Start a new topic

SDL Studio 2017 R1’s new feature called ‘LookAhead’.

So the new Studio 2017 R1 has a new feature they call ‘LookAhead’. 

If I'm not mistaken, I suggested exactly such a feature, years ago, to various CAT tool developers. No time to look for my old posts/emails, but my idea was exactly what they have finally implemented: some form of 2-3 segments back, 2-3 segments forward lookup caching system, to speed up quickly switching between the next few segments, either up or down. Seems extremely logical and obvious, once someone finally implements it.

Count Beijer III oh Hastings

Hi Michael,

See the menu Translate > Preliminary matching actions. For example, Preliminary matching from current segment and then check the speed of going to the next segment. :)

1 person likes this

Interesting, I didn't know that you could access Preliminary matching from there. Will play around with a bit!

I don't know how much you know about the new LookAhead feature yet, but how would you say your implementation (which obviously predates SDL's version by many years) compares?

I haven't compared any other CAT tool features to CT but based on your description, the LookAhead feature goes head 2-3 segments while CafeTran's Preliminary matching races with your pace all the way ahead in the background.

More info on LookAhead feature:

Now, when you move to the next segment, Studio performs a look-up on the following two segments while you’re working on the active segment.

The effect of LookAhead? Almost instantaneous results every time you change segments. This is particularly relevant for cloud-based resources (SDL Language Cloud and other MT providers) and also for local resources (Slate Desktop engine and big local TMs with fragment recall and fuzzy match repair enabled).

So, if I understand this correctly, what SDL calls LookAhead (they are never short of attractive marketing lingo), is already available in CafeTran?

If so, it would be interesting to give this more publicity.

(Warning, am dictating this!) I'm not 100% sure about this, but I think that one difference might be that, although it is called lookahead, it also looks back, which I don't think that the CT version does. That is, when translating a particular segment, LookAhead is actually simultaneously looking back one or two segments, just in case you want to go back one or two segments.

 which I don't think that the CT version does

CT version races forward (either from the current segment or from the start). It caches everything so when you move back, you can still access the cached result fast.

In another context, this looking behind your shoulders/along your nose technology (LBYSAYN) will also be the big challenge for all MT systems, to improve the quality of translation results.

EU funding has gone to research projects.


CT version races forward (either from the current segment or from the start). It caches everything so when you move back, you can still access the cached result fast.

However, as you move forward, translating segments, your resources (TMs/TBs) obviously change. I think LookAhead also takes this into account, whereas CT doesn't.

 CT doesn't

It does. When you add a new segment to the translation memory which has been doing Preliminary Matching, CT recognizes the change and updates the TM in real time showing the new match instead of the cached one.

Only make sure that the TM (e.g recalled from Total Recall) is not set read-only.

LookAhead is also about cloud services. Could also MT resources be looked up ahead a few segments and cached (prefetched) ? There is currently a noticeable delay, it would be nice if the results were a little quicker. I think CT already caches MT results (already queried segments in current session).

It seems to me that preliminary matching is not the same thing as LookAhead.

LookAhead (as I understand it) is a continuous process taking place in the background.

Preliminary matching (again as I understand it) is initially a one-off process, which means it's great if you're heading to the kitchen for a cup of tea, but not much good if you're in the middle of something, and also (presumably) wasteful if, for example, you're only going to translate a small portion of the file in that particular session. (?).

I have to say that to my mind LookAhead has the edge here - it's a smart concept for speeding up your work without slowing it down, if you get my meaning.

P.S. In fairness, CT is much less in need of this feature than the execrebly slow (in my admitedly underpowered VM) Studio.


Login to post a comment