Wittgentsein famously stated that “the meaning is the use”. Unfortunately no analytical explanation of this statement is provided in the posthumously published Philosophical Investigations. Wittgenstein’s point seems to be that language is not appropriately approached through logical or analytical means, or at least not in the analytic approach founded by Frege, espoused by the early Wittgenstein, and underlying logical treatments of meaning such as CCG. But I am an analytical person and I need to work with clear, perhaps even mathematically rigorous, statements to feel that I really understand something. So I am going to try to analyze “the meaning is the use”.
I want to start with a simple example.
Mary’s inheritance allowed her to retire. (1)
What is the meaning of “allow” in this sentence? Grammatically “allow” is a ditransitive verb. It has three roles which can be characterized as.
(the allower) allows (the allowee) to do (the action). (1)
If we attempt to write a formal meaning for “allow” we should include the idea that the allower has the ability to choose whether the allowee is, or is-not, allowed to do the action. A parent can allow, or not allow, their child to use the internet in the bedroom. But how does an inheritance allow (or grant permission) to Mary to retire? An inheritance is not an agent and cannot decide anything. Rather than strictly applying a definition we seem to be using (1) as an abbreviation for certain conclusions.
Before Mary received the inheritance she could not afford to retire. (2)
After Mary received the inheritance she could afford to retire. (3)
These days it should be clear that inference is soft and defeasible. So when I talk about “conclusions” I am not talking about strict logical implications. But it will be useful (the meaning is the use) to employ the language of logical entailment with the understanding that entailment is soft. There are various mathematically rigorous treatments of soft entailment, such as formulations in terms of probabilistic programs, but it seems useful here to leave any rigorous treatment unspecified.
Returning to the example, the issue is how is it that (1) entails (2) and (3) and nothing more. The idea here is that the invocation of the word “allow” in (1) is being used as a tool to more concisely represent (2) and (3). I will call this an instrumental use of a word. An instrumental use can be distinguished from a literal or compositional use because in an instrumental use the intended conclusion are not derived through classical logical inference.
One can try to convert all uses of a verb into literal or compositional uses by introducing a large set of different senses of the word, one sense for each possible use of the word. But it seems that language use is sufficiently creative that covering it by enumerating senses is infeasible. I will support this by giving examples.
Language of knowledge and belief. Consider
My GPS knows where Mary lives. (4)
We might paraphrase this as “there is an entry in some kind of database within my GPS for Mary’s address”. Or maybe “I can ask my GPS for Mary’s address it can tell me and can provide directions to that location”. The point here is that (4) is a very concise way of communicating what would otherwise be a very complex meaning. The work “know” is being used instrumentally for various conclusions typically associated with human behavior — what my GPS can do in response to certain requests.
Truth. An interesting question here is whether it is meaningful to assign (4) a truth value. Is it really true that my GPS knows where Mary lives? Here it seems useful to analyze the word “true” itself in instrumental terms. I can meaningfully say
My GPS does not know where Mary lives (5)
I can also construct a plan that does a case analysis on (4) as in
I think John’s GPS knows where Mary lives. But if not, we can always call Mary. (6)
It seems clear that (6) is meaningful and is treating (4) as if it has a truth value. The idea here is that the instrumental use of language produces meaningful statements that carry truth values.
Teleological language in biology. As another example we can consider the use of teleological language in biology — language that refers to the purpose of something. Consider
The purpose of the heart is to provide oxygen to the body by pumping blood. (7)
But what does “purpose” mean here? Contrary to what some mystics might believe, animals were not designed by a designer with organ purposes in mind. And yet (7) seems clearly true. When a new biological structure or chemical pathway is discovered it is natural to ask whether it has a purpose and what that purpose is. Instrumental uses of teleological terms produce statements which carry meaningful truth values.
Desires and wants in naive physics. Consider
The plug prevents the water from going down the drain. (8)
How do we think about the word “prevents” in this sentence. A sentence of the form “x prevented y from doing z” implies (softly) that y wants to do z and would do z if not for x preventing it from doing so. Here we a assigning a desire to the water. In spite of assigning desires to something as inanimate as water, (8) seems true.
The language of choice. Consider
Mary had a choice between attending Harvard or Yale. (9)
This means that Mary applied to both institutions and was accepted by both. This seems straightforward unless one considers the case where Mary might be either a real person or a deterministic computer. If Mary is a deterministic computer does (9) still mean only that Mary applied to both and was accepted to both? In that case (9) is still true even when Mary is deterministic. In any case the language of choice will be useful in some applications and not useful in others. It useful to talk about, and think in terms of, choices when designing a computer program for playing chess. The language of choice is not useful when designing a calculator.
The Language of Sentience. Now consider words like “believes”, “wants”, “likes” and “trying”. We have already discussed the sentence “My GPS knows where Mary lives”. Clearly this is not attributing sentience to my GPS. But what is required for sentience words to have truth values when interpreted in their fullest sense. I take the position that an AI is actually sentient simply when attributing “actual sentience” is the most useful thing to do — the meaning is the use and meaningful statements carry truth.
Machine Thoughts 