Interesting fact: the spurious highlights only occur when there is a heading apple. If I remove the heading or I make it apple bla bla or Apple, only correct instances of the keyphrase are highlighted.

Mystery solved

The current mechanism of highlighting keyphrases is as follows.

1. Search for all possible ways the words from the keyphrase occur in the text. In this example, this step will result in just apple.
2. Take sentences one by one and hang highlight-tags on words from the keyphrase in them. In this example, this step will result in

[
   "Maecenas <yoastmark class='yoast-text-mark'>apple</yoastmark> a auctor mi.", 
   "<yoastmark class='yoast-text-mark'>apple</yoastmark>",
   "Maecenas <yoastmark class='yoast-text-mark'>apple</yoastmark> a auctor mi."
]

3. Mark any instances of these sentences in the text. This is the moment when it goes wrong: The word "ametapple" is considered to include the sentence
   "<yoastmark class='yoast-text-mark'>apple</yoastmark>".

How bad is that?

The problem occurs if the following two conditions are fulfilled:

1. There exists a sentence with keyphrase words in the text.
2. There exists a word in the text that includes this sentence entirely (case-sensitive).

It's hard to believe that this happens with a regular text. However, if the keyphrase includes somewhat shorter (non-function) words and if one / some of these words are used alone in a sentence (for instance, in a heading), there is a risk of a problem demonstrated by the issue.

We at Team Lingo believe it's an edge case, which is unlikely to bother a lot of users and which can wait for the new Tree Parser to solve it. @moorscode do you agree?