Abstract: In this squib, I explore the relation between pronominal agreement and bound variable anaphora in the framework of Collins and Postal 2012.
Examples such as the following were pointed out in Collins and Postal 2012:
I am a person who likes to take care of myself/himself.
This sentence shows that the reflexive pronoun in the relative clause can be either 1SG or 3SG. Collins and Postal (2012: vii) claim that there are no truth conditional differences between the two variants.
Given this background, consider the paradigm in 2 and 3:
I am the only person who takes care of my mother.
This sentence is ambiguous (similar to related facts pointed out by Barbara Partee and many people). It can either mean that there are no other people who take care of their mothers (bound variable interpretation), or it can mean that there are no other people who take care of my mother in particular (non-bound variable interpretation).
But now consider:
I am the only person who takes care of his mother.
Putting aside the irrelevant reading where his refers to some particular person in the discourse other than me, 3 can only mean that there are no other people who take care of their mothers. It cannot mean that there are no other people who take care of my mother.
Therefore, there is an important asymmetry between 2 and 3. Contrary to the claim in Collins and Postal 2012, there is a clear truth conditional difference between 2 and 3.
The easiest way to explain this set of facts is to say that in 2, my can either have who as an antecedent (yielding the bound variable reading) or the subject of the matrix clause I as an antecedent (yielding the non-bound variable reading). When my has who as an antecedent, it gets its phi-features from the matrix subject which is a source of phi-features (in the technical sense of source in Collins and Postal 2012).
In 3, his can only take who as an antecedent, not the matrix subject, and so only the bound variable reading is allowed. The crucial assumption here is that the phi-features of his reflect its antecedent in this case.
Curiously, this asymmetry does not hold for the following examples:
Only the present authors take care of our mothers.
Only the present authors take care of their mothers.
To me, both 4 and 5 allow both bound variable and non-bound variable interpretations. In particular, 5 can have an interpretation where nobody else takes care of the mothers of the present authors.
In this case the ambiguity of 5 is explained by saying that their can either have only the present authors or the present authors as antecedent, yielding the ambiguity. Note that only the present authors and the present authors both can give rise to 3PL pronominal agreement.
To explain the ambiguity in 4, all that needs to be said is that both only the present authors and the present authors can give rise to 1PL pronominal agreement as well.
Going back to 1, adding the focus particle only brings out the two readings:
I am the only person who takes care of myself/himself.
Indeed, it seems in this case that the non-bound variable interpretation with myself is difficult. I would attribute this to the fact that because myself is a reflexive pronoun, it must take who as an antecedent.
This means that there are two independent ways to force who to be the antecedent of the relevant pronoun: (a) pronominal agreement (3SG) or (b) using a reflexive pronoun. In either of these cases, who must be the antecedent, forcing the bound variable interpretation.
Conversely, if there were a way to force the object pronoun in examples like 6 to take the matrix subject as antecedent, there should be no bound variable interpretation. That is exactly what is found:
I am the only person who takes care of me.
This seems to force an interpretation where there are no other people who take care of me (the non-bound variable interpretation). But that follows from the above account since in this case the antecedent of me cannot be who (because of Principle B of the Binding Theory).
Lastly consider the possible interpretations of 8:
I am the only person who takes care of him.
This cannot have a bound variable interpretation (since him cannot take who as antecedent, because of Principle B). It is also blocked from having the relevant non-bound variable interpretation (where him refers to the speaker), because if the matrix subject where the antecedent, there should be 1SG pronominal agreement. The only possible interpretation is that him refers to some particular person in the discourse.
So the complex facts in (1-8) follow neatly from the system of pronominal agreement given in Collins and Postal 2012, thereby strongly supporting the analyses presented there.