So, here is how we answered them:

How do we identify what comments were made on what paragraph?

We developed an engine in our content app, which assigns unique IDs to each content type on which annotations can be created by the user before it is saved to the database. For every article, these IDs start from 1. The obvious limitation being that the we cannot have annotations on a page where more than one database entries are being displayed. I think that is okay, for annotations on blurb listing pages would be absurd for me. Anytime the content gets updated, the previous IDs remain intact (unless the content is deleted). Even in the event of content deletion, the ID is never reused (it is a linear list which constantly increments). The annotations are saved with keys to their article and its paragraph ID.

What if that paragraph or content is deleted?

The answer lies not in how we save annotations, but in how we display it. Annotations are displayed in two places on a page. One is where they belong, on the margins of the text. Next, they are also collected down in a separate tab adjacent to the comments tab. If there are some orphaned annotations, you can find them in this annotation bucket. Plus, if you want to treat them as regular comments, so be it, they are here, at the bottom of the page.

What if the author of the article wants to respond but only let the author of annotation know?

Since we have already proposed that we would want to maximize tunability of privacy in our projects, scalability of such flexibility is paramount. But we have to make tradeoffs to make something work, and since we value your privacy more than anything, we have found our answer in dealing with this problem outside the context of annotations. But to understand the rationale of our solution, you will have to understand what annotations mean to us. Annotations are personal interjections, or ponderings that happen while reading an article. They are not exactly debatable, but can be a short feedback of sorts to authors, something said by a person to a person, and not broadcasted to everyone. Thus, annotations that are aimed at starting conversations must result into conversations, but not on the post, but in the privacy of message boxes. Say, a person made a public annotation on an article, and somebody wanted to respond to it. He can respond to it via a message. This is possible rather easily because the other person too is a member (since annotations are for signed up users). This can lead to great friendships and establishing novel social study circles while keeping the article relatively clean of the informal chatter that might then take place. The concept of group messaging shall allow more people to join in the conversation should the two (original people in conversation) oblige unanimously. Unanimous vote is required because even one person who does not oblige will be offended, and if that is so, we have a problem.