Adding words to our account of the communicative landscape does not fundamentally change what we’ve observed about inference and communication. Just as I make inferential estimates of what time in the morning it is (speaking of which, I need to find my sleep mask soon), or from my beloved wife’s mimed gestures when babies are sleeping, so I make inferential estimates of the most likely sense for the words she speaks or writes. There’s no “inner” or “real” meaning in the words; they’re a gesture, a verbal gesture, with the same status as a finger held to her lips, or a flat hand raised above her shoulder.
But that’s the second key element in the picture: when Margaret (or anyone, but we’re talking about Margaret now) speaks or writes words, they are words she has chosen based on her estimate (as speaker) of what I am most likely to infer from them. Again, there’s no intrinsic meaning at stake; she produces words calculated to elicit from me the results she wants. If she wants me to go to the grocery store to obtain food for dinner, she says, “Sweetheart, would you go to Tesco for a couple of things?” and it’s a pretty safe bet that I will in fact satisfy her desire. Were she to aim at the same effect by saying “Rapidly piddlepot strumming Hanover peace pudding mouse rumpling cuddly corridor cabinets?”, we may safely predict that the results would be different. Linguistic communication, on this account, is not sui generis nor paradigmatic for other modes of communication; it is continuous with other communicative modes, albeit in an extraordinarily precise, rule-governed way. It would be a dire mistake, however, to leap from “atypically precise” to “intrinsically precise” in order to amp up the degree of certainty that our inference can provide. We may be able often to recognise “time to wake up”, but that doesn’t entail our capacity really to ascertain that it’s 6:47.