Bertrand Meyer's technology+ blog

I have been conducting interesting conversations with a two-something child who has not quite mastered the speaker-dependent [1] personal pronouns. He says things like “Where is your mom?”, when he actually means to ask about his mother, not mine; from hearing people tell him things like “your mom is coming”, he is clearly taking “your mom” as some kind of paraphrase for “mom”. He approaches people and says “I want to help me”, obviously from having heard, from others, “would you come and help me?”.

I am not concerned about young Tim (let us call him that), who sooner or later will get the point. But let us assume that someone, call him Tom, is determined to explain to Tim the concepts of “you” and “me”. Not so easy! In fact, come to think of it, fairly tricky, treacherous even. We all learned the concept at some point; I wish I remembered how it happened for me [2].

The difficulty is that someone, whether Tom or a third party, has to provide the explanation, a circumstance that messes up the explanation itself since it contains speaker-dependent terms. (Did I hear the word “Heisenberg”? If so I will ignore it.) Tom might try something like the following explanation — let us call it /A/  — but it will not work:

If Tom says me it means Tom, but if Tim says me it means Tim.

If Tim says you it means Tom, but if Tom says you it means Tim.

The reason it does not work is lack of quoting, but before I get to this let me mention that a mathematical model is not hard to come by. From a two-element set Names = {Tim, Tom} to itself, there are four total functions. Of these, two are constant functions. The other two are non-constant; one, me, is the identity function on Names, and you is the other. They are the functions me = {<Tim, Tim>, <Tom, Tom>} and you = {<Tim, Tom>, <Tom, Tim>}. I is just a synonym for me (the accusative in the latter being one of the few remnants of declension in English.)

Let us call this explanation /B/.  It can be illustrated as follows:

We can also write these functions (using for conciseness [c: A \ d: B … \ Z] for the conditional expression if c then A elseif d then B … else Z end) as the following version /C/:

me = λ n | [n = Tim: Tim \ Tom]

you = λ n | [n = Tim: Tom \ Tim]

In practice, however, this modeling of the concepts “you” and “me” as functions in Names → Names is not sufficient to define them properly, for example to turn the informal explanation /A/ into one that makes sense. The problem is that we are dealing with a larger set of words than just Names: the set   Words = {Tim, Tom, Me, You} with four elements. The meaning of a sentence that may include any of them is not absolute, but depends on who is uttering the sentence. When I say “me” and you say “me” the meaning is different, although when I say “you” and you say “me” the meaning is the same.

To interpret a sentence, then, we need a second-level function: a function meaning not in Words → Words but in Names → Words → Names, defined as follows:

meaning = λ n  |  λ w  |     [w ∈ Names: w    \    w = Me: me (n)    \    you (n)]

Let us call this definition /D/. Examples of its application include:

• meaning (Tim) (Tom) = Tom. (First case: no speaker-dependency).
• meaning (Tim) (Me) = Tim. (Second case, expresses the part of /A/ that said “if Tim says me it means Tim”.)
• meaning (Tim) (You) = Tom. (Third case, expresses the part of /A/ that said “if Tim says you it means Tom”.)

/D/ is the most accurate definition of speaker-dependent pronouns and I may try it with the real Tim at the next opportunity (as soon as I have taught him lambda calculus).

/D/ also helps us understand what does not quite work in /A/, the first and perhaps most intuitive attempt. With explanations such as  “if Tim says you it means Tom” we are actually trying to express the need for making meanings speaker-dependent, that is to say, explain that the evaluation of a sentence must take into account who utters it. But this attempt fails because we need a general rule which will apply to all sentences, and the explanation itself is a sentence. To evaluate the sentence “if Tim says you it means Tom” we need to evaluate “you” which figures in it; that evaluation would yield either Tim or Tom depending on who is speaking. But because the sentence is itself part a definition of “you”, we do not want in this case to evaluate “you”.

We could write you in straight quotes as in programming languages:

if Tim says “you” it means Tom

but that is not the form of quoting we need. In a programming language, “you” is a character string, made of the characters ‘y‘, ‘o‘ and ‘u‘ in that order. Here, the makeup of this string in terms of its characters is completely irrelevant. The form of quoting that we need must achieve something else: preventing, in the evaluation of a formula, the evaluation of one of its parts.

John McCarthy’s Lisp language, coming out as early as 1959, showed how to handle this issue correctly. (This contribution is one of many from Lisp, including garbage collection and the notion of functional programming — not bad for a single language. See my 2011 obituary of John McCarthy on this blog [3].) One of the reasons Lisp needed a quoting mechanism is that it was designed to be self-describing and specifically self-evaluating, a spectacularly new concept at the time. McCarthy’s original book on Lisp (LISP 1.5 Programmer’s Manual, MIT Press, 1962) contains a dazzling interpreter of Lisp, written in Lisp and taking up no more than half a page. The interpreter is based on eval, a function (in the sense of a Lisp function, defined by a Lisp expression) which evaluates any Lisp program. You cannot define eval unless you have some way to prevent evaluation of part of an expression.

The convention is simple. For any expression x we may define another expression ‘x — which we can read aloud as “QUOTED x”. Then:

• Evaluating x yields the value of x.
• Evaluating ‘x yields x.

The original Lisp notation was (QUOTE x), which clearly conveys the idea: QUOTE is a function — in Lisp, function application f (x) is written (f x) — whose value, when applied to an argument, is that argument. (Not the value of the argument: the argument itself.)

With such a quote notation, explanations in the style of /A/ can be made meaningful; for example (version /E/):

If Tom says ‘me it means Tom, but if Tim says ‘me it means Tim.

This should be said with the quote spoken out loud: “QUOTED me”.

We’ll see if Tim sees what /E/ means. At least, you see what you mean. I mean, I see what I  mean. Sorry, I meant that I see what you mean. Oh well. We see what we mean.

Notes and references

[1] I really want to call them “subjective personal pronouns”, in the ordinary sense of “subjective” opposed to “objective” as in “a subjective assessment”, but in grammatical terminology “subjective pronoun” has a well-accepted but entirely different meaning, that of a pronoun used as the subject of a sentence, like “she” rather than “her”. Hence “speaker-dependent pronoun”. Linguists also use the term “indexical pronouns”, as in the reference cited in note [2].

[2] Linguists investigating language learning have studied the matter. A recent article (2015) is 2-Year-Olds’ Comprehension of Personal Pronouns by M. Moyer, K. Harrigan, V. Hacquard and J. Lidz (in Supplement to the proceedings of the 39th Boston University Conference on Language Development, eds E. Grillo, K. Jepson and M. LaMendola), available here. The article includes a literature review. It acknowledges the you-and-me problem (first- and second-person pronouns) although it suggests that third-person pronouns (“he”, “she”) may be even more of a challenge to infants. Two quotes (with bibliographic references removed):

• The evidence from previous studies examining two year olds’ competence is mixed, with some studies finding children succeeding with pronouns as early as 22-24 months, others finding success much later, and some with children succeeding at a range of ages.
• Most [studies] suggest a production/comprehension asymmetry… Production studies suggest that children begin producing pronouns around 15-18 months, starting with first person, then second person, and finally third person pronouns. These early productions typically include some errors where the child reverses first and second person, but on the whole consistent errors are few . On the other hand, comprehension data focusing on speech addressed to the child suggest that children first understand second person pronouns, then first person, and finally third person.

[3] My article on John McCarthy can be found here.