT results.


This is particularly concerning, as this sort of bait and switch is very easy to overlook when translating at speed.




This is a hit, not a fuzzy. 


You can stop this display of hits in the TM options. 

OK, but it's still an erroneous hit.

And: what is a hit? (i.e. what's the difference between a fuzzy and a hit?)

Fragment Hits - htps://cafetran.freshdesk.com/support/solutions/articles/6000088087-fragment-hits

Check Preferences > Memory if you want to fine-tune Hits.


To disable Hits on a specific TM, just select Matching type: Fuzzy (instead of Fuzzy and hits) in the TM options.

Some hits are a miss. That's normal. You need to fine-tune them or may prefer to disable them altogether.

Hi Jean,

many thanks for taking the time to explain this to me. I had misunderstood the terminology here.

.

Nonetheless this still looks like a bug. I have "Subsegment hits threshold" and "Subsegment to auto threshold" set to the default values of 3 and 7 respectively, but this false hit doesn't meet those criteria, and anyway why does CT decide that the match for DE:Neuromonitoring is EN:neurostimulator rather than the vastly more consistent match EN:neuromonitoring? Surely this is a bug in the matching algorithm?

Jeremy


This question is probably for Igor, but the answer may be found in the relevant hits.


Hover your mouse above the numbers next to the hit to check the relevant occurrences.


If Hit accuracy is not as good as expected, I suggest you set the "Subsegment hits threshold" and "Subsegment to auto threshold" higher.


This should improve the quality of subsegment matches, although they will be fewer in number.


On a personal note: I'm currently testing using "Fuzzy" instead of "Fuzzy & Hits" in most attached TMs. Disabling subsegment analysis speeds up the matching process and is less heavy on computer resources. I mostly don't use Auto-assembling, Hits are less useful if you don't seek to take advantage of this feature.

Thanks for the tip Jean – I'm not sure why fuzzy and hits was active on some of my memories (this setting is stored in the HTML), I've never consciously activated it. Turning it off seems to have restored some sense of normality. (And no, I don't get much use out of auto-assmbling either.)


Login to post a comment