My attempted answer: 77%
How I got there:
- start by taking the "L2 speakers" column from wikipedia https://en.wikipedia.org/wiki/List_of_languages_by_total_number_of_speakers
- assume that every "conversation between two people with different native languages" is between two people selected at random. So for each language, we square the number of L2 speakers to get the number of possible pairs of people that could converse in this L2 language.
- divide by the sum of the squares to normalize
e.g. the two largest listed are English at 1 billion L2 speakers and Modern Standard Arabic at 275 million L2 speakers. So, for English, there are 1 billion * 1 billion = 10^18 possible conversational pairs possible, while for Modern Standard Arabic there are 275 million * 275 million = 7.5*10^16 pairs.
Note that this means that conversations are much more skewed towards the most-spoken languages than the raw numbers of speakers.
Does anyone have an alternate/better approach? E.g. I think naive squaring is probably too simplistic, since people will tend to cluster geographically by their L2 languages.
Note: not sure if this is the ideal forum for this question, but it seems similar in spirit to some existing questions like What languages together maximize the number of people you can speak with?