Indexiphors: notes on embedded indexicals, shifty agreement, and logophoricity

Over the past 15 years there has been an explosion in the number of languages reported to allow indexical shift. Across this literature, the argument for indexical shifting is typically quite simple: (Premise 1) Indexical n has non-utterance-based reference in embedded clause c of language L. (Premise 2) Clause c cannot be a quote (e.g. because it is permeable to wh-movement or NPI licensing, or contains descriptions de re). (Conclusion) Therefore, Language L allows indexical shift. Is this form of argument valid? For many cases discussed in the recent literature: only if we take 1st/2nd person agreement to always indicate that the agreement controller is actually an indexical. But there is reason to doubt that premise, both from the theoretical literature (Anand 2006) and the typological/descriptive literature (Culy 1994, Curnow 2002). Elements that may control what is canonically 1st person agreement include not only 1st person indexicals but also a class of elements I will call indexiphors (sometimes also called 'first person logophors'). An indexiphor is like a logophor and unlike an indexical in that it must be bound by an OP-log in the left periphery of an embedded clause; it is like an indexical in the agreement that it controls.
In this paper, I argue (following Anand 2006) that indexiphors and true shifted indexicals are both possible in natural language, with distinct empirical profiles. (E.g. indexicals but not indexiphors obey Shift Together; indexiphors but not indexicals obey De Re Blocking.) Unknotting the two takes us a significant distance toward a constrained typology of indexical shift and "shifty agreement" crosslinguistically. Languages discussed include Amharic and Mishar Tatar.