Wow thank you so much for making a video Paul! Interesting approach, but as you say, it would need to work on a whole powerpoint file. I also have a feeling manual processing would still be necessary as the judgment of whether a line break is mid-sentence or not requires human understanding of the language. Automated approaches tend to produce a lot of false positives which then have to be fixed manually, taking more time than just doing it manually from the start.

James Plastow 

James Plastow said:

I also have a feeling manual processing would still be necessary as the judgment of whether a line break is mid-sentence or not requires human understanding of the language

I'd be very interested to see an example of where you think this would be required?  Just curiosity.  If you can create a small sample file that you know an automated approach gets wrong I'd love to put it to the test.  I am consistently surprised at the capabilities of this sort of technology.

Well, when a client makes a PowerPoint presentation they will sometimes use a line break to put separate words/phrases/sentences one above another in a text box (full stops may or may not be present and clients are very inconsistent in their use of full stops) and sometimes use a line break to split a phrase/sentence into two or more rows in order to fit the space available on the slide, and they will do both of these things many times in a presentation. I don't see how any automated tool could make a correct distinction between these two cases reliably without being able to understand the actual meaning of the language.

E.g. 1: A phrase split into two lines:

Introduction to

FY2022 Strategy

(Japanese sentence structure would be reversed)

E.g. 2: Two separate phrases in one text box

FY2022 Strategy

FY2022 Results

James Plastow 

ok - here's something to think about:

I played with the prompt a little until I got it right.

"Please join sentences if it makes sense to do so, keeping in mind the context and meaning of each sentence. If two sentences are closely related and form a complete idea when combined, join them. Otherwise, keep them separate. Do not change the content in any way, only address whether or not they should be joined. Here is the text:"

This successfully helped the engine to "make a correct distinction between these two cases".  The advancement in this space is quite remarkable and whilst not perfect I think this is another example of how we are going to see changes in our technology over the coming years that are more intuitive and based on some understanding of the information it receives.  I'm not saying my prompt would now solve every text in a powerpoint but it is food for thought!

OK, that is pretty mind-blowing...