For those wondering about the article -https://www.nature.com/articles/s41576-019-0150-2

As per the bold statement, it is basically trying to point out to you that a considerable chunk of the genome consists of relatively large transcript sizes that may exceed the read length of the short read platforms. Imagine you have a transcript for a gene that is 1000 bases long but your read length is 150 bases (150x2 or sometimes for RNA 75 or 75x2) - as with Illumina devices) but with pair-end method you can generate 300-400ish base reads. How can you cover all the exons (assuming cDNA input) when your reads are shorter than your input cDNA...well the answer is you can't ascertain the whole molecule but parts of it. If you could sequence the transcript end-to-end, you'd be more confident which transcript you saw. Have a look at the first picture in this link(https://www.pacb.com/blog/the-value-of-sequencing-full-length-rna-transcripts/), if you only read the first transcript (the rectangle on the left), how can you be sure which transcript that read came from? The transcript with 3 exons, 4 exons or 5 exons? Answer: you do not know unless you can gather enough unique evidence to differentiate it enough. This concept is briefly discussed in figure 1C of your article. It is basically a known caveat of short read platforms.

One specific limitation mentioned is the difficulty in accurately identifying and quantifying isoforms expressed from a gene, especially for very long or highly variable transcript isoforms. This is particularly relevant in the human transcriptome, where transcripts can vary greatly in length, ranging from 186 base pairs to 109 kilobases. The presence of such long and variable isoforms poses challenges for proper annotation and quantification.