Must linguistic tokens be physical?

In case you've ever wondered about the question in the title of this post, you can stop wondering: the answer is negative. Here is a simple case. Suppose I want to communicate with you. There exist a million non-physical angels, numbered in some natural way (maybe from least to most wise). Both you and I are empaths of a special sort: we can both feel the angels' emotions and impress emotions on the angels. So now I communicate with you as follows. I take what I want to say to you, and translate it into a binary bit sequence (say, using ASCII). It had better have no more than a million bits. I then take this binary bit string and encode it in the angels' emotions: if the nth bit is one, I make the nth angel happy, and if the nth bit is zero, I sadden the nth angel. You then read off the angels' emotions using your emphathic skills, and so you know what bit string I was communicating to you, and you can decode it. Eventually, the binary encoding and decoding becomes second nature, and we're talking by making angels sadder or happier. The linguistic expression—utterance or inscription—then is something entirely non-physical, a sequence of angelic emotions.

This example shows that there is very little in the way of restrictions on what could count as a linguistic token. Linguistic tokens do not, for instance, need physical realization. What is needed for communication is the ability on the speaker's part to bring about that there is a linguistic token and an ability on the listener's part to identify some of the linguistically relevant features of the token.

It need not even be the case that something positive has to be caused for communication. Maybe you and I could adopt a convention that whenever I am silent in your company, I am asserting: "I am now happy." So, if I'm not happy, I need to chatter all the time.