This is currently not available in the app out-of-the-box – but If you need this functionality, you could try a userscript solution, similar to the one described on this thread. Since the highlights are implemented as <mark> elements (and the only elements of that type), they’ll be easy to target.
But depending on what you’re trying to do, you might also want to consider shortening your texts instead. Of course, every use case is different and sometimes you just can’t avoid dealing with very long texts and contexts. However, I’d say that in 90% of the cases, coming across an issue like this is a clear indicator that your texts are too long and that you should consider truncating them. Some reasons for that:
- If you’re interested in auto-scrolling, you’re likely not even interested in the full context and just want to see the entity straight away. So it’s often more efficient to only show a short excerpt containing the match. If you do come across a task with insufficient context, you can always click “ignore” and reannotate it later.
- If you’re planning on training an NER model on the data and you do need this much context to make a decision of whether a label applies to a text or not, the model is also much less likely to learn anything meaningful from the data. If you’re able to make the annotation decisions from very short, surrounding context, you’ll likely see much better results.
- The annotation UI is significantly less effective if the annotator has to read a long text before making a decision. Even if you get to focus on the highlighted span first, it’ll still be surrounded by a wall of text and it’s much easier to get distracted.