From Francois Recanati’s Literalism and Contextualism: Some Varieties:

Let us start with a simple example in which modulation is required to overcome a semantic mismatch:

John hears the piano.

The verb ‘hear’ arguably denotes a relation between sentient organisms and sounds. Only sounds can be heard. Since a piano is not a sound, but a musical instrument, some adjustment is needed to make sense of ‘hear the piano’: either the noun-phrase ‘the piano’ must be given a metonymical interpretation, so that it stands for the sounds emitted by the piano; or (more plausibly) the verb ‘hear’ itself must be understood, not in its basic sense, but in a derived sense resulting from semantic transfer. An object is heard in the derived sense whenever the sound it emits is heard in the literal, basic sense.

This is a kind of argument for contextualism I always dislike. Doesn’t the data sentence count itself as a strong argument against the claim that only sounds can be heard? What is the evidence for that claim? It certainly doesn’t spring out from the usage data for ‘hears’. If Recanati is right then (1) should, I think, be a garden path sentence:

(1) John heard the voices and the piano.

Initially we are led to interpret ‘heard’ literally, but then we have to reanalyse it when we learn that ‘heard’ is meant to apply to ‘the piano’. But (1) certainly doesn’t feel like a garden path sentence. This is consistent with Recanati’s theory, provided we assume that there isn’t even a default preference for literally interpreting sentences. But then what constrains what we say about what can really be heard? Why not say that hears is a relation between a sentient being and a deity? We only ‘hear’ sounds in a derived sense. There’s simply got to be constraints, and I don’t know what they are meant to be from reading Recanati.

By the way, what’s the argument that only sentient beings can hear things. Can’t an alarm system hear the approaching burglar?

(Postscript: There might be some misleading implicatures there about how broad the argument against appeals to metonymy is meant to be. I really just mean to be attacking the argument about ‘hears’, which plays an important role I think in Recanati’s paper, and which I don’t think is supported by a great deal of evidence. What I need to do, before saying anything more detailed, is read Geoff Nunberg’s papers here and here on transferred meaning. Maybe I’ll post on that tomorrow.)